* Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
* Author: Mikhail Pershin <mike.pershin@intel.com>
*/
+/**
+ * The Logical Object Device (LOD) layer manages access to striped
+ * objects (both regular files and directories). It implements the DT
+ * device and object APIs and is responsible for creating, storing,
+ * and loading striping information as an extended attribute of the
+ * underlying OSD object. LOD is the server side analog of the LOV and
+ * LMV layers on the client side.
+ *
+ * Metadata LU object stack (layers of the same compound LU object,
+ * all have the same FID):
+ *
+ * MDT
+ * | MD API
+ * MDD
+ * | DT API
+ * LOD
+ * / \ DT API
+ * OSD OSP
+ *
+ * During LOD object initialization the localness or remoteness of the
+ * object FID dictates the choice between OSD and OSP.
+ *
+ * An LOD object (file or directory) with N stripes (each has a
+ * different FID):
+ *
+ * LOD
+ * |
+ * +---+---+---+...+
+ * | | | | |
+ * S0 S1 S2 S3 S(N-1) OS[DP] objects, seen as DT objects by LOD
+ *
+ * When upper layers must access an object's stripes (which are
+ * themselves OST or MDT LU objects) LOD finds these objects by their
+ * FIDs and stores them as an array of DT object pointers on the
+ * object. Declarations and operations on LOD objects are received by
+ * LOD (as DT object operations) and performed on the underlying
+ * OS[DP] object and (as needed) on the stripes. From the perspective
+ * of LOD, a stripe-less file (created by mknod() or open with
+ * O_LOV_DELAY_CREATE) is an object which does not yet have stripes,
+ * while a non-striped directory (created by mkdir()) is an object
+ * which will never have stripes.
+ *
+ * The LOD layer also implements a small subset of the OBD device API
+ * to support MDT stack initialization and finalization (an MDD device
+ * connects and disconnects itself to and from the underlying LOD
+ * device), and pool management. In turn LOD uses the OBD device API
+ * to connect it self to the underlying OSD, and to connect itself to
+ * OSP devices representing the MDTs and OSTs that bear the stripes of
+ * its objects.
+ */
#define DEBUG_SUBSYSTEM S_MDS
#include "lod_internal.h"
-/**
- * Lookup MDT/OST index \a tgt by FID \a fid.
+/*
+ * Lookup target by FID.
+ *
+ * Lookup MDT/OST target index by FID. Type of the target can be
+ * specific or any.
*
- * \param lod LOD to be lookup at.
- * \param fid FID of object to find MDT/OST.
- * \param tgt MDT/OST index to return.
- * \param type indidcate the FID is on MDS or OST.
+ * \param[in] env LU environment provided by the caller
+ * \param[in] lod lod device
+ * \param[in] fid FID
+ * \param[out] tgt result target index
+ * \param[in] type expected type of the target:
+ * LU_SEQ_RANGE_{MDT,OST,ANY}
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
**/
int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
const struct lu_fid *fid, __u32 *tgt, int *type)
int rc = 0;
ENTRY;
- LASSERTF(fid_is_sane(fid), "Invalid FID "DFID"\n", PFID(fid));
+ if (!fid_is_sane(fid)) {
+ CERROR("%s: invalid FID "DFID"\n", lod2obd(lod)->obd_name,
+ PFID(fid));
+ RETURN(-EIO);
+ }
if (fid_is_idif(fid)) {
*tgt = fid_idif_ost_idx(fid);
static struct lu_device *lod_device_fini(const struct lu_env *env,
struct lu_device *d);
+/**
+ * Implementation of lu_device_operations::ldo_object_alloc() for LOD
+ *
+ * Allocates and initializes LOD's slice in the given object.
+ *
+ * see include/lu_object.h for the details.
+ */
struct lu_object *lod_object_alloc(const struct lu_env *env,
const struct lu_object_header *hdr,
struct lu_device *dev)
RETURN(lu_obj);
}
+/**
+ * Cleanup table of target's descriptors.
+ *
+ * The function goes through all the targets in the given table
+ * and apply given configuration command on to the targets.
+ * Used to cleanup the targets at unmount.
+ *
+ * \param[in] env LU environment provided by the caller
+ * \param[in] lod lod device
+ * \param[in] ltd target's table to go through
+ * \param[in] lcfg configuration command to apply
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ **/
static int lod_cleanup_desc_tgts(const struct lu_env *env,
struct lod_device *lod,
struct lod_tgt_descs *ltd,
{
struct lu_device *next;
int rc = 0;
- int i;
+ unsigned int i;
lod_getref(ltd);
if (ltd->ltd_tgts_size <= 0) {
return rc;
}
+/**
+ * Extract MDT target index from a device name.
+ *
+ * a helper function to extract index from the given device name
+ * like "fsname-MDTxxxx-mdtlov"
+ *
+ * \param[in] lodname device name
+ * \param[out] index extracted index
+ *
+ * \retval 0 on success
+ * \retval -EINVAL if the name is invalid
+ */
static int lodname2mdt_index(char *lodname, long *index)
{
char *ptr, *tmp;
- /* The lodname suppose to be fsname-MDTxxxx-mdtlov */
ptr = strrchr(lodname, '-');
if (ptr == NULL) {
CERROR("invalid MDT index in '%s'\n", lodname);
}
/**
- * Procss config log on LOD
- * \param env environment info
- * \param dev lod device
- * \param lcfg config log
+ * Implementation of lu_device_operations::ldo_process_config() for LOD
+ *
+ * The method is called by the configuration subsystem during setup,
+ * cleanup and when the configuration changes. The method processes
+ * few specific commands like adding/removing the targets, changing
+ * the runtime parameters.
+
+ * \param[in] env LU environment provided by the caller
+ * \param[in] dev lod device
+ * \param[in] lcfg configuration command to apply
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ *
+ * The examples are below.
*
- * Add osc config log,
+ * Add osc config log:
* marker 20 (flags=0x01, v2.2.49.56) lustre-OST0001 'add osc'
* add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nidxxx
* attach 0:lustre-OST0001-osc-MDT0001 1:osc 2:lustre-MDT0001-mdtlov_UUID
* lov_modify_tgts add 0:lustre-MDT0001-mdtlov 1:lustre-OST0001_UUID 2:1 3:1
* marker 20 (flags=0x02, v2.2.49.56) lustre-OST0001 'add osc'
*
- * Add mdc config log
+ * Add mdc config log:
* marker 10 (flags=0x01, v2.2.49.56) lustre-MDT0000 'add osp'
* add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:nid
* attach 0:lustre-MDT0000-osp-MDT0001 1:osp 2:lustre-MDT0001-mdtlov_UUID
* setup 0:lustre-MDT0000-osp-MDT0001 1:lustre-MDT0000_UUID 2:nid
* modify_mdc_tgts add 0:lustre-MDT0001 1:lustre-MDT0000_UUID 2:0 3:1
* marker 10 (flags=0x02, v2.2.49.56) lustre-MDT0000_UUID 'add osp'
- **/
+ */
static int lod_process_config(const struct lu_env *env,
struct lu_device *dev,
struct lustre_cfg *lcfg)
case LCFG_PARAM: {
struct obd_device *obd = lod2obd(lod);
- rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
- lcfg, obd);
+ rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
+ lcfg, obd);
if (rc > 0)
rc = 0;
GOTO(out, rc);
RETURN(rc);
}
+/**
+ * Implementation of lu_device_operations::ldo_recovery_complete() for LOD
+ *
+ * The method is called once the recovery is complete. This implementation
+ * distributes the notification to all the known targets.
+ *
+ * see include/lu_object.h for the details
+ */
static int lod_recovery_complete(const struct lu_env *env,
struct lu_device *dev)
{
struct lod_device *lod = lu2lod_dev(dev);
struct lu_device *next = &lod->lod_child->dd_lu_dev;
- int i, rc;
+ unsigned int i;
+ int rc;
ENTRY;
LASSERT(lod->lod_recovery_completed == 0);
RETURN(rc);
}
+/**
+ * Implementation of lu_device_operations::ldo_prepare() for LOD
+ *
+ * see include/lu_object.h for the details.
+ */
static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
struct lu_device *cdev)
{
.ldo_prepare = lod_prepare,
};
+/**
+ * Implementation of dt_device_operations::dt_root_get() for LOD
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_root_get(const struct lu_env *env,
struct dt_device *dev, struct lu_fid *f)
{
return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
}
+/**
+ * Implementation of dt_device_operations::dt_statfs() for LOD
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_statfs(const struct lu_env *env,
struct dt_device *dev, struct obd_statfs *sfs)
{
return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
}
+/**
+ * Implementation of dt_device_operations::dt_trans_create() for LOD
+ *
+ * Creates a transaction using local (to this node) OSD.
+ *
+ * see include/dt_object.h for the details.
+ */
static struct thandle *lod_trans_create(const struct lu_env *env,
struct dt_device *dev)
{
return th;
}
+/**
+ * Implementation of dt_device_operations::dt_trans_start() for LOD
+ *
+ * Starts the set of local transactions using the targets involved
+ * in declare phase. Initial support for the distributed transactions.
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_trans_start(const struct lu_env *env, struct dt_device *dev,
struct thandle *th)
{
return dt_trans_start(env, lod->lod_child, th);
}
+/**
+ * Implementation of dt_device_operations::dt_trans_stop() for LOD
+ *
+ * Stops the set of local transactions using the targets involved
+ * in declare phase. Initial support for the distributed transactions.
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_trans_stop(const struct lu_env *env, struct dt_device *dt,
struct thandle *th)
{
RETURN(rc);
}
+/**
+ * Implementation of dt_device_operations::dt_conf_get() for LOD
+ *
+ * Currently returns the configuration provided by the local OSD.
+ *
+ * see include/dt_object.h for the details.
+ */
static void lod_conf_get(const struct lu_env *env,
const struct dt_device *dev,
struct dt_device_param *param)
dt_conf_get(env, dt2lod_dev((struct dt_device *)dev)->lod_child, param);
}
+/**
+ * Implementation of dt_device_operations::dt_sync() for LOD
+ *
+ * Syncs all known OST targets. Very very expensive and used
+ * rarely by LFSCK now. Should not be used in general.
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_sync(const struct lu_env *env, struct dt_device *dev)
{
struct lod_device *lod = dt2lod_dev(dev);
struct lod_ost_desc *ost;
- int rc = 0, i;
+ unsigned int i;
+ int rc = 0;
ENTRY;
lod_getref(&lod->lod_ost_descs);
RETURN(rc);
}
+/**
+ * Implementation of dt_device_operations::dt_ro() for LOD
+ *
+ * Turns local OSD read-only, used for the testing only.
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_ro(const struct lu_env *env, struct dt_device *dev)
{
return dt_ro(env, dt2lod_dev(dev)->lod_child);
}
+/**
+ * Implementation of dt_device_operations::dt_commit_async() for LOD
+ *
+ * Asks local OSD to commit sooner.
+ *
+ * see include/dt_object.h for the details.
+ */
static int lod_commit_async(const struct lu_env *env, struct dt_device *dev)
{
return dt_commit_async(env, dt2lod_dev(dev)->lod_child);
}
+/**
+ * Not used
+ */
static int lod_init_capa_ctxt(const struct lu_env *env, struct dt_device *dev,
int mode, unsigned long timeout,
__u32 alg, struct lustre_capa_key *keys)
.dt_init_capa_ctxt = lod_init_capa_ctxt,
};
+/**
+ * Connect to a local OSD.
+ *
+ * Used to connect to the local OSD at mount. OSD name is taken from the
+ * configuration command passed. This connection is used to identify LU
+ * site and pin the OSD from early removal.
+ *
+ * \param[in] env LU environment provided by the caller
+ * \param[in] lod lod device
+ * \param[in] cfg configuration command to apply
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ **/
static int lod_connect_to_osd(const struct lu_env *env, struct lod_device *lod,
struct lustre_cfg *cfg)
{
RETURN(rc);
}
+/**
+ * Allocate and initialize target table.
+ *
+ * A helper function to initialize the target table and allocate
+ * a bitmap of the available targets.
+ *
+ * \param[in] ltd target's table to initialize
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ **/
static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
{
mutex_init(<d->ltd_mutex);
return 0;
}
+/**
+ * Initialize LOD device at setup.
+ *
+ * Initializes the given LOD device using the original configuration command.
+ * The function initiates a connection to the local OSD and initializes few
+ * internal structures like pools, target tables, etc.
+ *
+ * \param[in] env LU environment provided by the caller
+ * \param[in] lod lod device
+ * \param[in] ldt not used
+ * \param[in] cfg configuration command
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ **/
static int lod_init0(const struct lu_env *env, struct lod_device *lod,
struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
RETURN(rc);
}
+/**
+ * Implementation of lu_device_type_operations::ldto_device_free() for LOD
+ *
+ * Releases the memory allocated for LOD device.
+ *
+ * see include/lu_object.h for the details.
+ */
static struct lu_device *lod_device_free(const struct lu_env *env,
struct lu_device *lu)
{
RETURN(next);
}
+/**
+ * Implementation of lu_device_type_operations::ldto_device_alloc() for LOD
+ *
+ * Allocates LOD device and calls the helpers to initialize it.
+ *
+ * see include/lu_object.h for the details.
+ */
static struct lu_device *lod_device_alloc(const struct lu_env *env,
struct lu_device_type *type,
struct lustre_cfg *lcfg)
return lu_dev;
}
+/**
+ * Implementation of lu_device_type_operations::ldto_device_fini() for LOD
+ *
+ * Releases the internal resources used by LOD device.
+ *
+ * see include/lu_object.h for the details.
+ */
static struct lu_device *lod_device_fini(const struct lu_env *env,
struct lu_device *d)
{
RETURN(NULL);
}
-/*
- * we use exports to track all LOD users
- */
+/**
+ * Implementation of obd_ops::o_connect() for LOD
+ *
+ * Used to track all the users of this specific LOD device,
+ * so the device stays up until the last user disconnected.
+ *
+ * \param[in] env LU environment provided by the caller
+ * \param[out] exp export the caller will be using to access LOD
+ * \param[in] obd OBD device representing LOD device
+ * \param[in] cluuid unique identifier of the caller
+ * \param[in] data not used
+ * \param[in] localdata not used
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ **/
static int lod_obd_connect(const struct lu_env *env, struct obd_export **exp,
struct obd_device *obd, struct obd_uuid *cluuid,
struct obd_connect_data *data, void *localdata)
RETURN(0);
}
-/*
- * once last export (we don't count self-export) disappeared
- * lod can be released
- */
+/**
+ *
+ * Implementation of obd_ops::o_disconnect() for LOD
+ *
+ * When the caller doesn't need to use this LOD instance, it calls
+ * obd_disconnect() and LOD releases corresponding export/reference count.
+ * Once all the users gone, LOD device is released.
+ *
+ * \param[in] exp export provided to the caller in obd_connect()
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ **/
static int lod_obd_disconnect(struct obd_export *exp)
{
struct obd_device *obd = exp->exp_obd;
info->lti_ea_store = NULL;
info->lti_ea_store_size = 0;
}
+ lu_buf_free(&info->lti_linkea_buf);
OBD_FREE_PTR(info);
}
.ldt_ctx_tags = LCT_MD_THREAD,
};
+/**
+ * Implementation of obd_ops::o_get_info() for LOD
+ *
+ * Currently, there is only one supported key: KEY_OSP_CONNECTED , to provide
+ * the caller binary status whether LOD has seen connection to any OST target.
+ *
+ * \param[in] env LU environment provided by the caller
+ * \param[in] exp export of the caller
+ * \param[in] keylen len of the key
+ * \param[in] key the key
+ * \param[in] vallen not used
+ * \param[in] val not used
+ * \param[in] lsm not used
+ *
+ * \retval 0 if a connection was seen
+ * \retval -EAGAIN if LOD isn't running yet or no
+ * connection has been seen yet
+ * \retval -EINVAL if not supported key is requested
+ **/
static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
__u32 keylen, void *key, __u32 *vallen, void *val,
struct lov_stripe_md *lsm)
struct obd_device *obd = exp->exp_obd;
struct lod_device *d;
struct lod_ost_desc *ost;
- int i, rc = 1;
+ unsigned int i;
+ int rc = 1;
if (!obd->obd_set_up || obd->obd_stopping)
RETURN(-EAGAIN);
return rc;
rc = class_register_type(&lod_obd_device_ops, NULL, true, NULL,
-#ifndef HAVE_ONLY_PROCFS_SEQ
- NULL,
-#endif
LUSTRE_LOD_NAME, &lod_device_type);
if (rc) {
lu_kmem_fini(lod_caches);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff