X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosp%2Fosp_object.c;h=2835bbb6a66fbff63b156600e93d162d65685a6d;hp=78e2250c95ecf84e7d11d98937994b27d99f36bc;hb=846dd0cb79fc309703afdaae7622e4ddeb0e2e49;hpb=0209add4a5099817111c8576afe930d1e2daef03 diff --git a/lustre/osp/osp_object.c b/lustre/osp/osp_object.c index 78e2250..2835bbb 100644 --- a/lustre/osp/osp_object.c +++ b/lustre/osp/osp_object.c @@ -27,15 +27,63 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2014, Intel Corporation. */ /* - * This file is part of Lustre, http://www.lustre.org/ - * Lustre is a trademark of Sun Microsystems, Inc. - * * lustre/osp/osp_object.c * - * Lustre OST Proxy Device + * Lustre OST Proxy Device (OSP) is the agent on the local MDT for the OST + * or remote MDT. + * + * OSP object attributes cache + * --------------------------- + * OSP object is the stub of the remote OST-object or MDT-object. Both the + * attribute and the extended attributes are stored on the peer side remotely. + * It is inefficient to send RPC to peer to fetch those attributes when every + * get_attr()/get_xattr() called. For a large system, the LFSCK synchronous + * mode scanning is prohibitively inefficient. + * + * So the OSP maintains the OSP object attributes cache to cache some + * attributes on the local MDT. The cache is organized against the OSP + * object as follows: + * + * struct osp_xattr_entry { + * struct list_head oxe_list; + * atomic_t oxe_ref; + * void *oxe_value; + * int oxe_buflen; + * int oxe_namelen; + * int oxe_vallen; + * unsigned int oxe_exist:1, + * oxe_ready:1; + * char oxe_buf[0]; + * }; + * + * struct osp_object_attr { + * struct lu_attr ooa_attr; + * struct list_head ooa_xattr_list; + * }; + * + * struct osp_object { + * ... + * struct osp_object_attr *opo_ooa; + * spinlock_t opo_lock; + * ... + * }; + * + * The basic attributes, such as owner/mode/flags, are stored in the + * osp_object_attr::ooa_attr. The extended attributes will be stored + * as osp_xattr_entry. Every extended attribute has an independent + * osp_xattr_entry, and all the osp_xattr_entry are linked into the + * osp_object_attr::ooa_xattr_list. The OSP object attributes cache + * is protected by the osp_object::opo_lock. + * + * Not all OSP objects have an attributes cache because maintaining + * the cache requires some resources. Currently, the OSP object + * attributes cache will be initialized when the attributes or the + * extended attributes are pre-fetched via osp_declare_attr_get() + * or osp_declare_xattr_get(). That is usually for LFSCK purpose, + * but it also can be shared by others. * * Author: Alex Zhuravlev * Author: Mikhail Pershin @@ -45,13 +93,28 @@ #include "osp_internal.h" +static inline __u32 osp_dev2node(struct osp_device *osp) +{ + return osp->opd_storage->dd_lu_dev.ld_site->ld_seq_site->ss_node_id; +} + static inline bool is_ost_obj(struct lu_object *lo) { return !lu2osp_dev(lo->lo_dev)->opd_connect_mdt; } +/** + * Assign FID to the OST object. + * + * This function will assign the FID to the OST object of a striped file. + * + * \param[in] env pointer to the thread context + * \param[in] d pointer to the OSP device + * \param[in] o pointer to the OSP object that the FID will be + * assigned to + */ static void osp_object_assign_fid(const struct lu_env *env, - struct osp_device *d, struct osp_object *o) + struct osp_device *d, struct osp_object *o) { struct osp_thread_info *osi = osp_env_info(env); @@ -64,6 +127,14 @@ static void osp_object_assign_fid(const struct lu_env *env, lu_object_assign_fid(env, &o->opo_obj.do_lu, &osi->osi_fid); } +/** + * Initialize the OSP object attributes cache. + * + * \param[in] obj pointer to the OSP object + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_oac_init(struct osp_object *obj) { struct osp_object_attr *ooa; @@ -85,53 +156,97 @@ static int osp_oac_init(struct osp_object *obj) return 0; } +/** + * Find the named extended attribute in the OSP object attributes cache. + * + * The caller should take the osp_object::opo_lock before calling + * this function. + * + * \param[in] ooa pointer to the OSP object attributes cache + * \param[in] name the name of the extended attribute + * \param[in] namelen the name length of the extended attribute + * + * \retval pointer to the found extended attribute entry + * \retval NULL if the specified extended attribute is not + * in the cache + */ static struct osp_xattr_entry * osp_oac_xattr_find_locked(struct osp_object_attr *ooa, - const char *name, int namelen, bool unlink) + const char *name, size_t namelen) { struct osp_xattr_entry *oxe; list_for_each_entry(oxe, &ooa->ooa_xattr_list, oxe_list) { if (namelen == oxe->oxe_namelen && - strncmp(name, oxe->oxe_buf, namelen) == 0) { - if (unlink) - list_del_init(&oxe->oxe_list); - else - atomic_inc(&oxe->oxe_ref); - + strncmp(name, oxe->oxe_buf, namelen) == 0) return oxe; - } } return NULL; } +/** + * Find the named extended attribute in the OSP object attributes cache. + * + * Call osp_oac_xattr_find_locked() with the osp_object::opo_lock held. + * + * \param[in] obj pointer to the OSP object + * \param[in] name the name of the extended attribute + * \param[in] unlink true if the extended attribute entry is to be removed + * from the cache + * + * \retval pointer to the found extended attribute entry + * \retval NULL if the specified extended attribute is not + * in the cache + */ static struct osp_xattr_entry *osp_oac_xattr_find(struct osp_object *obj, - const char *name) + const char *name, bool unlink) { struct osp_xattr_entry *oxe = NULL; spin_lock(&obj->opo_lock); - if (obj->opo_ooa != NULL) + if (obj->opo_ooa != NULL) { oxe = osp_oac_xattr_find_locked(obj->opo_ooa, name, - strlen(name), false); + strlen(name)); + if (oxe != NULL) { + if (unlink) + list_del_init(&oxe->oxe_list); + else + atomic_inc(&oxe->oxe_ref); + } + } spin_unlock(&obj->opo_lock); return oxe; } +/** + * Find the named extended attribute in the OSP object attributes cache. + * + * If it is not in the cache, then add an empty entry (that will be + * filled later) to cache with the given name. + * + * \param[in] obj pointer to the OSP object + * \param[in] name the name of the extended attribute + * \param[in] len the length of the extended attribute value + * + * \retval pointer to the found or new-created extended + * attribute entry + * \retval NULL if the specified extended attribute is not in the + * cache or fail to add new empty entry to the cache. + */ static struct osp_xattr_entry * -osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, int len) +osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, size_t len) { struct osp_object_attr *ooa = obj->opo_ooa; struct osp_xattr_entry *oxe; struct osp_xattr_entry *tmp = NULL; - int namelen = strlen(name); - int size = sizeof(*oxe) + namelen + 1 + len; + size_t namelen = strlen(name); + size_t size = sizeof(*oxe) + namelen + 1 + len; LASSERT(ooa != NULL); - oxe = osp_oac_xattr_find(obj, name); + oxe = osp_oac_xattr_find(obj, name, false); if (oxe != NULL) return oxe; @@ -148,9 +263,11 @@ osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, int len) atomic_set(&oxe->oxe_ref, 2); spin_lock(&obj->opo_lock); - tmp = osp_oac_xattr_find_locked(ooa, name, namelen, false); + tmp = osp_oac_xattr_find_locked(ooa, name, namelen); if (tmp == NULL) list_add_tail(&oxe->oxe_list, &ooa->ooa_xattr_list); + else + atomic_inc(&tmp->oxe_ref); spin_unlock(&obj->opo_lock); if (tmp != NULL) { @@ -161,16 +278,31 @@ osp_oac_xattr_find_or_add(struct osp_object *obj, const char *name, int len) return oxe; } +/** + * Add the given extended attribute to the OSP object attributes cache. + * + * If there is an old extended attributed entry with the same name, + * remove it from the cache and return it via the parameter \a poxe. + * + * \param[in] obj pointer to the OSP object + * \param[in,out] poxe double pointer to the OSP object extended attribute + * entry: the new extended attribute entry is transfered + * via such pointer target, and if old the extended + * attribute entry exists, then it will be returned back + * via such pointer target. + * \param[in] len the length of the (new) extended attribute value + * + * \retval pointer to the new extended attribute entry + * \retval NULL for failure cases. + */ static struct osp_xattr_entry * osp_oac_xattr_replace(struct osp_object *obj, - struct osp_xattr_entry **poxe, int len) + struct osp_xattr_entry **poxe, size_t len) { struct osp_object_attr *ooa = obj->opo_ooa; - struct osp_xattr_entry *old = *poxe; struct osp_xattr_entry *oxe; - struct osp_xattr_entry *tmp = NULL; - int namelen = old->oxe_namelen; - int size = sizeof(*oxe) + namelen + 1 + len; + size_t namelen = (*poxe)->oxe_namelen; + size_t size = sizeof(*oxe) + namelen + 1 + len; LASSERT(ooa != NULL); @@ -181,22 +313,29 @@ osp_oac_xattr_replace(struct osp_object *obj, INIT_LIST_HEAD(&oxe->oxe_list); oxe->oxe_buflen = size; oxe->oxe_namelen = namelen; - memcpy(oxe->oxe_buf, old->oxe_buf, namelen); + memcpy(oxe->oxe_buf, (*poxe)->oxe_buf, namelen); oxe->oxe_value = oxe->oxe_buf + namelen + 1; /* One ref is for the caller, the other is for the entry on the list. */ atomic_set(&oxe->oxe_ref, 2); spin_lock(&obj->opo_lock); - tmp = osp_oac_xattr_find_locked(ooa, oxe->oxe_buf, namelen, true); + *poxe = osp_oac_xattr_find_locked(ooa, oxe->oxe_buf, namelen); + LASSERT(*poxe != NULL); + + list_del_init(&(*poxe)->oxe_list); list_add_tail(&oxe->oxe_list, &ooa->ooa_xattr_list); spin_unlock(&obj->opo_lock); - *poxe = tmp; - LASSERT(tmp != NULL); - return oxe; } +/** + * Release reference from the OSP object extended attribute entry. + * + * If it is the last reference, then free the entry. + * + * \param[in] oxe pointer to the OSP object extended attribute entry. + */ static inline void osp_oac_xattr_put(struct osp_xattr_entry *oxe) { if (atomic_dec_and_test(&oxe->oxe_ref)) { @@ -206,8 +345,25 @@ static inline void osp_oac_xattr_put(struct osp_xattr_entry *oxe) } } +/** + * Parse the OSP object attribute from the RPC reply. + * + * If the attribute is valid, then it will be added to the OSP object + * attributes cache. + * + * \param[in] env pointer to the thread context + * \param[in] reply pointer to the RPC reply + * \param[in] req pointer to the RPC request + * \param[out] attr pointer to buffer to hold the output attribute + * \param[in] obj pointer to the OSP object + * \param[in] index the index of the attribute buffer in the reply + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_get_attr_from_reply(const struct lu_env *env, struct object_update_reply *reply, + struct ptlrpc_request *req, struct lu_attr *attr, struct osp_object *obj, int index) { @@ -225,7 +381,10 @@ static int osp_get_attr_from_reply(const struct lu_env *env, if (rbuf->lb_len != sizeof(*wobdo)) return -EPROTO; - obdo_le_to_cpu(wobdo, wobdo); + LASSERT(req != NULL); + if (ptlrpc_req_need_swab(req)) + lustre_swab_obdo(wobdo); + lustre_get_wire_obdo(NULL, lobdo, wobdo); spin_lock(&obj->opo_lock); if (obj->opo_ooa != NULL) { @@ -242,8 +401,26 @@ static int osp_get_attr_from_reply(const struct lu_env *env, return 0; } +/** + * Interpreter function for getting OSP object attribute asynchronously. + * + * Called to interpret the result of an async mode RPC for getting the + * OSP object attribute. + * + * \param[in] env pointer to the thread context + * \param[in] reply pointer to the RPC reply + * \param[in] req pointer to the RPC request + * \param[in] obj pointer to the OSP object + * \param[out] data pointer to buffer to hold the output attribute + * \param[in] index the index of the attribute buffer in the reply + * \param[in] rc the result for handling the RPC + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_attr_get_interpterer(const struct lu_env *env, struct object_update_reply *reply, + struct ptlrpc_request *req, struct osp_object *obj, void *data, int index, int rc) { @@ -255,7 +432,8 @@ static int osp_attr_get_interpterer(const struct lu_env *env, osp2lu_obj(obj)->lo_header->loh_attr |= LOHA_EXISTS; obj->opo_non_exist = 0; - return osp_get_attr_from_reply(env, reply, NULL, obj, index); + return osp_get_attr_from_reply(env, reply, req, NULL, obj, + index); } else { if (rc == -ENOENT) { osp2lu_obj(obj)->lo_header->loh_attr &= ~LOHA_EXISTS; @@ -270,12 +448,28 @@ static int osp_attr_get_interpterer(const struct lu_env *env, return 0; } +/** + * Implement OSP layer dt_object_operations::do_declare_attr_get() interface. + * + * Declare that the caller will get attribute from the specified OST object. + * + * This function adds an Object Unified Target (OUT) sub-request to the per-OSP + * based shared asynchronous request queue. The osp_attr_get_interpterer() + * is registered as the interpreter function to handle the result of this + * sub-request. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt, struct lustre_capa *capa) { struct osp_object *obj = dt2osp_obj(dt); struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); - struct dt_update_request *update; int rc = 0; if (obj->opo_ooa == NULL) { @@ -285,19 +479,32 @@ static int osp_declare_attr_get(const struct lu_env *env, struct dt_object *dt, } mutex_lock(&osp->opd_async_requests_mutex); - update = osp_find_or_create_async_update_request(osp); - if (IS_ERR(update)) - rc = PTR_ERR(update); - else - rc = osp_insert_async_update(env, update, OUT_ATTR_GET, obj, 0, - NULL, NULL, - &obj->opo_ooa->ooa_attr, - osp_attr_get_interpterer); + rc = osp_insert_async_request(env, OUT_ATTR_GET, obj, 0, NULL, NULL, + &obj->opo_ooa->ooa_attr, + osp_attr_get_interpterer); mutex_unlock(&osp->opd_async_requests_mutex); return rc; } +/** + * Implement OSP layer dt_object_operations::do_attr_get() interface. + * + * Get attribute from the specified MDT/OST object. + * + * If the attribute is in the OSP object attributes cache, then return + * the cached attribute directly. Otherwise it will trigger an OUT RPC + * to the peer to get the attribute synchronously, if successful, add it + * to the OSP attributes cache. (\see lustre/osp/osp_trans.c for OUT RPC.) + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[out] attr pointer to the buffer to hold the output attribute + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ int osp_attr_get(const struct lu_env *env, struct dt_object *dt, struct lu_attr *attr, struct lustre_capa *capa) { @@ -324,12 +531,12 @@ int osp_attr_get(const struct lu_env *env, struct dt_object *dt, spin_unlock(&obj->opo_lock); } - update = out_create_update_req(dev); + update = dt_update_request_create(dev); if (IS_ERR(update)) RETURN(PTR_ERR(update)); - rc = out_insert_update(env, update, OUT_ATTR_GET, - lu_object_fid(&dt->do_lu), 0, NULL, NULL); + rc = out_attr_get_pack(env, &update->dur_buf, + lu_object_fid(&dt->do_lu)); if (rc != 0) { CERROR("%s: Insert update error "DFID": rc = %d\n", dev->dd_lu_dev.ld_obd->obd_name, @@ -360,42 +567,34 @@ int osp_attr_get(const struct lu_env *env, struct dt_object *dt, if (reply == NULL || reply->ourp_magic != UPDATE_REPLY_MAGIC) GOTO(out, rc = -EPROTO); - rc = osp_get_attr_from_reply(env, reply, attr, obj, 0); + rc = osp_get_attr_from_reply(env, reply, req, attr, obj, 0); if (rc != 0) GOTO(out, rc); - if (!is_ost_obj(&dt->do_lu)) { - if (attr->la_flags == 1) - obj->opo_empty = 0; - else - obj->opo_empty = 1; - } - GOTO(out, rc = 0); out: if (req != NULL) ptlrpc_req_finished(req); - out_destroy_update_req(update); + dt_update_request_destroy(update); return rc; } -static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_attr *attr, struct thandle *th) +static int __osp_attr_set(const struct lu_env *env, struct dt_object *dt, + const struct lu_attr *attr, struct thandle *th) { struct osp_device *d = lu2osp_dev(dt->do_lu.lo_dev); struct osp_object *o = dt2osp_obj(dt); struct lu_attr *la; int rc = 0; - ENTRY; /* * Usually we don't allow server stack to manipulate size * but there is a special case when striping is created - * late, after stripless file got truncated to non-zero. + * late, after stripeless file got truncated to non-zero. * * In this case we do the following: * @@ -424,10 +623,9 @@ static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } - if (o->opo_new) { + if (o->opo_new) /* no need in logging for new objects being created */ RETURN(0); - } if (!(attr->la_valid & (LA_UID | LA_GID))) RETURN(0); @@ -445,6 +643,7 @@ static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, if (rc != 0 || o->opo_ooa == NULL) RETURN(rc); + /* Update the OSP object attributes cache. */ la = &o->opo_ooa->ooa_attr; spin_lock(&o->opo_lock); if (attr->la_valid & LA_UID) { @@ -461,15 +660,94 @@ static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, RETURN(0); } +/** + * Implement OSP layer dt_object_operations::do_declare_attr_set() interface. + * XXX: NOT prepare set_{attr,xattr} RPC for remote transaction. + * + * According to our current transaction/dt_object_lock framework (to make + * the cross-MDTs modification for DNE1 to be workable), the transaction + * sponsor will start the transaction firstly, then try to acquire related + * dt_object_lock if needed. Under such rules, if we want to prepare the + * set_{attr,xattr} RPC in the RPC declare phase, then related attr/xattr + * should be known without dt_object_lock. But such condition maybe not + * true for some remote transaction case. For example: + * + * For linkEA repairing (by LFSCK) case, before the LFSCK thread obtained + * the dt_object_lock on the target MDT-object, it cannot know whether + * the MDT-object has linkEA or not, neither invalid or not. + * + * Since the LFSCK thread cannot hold dt_object_lock before the (remote) + * transaction start (otherwise there will be some potential deadlock), + * it cannot prepare related RPC for repairing during the declare phase + * as other normal transactions do. + * + * To resolve the trouble, we will make OSP to prepare related RPC + * (set_attr/set_xattr/del_xattr) after remote transaction started, + * and trigger the remote updating (RPC sending) when trans_stop. + * Then the up layer users, such as LFSCK, can follow the general + * rule to handle trans_start/dt_object_lock for repairing linkEA + * inconsistency without distinguishing remote MDT-object. + * + * In fact, above solution for remote transaction should be the normal + * model without considering DNE1. The trouble brought by DNE1 will be + * resolved in DNE2. At that time, this patch can be removed. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] attr pointer to the attribute to be set + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ +static int osp_declare_attr_set(const struct lu_env *env, struct dt_object *dt, + const struct lu_attr *attr, struct thandle *th) +{ + int rc = 0; + + if (!is_only_remote_trans(th)) + rc = __osp_attr_set(env, dt, attr, th); + + return rc; +} + +/** + * Implement OSP layer dt_object_operations::do_attr_set() interface. + * + * Set attribute to the specified OST object. + * + * If the transaction is a remote transaction, then related modification + * sub-request has been added in the declare phase and related OUT RPC + * has been triggered at transaction start. Otherwise it will generate + * a MDS_SETATTR64_REC record in the llog. There is a dedicated thread + * to handle the llog asynchronously. + * + * If the attribute entry exists in the OSP object attributes cache, + * then update the cached attribute according to given attribute. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] attr pointer to the attribute to be set + * \param[in] th pointer to the transaction handler + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_attr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_attr *attr, struct thandle *th, struct lustre_capa *capa) { struct osp_object *o = dt2osp_obj(dt); int rc = 0; - ENTRY; + if (is_only_remote_trans(th)) { + rc = __osp_attr_set(env, dt, attr, th); + if (rc != 0) + RETURN(rc); + } + /* we're interested in uid/gid changes only */ if (!(attr->la_valid & (LA_UID | LA_GID))) RETURN(0); @@ -498,8 +776,26 @@ static int osp_attr_set(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } +/** + * Interpreter function for getting OSP object extended attribute asynchronously + * + * Called to interpret the result of an async mode RPC for getting the + * OSP object extended attribute. + * + * \param[in] env pointer to the thread context + * \param[in] reply pointer to the RPC reply + * \param[in] req pointer to the RPC request + * \param[in] obj pointer to the OSP object + * \param[out] data pointer to OSP object attributes cache + * \param[in] index the index of the attribute buffer in the reply + * \param[in] rc the result for handling the RPC + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_xattr_get_interpterer(const struct lu_env *env, struct object_update_reply *reply, + struct ptlrpc_request *req, struct osp_object *obj, void *data, int index, int rc) { @@ -510,7 +806,7 @@ static int osp_xattr_get_interpterer(const struct lu_env *env, LASSERT(ooa != NULL); if (rc == 0) { - int len = sizeof(*oxe) + oxe->oxe_namelen + 1; + size_t len = sizeof(*oxe) + oxe->oxe_namelen + 1; rc = object_update_result_data_get(reply, rbuf, index); if (rc < 0 || rbuf->lb_len > (oxe->oxe_buflen - len)) { @@ -544,15 +840,33 @@ static int osp_xattr_get_interpterer(const struct lu_env *env, return 0; } +/** + * Implement OSP dt_object_operations::do_declare_xattr_get() interface. + * + * Declare that the caller will get extended attribute from the specified + * OST object. + * + * This function will add an OUT_XATTR_GET sub-request to the per OSP + * based shared asynchronous request queue with the interpreter function: + * osp_xattr_get_interpterer(). + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[out] buf pointer to the lu_buf to hold the extended attribute + * \param[in] name the name for the expected extended attribute + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt, struct lu_buf *buf, const char *name, struct lustre_capa *capa) { struct osp_object *obj = dt2osp_obj(dt); struct osp_device *osp = lu2osp_dev(dt->do_lu.lo_dev); - struct dt_update_request *update; struct osp_xattr_entry *oxe; - int namelen = strlen(name); + __u16 namelen = strlen(name); int rc = 0; LASSERT(buf != NULL); @@ -573,39 +887,58 @@ static int osp_declare_xattr_get(const struct lu_env *env, struct dt_object *dt, return -ENOMEM; mutex_lock(&osp->opd_async_requests_mutex); - update = osp_find_or_create_async_update_request(osp); - if (IS_ERR(update)) { - rc = PTR_ERR(update); + rc = osp_insert_async_request(env, OUT_XATTR_GET, obj, 1, + &namelen, (const void **)&name, oxe, + osp_xattr_get_interpterer); + if (rc != 0) { mutex_unlock(&osp->opd_async_requests_mutex); osp_oac_xattr_put(oxe); } else { - rc = osp_insert_async_update(env, update, OUT_XATTR_GET, obj, - 1, &namelen, &name, oxe, - osp_xattr_get_interpterer); - if (rc != 0) { + struct dt_update_request *update; + + /* XXX: Currently, we trigger the batched async OUT + * RPC via dt_declare_xattr_get(). It is not + * perfect solution, but works well now. + * + * We will improve it in the future. */ + update = osp->opd_async_requests; + if (update != NULL && update->dur_buf.ub_req != NULL && + update->dur_buf.ub_req->ourq_count > 0) { + osp->opd_async_requests = NULL; mutex_unlock(&osp->opd_async_requests_mutex); - osp_oac_xattr_put(oxe); + rc = osp_unplug_async_request(env, osp, update); } else { - /* XXX: Currently, we trigger the batched async OUT - * RPC via dt_declare_xattr_get(). It is not - * perfect solution, but works well now. - * - * We will improve it in the future. */ - update = osp->opd_async_requests; - if (update != NULL && update->dur_req != NULL && - update->dur_req->ourq_count > 0) { - osp->opd_async_requests = NULL; - mutex_unlock(&osp->opd_async_requests_mutex); - rc = osp_unplug_async_update(env, osp, update); - } else { - mutex_unlock(&osp->opd_async_requests_mutex); - } + mutex_unlock(&osp->opd_async_requests_mutex); } } return rc; } +/** + * Implement OSP layer dt_object_operations::do_xattr_get() interface. + * + * Get extended attribute from the specified MDT/OST object. + * + * If the extended attribute is in the OSP object attributes cache, then + * return the cached extended attribute directly. Otherwise it will get + * the extended attribute synchronously, if successful, add it to the OSP + * attributes cache. (\see lustre/osp/osp_trans.c for OUT RPC.) + * + * There is a race condition: some other thread has added the named extended + * attributed entry to the OSP object attributes cache during the current + * OUT_XATTR_GET handling. If such case happens, the OSP will replace the + * (just) existing extended attribute entry with the new replied one. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[out] buf pointer to the lu_buf to hold the extended attribute + * \param[in] name the name for the expected extended attribute + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ int osp_xattr_get(const struct lu_env *env, struct dt_object *dt, struct lu_buf *buf, const char *name, struct lustre_capa *capa) @@ -619,17 +952,27 @@ int osp_xattr_get(const struct lu_env *env, struct dt_object *dt, struct object_update_reply *reply; struct osp_xattr_entry *oxe = NULL; const char *dname = dt->do_lu.lo_dev->ld_obd->obd_name; - int namelen; int rc = 0; ENTRY; LASSERT(buf != NULL); LASSERT(name != NULL); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NETWORK) && + osp->opd_index == cfs_fail_val) { + if (is_ost_obj(&dt->do_lu)) { + if (osp_dev2node(osp) == cfs_fail_val) + RETURN(-ENOTCONN); + } else { + if (strcmp(name, XATTR_NAME_LINK) == 0) + RETURN(-ENOTCONN); + } + } + if (unlikely(obj->opo_non_exist)) RETURN(-ENOENT); - oxe = osp_oac_xattr_find(obj, name); + oxe = osp_oac_xattr_find(obj, name, false); if (oxe != NULL) { spin_lock(&obj->opo_lock); if (oxe->oxe_ready) { @@ -655,22 +998,25 @@ unlock: spin_unlock(&obj->opo_lock); } - update = out_create_update_req(dev); + update = dt_update_request_create(dev); if (IS_ERR(update)) GOTO(out, rc = PTR_ERR(update)); - namelen = strlen(name) + 1; - rc = out_insert_update(env, update, OUT_XATTR_GET, - lu_object_fid(&dt->do_lu), 1, &namelen, &name); + rc = out_xattr_get_pack(env, &update->dur_buf, + lu_object_fid(&dt->do_lu), name); if (rc != 0) { CERROR("%s: Insert update error "DFID": rc = %d\n", dname, PFID(lu_object_fid(&dt->do_lu)), rc); - GOTO(out, rc); } rc = out_remote_sync(env, osp->opd_obd->u.cli.cl_import, update, &req); if (rc != 0) { + if (rc == -ENOENT) { + dt->do_lu.lo_header->loh_attr &= ~LOHA_EXISTS; + obj->opo_non_exist = 1; + } + if (obj->opo_ooa == NULL) GOTO(out, rc); @@ -746,7 +1092,7 @@ unlock: "cache for "DFID": rc = %d\n", dname, name, PFID(lu_object_fid(&dt->do_lu)), rc); spin_lock(&obj->opo_lock); - oxe->oxe_ready = 0; + old->oxe_ready = 0; spin_unlock(&obj->opo_lock); GOTO(out, rc); @@ -770,7 +1116,7 @@ out: ptlrpc_req_finished(req); if (update != NULL && !IS_ERR(update)) - out_destroy_update_req(update); + dt_update_request_destroy(update); if (oxe != NULL) osp_oac_xattr_put(oxe); @@ -778,47 +1124,41 @@ out: return rc; } -int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, - const struct lu_buf *buf, const char *name, - int flag, struct thandle *th) +static int __osp_xattr_set(const struct lu_env *env, struct dt_object *dt, + const struct lu_buf *buf, const char *name, + int flag, struct thandle *th) { - struct osp_object *o = dt2osp_obj(dt); + struct osp_object *o = dt2osp_obj(dt); struct dt_update_request *update; - struct lu_fid *fid; struct osp_xattr_entry *oxe; - int sizes[3] = {strlen(name), buf->lb_len, - sizeof(int)}; - char *bufs[3] = {(char *)name, (char *)buf->lb_buf }; int rc; + ENTRY; LASSERT(buf->lb_len > 0 && buf->lb_buf != NULL); - update = out_find_create_update_loc(th, dt); + update = dt_update_request_find_or_create(th, dt); if (IS_ERR(update)) { CERROR("%s: Get OSP update buf failed "DFID": rc = %d\n", dt->do_lu.lo_dev->ld_obd->obd_name, PFID(lu_object_fid(&dt->do_lu)), (int)PTR_ERR(update)); - return PTR_ERR(update); + RETURN(PTR_ERR(update)); } - flag = cpu_to_le32(flag); - bufs[2] = (char *)&flag; - - fid = (struct lu_fid *)lu_object_fid(&dt->do_lu); - rc = out_insert_update(env, update, OUT_XATTR_SET, fid, - ARRAY_SIZE(sizes), sizes, (const char **)bufs); + rc = out_xattr_set_pack(env, &update->dur_buf, + lu_object_fid(&dt->do_lu), + buf, name, flag, update->dur_batchid); if (rc != 0 || o->opo_ooa == NULL) - return rc; + RETURN(rc); oxe = osp_oac_xattr_find_or_add(o, name, buf->lb_len); if (oxe == NULL) { - CWARN("%s: Fail to add xattr (%s) to cache for "DFID - ": rc = %d\n", dt->do_lu.lo_dev->ld_obd->obd_name, - name, PFID(lu_object_fid(&dt->do_lu)), rc); + CWARN("%s: Fail to add xattr (%s) to cache for "DFID, + dt->do_lu.lo_dev->ld_obd->obd_name, + name, PFID(lu_object_fid(&dt->do_lu))); - return 0; + RETURN(0); } if (oxe->oxe_buflen - oxe->oxe_namelen - 1 < buf->lb_len) { @@ -829,14 +1169,14 @@ int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, osp_oac_xattr_put(oxe); oxe = tmp; if (tmp == NULL) { - CWARN("%s: Fail to update xattr (%s) to cache for "DFID - ": rc = %d\n", dt->do_lu.lo_dev->ld_obd->obd_name, - name, PFID(lu_object_fid(&dt->do_lu)), rc); + CWARN("%s: Fail to update xattr (%s) to cache for "DFID, + dt->do_lu.lo_dev->ld_obd->obd_name, + name, PFID(lu_object_fid(&dt->do_lu))); spin_lock(&o->opo_lock); - oxe->oxe_ready = 0; + old->oxe_ready = 0; spin_unlock(&o->opo_lock); - return 0; + RETURN(0); } /* Drop the ref for entry on list. */ @@ -851,19 +1191,208 @@ int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, spin_unlock(&o->opo_lock); osp_oac_xattr_put(oxe); - return 0; + RETURN(0); +} + +/** + * Implement OSP layer dt_object_operations::do_declare_xattr_set() interface. + * + * Declare that the caller will set extended attribute to the specified + * MDT/OST object. + * + * This function will add an OUT_XATTR_SET sub-request to the per + * OSP-transaction based request queue which will be flushed when + * the transaction starts. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] buf pointer to the lu_buf to hold the extended attribute + * \param[in] name the name of the extended attribute to be set + * \param[in] flag to indicate the detailed set operation: LU_XATTR_CREATE + * or LU_XATTR_REPLACE or others + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ +int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt, + const struct lu_buf *buf, const char *name, + int flag, struct thandle *th) +{ + int rc = 0; + + /* Please check the comment in osp_attr_set() for handling + * remote transaction. */ + if (!is_only_remote_trans(th)) + rc = __osp_xattr_set(env, dt, buf, name, flag, th); + + return rc; } +/** + * Implement OSP layer dt_object_operations::do_xattr_set() interface. + * + * Set extended attribute to the specified MDT/OST object. + * + * The real modification sub-request has been added in the declare phase + * and related (OUT) RPC has been triggered when transaction start. + * + * If the OSP attributes cache is initialized, then check whether the name + * extended attribute entry exists in the cache or not. If yes, replace it; + * otherwise, add the extended attribute to the cache. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] buf pointer to the lu_buf to hold the extended attribute + * \param[in] name the name of the extended attribute to be set + * \param[in] fl to indicate the detailed set operation: LU_XATTR_CREATE + * or LU_XATTR_REPLACE or others + * \param[in] th pointer to the transaction handler + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ int osp_xattr_set(const struct lu_env *env, struct dt_object *dt, const struct lu_buf *buf, const char *name, int fl, struct thandle *th, struct lustre_capa *capa) { + int rc = 0; + CDEBUG(D_INFO, "xattr %s set object "DFID"\n", name, PFID(&dt->do_lu.lo_header->loh_fid)); + /* Please check the comment in osp_attr_set() for handling + * remote transaction. */ + if (is_only_remote_trans(th)) + rc = __osp_xattr_set(env, dt, buf, name, fl, th); + + return rc; +} + +static int __osp_xattr_del(const struct lu_env *env, struct dt_object *dt, + const char *name, struct thandle *th) +{ + struct dt_update_request *update; + const struct lu_fid *fid; + struct osp_object *o = dt2osp_obj(dt); + struct osp_xattr_entry *oxe; + int rc; + + update = dt_update_request_find_or_create(th, dt); + if (IS_ERR(update)) + return PTR_ERR(update); + + fid = lu_object_fid(&dt->do_lu); + + rc = out_xattr_del_pack(env, &update->dur_buf, fid, name, + update->dur_batchid); + + if (rc != 0 || o->opo_ooa == NULL) + return rc; + + oxe = osp_oac_xattr_find(o, name, true); + if (oxe != NULL) + /* Drop the ref for entry on list. */ + osp_oac_xattr_put(oxe); + return 0; } +/** + * Implement OSP layer dt_object_operations::do_declare_xattr_del() interface. + * + * Declare that the caller will delete extended attribute on the specified + * MDT/OST object. + * + * This function will add an OUT_XATTR_DEL sub-request to the per + * OSP-transaction based request queue which will be flushed when + * transaction start. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] name the name of the extended attribute to be set + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ +int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt, + const char *name, struct thandle *th) +{ + int rc = 0; + + /* Please check the comment in osp_attr_set() for handling + * remote transaction. */ + if (!is_only_remote_trans(th)) + rc = __osp_xattr_del(env, dt, name, th); + + return rc; +} + +/** + * Implement OSP layer dt_object_operations::do_xattr_del() interface. + * + * Delete extended attribute on the specified MDT/OST object. + * + * The real modification sub-request has been added in the declare phase + * and related (OUT) RPC has been triggered when transaction start. + * + * If the name extended attribute entry exists in the OSP attributes + * cache, then remove it from the cache. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] name the name of the extended attribute to be set + * \param[in] th pointer to the transaction handler + * \param[in] capa the capability for this operation + * + * \retval 0 for success + * \retval negative error number on failure + */ +int osp_xattr_del(const struct lu_env *env, struct dt_object *dt, + const char *name, struct thandle *th, + struct lustre_capa *capa) +{ + int rc = 0; + + CDEBUG(D_INFO, "xattr %s del object "DFID"\n", name, + PFID(&dt->do_lu.lo_header->loh_fid)); + + /* Please check the comment in osp_attr_set() for handling + * remote transaction. */ + if (is_only_remote_trans(th)) + rc = __osp_xattr_del(env, dt, name, th); + + return rc; +} + +/** + * Implement OSP layer dt_object_operations::do_declare_create() interface. + * + * Declare that the caller will create the OST object. + * + * If the transaction is a remote transaction (please refer to the + * comment of osp_trans_create() for remote transaction), then the FID + * for the OST object has been assigned already, and will be handled + * as create (remote) MDT object via osp_md_declare_object_create(). + * This function is usually used for LFSCK to re-create the lost OST + * object. Otherwise, if it is not replay case, the OSP will reserve + * pre-created object for the subsequent create operation; if the MDT + * side cached pre-created objects are less than some threshold, then + * it will wakeup the pre-create thread. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] attr the attribute for the object to be created + * \param[in] hint pointer to the hint for creating the object, such as + * the parent object + * \param[in] dof pointer to the dt_object_format for help the creation + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_declare_object_create(const struct lu_env *env, struct dt_object *dt, struct lu_attr *attr, @@ -943,6 +1472,30 @@ static int osp_declare_object_create(const struct lu_env *env, RETURN(rc); } +/** + * Implement OSP layer dt_object_operations::do_create() interface. + * + * Create the OST object. + * + * For remote transaction case, the real create sub-request has been + * added in the declare phase and related (OUT) RPC has been triggered + * at transaction start. Here, like creating (remote) MDT object, the + * OSP will mark the object existence via osp_md_object_create(). + * + * For non-remote transaction case, the OSP will assign FID to the + * object to be created, and update last_used Object ID (OID) file. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] attr the attribute for the object to be created + * \param[in] hint pointer to the hint for creating the object, such as + * the parent object + * \param[in] dof pointer to the dt_object_format for help the creation + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, struct lu_attr *attr, struct dt_allocation_hint *hint, @@ -967,7 +1520,7 @@ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, o->opo_non_exist = 0; if (o->opo_reserved) { - /* regular case, fid is assigned holding trunsaction open */ + /* regular case, fid is assigned holding transaction open */ osp_object_assign_fid(env, d, o); } @@ -1013,7 +1566,7 @@ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, d->opd_gap_count = 0; spin_unlock(&d->opd_pre_lock); - CDEBUG(D_HA, "Writting gap "DFID"+%d in llog\n", + CDEBUG(D_HA, "Writing gap "DFID"+%d in llog\n", PFID(&d->opd_gap_start_fid), count); /* real gap handling is disabled intil ORI-692 will be * fixed, now we only report gaps */ @@ -1040,6 +1593,21 @@ static int osp_object_create(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } +/** + * Implement OSP layer dt_object_operations::do_declare_destroy() interface. + * + * Declare that the caller will destroy the specified OST object. + * + * The OST object destroy will be handled via llog asynchronously. This + * function will declare the credits for generating MDS_UNLINK64_REC llog. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object to be destroyed + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ int osp_declare_object_destroy(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { @@ -1056,6 +1624,23 @@ int osp_declare_object_destroy(const struct lu_env *env, RETURN(rc); } +/** + * Implement OSP layer dt_object_operations::do_destroy() interface. + * + * Destroy the specified OST object. + * + * The OSP generates a MDS_UNLINK64_REC record in the llog. There + * will be some dedicated thread to handle the llog asynchronously. + * + * It also marks the object as non-cached. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object to be destroyed + * \param[in] th pointer to the transaction handler + * + * \retval 0 for success + * \retval negative error number on failure + */ int osp_object_destroy(const struct lu_env *env, struct dt_object *dt, struct thandle *th) { @@ -1077,21 +1662,6 @@ int osp_object_destroy(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } -struct osp_orphan_it { - int ooi_pos0; - int ooi_pos1; - int ooi_pos2; - int ooi_total_npages; - int ooi_valid_npages; - unsigned int ooi_swab:1; - __u64 ooi_next; - struct dt_object *ooi_obj; - struct lu_orphan_ent *ooi_ent; - struct page *ooi_cur_page; - struct lu_idxpage *ooi_cur_idxpage; - struct page **ooi_pages; -}; - static int osp_orphan_index_lookup(const struct lu_env *env, struct dt_object *dt, struct dt_rec *rec, @@ -1138,30 +1708,44 @@ static int osp_orphan_index_delete(const struct lu_env *env, return -EOPNOTSUPP; } -static struct dt_it *osp_orphan_it_init(const struct lu_env *env, - struct dt_object *dt, - __u32 attr, - struct lustre_capa *capa) +/** + * Initialize the OSP layer index iteration. + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the index object to be iterated + * \param[in] attr unused + * \param[in] capa the capability for this operation + * + * \retval pointer to the iteration structure + * \retval negative error number on failure + */ +struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt, + __u32 attr, struct lustre_capa *capa) { - struct osp_orphan_it *it; + struct osp_it *it; OBD_ALLOC_PTR(it); if (it == NULL) return ERR_PTR(-ENOMEM); - it->ooi_pos2 = -1; + it->ooi_pos_ent = -1; it->ooi_obj = dt; return (struct dt_it *)it; } -static void osp_orphan_it_fini(const struct lu_env *env, - struct dt_it *di) +/** + * Finalize the OSP layer index iteration. + * + * \param[in] env pointer to the thread context + * \param[in] di pointer to the iteration structure + */ +void osp_it_fini(const struct lu_env *env, struct dt_it *di) { - struct osp_orphan_it *it = (struct osp_orphan_it *)di; - struct page **pages = it->ooi_pages; - int npages = it->ooi_total_npages; - int i; + struct osp_it *it = (struct osp_it *)di; + struct page **pages = it->ooi_pages; + int npages = it->ooi_total_npages; + int i; if (pages != NULL) { for (i = 0; i < npages; i++) { @@ -1178,8 +1762,19 @@ static void osp_orphan_it_fini(const struct lu_env *env, OBD_FREE_PTR(it); } -static int osp_orphan_it_fetch(const struct lu_env *env, - struct osp_orphan_it *it) +/** + * Get more records for the iteration from peer. + * + * The new records will be filled in an array of pages. The OSP side + * allows 1MB bulk data to be transfered. + * + * \param[in] env pointer to the thread context + * \param[in] it pointer to the iteration structure + * + * \retval 0 for success + * \retval negative error number on failure + */ +static int osp_it_fetch(const struct lu_env *env, struct osp_it *it) { struct lu_device *dev = it->ooi_obj->do_lu.lo_dev; struct osp_device *osp = lu2osp_dev(dev); @@ -1219,7 +1814,27 @@ static int osp_orphan_it_fetch(const struct lu_env *env, RETURN(rc); } - req->rq_request_portal = OST_IDX_PORTAL; + req->rq_request_portal = OUT_PORTAL; + ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO); + memset(ii, 0, sizeof(*ii)); + if (fid_is_last_id(lu_object_fid(&it->ooi_obj->do_lu))) { + /* LFSCK will iterate orphan object[FID_SEQ_LAYOUT_BTREE, + * ost_index, 0] with LAST_ID FID, so it needs to replace + * the FID with orphan FID here */ + ii->ii_fid.f_seq = FID_SEQ_LAYOUT_RBTREE; + ii->ii_fid.f_oid = osp->opd_index; + ii->ii_fid.f_ver = 0; + ii->ii_flags = II_FL_NOHASH; + } else { + ii->ii_fid = *lu_object_fid(&it->ooi_obj->do_lu); + ii->ii_flags = II_FL_NOHASH | II_FL_NOKEY | II_FL_VARKEY | + II_FL_VARREC; + } + ii->ii_magic = IDX_INFO_MAGIC; + ii->ii_count = npages * LU_PAGE_COUNT; + ii->ii_hash_start = it->ooi_next; + ii->ii_attrs = osp_dev2node(osp); + ptlrpc_at_set_req_timeout(req); desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK, @@ -1232,18 +1847,6 @@ static int osp_orphan_it_fetch(const struct lu_env *env, for (i = 0; i < npages; i++) ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE); - ii = req_capsule_client_get(&req->rq_pill, &RMF_IDX_INFO); - memset(ii, 0, sizeof(*ii)); - ii->ii_fid.f_seq = FID_SEQ_LAYOUT_RBTREE; - ii->ii_fid.f_oid = osp->opd_index; - ii->ii_fid.f_ver = 0; - ii->ii_magic = IDX_INFO_MAGIC; - ii->ii_flags = II_FL_NOHASH; - ii->ii_count = npages * LU_PAGE_COUNT; - ii->ii_hash_start = it->ooi_next; - ii->ii_attrs = - osp->opd_storage->dd_lu_dev.ld_site->ld_seq_site->ss_node_id; - ptlrpc_request_set_replen(req); rc = ptlrpc_queue_wait(req); if (rc != 0) @@ -1253,6 +1856,7 @@ static int osp_orphan_it_fetch(const struct lu_env *env, req->rq_bulk->bd_nob_transferred); if (rc < 0) GOTO(out, rc); + rc = 0; ii = req_capsule_server_get(&req->rq_pill, &RMF_IDX_INFO); if (ii->ii_magic != IDX_INFO_MAGIC) @@ -1272,22 +1876,36 @@ static int osp_orphan_it_fetch(const struct lu_env *env, it->ooi_next = ii->ii_hash_end; - GOTO(out, rc = 0); - out: ptlrpc_req_finished(req); return rc; } -static int osp_orphan_it_next(const struct lu_env *env, - struct dt_it *di) +/** + * Move the iteration cursor to the next lu_page. + * + * One system page (PAGE_SIZE) may contain multiple lu_page (4KB), + * that depends on the LU_PAGE_COUNT. If it is not the last lu_page + * in current system page, then move the iteration cursor to the next + * lu_page in current system page. Otherwise, if there are more system + * pages in the cache, then move the iteration cursor to the next system + * page. If all the cached records (pages) have been iterated, then fetch + * more records via osp_it_fetch(). + * + * \param[in] env pointer to the thread context + * \param[in] di pointer to the iteration structure + * + * \retval 0 for success + * \retval negative error number on failure + */ +int osp_it_next_page(const struct lu_env *env, struct dt_it *di) { - struct osp_orphan_it *it = (struct osp_orphan_it *)di; - struct lu_idxpage *idxpage; + struct osp_it *it = (struct osp_it *)di; + struct lu_idxpage *idxpage; struct page **pages; - int rc; - int i; + int rc; + int i; ENTRY; again2: @@ -1296,23 +1914,17 @@ again2: if (idxpage->lip_nr == 0) RETURN(1); - it->ooi_pos2++; - if (it->ooi_pos2 < idxpage->lip_nr) { - it->ooi_ent = - (struct lu_orphan_ent *)idxpage->lip_entries + - it->ooi_pos2; - if (it->ooi_swab) - lustre_swab_orphan_ent(it->ooi_ent); + if (it->ooi_pos_ent < idxpage->lip_nr) { + CDEBUG(D_INFO, "ooi_pos %d nr %d\n", + (int)it->ooi_pos_ent, (int)idxpage->lip_nr); RETURN(0); } - it->ooi_cur_idxpage = NULL; - it->ooi_pos1++; - + it->ooi_pos_lu_page++; again1: - if (it->ooi_pos1 < LU_PAGE_COUNT) { + if (it->ooi_pos_lu_page < LU_PAGE_COUNT) { it->ooi_cur_idxpage = (void *)it->ooi_cur_page + - LU_PAGE_SIZE * it->ooi_pos1; + LU_PAGE_SIZE * it->ooi_pos_lu_page; if (it->ooi_swab) lustre_swab_lip_header(it->ooi_cur_idxpage); if (it->ooi_cur_idxpage->lip_magic != LIP_MAGIC) { @@ -1323,24 +1935,25 @@ again1: "%d/%d while read layout orphan index\n", osp->opd_obd->obd_name, it->ooi_cur_idxpage->lip_magic, - LIP_MAGIC, it->ooi_pos0, it->ooi_pos1); + LIP_MAGIC, it->ooi_pos_page, + it->ooi_pos_lu_page); /* Skip this lu_page next time. */ - it->ooi_pos2 = idxpage->lip_nr - 1; + it->ooi_pos_ent = idxpage->lip_nr - 1; RETURN(-EINVAL); } - it->ooi_pos2 = -1; + it->ooi_pos_ent = -1; goto again2; } kunmap(it->ooi_cur_page); it->ooi_cur_page = NULL; - it->ooi_pos0++; + it->ooi_pos_page++; again0: pages = it->ooi_pages; - if (it->ooi_pos0 < it->ooi_valid_npages) { - it->ooi_cur_page = kmap(pages[it->ooi_pos0]); - it->ooi_pos1 = 0; + if (it->ooi_pos_page < it->ooi_valid_npages) { + it->ooi_cur_page = kmap(pages[it->ooi_pos_page]); + it->ooi_pos_lu_page = 0; goto again1; } @@ -1350,7 +1963,7 @@ again0: } OBD_FREE(pages, it->ooi_total_npages * sizeof(*pages)); - it->ooi_pos0 = 0; + it->ooi_pos_page = 0; it->ooi_total_npages = 0; it->ooi_valid_npages = 0; it->ooi_swab = 0; @@ -1363,30 +1976,72 @@ again0: if (it->ooi_next == II_END_OFF) RETURN(1); - rc = osp_orphan_it_fetch(env, it); + rc = osp_it_fetch(env, it); if (rc == 0) goto again0; RETURN(rc); } -static int osp_orphan_it_get(const struct lu_env *env, - struct dt_it *di, - const struct dt_key *key) +/** + * Move the iteration cursor to the next record. + * + * If there are more records in the lu_page, then move the iteration + * cursor to the next record directly. Otherwise, move the iteration + * cursor to the record in the next lu_page via osp_it_next_page() + * + * \param[in] env pointer to the thread context + * \param[in] di pointer to the iteration structure + * + * \retval 0 for success + * \retval negative error number on failure + */ +int osp_orphan_it_next(const struct lu_env *env, struct dt_it *di) +{ + struct osp_it *it = (struct osp_it *)di; + struct lu_idxpage *idxpage; + int rc; + ENTRY; + +again: + idxpage = it->ooi_cur_idxpage; + if (idxpage != NULL) { + if (idxpage->lip_nr == 0) + RETURN(1); + + it->ooi_pos_ent++; + if (it->ooi_pos_ent < idxpage->lip_nr) { + it->ooi_ent = + (struct lu_orphan_ent *)idxpage->lip_entries + + it->ooi_pos_ent; + if (it->ooi_swab) + lustre_swab_orphan_ent(it->ooi_ent); + RETURN(0); + } + } + + rc = osp_it_next_page(env, di); + if (rc == 0) + goto again; + + RETURN(rc); +} + +int osp_it_get(const struct lu_env *env, struct dt_it *di, + const struct dt_key *key) { - return -ENOSYS; + return 1; } -static void osp_orphan_it_put(const struct lu_env *env, - struct dt_it *di) +void osp_it_put(const struct lu_env *env, struct dt_it *di) { } -static struct dt_key *osp_orphan_it_key(const struct lu_env *env, - const struct dt_it *di) +struct dt_key *osp_orphan_it_key(const struct lu_env *env, + const struct dt_it *di) { - struct osp_orphan_it *it = (struct osp_orphan_it *)di; - struct lu_orphan_ent *ent = it->ooi_ent; + struct osp_it *it = (struct osp_it *)di; + struct lu_orphan_ent *ent = (struct lu_orphan_ent *)it->ooi_ent; if (likely(ent != NULL)) return (struct dt_key *)(&ent->loe_key); @@ -1394,19 +2049,16 @@ static struct dt_key *osp_orphan_it_key(const struct lu_env *env, return NULL; } -static int osp_orphan_it_key_size(const struct lu_env *env, - const struct dt_it *di) +int osp_orphan_it_key_size(const struct lu_env *env, const struct dt_it *di) { return sizeof(struct lu_fid); } -static int osp_orphan_it_rec(const struct lu_env *env, - const struct dt_it *di, - struct dt_rec *rec, - __u32 attr) +int osp_orphan_it_rec(const struct lu_env *env, const struct dt_it *di, + struct dt_rec *rec, __u32 attr) { - struct osp_orphan_it *it = (struct osp_orphan_it *)di; - struct lu_orphan_ent *ent = it->ooi_ent; + struct osp_it *it = (struct osp_it *)di; + struct lu_orphan_ent *ent = (struct lu_orphan_ent *)it->ooi_ent; if (likely(ent != NULL)) { *(struct lu_orphan_rec *)rec = ent->loe_rec; @@ -1416,26 +2068,30 @@ static int osp_orphan_it_rec(const struct lu_env *env, return -EINVAL; } -static __u64 osp_orphan_it_store(const struct lu_env *env, - const struct dt_it *di) +__u64 osp_it_store(const struct lu_env *env, const struct dt_it *di) { - struct osp_orphan_it *it = (struct osp_orphan_it *)di; + struct osp_it *it = (struct osp_it *)di; return it->ooi_next; } /** - * \retval +1: locate to the exactly position - * \retval 0: cannot locate to the exactly position, - * call next() to move to a valid position. - * \retval -ve: on error + * Locate the iteration cursor to the specified position (cookie). + * + * \param[in] env pointer to the thread context + * \param[in] di pointer to the iteration structure + * \param[in] hash the specified position + * + * \retval positive number for locating to the exactly position + * or the next + * \retval 0 for arriving at the end of the iteration + * \retval negative error number on failure */ -static int osp_orphan_it_load(const struct lu_env *env, - const struct dt_it *di, - __u64 hash) +int osp_orphan_it_load(const struct lu_env *env, const struct dt_it *di, + __u64 hash) { - struct osp_orphan_it *it = (struct osp_orphan_it *)di; - int rc; + struct osp_it *it = (struct osp_it *)di; + int rc; it->ooi_next = hash; rc = osp_orphan_it_next(env, (struct dt_it *)di); @@ -1448,9 +2104,8 @@ static int osp_orphan_it_load(const struct lu_env *env, return rc; } -static int osp_orphan_it_key_rec(const struct lu_env *env, - const struct dt_it *di, - void *key_rec) +int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di, + void *key_rec) { return 0; } @@ -1462,33 +2117,46 @@ static const struct dt_index_operations osp_orphan_index_ops = { .dio_declare_delete = osp_orphan_index_declare_delete, .dio_delete = osp_orphan_index_delete, .dio_it = { - .init = osp_orphan_it_init, - .fini = osp_orphan_it_fini, + .init = osp_it_init, + .fini = osp_it_fini, .next = osp_orphan_it_next, - .get = osp_orphan_it_get, - .put = osp_orphan_it_put, + .get = osp_it_get, + .put = osp_it_put, .key = osp_orphan_it_key, .key_size = osp_orphan_it_key_size, .rec = osp_orphan_it_rec, - .store = osp_orphan_it_store, + .store = osp_it_store, .load = osp_orphan_it_load, - .key_rec = osp_orphan_it_key_rec, + .key_rec = osp_it_key_rec, } }; +/** + * Implement OSP layer dt_object_operations::do_index_try() interface. + * + * Negotiate the index type. + * + * If the target index is an IDIF object, then use osp_orphan_index_ops. + * Otherwise, assign osp_md_index_ops to the dt_object::do_index_ops. + * (\see lustre/include/lustre_fid.h for IDIF.) + * + * \param[in] env pointer to the thread context + * \param[in] dt pointer to the OSP layer dt_object + * \param[in] feat unused + * + * \retval 0 for success + */ static int osp_index_try(const struct lu_env *env, struct dt_object *dt, const struct dt_index_features *feat) { const struct lu_fid *fid = lu_object_fid(&dt->do_lu); - if (fid_is_last_id(fid) && fid_is_idif(fid)) { + if (fid_is_last_id(fid) && fid_is_idif(fid)) dt->do_index_ops = &osp_orphan_index_ops; - - return 0; - } - - return -EINVAL; + else + dt->do_index_ops = &osp_md_index_ops; + return 0; } struct dt_object_operations osp_obj_ops = { @@ -1507,6 +2175,21 @@ struct dt_object_operations osp_obj_ops = { .do_index_try = osp_index_try, }; +/** + * Implement OSP layer lu_object_operations::loo_object_init() interface. + * + * Initialize the object. + * + * If it is a remote MDT object, then call do_attr_get() to fetch + * the attribute from the peer. + * + * \param[in] env pointer to the thread context + * \param[in] o pointer to the OSP layer lu_object + * \param[in] conf unused + * + * \retval 0 for success + * \retval negative error number on failure + */ static int osp_object_init(const struct lu_env *env, struct lu_object *o, const struct lu_object_conf *conf) { @@ -1538,6 +2221,17 @@ static int osp_object_init(const struct lu_env *env, struct lu_object *o, RETURN(rc); } +/** + * Implement OSP layer lu_object_operations::loo_object_free() interface. + * + * Finalize the object. + * + * If the OSP object has attributes cache, then destroy the cache. + * Free the object finally. + * + * \param[in] env pointer to the thread context + * \param[in] o pointer to the OSP layer lu_object + */ static void osp_object_free(const struct lu_env *env, struct lu_object *o) { struct osp_object *obj = lu2osp_obj(o); @@ -1557,7 +2251,7 @@ static void osp_object_free(const struct lu_env *env, struct lu_object *o) count = atomic_read(&oxe->oxe_ref); LASSERTF(count == 1, "Still has %d users on the xattr entry %.*s\n", - count - 1, oxe->oxe_namelen, oxe->oxe_buf); + count-1, (int)oxe->oxe_namelen, oxe->oxe_buf); OBD_FREE(oxe, oxe->oxe_buflen); } @@ -1566,6 +2260,17 @@ static void osp_object_free(const struct lu_env *env, struct lu_object *o) OBD_SLAB_FREE_PTR(obj, osp_object_kmem); } +/** + * Implement OSP layer lu_object_operations::loo_object_release() interface. + * + * Cleanup (not free) the object. + * + * If it is a reserved object but failed to be created, or it is an OST + * object, then mark the object as non-cached. + * + * \param[in] env pointer to the thread context + * \param[in] o pointer to the OSP layer lu_object + */ static void osp_object_release(const struct lu_env *env, struct lu_object *o) { struct osp_object *po = lu2osp_obj(o);