#include <linux/pagemap.h>
#include <asm/div64.h>
#include <linux/seq_file.h>
+#include <linux/namei.h>
#else
#include <liblustre.h>
#endif
#include <linux/lprocfs_status.h>
#include <linux/lustre_fsfilt.h>
#include <linux/obd_lmv.h>
+#include <linux/lustre_lite.h>
#include "lmv_internal.h"
+/* not defined for liblustre building */
+#if !defined(ATOMIC_INIT)
+#define ATOMIC_INIT(val) { (val) }
+#endif
+
/* object cache. */
kmem_cache_t *obj_cache;
atomic_t obj_cache_count = ATOMIC_INIT(0);
CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
- if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
+ if (obd_uuid_equals(uuid, &tgt->uuid))
break;
}
/* this is fake connect function. Its purpose is to initialize lmv and say
* caller that everything is okay. Real connection will be performed later. */
static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
- struct obd_uuid *cluuid, unsigned long flags)
+ struct obd_uuid *cluuid, struct obd_connect_data *data,
+ unsigned long flags)
{
#ifdef __KERNEL__
struct proc_dir_entry *lmv_proc_dir;
lmv->cluuid = *cluuid;
lmv->connect_flags = flags;
sema_init(&lmv->init_sem, 1);
+ if (data)
+ memcpy(&lmv->conn_data, data, sizeof(*data));
#ifdef __KERNEL__
lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
GOTO(out_disc, rc = -EINVAL);
}
- rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid,
+ rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid, &lmv->conn_data,
lmv->connect_flags);
if (rc) {
CERROR("target %s connect error %d\n",
mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
+ if (mdc_obd)
+ mdc_obd->obd_no_recov = obd->obd_no_recov;
+
#ifdef __KERNEL__
if (lmv_proc_dir) {
struct proc_dir_entry *mdc_symlink;
}
}
#endif
- if (obd->obd_no_recov) {
- if (mdc_obd)
- mdc_obd->obd_no_recov = 1;
- }
CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
lmv->tgts[i].ltd_exp->exp_obd->obd_name,
lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
struct lmv_obd *lmv = &obd->u.lmv;
ENTRY;
- if (lcfg->lcfg_inllen1 < 1) {
+ if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
CERROR("LMV setup requires a descriptor\n");
RETURN(-EINVAL);
}
- if (lcfg->lcfg_inllen2 < 1) {
+ if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) {
CERROR("LMV setup requires an MDT UUID list\n");
RETURN(-EINVAL);
}
- desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
- if (sizeof(*desc) > lcfg->lcfg_inllen1) {
+ desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
+ if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
CERROR("descriptor size wrong: %d > %d\n",
- (int)sizeof(*desc), lcfg->lcfg_inllen1);
+ (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
RETURN(-EINVAL);
}
- uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
- if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
+ uuids = (struct obd_uuid *)lustre_cfg_buf(lcfg, 2);
+ if (sizeof(*uuids) * desc->ld_tgt_count != LUSTRE_CFG_BUFLEN(lcfg, 2)) {
CERROR("UUID array size wrong: %u * %u != %u\n",
- sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
+ sizeof(*uuids), desc->ld_tgt_count, LUSTRE_CFG_BUFLEN(lcfg, 2));
RETURN(-EINVAL);
}
RETURN(-ENOMEM);
for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
- if (lmv->tgts[i].ltd_exp == NULL) {
- CWARN("%s: NULL export for %d\n", obd->obd_name, i);
+ if (lmv->tgts[i].ltd_exp == NULL)
continue;
- }
rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age);
if (rc) {
}
static int lmv_getattr(struct obd_export *exp, struct lustre_id *id,
- __u64 valid, unsigned int ea_size,
- struct ptlrpc_request **request)
+ __u64 valid, const char *ea_name, int ea_namelen,
+ unsigned int ea_size, struct ptlrpc_request **request)
{
struct obd_device *obd = exp->exp_obd;
struct lmv_obd *lmv = &obd->u.lmv;
LASSERT(i < lmv->desc.ld_tgt_count);
+
rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid,
- ea_size, request);
+ ea_name, ea_namelen, ea_size, request);
if (rc)
RETURN(rc);
md.mea = NULL;
mealen = MEA_SIZE_LMV(lmv);
- valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
+ valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA;
/* time to update mea of parent id */
rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp,
- id, valid, mealen, &req);
+ id, valid, NULL, 0, mealen, &req);
if (rc) {
CERROR("md_getattr() failed, error %d\n", rc);
GOTO(cleanup, rc);
cb_compl, cb_blocking, cb_data);
CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n",
- OLID4(&mea->mea_ids[i]), rc, it->d.lustre.it_status);
+ OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status);
if (rc)
GOTO(cleanup, rc);
- if (it->d.lustre.it_data) {
+ if (LUSTRE_IT(it)->it_data) {
struct ptlrpc_request *req;
- req = (struct ptlrpc_request *)it->d.lustre.it_data;
+ req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data;
ptlrpc_req_finished(req);
}
- if (it->d.lustre.it_status)
- GOTO(cleanup, rc = it->d.lustre.it_status);
+ if (LUSTRE_IT(it)->it_status)
+ GOTO(cleanup, rc = LUSTRE_IT(it)->it_status);
}
OBD_FREE(data2, sizeof(*data2));
return rc;
}
+int lmv_enqueue_remote(struct obd_export *exp, int lock_type,
+ struct lookup_intent *it, int lock_mode,
+ struct mdc_op_data *data, struct lustre_handle *lockh,
+ void *lmm, int lmmsize, ldlm_completion_callback cb_compl,
+ ldlm_blocking_callback cb_blocking, void *cb_data)
+{
+ struct ptlrpc_request *req = LUSTRE_IT(it)->it_data;
+ struct obd_device *obd = exp->exp_obd;
+ struct lmv_obd *lmv = &obd->u.lmv;
+ struct lustre_handle plock;
+ struct mdc_op_data rdata;
+ struct mds_body *body = NULL;
+ int rc = 0, pmode;
+ ENTRY;
+
+ body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
+ LASSERT(body != NULL);
+
+ if (!(body->valid & OBD_MD_MDS))
+ RETURN(0);
+
+ CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n",
+ LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1));
+
+ /* we got LOOKUP lock, but we really need attrs */
+ pmode = LUSTRE_IT(it)->it_lock_mode;
+ LASSERT(pmode != 0);
+ memcpy(&plock, lockh, sizeof(plock));
+ LUSTRE_IT(it)->it_lock_mode = 0;
+ LUSTRE_IT(it)->it_data = NULL;
+ LASSERT((body->valid & OBD_MD_FID) != 0);
+
+ memcpy(&rdata, data, sizeof(rdata));
+ rdata.id1 = body->id1;
+ rdata.name = NULL;
+ rdata.namelen = 0;
+
+ LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE;
+ ptlrpc_req_finished(req);
+
+ rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp,
+ lock_type, it, lock_mode, &rdata, lockh, lmm,
+ lmmsize, cb_compl, cb_blocking, cb_data);
+ ldlm_lock_decref(&plock, pmode);
+ RETURN(rc);
+}
+
int lmv_enqueue(struct obd_export *exp, int lock_type,
struct lookup_intent *it, int lock_mode,
struct mdc_op_data *data, struct lustre_handle *lockh,
rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp,
lock_type, it, lock_mode, data, lockh, lmm,
lmmsize, cb_compl, cb_blocking, cb_data);
+ if (rc == 0 && it->it_op == IT_OPEN)
+ rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode,
+ data, lockh, lmm, lmmsize,
+ cb_compl, cb_blocking, cb_data);
RETURN(rc);
}
struct lustre_id rid = *id;
struct mds_body *body;
struct lmv_obj *obj;
- __u64 old_valid;
ENTRY;
rc = lmv_check_connect(obd);
CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n",
namelen, filename, OLID4(id), OLID4(&rid));
- old_valid = valid;
-
- /*
- * here should be applied OBD_MD_FID to ->valid, because otherwise,
- * mds_getattr_lock() will not fetch fid component of lustre_id and
- * thus, next call to md_getattr_lock() will be performed to wrong mds.
- */
- if (!(old_valid & OBD_MD_FID))
- valid |= OBD_MD_FID;
-
- rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
- &rid, filename, namelen, valid,
+ rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
+ &rid, filename, namelen,
+ valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID,
ea_size, request);
if (rc == 0) {
/*
*/
body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
LASSERT(body != NULL);
+ LASSERT((body->valid & OBD_MD_FID) != 0
+ || body->valid == OBD_MD_FLID);
+
if (body->valid & OBD_MD_MDS) {
struct ptlrpc_request *req = NULL;
rid = body->id1;
CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid));
- /*
- * turning OBD_MD_FID fetching off, as we already have
- * full lustre_id and do need to fetch fid component
- * again. This will help to make thing slightly faster.
- */
- if (!(old_valid & OBD_MD_FID))
- valid &= ~OBD_MD_FID;
-
rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp,
&rid, NULL, 1, valid, ea_size, &req);
ptlrpc_req_finished(*request);
body = lustre_msg_buf((*request)->rq_repmsg, 0,
sizeof(*body));
LASSERT(body != NULL);
+ LASSERT((body->valid & OBD_MD_FID) != 0);
LASSERT(id_group(&body->id1) == id_group(&data->id1));
}
}
}
struct obd_device *lmv_get_real_obd(struct obd_export *exp,
- char *name, int len)
+ struct lustre_id *id)
{
struct obd_device *obd = exp->exp_obd;
struct lmv_obd *lmv = &obd->u.lmv;
rc = lmv_check_connect(obd);
if (rc)
RETURN(ERR_PTR(rc));
- obd = lmv->tgts[0].ltd_exp->exp_obd;
+ obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd;
EXIT;
return obd;
}
int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
+ void *acl, int acl_size,
struct lov_stripe_md **ea, struct obd_trans_info *oti)
{
struct obd_device *obd = exp->exp_obd;
LASSERT(ea == NULL);
LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
- rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp,
- oa, &obj_mdp, oti);
+ rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa,
+ acl, acl_size, &obj_mdp, oti);
RETURN(rc);
}
* values for "master" object, as it will be used.
*/
int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
+ void *acl, int acl_size,
struct lov_stripe_md **ea, struct obd_trans_info *oti)
{
struct obd_device *obd = exp->exp_obd;
LASSERT(oa != NULL);
if (ea == NULL) {
- rc = lmv_obd_create_single(exp, oa, NULL, oti);
+ rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti);
if (rc)
CERROR("Can't create object, rc = %d\n", rc);
RETURN(rc);
}
+ /* acl is only suppied when mds create single remote obj */
+ LASSERT(acl == NULL && acl_size == 0);
+
if (*ea == NULL) {
rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
if (rc < 0) {
oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE |
OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
- rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
+ rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0,
+ &obj_mdp, oti);
if (rc) {
CERROR("obd_create() failed on MDT target %d, "
"error %d\n", c, rc);
struct lmv_desc *desc_ret = val;
*desc_ret = lmv->desc;
RETURN(0);
+ } else if (keylen == strlen("remote_flag") &&
+ !strcmp(key, "remote_flag")) {
+ struct lmv_tgt_desc *tgts;
+ int i;
+
+ LASSERT(*vallen == sizeof(__u32));
+ for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count;
+ i++, tgts++) {
+
+ /* all tgts should be connected when this get called. */
+ if (!tgts || !tgts->ltd_exp) {
+ CERROR("target not setup?\n");
+ continue;
+ }
+
+ if (!obd_get_info(tgts->ltd_exp, keylen, key,
+ vallen, val))
+ RETURN(0);
+ }
+ RETURN(-EINVAL);
}
CDEBUG(D_IOCTL, "invalid key\n");
lmv_set_timeouts(obd);
RETURN(0);
}
-
+
+ /* maybe this could be default */
+ if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) ||
+ (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) {
+ struct lmv_tgt_desc *tgt;
+ struct obd_export *exp;
+ int rc = 0, err, i;
+
+ spin_lock(&lmv->lmv_lock);
+ for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count;
+ i++, tgt++) {
+ exp = tgt->ltd_exp;
+ /* during setup time the connections to mdc might
+ * haven't been established.
+ */
+ if (exp == NULL) {
+ struct obd_device *tgt_obd;
+
+ tgt_obd = class_find_client_obd(&tgt->uuid,
+ LUSTRE_MDC_NAME,
+ &obd->obd_uuid);
+ if (!tgt_obd) {
+ CERROR("can't set info %s, "
+ "device %s not attached?\n",
+ (char *) key, tgt->uuid.uuid);
+ rc = -EINVAL;
+ continue;
+ }
+ exp = tgt_obd->obd_self_export;
+ }
+
+ err = obd_set_info(exp, keylen, key, vallen, val);
+ if (!rc)
+ rc = err;
+ }
+ spin_unlock(&lmv->lmv_lock);
+
+ RETURN(rc);
+ }
+
RETURN(-EINVAL);
}
lsmp = (struct mea *)lsm;
meap = (struct mea *)*lmmp;
+ if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
+ lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
+ RETURN(-EINVAL);
+
meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
meap->mea_count = cpu_to_le32(lsmp->mea_count);
meap->mea_master = cpu_to_le32(lsmp->mea_master);
RETURN(mea_size);
}
-int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
- struct lov_mds_md *disk_src, int mdsize)
+int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
+ struct lov_mds_md *lmm, int lmm_size)
{
struct obd_device *obd = class_exp2obd(exp);
- struct mea **tmea = (struct mea **)mem_tgt;
- struct mea *mea = (struct mea *)disk_src;
+ struct mea **tmea = (struct mea **)lsmp;
+ struct mea *mea = (struct mea *)lmm;
struct lmv_obd *lmv = &obd->u.lmv;
- int mea_size, i;
+ int mea_size, i, rc = 0;
+ __u32 magic;
ENTRY;
- mea_size = sizeof(struct lustre_id) *
+ mea_size = sizeof(struct lustre_id) *
lmv->desc.ld_tgt_count + sizeof(struct mea);
- if (mem_tgt == NULL)
+
+ if (lsmp == NULL)
return mea_size;
- if (*mem_tgt != NULL && disk_src == NULL) {
+ if (*lsmp != NULL && lmm == NULL) {
OBD_FREE(*tmea, mea_size);
RETURN(0);
}
- LASSERT(mea_size == mdsize);
+ LASSERT(mea_size == lmm_size);
OBD_ALLOC(*tmea, mea_size);
if (*tmea == NULL)
RETURN(-ENOMEM);
- if (!disk_src)
+ if (!lmm)
RETURN(mea_size);
- (*tmea)->mea_magic = le32_to_cpu(mea->mea_magic);
+ if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
+ mea->mea_magic == MEA_MAGIC_ALL_CHARS)
+ {
+ magic = le32_to_cpu(mea->mea_magic);
+ } else {
+ struct mea_old *old = (struct mea_old *)lmm;
+
+ mea_size = sizeof(struct lustre_id) * old->mea_count +
+ sizeof(struct mea_old);
+
+ if (old->mea_count > 256 || old->mea_master > 256 ||
+ lmm_size < mea_size || old->mea_master > old->mea_count) {
+ CWARN("bad MEA: count %u, master %u, size %u\n",
+ old->mea_count, old->mea_master, mea_size);
+ GOTO(out_free_mea, rc = -EINVAL);
+ }
+ magic = MEA_MAGIC_LAST_CHAR;
+ }
+
+ (*tmea)->mea_magic = magic;
(*tmea)->mea_count = le32_to_cpu(mea->mea_count);
(*tmea)->mea_master = le32_to_cpu(mea->mea_master);
- for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
+ for (i = 0; i < (*tmea)->mea_count; i++) {
(*tmea)->mea_ids[i] = mea->mea_ids[i];
id_le_to_cpu(&(*tmea)->mea_ids[i]);
}
-
RETURN(mea_size);
+
+out_free_mea:
+ OBD_FREE(*tmea, mea_size);
+ return rc;
}
int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,