X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=cf82eb024ae72f2a40f723bca551eb69bc437b61;hp=8d4b2b5329c4edd55f1f905e01dc77539645abf5;hb=8daba6a7381a2fb8cc933f7e9486f60e659465d4;hpb=0bdd2c864c26e6f48c57505aa6e4143bf1497833 diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 8d4b2b5..cf82eb0 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2010, 2013, Intel Corporation. + * Copyright (c) 2010, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -53,6 +53,7 @@ * struct OBD_{ALLOC,FREE}*() */ #include +#include /* struct ptlrpc_request */ #include /* struct obd_export */ @@ -67,8 +68,8 @@ #include #include #include -#include #include +#include mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_MINMODE] = MDL_MINMODE, @@ -135,21 +136,25 @@ int mdt_get_disposition(struct ldlm_reply *rep, int flag) } void mdt_clear_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, int flag) { - if (info) - info->mti_opdata &= ~flag; - if (rep) - rep->lock_policy_res1 &= ~flag; + if (info) { + info->mti_opdata &= ~flag; + tgt_opdata_clear(info->mti_env, flag); + } + if (rep) + rep->lock_policy_res1 &= ~flag; } void mdt_set_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, int flag) { - if (info) - info->mti_opdata |= flag; - if (rep) - rep->lock_policy_res1 |= flag; + if (info) { + info->mti_opdata |= flag; + tgt_opdata_set(info->mti_env, flag); + } + if (rep) + rep->lock_policy_res1 |= flag; } void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm) @@ -160,27 +165,28 @@ void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm) lh->mlh_type = MDT_REG_LOCK; } -void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm, - const char *name, int namelen) +void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode, + const struct lu_name *lname) { - lh->mlh_reg_mode = lm; - lh->mlh_rreg_mode = lm; - lh->mlh_type = MDT_PDO_LOCK; + lh->mlh_reg_mode = lock_mode; + lh->mlh_rreg_mode = lock_mode; + lh->mlh_type = MDT_PDO_LOCK; - if (name != NULL && (name[0] != '\0')) { - LASSERT(namelen > 0); - lh->mlh_pdo_hash = full_name_hash(name, namelen); + if (lu_name_is_valid(lname)) { + lh->mlh_pdo_hash = full_name_hash(lname->ln_name, + lname->ln_namelen); /* XXX Workaround for LU-2856 - * Zero is a valid return value of full_name_hash, but several - * users of mlh_pdo_hash assume a non-zero hash value. We - * therefore map zero onto an arbitrary, but consistent - * value (1) to avoid problems further down the road. */ - if (unlikely(!lh->mlh_pdo_hash)) + * + * Zero is a valid return value of full_name_hash, but + * several users of mlh_pdo_hash assume a non-zero + * hash value. We therefore map zero onto an + * arbitrary, but consistent value (1) to avoid + * problems further down the road. */ + if (unlikely(lh->mlh_pdo_hash == 0)) lh->mlh_pdo_hash = 1; - } else { - LASSERT(namelen == 0); - lh->mlh_pdo_hash = 0ull; - } + } else { + lh->mlh_pdo_hash = 0; + } } static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o, @@ -258,7 +264,7 @@ static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o EXIT; } -int mdt_getstatus(struct tgt_session_info *tsi) +static int mdt_getstatus(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_device *mdt = info->mti_mdt; @@ -275,15 +281,15 @@ int mdt_getstatus(struct tgt_session_info *tsi) GOTO(out, rc = err_serious(-ENOMEM)); repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - repbody->fid1 = mdt->mdt_md_root_fid; - repbody->valid |= OBD_MD_FLID; + repbody->mbo_fid1 = mdt->mdt_md_root_fid; + repbody->mbo_valid |= OBD_MD_FLID; if (tsi->tsi_tgt->lut_mds_capa && exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { struct mdt_object *root; struct lustre_capa *capa; - root = mdt_object_find(info->mti_env, mdt, &repbody->fid1); + root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1); if (IS_ERR(root)) GOTO(out, rc = PTR_ERR(root)); @@ -294,7 +300,7 @@ int mdt_getstatus(struct tgt_session_info *tsi) 0); mdt_object_put(info->mti_env, root); if (rc == 0) - repbody->valid |= OBD_MD_FLMDSCAPA; + repbody->mbo_valid |= OBD_MD_FLMDSCAPA; } EXIT; out: @@ -302,7 +308,7 @@ out: return rc; } -int mdt_statfs(struct tgt_session_info *tsi) +static int mdt_statfs(struct tgt_session_info *tsi) { struct ptlrpc_request *req = tgt_ses_req(tsi); struct mdt_thread_info *info = tsi2mdt_info(tsi); @@ -356,100 +362,159 @@ out: RETURN(rc); } -/** - * Pack SOM attributes into the reply. - * Call under a DLM UPDATE lock. +#ifdef CONFIG_FS_POSIX_ACL +/* + * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap. + * + * \param info thread info object + * \param repbody reply to pack ACLs into + * \param o mdt object of file to examine + * \param nodemap nodemap of client to reply to + * \retval 0 success + * \retval -errno error getting or parsing ACL from disk */ -static void mdt_pack_size2body(struct mdt_thread_info *info, - struct mdt_object *mo) +int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody, + struct mdt_object *o, struct lu_nodemap *nodemap) { - struct mdt_body *b; - struct md_attr *ma = &info->mti_attr; - - LASSERT(ma->ma_attr.la_valid & LA_MODE); - b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - - /* Check if Size-on-MDS is supported, if this is a regular file, - * if SOM is enabled on the object and if SOM cache exists and valid. - * Otherwise do not pack Size-on-MDS attributes to the reply. */ - if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) || - !S_ISREG(ma->ma_attr.la_mode) || - !mdt_object_is_som_enabled(mo) || - !(ma->ma_valid & MA_SOM)) - return; + const struct lu_env *env = info->mti_env; + struct md_object *next = mdt_object_child(o); + struct lu_buf *buf = &info->mti_buf; + int rc; - b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; - b->size = ma->ma_som->msd_size; - b->blocks = ma->ma_som->msd_blocks; + buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL); + buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL, + RCL_SERVER); + if (buf->lb_len == 0) + return 0; + + rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS); + if (rc < 0) { + if (rc == -ENODATA) { + repbody->mbo_aclsize = 0; + repbody->mbo_valid |= OBD_MD_FLACL; + rc = 0; + } else if (rc == -EOPNOTSUPP) { + rc = 0; + } else { + CERROR("%s: unable to read "DFID" ACL: rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); + } + } else { + rc = nodemap_map_acl(nodemap, buf->lb_buf, + rc, NODEMAP_FS_TO_CLIENT); + /* if all ACLs mapped out, rc is still >= 0 */ + if (rc < 0) { + CERROR("%s: nodemap_map_acl unable to parse "DFID + " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o)), rc); + } else { + repbody->mbo_aclsize = rc; + repbody->mbo_valid |= OBD_MD_FLACL; + rc = 0; + } + } + return rc; } +#endif void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, const struct lu_attr *attr, const struct lu_fid *fid) { - struct md_attr *ma = &info->mti_attr; - - LASSERT(ma->ma_valid & MA_INODE); - - b->atime = attr->la_atime; - b->mtime = attr->la_mtime; - b->ctime = attr->la_ctime; - b->mode = attr->la_mode; - b->size = attr->la_size; - b->blocks = attr->la_blocks; - b->uid = attr->la_uid; - b->gid = attr->la_gid; - b->flags = attr->la_flags; - b->nlink = attr->la_nlink; - b->rdev = attr->la_rdev; - - /*XXX should pack the reply body according to lu_valid*/ - b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID | - OBD_MD_FLGID | OBD_MD_FLTYPE | - OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS | - OBD_MD_FLATIME | OBD_MD_FLMTIME ; - - if (!S_ISREG(attr->la_mode)) { - b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV; + struct md_attr *ma = &info->mti_attr; + struct obd_export *exp = info->mti_exp; + struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap; + + LASSERT(ma->ma_valid & MA_INODE); + + if (attr->la_valid & LA_ATIME) { + b->mbo_atime = attr->la_atime; + b->mbo_valid |= OBD_MD_FLATIME; + } + if (attr->la_valid & LA_MTIME) { + b->mbo_mtime = attr->la_mtime; + b->mbo_valid |= OBD_MD_FLMTIME; + } + if (attr->la_valid & LA_CTIME) { + b->mbo_ctime = attr->la_ctime; + b->mbo_valid |= OBD_MD_FLCTIME; + } + if (attr->la_valid & LA_FLAGS) { + b->mbo_flags = attr->la_flags; + b->mbo_valid |= OBD_MD_FLFLAGS; + } + if (attr->la_valid & LA_NLINK) { + b->mbo_nlink = attr->la_nlink; + b->mbo_valid |= OBD_MD_FLNLINK; + } + if (attr->la_valid & LA_UID) { + b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_FS_TO_CLIENT, + attr->la_uid); + b->mbo_valid |= OBD_MD_FLUID; + } + if (attr->la_valid & LA_GID) { + b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_FS_TO_CLIENT, + attr->la_gid); + b->mbo_valid |= OBD_MD_FLGID; + } + b->mbo_mode = attr->la_mode; + if (attr->la_valid & LA_MODE) + b->mbo_valid |= OBD_MD_FLMODE; + if (attr->la_valid & LA_TYPE) + b->mbo_valid |= OBD_MD_FLTYPE; + + if (fid != NULL) { + b->mbo_fid1 = *fid; + b->mbo_valid |= OBD_MD_FLID; + CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n", + PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid); + } + + if (info != NULL) + mdt_body_reverse_idmap(info, b); + + if (!(attr->la_valid & LA_TYPE)) + return; + + b->mbo_rdev = attr->la_rdev; + b->mbo_size = attr->la_size; + b->mbo_blocks = attr->la_blocks; + + if (!S_ISREG(attr->la_mode)) { + b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV; } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) { - /* means no objects are allocated on osts. */ - LASSERT(!(ma->ma_valid & MA_LOV)); - /* just ignore blocks occupied by extend attributes on MDS */ - b->blocks = 0; - /* if no object is allocated on osts, the size on mds is valid. b=22272 */ - b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + /* means no objects are allocated on osts. */ + LASSERT(!(ma->ma_valid & MA_LOV)); + /* just ignore blocks occupied by extend attributes on MDS */ + b->mbo_blocks = 0; + /* if no object is allocated on osts, the size on mds is valid. + * b=22272 */ + b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL && ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) { /* A released file stores its size on MDS. */ /* But return 1 block for released file, unless tools like tar * will consider it fully sparse. (LU-3864) */ - if (unlikely(b->size == 0)) - b->blocks = 0; + if (unlikely(b->mbo_size == 0)) + b->mbo_blocks = 0; else - b->blocks = 1; - b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + b->mbo_blocks = 1; + b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; } - if (fid) { - b->fid1 = *fid; - b->valid |= OBD_MD_FLID; - CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n", - PFID(fid), b->nlink, b->mode, b->size); - } - - if (info) - mdt_body_reverse_idmap(info, b); - - if (fid != NULL && (b->valid & OBD_MD_FLSIZE)) - CDEBUG(D_VFSTRACE, DFID": returning size %llu\n", - PFID(fid), (unsigned long long)b->size); + if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE)) + CDEBUG(D_VFSTRACE, DFID": returning size %llu\n", + PFID(fid), (unsigned long long)b->mbo_size); } static inline int mdt_body_has_lov(const struct lu_attr *la, - const struct mdt_body *body) + const struct mdt_body *body) { - return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) || - (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) ); + return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) || + (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA)); } void mdt_client_compatibility(struct mdt_thread_info *info) @@ -477,8 +542,41 @@ void mdt_client_compatibility(struct mdt_thread_info *info) EXIT; } -static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o, - char *name) +static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, + struct mdt_object *o) +{ + const struct lu_env *env = info->mti_env; + int rc, rc2; + + rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, + XATTR_NAME_LOV); + + if (rc == -ENODATA) + rc = 0; + + if (rc < 0) + goto out; + + /* Is it a directory? Let's check for the LMV as well */ + if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) { + rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, + XATTR_NAME_LMV); + + if (rc2 == -ENODATA) + rc2 = mo_xattr_get(env, mdt_object_child(o), + &LU_BUF_NULL, + XATTR_NAME_DEFAULT_LMV); + + if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc)) + rc = rc2; + } + +out: + return rc; +} + +int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o, + const char *name) { const struct lu_env *env = info->mti_env; int rc; @@ -516,43 +614,83 @@ static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o, RETURN(rc); } -int mdt_attr_get_lov(struct mdt_thread_info *info, - struct mdt_object *o, struct md_attr *ma) +int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o, + struct md_attr *ma, const char *name) { struct md_object *next = mdt_object_child(o); struct lu_buf *buf = &info->mti_buf; int rc; - buf->lb_buf = ma->ma_lmm; - buf->lb_len = ma->ma_lmm_size; - rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV); + if (strcmp(name, XATTR_NAME_LOV) == 0) { + buf->lb_buf = ma->ma_lmm; + buf->lb_len = ma->ma_lmm_size; + LASSERT(!(ma->ma_valid & MA_LOV)); + } else if (strcmp(name, XATTR_NAME_LMV) == 0) { + buf->lb_buf = ma->ma_lmv; + buf->lb_len = ma->ma_lmv_size; + LASSERT(!(ma->ma_valid & MA_LMV)); + } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) { + buf->lb_buf = ma->ma_lmv; + buf->lb_len = ma->ma_lmv_size; + LASSERT(!(ma->ma_valid & MA_LMV_DEF)); + } else { + return -EINVAL; + } + + rc = mo_xattr_get(info->mti_env, next, buf, name); if (rc > 0) { - ma->ma_lmm_size = rc; - ma->ma_valid |= MA_LOV; + +got: + if (strcmp(name, XATTR_NAME_LOV) == 0) { + if (info->mti_big_lmm_used) + ma->ma_lmm = info->mti_big_lmm; + + /* NOT return LOV EA with hole to old client. */ + if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) & + LOV_PATTERN_F_HOLE) && + !(exp_connect_flags(info->mti_exp) & + OBD_CONNECT_LFSCK)) { + return -EIO; + } else { + ma->ma_lmm_size = rc; + ma->ma_valid |= MA_LOV; + } + } else if (strcmp(name, XATTR_NAME_LMV) == 0) { + if (info->mti_big_lmm_used) + ma->ma_lmv = info->mti_big_lmm; + + ma->ma_lmv_size = rc; + ma->ma_valid |= MA_LMV; + } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) { + ma->ma_lmv_size = rc; + ma->ma_valid |= MA_LMV_DEF; + } + + /* Update mdt_max_mdsize so all clients will be aware that */ + if (info->mti_mdt->mdt_max_mdsize < rc) + info->mti_mdt->mdt_max_mdsize = rc; + rc = 0; } else if (rc == -ENODATA) { /* no LOV EA */ rc = 0; } else if (rc == -ERANGE) { - rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV); + /* Default LMV has fixed size, so it must be able to fit + * in the original buffer */ + if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) + return rc; + rc = mdt_big_xattr_get(info, o, name); if (rc > 0) { info->mti_big_lmm_used = 1; - ma->ma_valid |= MA_LOV; - ma->ma_lmm = info->mti_big_lmm; - ma->ma_lmm_size = rc; - /* update mdt_max_mdsize so all clients - * will be aware about that */ - if (info->mti_mdt->mdt_max_mdsize < rc) - info->mti_mdt->mdt_max_mdsize = rc; - rc = 0; + goto got; } } return rc; } -int mdt_attr_get_pfid(struct mdt_thread_info *info, - struct mdt_object *o, struct lu_fid *pfid) +static int mdt_attr_get_pfid(struct mdt_thread_info *info, + struct mdt_object *o, struct lu_fid *pfid) { struct lu_buf *buf = &info->mti_buf; struct link_ea_header *leh; @@ -604,13 +742,17 @@ int mdt_attr_get_complex(struct mdt_thread_info *info, const struct lu_env *env = info->mti_env; struct md_object *next = mdt_object_child(o); struct lu_buf *buf = &info->mti_buf; - u32 mode = lu_object_attr(&next->mo_lu); int need = ma->ma_need; int rc = 0, rc2; + u32 mode; ENTRY; ma->ma_valid = 0; + if (mdt_object_exists(o) == 0) + GOTO(out, rc = -ENOENT); + mode = lu_object_attr(&next->mo_lu); + if (need & MA_INODE) { ma->ma_need = MA_INODE; rc = mo_attr_get(env, next, ma); @@ -628,36 +770,21 @@ int mdt_attr_get_complex(struct mdt_thread_info *info, } if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) { - rc = mdt_attr_get_lov(info, o, ma); + rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV); if (rc) GOTO(out, rc); } if (need & MA_LMV && S_ISDIR(mode)) { - buf->lb_buf = ma->ma_lmv; - buf->lb_len = ma->ma_lmv_size; - rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV); - if (rc2 > 0) { - ma->ma_lmv_size = rc2; - ma->ma_valid |= MA_LMV; - } else if (rc2 == -ENODATA) { - /* no LMV EA */ - ma->ma_lmv_size = 0; - } else - GOTO(out, rc = rc2); + rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV); + if (rc != 0) + GOTO(out, rc); } - if (need & MA_SOM && S_ISREG(mode)) { - buf->lb_buf = info->mti_xattr_buf; - buf->lb_len = sizeof(info->mti_xattr_buf); - CLASSERT(sizeof(struct som_attrs) <= - sizeof(info->mti_xattr_buf)); - rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM); - rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som); - if (rc2 == 0) - ma->ma_valid |= MA_SOM; - else if (rc2 < 0 && rc2 != -ENODATA) - GOTO(out, rc = rc2); + if (need & MA_LMV_DEF && S_ISDIR(mode)) { + rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV); + if (rc != 0) + GOTO(out, rc); } if (need & MA_HSM && S_ISREG(mode)) { @@ -698,65 +825,82 @@ out: static int mdt_getattr_internal(struct mdt_thread_info *info, struct mdt_object *o, int ma_need) { - struct md_object *next = mdt_object_child(o); - const struct mdt_body *reqbody = info->mti_body; - struct ptlrpc_request *req = mdt_info_req(info); - struct md_attr *ma = &info->mti_attr; - struct lu_attr *la = &ma->ma_attr; - struct req_capsule *pill = info->mti_pill; - const struct lu_env *env = info->mti_env; - struct mdt_body *repbody; - struct lu_buf *buffer = &info->mti_buf; - int rc; - int is_root; - ENTRY; + struct md_object *next = mdt_object_child(o); + const struct mdt_body *reqbody = info->mti_body; + struct ptlrpc_request *req = mdt_info_req(info); + struct md_attr *ma = &info->mti_attr; + struct lu_attr *la = &ma->ma_attr; + struct req_capsule *pill = info->mti_pill; + const struct lu_env *env = info->mti_env; + struct mdt_body *repbody; + struct lu_buf *buffer = &info->mti_buf; + struct obd_export *exp = info->mti_exp; + struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap; + int rc; + int is_root; + ENTRY; - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) - RETURN(err_serious(-ENOMEM)); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) + RETURN(err_serious(-ENOMEM)); - repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); + repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); - ma->ma_valid = 0; + ma->ma_valid = 0; if (mdt_object_remote(o)) { /* This object is located on remote node.*/ - /* Return -EIO for old client */ + /* Return -ENOTSUPP for old client */ if (!mdt_is_dne_client(req->rq_export)) - GOTO(out, rc = -EIO); + GOTO(out, rc = -ENOTSUPP); - repbody->fid1 = *mdt_object_fid(o); - repbody->valid = OBD_MD_FLID | OBD_MD_MDS; + repbody->mbo_fid1 = *mdt_object_fid(o); + repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS; GOTO(out, rc = 0); } - buffer->lb_len = reqbody->eadatasize; - if (buffer->lb_len > 0) + if (reqbody->mbo_eadatasize > 0) { buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD); - else + if (buffer->lb_buf == NULL) + GOTO(out, rc = -EPROTO); + buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD, + RCL_SERVER); + } else { buffer->lb_buf = NULL; + buffer->lb_len = 0; + ma_need &= ~(MA_LOV | MA_LMV); + CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n", + mdt_obd_name(info->mti_mdt), + req->rq_export->exp_client_uuid.uuid); + } - /* If it is dir object and client require MEA, then we got MEA */ - if (S_ISDIR(lu_object_attr(&next->mo_lu)) && - reqbody->valid & OBD_MD_MEA) { - /* Assumption: MDT_MD size is enough for lmv size. */ - ma->ma_lmv = buffer->lb_buf; - ma->ma_lmv_size = buffer->lb_len; - ma->ma_need = MA_LMV | MA_INODE; - } else { - ma->ma_lmm = buffer->lb_buf; - ma->ma_lmm_size = buffer->lb_len; - ma->ma_need = MA_LOV | MA_INODE | MA_HSM; - } + /* If it is dir object and client require MEA, then we got MEA */ + if (S_ISDIR(lu_object_attr(&next->mo_lu)) && + (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) { + /* Assumption: MDT_MD size is enough for lmv size. */ + ma->ma_lmv = buffer->lb_buf; + ma->ma_lmv_size = buffer->lb_len; + ma->ma_need = MA_INODE; + if (ma->ma_lmv_size > 0) { + if (reqbody->mbo_valid & OBD_MD_MEA) + ma->ma_need |= MA_LMV; + else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) + ma->ma_need |= MA_LMV_DEF; + } + } else { + ma->ma_lmm = buffer->lb_buf; + ma->ma_lmm_size = buffer->lb_len; + ma->ma_need = MA_INODE | MA_HSM; + if (ma->ma_lmm_size > 0) + ma->ma_need |= MA_LOV; + } if (S_ISDIR(lu_object_attr(&next->mo_lu)) && - reqbody->valid & OBD_MD_FLDIREA && + reqbody->mbo_valid & OBD_MD_FLDIREA && lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) { /* get default stripe info for this dir. */ ma->ma_need |= MA_LOV_DEF; } ma->ma_need |= ma_need; - if (ma->ma_need & MA_SOM) - ma->ma_som = &info->mti_u.som.data; rc = mdt_attr_get_complex(info, o, ma); if (unlikely(rc)) { @@ -769,8 +913,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, /* if file is released, check if a restore is running */ if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) && mdt_hsm_restore_is_running(info, mdt_object_fid(o))) { - repbody->t_state = MS_RESTORE; - repbody->valid |= OBD_MD_TSTATE; + repbody->mbo_t_state = MS_RESTORE; + repbody->mbo_valid |= OBD_MD_TSTATE; } is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid); @@ -789,7 +933,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, root = mdt_object_find(env, mdt, &rootfid); if (IS_ERR(root)) RETURN(PTR_ERR(root)); - rc = mdt_attr_get_lov(info, root, ma); + rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV); mdt_object_put(info->mti_env, root); if (unlikely(rc)) { CERROR("%s: getattr error for "DFID": rc = %d\n", @@ -807,24 +951,38 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, if (mdt_body_has_lov(la, reqbody)) { if (ma->ma_valid & MA_LOV) { LASSERT(ma->ma_lmm_size); - mdt_dump_lmm(D_INFO, ma->ma_lmm); - repbody->eadatasize = ma->ma_lmm_size; - if (S_ISDIR(la->la_mode)) - repbody->valid |= OBD_MD_FLDIREA; - else - repbody->valid |= OBD_MD_FLEASIZE; - } - if (ma->ma_valid & MA_LMV) { - LASSERT(S_ISDIR(la->la_mode)); - repbody->eadatasize = ma->ma_lmv_size; - repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA); + repbody->mbo_eadatasize = ma->ma_lmm_size; + if (S_ISDIR(la->la_mode)) + repbody->mbo_valid |= OBD_MD_FLDIREA; + else + repbody->mbo_valid |= OBD_MD_FLEASIZE; + mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid); } + if (ma->ma_valid & MA_LMV) { + /* Return -ENOTSUPP for old client */ + if (!mdt_is_striped_client(req->rq_export)) + RETURN(-ENOTSUPP); + + LASSERT(S_ISDIR(la->la_mode)); + mdt_dump_lmv(D_INFO, ma->ma_lmv); + repbody->mbo_eadatasize = ma->ma_lmv_size; + repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA); + } + if (ma->ma_valid & MA_LMV_DEF) { + /* Return -ENOTSUPP for old client */ + if (!mdt_is_striped_client(req->rq_export)) + RETURN(-ENOTSUPP); + LASSERT(S_ISDIR(la->la_mode)); + repbody->mbo_eadatasize = ma->ma_lmv_size; + repbody->mbo_valid |= (OBD_MD_FLDIREA | + OBD_MD_DEFAULT_MEA); + } } else if (S_ISLNK(la->la_mode) && - reqbody->valid & OBD_MD_LINKNAME) { + reqbody->mbo_valid & OBD_MD_LINKNAME) { buffer->lb_buf = ma->ma_lmm; /* eadatasize from client includes NULL-terminator, so * there is no need to read it */ - buffer->lb_len = reqbody->eadatasize - 1; + buffer->lb_len = reqbody->mbo_eadatasize - 1; rc = mo_readlink(env, next, buffer); if (unlikely(rc <= 0)) { CERROR("%s: readlink failed for "DFID": rc = %d\n", @@ -836,16 +994,16 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO)) rc -= 2; - repbody->valid |= OBD_MD_LINKNAME; + repbody->mbo_valid |= OBD_MD_LINKNAME; /* we need to report back size with NULL-terminator * because client expects that */ - repbody->eadatasize = rc + 1; - if (repbody->eadatasize != reqbody->eadatasize) + repbody->mbo_eadatasize = rc + 1; + if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize) CDEBUG(D_INODE, "%s: Read shorter symlink %d " "on "DFID ", expected %d\n", mdt_obd_name(info->mti_mdt), rc, PFID(mdt_object_fid(o)), - reqbody->eadatasize - 1); + reqbody->mbo_eadatasize - 1); /* NULL terminate */ ((char *)ma->ma_lmm)[rc] = 0; @@ -859,62 +1017,37 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } } - if (reqbody->valid & OBD_MD_FLMODEASIZE) { - repbody->max_cookiesize = 0; - repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize; - repbody->valid |= OBD_MD_FLMODEASIZE; - CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & " - "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize, - repbody->max_cookiesize); - } + if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) { + repbody->mbo_max_cookiesize = 0; + repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize; + repbody->mbo_valid |= OBD_MD_FLMODEASIZE; + CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & " + "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize, + repbody->mbo_max_cookiesize); + } - if (exp_connect_rmtclient(info->mti_exp) && - reqbody->valid & OBD_MD_FLRMTPERM) { - void *buf = req_capsule_server_get(pill, &RMF_ACL); + if (exp_connect_rmtclient(info->mti_exp) && + reqbody->mbo_valid & OBD_MD_FLRMTPERM) { + void *buf = req_capsule_server_get(pill, &RMF_ACL); - /* mdt_getattr_lock only */ - rc = mdt_pack_remote_perm(info, o, buf); - if (rc) { - repbody->valid &= ~OBD_MD_FLRMTPERM; - repbody->aclsize = 0; - RETURN(rc); - } else { - repbody->valid |= OBD_MD_FLRMTPERM; - repbody->aclsize = sizeof(struct mdt_remote_perm); - } - } + /* mdt_getattr_lock only */ + rc = mdt_pack_remote_perm(info, o, buf); + if (rc) { + repbody->mbo_valid &= ~OBD_MD_FLRMTPERM; + repbody->mbo_aclsize = 0; + RETURN(rc); + } else { + repbody->mbo_valid |= OBD_MD_FLRMTPERM; + repbody->mbo_aclsize = sizeof(struct mdt_remote_perm); + } + } #ifdef CONFIG_FS_POSIX_ACL else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) && - (reqbody->valid & OBD_MD_FLACL)) { - buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL); - buffer->lb_len = req_capsule_get_size(pill, - &RMF_ACL, RCL_SERVER); - if (buffer->lb_len > 0) { - rc = mo_xattr_get(env, next, buffer, - XATTR_NAME_ACL_ACCESS); - if (rc < 0) { - if (rc == -ENODATA) { - repbody->aclsize = 0; - repbody->valid |= OBD_MD_FLACL; - rc = 0; - } else if (rc == -EOPNOTSUPP) { - rc = 0; - } else { - CERROR("%s: unable to read "DFID - " ACL: rc = %d\n", - mdt_obd_name(info->mti_mdt), - PFID(mdt_object_fid(o)), rc); - } - } else { - repbody->aclsize = rc; - repbody->valid |= OBD_MD_FLACL; - rc = 0; - } - } - } + (reqbody->mbo_valid & OBD_MD_FLACL)) + rc = mdt_pack_acl2body(info, repbody, o, nodemap); #endif - if (reqbody->valid & OBD_MD_FLMDSCAPA && + if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA && info->mti_mdt->mdt_lut.lut_mds_capa && exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) { struct lustre_capa *capa; @@ -925,7 +1058,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, rc = mo_capa_get(env, next, capa, 0); if (rc) RETURN(rc); - repbody->valid |= OBD_MD_FLMDSCAPA; + repbody->mbo_valid |= OBD_MD_FLMDSCAPA; } out: @@ -963,25 +1096,24 @@ static int mdt_renew_capa(struct mdt_thread_info *info) *capa = *c; rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1); if (rc == 0) - body->valid |= OBD_MD_FLOSSCAPA; + body->mbo_valid |= OBD_MD_FLOSSCAPA; RETURN(rc); } -int mdt_getattr(struct tgt_session_info *tsi) +static int mdt_getattr(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_object *obj = info->mti_object; struct req_capsule *pill = info->mti_pill; struct mdt_body *reqbody; struct mdt_body *repbody; - mode_t mode; int rc, rc2; ENTRY; reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY); LASSERT(reqbody); - if (reqbody->valid & OBD_MD_FLOSSCAPA) { + if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) { rc = req_capsule_server_pack(pill); if (unlikely(rc)) RETURN(err_serious(rc)); @@ -992,13 +1124,36 @@ int mdt_getattr(struct tgt_session_info *tsi) LASSERT(obj != NULL); LASSERT(lu_object_assert_exists(&obj->mot_obj)); - mode = lu_object_attr(&obj->mot_obj); + /* Unlike intent case where we need to pre-fill out buffers early on + * in intent policy for ldlm reasons, here we can have a much better + * guess at EA size by just reading it from disk. + * Exceptions are readdir and (missing) directory striping */ + /* Readlink */ + if (reqbody->mbo_valid & OBD_MD_LINKNAME) { + /* No easy way to know how long is the symlink, but it cannot + * be more than PATH_MAX, so we allocate +1 */ + rc = PATH_MAX + 1; + + /* A special case for fs ROOT: getattr there might fetch + * default EA for entire fs, not just for this dir! + */ + } else if (lu_fid_eq(mdt_object_fid(obj), + &info->mti_mdt->mdt_md_root_fid) && + (reqbody->mbo_valid & OBD_MD_FLDIREA) && + (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) == + MDS_GETATTR)) { + /* Should the default strping be bigger, mdt_fix_reply + * will reallocate */ + rc = DEF_REP_MD_SIZE; + } else { + /* Read the actual EA size from disk */ + rc = mdt_attr_get_eabuf_size(info, obj); + } + + if (rc < 0) + GOTO(out, rc = err_serious(rc)); - /* old clients may not report needed easize, use max value then */ - req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, - reqbody->eadatasize == 0 ? - info->mti_mdt->mdt_max_mdsize : - reqbody->eadatasize); + req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc); rc = req_capsule_server_pack(pill); if (unlikely(rc != 0)) @@ -1006,25 +1161,25 @@ int mdt_getattr(struct tgt_session_info *tsi) repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); LASSERT(repbody != NULL); - repbody->eadatasize = 0; - repbody->aclsize = 0; + repbody->mbo_eadatasize = 0; + repbody->mbo_aclsize = 0; - if (reqbody->valid & OBD_MD_FLRMTPERM) - rc = mdt_init_ucred(info, reqbody); - else - rc = mdt_check_ucred(info); - if (unlikely(rc)) - GOTO(out_shrink, rc); + if (reqbody->mbo_valid & OBD_MD_FLRMTPERM) + rc = mdt_init_ucred(info, reqbody); + else + rc = mdt_check_ucred(info); + if (unlikely(rc)) + GOTO(out_shrink, rc); - info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF); + info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF); - /* - * Don't check capability at all, because rename might getattr for - * remote obj, and at that time no capability is available. - */ - mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA); - rc = mdt_getattr_internal(info, obj, 0); - if (reqbody->valid & OBD_MD_FLRMTPERM) + /* + * Don't check capability at all, because rename might getattr for + * remote obj, and at that time no capability is available. + */ + mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA); + rc = mdt_getattr_internal(info, obj, 0); + if (reqbody->mbo_valid & OBD_MD_FLRMTPERM) mdt_exit_ucred(info); EXIT; out_shrink: @@ -1037,36 +1192,33 @@ out: return rc; } -int mdt_is_subdir(struct tgt_session_info *tsi) +/** + * Exchange MOF_LOV_CREATED flags between two objects after a + * layout swap. No assumption is made on whether o1 or o2 have + * created objects or not. + * + * \param[in,out] o1 First swap layout object + * \param[in,out] o2 Second swap layout object + */ +static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2) { - struct mdt_thread_info *info = tsi2mdt_info(tsi); - struct mdt_object *o = info->mti_object; - struct req_capsule *pill = info->mti_pill; - const struct mdt_body *body = info->mti_body; - struct mdt_body *repbody; - int rc; - ENTRY; + __u64 o1_flags; - LASSERT(o != NULL); + mutex_lock(&o1->mot_lov_mutex); + mutex_lock(&o2->mot_lov_mutex); - repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); + o1_flags = o1->mot_flags; + o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) | + (o2->mot_flags & MOF_LOV_CREATED); - /* - * We save last checked parent fid to @repbody->fid1 for remote - * directory case. - */ - LASSERT(fid_is_sane(&body->fid2)); - LASSERT(mdt_object_exists(o) && !mdt_object_remote(o)); - rc = mdo_is_subdir(info->mti_env, mdt_object_child(o), - &body->fid2, &repbody->fid1); - if (rc == 0 || rc == -EREMOTE) - repbody->valid |= OBD_MD_FLID; + o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) | + (o1_flags & MOF_LOV_CREATED); - mdt_thread_info_fini(info); - RETURN(rc); + mutex_unlock(&o2->mot_lov_mutex); + mutex_unlock(&o1->mot_lov_mutex); } -int mdt_swap_layouts(struct tgt_session_info *tsi) +static int mdt_swap_layouts(struct tgt_session_info *tsi) { struct mdt_thread_info *info; struct ptlrpc_request *req = tgt_ses_req(tsi); @@ -1088,26 +1240,30 @@ int mdt_swap_layouts(struct tgt_session_info *tsi) RETURN(-EOPNOTSUPP); info = tsi2mdt_info(tsi); + + if (info->mti_dlm_req != NULL) + ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP); + if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, &info->mti_body->fid1, + mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1, req_capsule_client_get(info->mti_pill, &RMF_CAPA1)); if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT)) - mdt_set_capainfo(info, 1, &info->mti_body->fid2, + mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2, req_capsule_client_get(info->mti_pill, &RMF_CAPA2)); o1 = info->mti_object; o = o2 = mdt_object_find(info->mti_env, info->mti_mdt, - &info->mti_body->fid2); + &info->mti_body->mbo_fid2); if (IS_ERR(o)) GOTO(out, rc = PTR_ERR(o)); if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */ GOTO(put, rc = -ENOENT); - rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2); + rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2); if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */ GOTO(put, rc); @@ -1147,7 +1303,11 @@ int mdt_swap_layouts(struct tgt_session_info *tsi) rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1), mdt_object_child(o2), msl->msl_flags); - GOTO(unlock2, rc); + if (rc < 0) + GOTO(unlock2, rc); + + mdt_swap_lov_flag(o1, o2); + unlock2: mdt_object_unlock(info, o2, lh2, rc); unlock1: @@ -1160,43 +1320,32 @@ out: } static int mdt_raw_lookup(struct mdt_thread_info *info, - struct mdt_object *parent, - const struct lu_name *lname, - struct ldlm_reply *ldlm_rep) + struct mdt_object *parent, + const struct lu_name *lname, + struct ldlm_reply *ldlm_rep) { - struct md_object *next = mdt_object_child(info->mti_object); - const struct mdt_body *reqbody = info->mti_body; - struct lu_fid *child_fid = &info->mti_tmp_fid1; - struct mdt_body *repbody; - int rc; - ENTRY; - - if (reqbody->valid != OBD_MD_FLID) - RETURN(0); - - LASSERT(!info->mti_cross_ref); + struct lu_fid *child_fid = &info->mti_tmp_fid1; + int rc; + ENTRY; - /* Only got the fid of this obj by name */ - fid_zero(child_fid); - rc = mdo_lookup(info->mti_env, next, lname, child_fid, - &info->mti_spec); -#if 0 - /* XXX is raw_lookup possible as intent operation? */ - if (rc != 0) { - if (rc == -ENOENT) - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG); - RETURN(rc); - } else - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); + LASSERT(!info->mti_cross_ref); + + /* Only got the fid of this obj by name */ + fid_zero(child_fid); + rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object), + lname, child_fid, &info->mti_spec); + if (rc == 0) { + struct mdt_body *repbody; + + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); + repbody->mbo_fid1 = *child_fid; + repbody->mbo_valid = OBD_MD_FLID; + mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); + } else if (rc == -ENOENT) { + mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG); + } - repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); -#endif - if (rc == 0) { - repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - repbody->fid1 = *child_fid; - repbody->valid = OBD_MD_FLID; - } - RETURN(1); + RETURN(rc); } /* @@ -1214,126 +1363,126 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, struct mdt_body *reqbody = NULL; struct mdt_object *parent = info->mti_object; struct mdt_object *child; - struct md_object *next = mdt_object_child(parent); struct lu_fid *child_fid = &info->mti_tmp_fid1; struct lu_name *lname = NULL; - const char *name = NULL; - int namelen = 0; struct mdt_lock_handle *lhp = NULL; struct ldlm_lock *lock; - struct ldlm_res_id *res_id; - int is_resent; - int ma_need = 0; - int rc; + bool is_resent; + bool try_layout; + int ma_need = 0; + int rc; + ENTRY; - ENTRY; + is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh); + LASSERT(ergo(is_resent, + lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)); - is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh); - LASSERT(ergo(is_resent, - lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)); + if (parent == NULL) + RETURN(-ENOENT); - LASSERT(parent != NULL); - name = req_capsule_client_get(info->mti_pill, &RMF_NAME); - if (name == NULL) - RETURN(err_serious(-EFAULT)); + if (info->mti_cross_ref) { + /* Only getattr on the child. Parent is on another node. */ + mdt_set_disposition(info, ldlm_rep, + DISP_LOOKUP_EXECD | DISP_LOOKUP_POS); + child = parent; + CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", " + "ldlm_rep = %p\n", + PFID(mdt_object_fid(child)), ldlm_rep); - namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME, - RCL_CLIENT) - 1; - if (!info->mti_cross_ref) { - /* - * XXX: Check for "namelen == 0" is for getattr by fid - * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name, - * that is the name must contain at least one character and - * the terminating '\0' - */ - if (namelen == 0) { - reqbody = req_capsule_client_get(info->mti_pill, - &RMF_MDT_BODY); - if (unlikely(reqbody == NULL)) - RETURN(err_serious(-EFAULT)); - - if (unlikely(!fid_is_sane(&reqbody->fid2))) - RETURN(err_serious(-EINVAL)); - - name = NULL; - CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", " - "ldlm_rep = %p\n", - PFID(mdt_object_fid(parent)), - PFID(&reqbody->fid2), ldlm_rep); - } else { - lname = mdt_name(info->mti_env, (char *)name, namelen); - CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, " - "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), - name, ldlm_rep); - } + rc = mdt_check_resent_lock(info, child, lhc); + if (rc < 0) { + RETURN(rc); + } else if (rc > 0) { + mdt_lock_handle_init(lhc); + mdt_lock_reg_init(lhc, LCK_PR); + + /* + * Object's name is on another MDS, no lookup or layout + * lock is needed here but update lock is. + */ + child_bits &= ~(MDS_INODELOCK_LOOKUP | + MDS_INODELOCK_LAYOUT); + child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE; + + rc = mdt_object_lock(info, child, lhc, child_bits, + MDT_LOCAL_LOCK); + if (rc < 0) + RETURN(rc); + } + + /* Finally, we can get attr for child. */ + if (!mdt_object_exists(child)) { + LU_OBJECT_DEBUG(D_INFO, info->mti_env, + &child->mot_obj, + "remote object doesn't exist.\n"); + mdt_object_unlock(info, child, lhc, 1); + RETURN(-ENOENT); + } + + mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA); + rc = mdt_getattr_internal(info, child, 0); + if (unlikely(rc != 0)) + mdt_object_unlock(info, child, lhc, 1); + + RETURN(rc); } - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD); - if (unlikely(!mdt_object_exists(parent)) && lname) { + lname = &info->mti_name; + mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON); + + if (lu_name_is_valid(lname)) { + CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", " + "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), + PNAME(lname), ldlm_rep); + } else { + reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); + if (unlikely(reqbody == NULL)) + RETURN(err_serious(-EPROTO)); + + *child_fid = reqbody->mbo_fid2; + + if (unlikely(!fid_is_sane(child_fid))) + RETURN(err_serious(-EINVAL)); + + CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", " + "ldlm_rep = %p\n", + PFID(mdt_object_fid(parent)), + PFID(&reqbody->mbo_fid2), ldlm_rep); + } + + mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD); + + if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) { LU_OBJECT_DEBUG(D_INODE, info->mti_env, &parent->mot_obj, "Parent doesn't exist!\n"); RETURN(-ESTALE); - } else if (!info->mti_cross_ref) { - LASSERTF(!mdt_object_remote(parent), - "Parent "DFID" is on remote server\n", - PFID(mdt_object_fid(parent))); } - if (lname) { - rc = mdt_raw_lookup(info, parent, lname, ldlm_rep); - if (rc != 0) { - if (rc > 0) - rc = 0; - RETURN(rc); - } - } - if (info->mti_cross_ref) { - /* Only getattr on the child. Parent is on another node. */ - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); - child = parent; - CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", " - "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep); - - if (is_resent) { - /* Do not take lock for resent case. */ - lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n", - lhc->mlh_reg_lh.cookie); - LASSERT(fid_res_name_eq(mdt_object_fid(child), - &lock->l_resource->lr_name)); - LDLM_LOCK_PUT(lock); - rc = 0; - } else { - mdt_lock_handle_init(lhc); - mdt_lock_reg_init(lhc, LCK_PR); + if (mdt_object_remote(parent)) { + CERROR("%s: parent "DFID" is on remote target\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(parent))); + RETURN(-EIO); + } - /* - * Object's name is on another MDS, no lookup lock is - * needed here but update is. - */ - child_bits &= ~MDS_INODELOCK_LOOKUP; - child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE; + if (lu_name_is_valid(lname)) { + /* Always allow to lookup ".." */ + if (unlikely(lname->ln_namelen == 2 && + lname->ln_name[0] == '.' && + lname->ln_name[1] == '.')) + info->mti_spec.sp_permitted = 1; - rc = mdt_object_lock(info, child, lhc, child_bits, - MDT_LOCAL_LOCK); - } - if (rc == 0) { - /* Finally, we can get attr for child. */ - mdt_set_capainfo(info, 0, mdt_object_fid(child), - BYPASS_CAPA); - rc = mdt_getattr_internal(info, child, 0); - if (unlikely(rc != 0)) - mdt_object_unlock(info, child, lhc, 1); - } - RETURN(rc); - } + if (info->mti_body->mbo_valid == OBD_MD_FLID) { + rc = mdt_raw_lookup(info, parent, lname, ldlm_rep); + + RETURN(rc); + } - if (lname) { - /* step 1: lock parent only if parent is a directory */ + /* step 1: lock parent only if parent is a directory */ if (S_ISDIR(lu_object_attr(&parent->mot_obj))) { - lhp = &info->mti_lh[MDT_LH_PARENT]; - mdt_lock_pdo_init(lhp, LCK_PR, name, namelen); + lhp = &info->mti_lh[MDT_LH_PARENT]; + mdt_lock_pdo_init(lhp, LCK_PR, lname); rc = mdt_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE, MDT_LOCAL_LOCK); @@ -1343,60 +1492,45 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, /* step 2: lookup child's fid by name */ fid_zero(child_fid); - rc = mdo_lookup(info->mti_env, next, lname, child_fid, - &info->mti_spec); + rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, + child_fid, &info->mti_spec); + if (rc == -ENOENT) + mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG); - if (rc != 0) { - if (rc == -ENOENT) - mdt_set_disposition(info, ldlm_rep, - DISP_LOOKUP_NEG); - GOTO(out_parent, rc); - } else - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); - } else { - *child_fid = reqbody->fid2; - mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); - } + if (rc != 0) + GOTO(out_parent, rc); + } - /* - *step 3: find the child object by fid & lock it. - * regardless if it is local or remote. - */ - child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid); + mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); + + /* + *step 3: find the child object by fid & lock it. + * regardless if it is local or remote. + * + *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e) + * set parent dir fid the same as child fid in getattr by fid case + * we should not lu_object_find() the object again, could lead + * to hung if there is a concurrent unlink destroyed the object. + */ + if (lu_fid_eq(mdt_object_fid(parent), child_fid)) { + mdt_object_get(info->mti_env, parent); + child = parent; + } else { + child = mdt_object_find(info->mti_env, info->mti_mdt, + child_fid); + } if (unlikely(IS_ERR(child))) GOTO(out_parent, rc = PTR_ERR(child)); - if (is_resent) { - /* Do not take lock for resent case. */ - lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n", - lhc->mlh_reg_lh.cookie); - - res_id = &lock->l_resource->lr_name; - if (!fid_res_name_eq(mdt_object_fid(child), - &lock->l_resource->lr_name)) { - LASSERTF(fid_res_name_eq(mdt_object_fid(parent), - &lock->l_resource->lr_name), - "Lock res_id: "DLDLMRES", fid: "DFID"\n", - PLDLMRES(lock->l_resource), - PFID(mdt_object_fid(parent))); - CWARN("Although resent, but still not get child lock" - "parent:"DFID" child:"DFID"\n", - PFID(mdt_object_fid(parent)), - PFID(mdt_object_fid(child))); - lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT); - LDLM_LOCK_PUT(lock); - GOTO(relock, 0); - } - LDLM_LOCK_PUT(lock); - rc = 0; - } else { - bool try_layout = false; -relock: + rc = mdt_check_resent_lock(info, child, lhc); + if (rc < 0) { + GOTO(out_child, rc); + } else if (rc > 0) { OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2); mdt_lock_handle_init(lhc); mdt_lock_reg_init(lhc, LCK_PR); + try_layout = false; if (!mdt_object_exists(child)) { LU_OBJECT_DEBUG(D_INODE, info->mti_env, @@ -1433,7 +1567,7 @@ relock: if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) && exp_connect_layout(info->mti_exp) && S_ISREG(lu_object_attr(&child->mot_obj)) && - ldlm_rep != NULL) { + !mdt_object_remote(child) && ldlm_rep != NULL) { /* try to grant layout lock for regular file. */ try_layout = true; } @@ -1453,6 +1587,10 @@ relock: ma_need |= MA_LOV; } } else { + /* Do not enqueue the UPDATE lock from MDT(cross-MDT), + * client will enqueue the lock to the remote MDT */ + if (mdt_object_remote(child)) + child_bits &= ~MDS_INODELOCK_UPDATE; rc = mdt_object_lock(info, child, lhc, child_bits, MDT_CROSS_LOCK); } @@ -1461,11 +1599,6 @@ relock: } lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - /* Get MA_SOM attributes if update lock is given. */ - if (lock && - lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE && - S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu))) - ma_need |= MA_SOM; /* finally, we can get attr for child. */ mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA); @@ -1474,15 +1607,12 @@ relock: mdt_object_unlock(info, child, lhc, 1); } else if (lock) { /* Debugging code. */ - res_id = &lock->l_resource->lr_name; LDLM_DEBUG(lock, "Returning lock to client"); LASSERTF(fid_res_name_eq(mdt_object_fid(child), &lock->l_resource->lr_name), "Lock res_id: "DLDLMRES", fid: "DFID"\n", PLDLMRES(lock->l_resource), PFID(mdt_object_fid(child))); - if (mdt_object_exists(child) && !mdt_object_remote(child)) - mdt_pack_size2body(info, child); } if (lock) LDLM_LOCK_PUT(lock); @@ -1497,7 +1627,7 @@ out_parent: } /* normal handler: should release the child lock */ -int mdt_getattr_name(struct tgt_session_info *tsi) +static int mdt_getattr_name(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD]; @@ -1511,9 +1641,9 @@ int mdt_getattr_name(struct tgt_session_info *tsi) repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(repbody != NULL); - info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF); - repbody->eadatasize = 0; - repbody->aclsize = 0; + info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF); + repbody->mbo_eadatasize = 0; + repbody->mbo_aclsize = 0; rc = mdt_init_ucred(info, reqbody); if (unlikely(rc)) @@ -1538,7 +1668,7 @@ out_shrink: static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg); -int mdt_set_info(struct tgt_session_info *tsi) +static int mdt_set_info(struct tgt_session_info *tsi) { struct ptlrpc_request *req = tgt_ses_req(tsi); char *key; @@ -1596,7 +1726,7 @@ int mdt_set_info(struct tgt_session_info *tsi) RETURN(rc); } -int mdt_readpage(struct tgt_session_info *tsi) +static int mdt_readpage(struct tgt_session_info *tsi) { struct mdt_thread_info *info = mdt_th_info(tsi->tsi_env); struct mdt_object *object = mdt_obj(tsi->tsi_corpus); @@ -1620,17 +1750,17 @@ int mdt_readpage(struct tgt_session_info *tsi) * reqbody->size contains offset of where to start to read and * reqbody->nlink contains number bytes to read. */ - rdpg->rp_hash = reqbody->size; - if (rdpg->rp_hash != reqbody->size) { - CERROR("Invalid hash: "LPX64" != "LPX64"\n", - rdpg->rp_hash, reqbody->size); - RETURN(-EFAULT); - } + rdpg->rp_hash = reqbody->mbo_size; + if (rdpg->rp_hash != reqbody->mbo_size) { + CERROR("Invalid hash: "LPX64" != "LPX64"\n", + rdpg->rp_hash, reqbody->mbo_size); + RETURN(-EFAULT); + } - rdpg->rp_attrs = reqbody->mode; + rdpg->rp_attrs = reqbody->mbo_mode; if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH) rdpg->rp_attrs |= LUDA_64BITHASH; - rdpg->rp_count = min_t(unsigned int, reqbody->nlink, + rdpg->rp_count = min_t(unsigned int, reqbody->mbo_nlink, exp_max_brw_size(tsi->tsi_exp)); rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; @@ -1685,7 +1815,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info, /* for replay (no_create) lmm is not needed, client has it already */ if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, - info->mti_rr.rr_eadatalen); + DEF_REP_MD_SIZE); /* llog cookies are always 0, the field is kept for compatibility */ if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) @@ -1700,8 +1830,8 @@ static int mdt_reint_internal(struct mdt_thread_info *info, if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) { repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); LASSERT(repbody); - repbody->eadatasize = 0; - repbody->aclsize = 0; + repbody->mbo_eadatasize = 0; + repbody->mbo_aclsize = 0; } OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10); @@ -1720,7 +1850,8 @@ static int mdt_reint_internal(struct mdt_thread_info *info, GOTO(out_ucred, rc = err_serious(rc)); if (mdt_check_resent(info, mdt_reconstruct, lhc)) { - rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg); + DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt."); + rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg); GOTO(out_ucred, rc); } rc = mdt_reint_rec(info, lhc); @@ -1761,7 +1892,7 @@ static long mdt_reint_opcode(struct ptlrpc_request *req, return opc; } -int mdt_reint(struct tgt_session_info *tsi) +static int mdt_reint(struct tgt_session_info *tsi) { long opc; int rc; @@ -1773,7 +1904,8 @@ int mdt_reint(struct tgt_session_info *tsi) [REINT_RENAME] = &RQF_MDS_REINT_RENAME, [REINT_OPEN] = &RQF_MDS_REINT_OPEN, [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR, - [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK + [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK, + [REINT_MIGRATE] = &RQF_MDS_REINT_RENAME }; ENTRY; @@ -1809,22 +1941,23 @@ static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt) /* this should sync this object */ static int mdt_object_sync(struct mdt_thread_info *info) { - struct md_object *next; - int rc; - ENTRY; + struct md_object *next; + int rc; + ENTRY; - if (!mdt_object_exists(info->mti_object)) { - CWARN("Non existing object "DFID"!\n", - PFID(mdt_object_fid(info->mti_object))); - RETURN(-ESTALE); - } - next = mdt_object_child(info->mti_object); - rc = mo_object_sync(info->mti_env, next); + if (!mdt_object_exists(info->mti_object)) { + CWARN("%s: non existing object "DFID": rc = %d\n", + mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(info->mti_object)), -ESTALE); + RETURN(-ESTALE); + } + next = mdt_object_child(info->mti_object); + rc = mo_object_sync(info->mti_env, next); - RETURN(rc); + RETURN(rc); } -int mdt_sync(struct tgt_session_info *tsi) +static int mdt_sync(struct tgt_session_info *tsi) { struct ptlrpc_request *req = tgt_ses_req(tsi); struct req_capsule *pill = tsi->tsi_pill; @@ -1836,7 +1969,7 @@ int mdt_sync(struct tgt_session_info *tsi) if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) RETURN(err_serious(-ENOMEM)); - if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) { + if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) { rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp)); } else { struct mdt_thread_info *info = tsi2mdt_info(tsi); @@ -1870,7 +2003,7 @@ int mdt_sync(struct tgt_session_info *tsi) * Handle quota control requests to consult current usage/limit, but also * to configure quota enforcement */ -int mdt_quotactl(struct tgt_session_info *tsi) +static int mdt_quotactl(struct tgt_session_info *tsi) { struct obd_export *exp = tsi->tsi_exp; struct req_capsule *pill = tsi->tsi_pill; @@ -1878,6 +2011,7 @@ int mdt_quotactl(struct tgt_session_info *tsi) int id, rc; struct mdt_device *mdt = mdt_exp2dev(exp); struct lu_device *qmt = mdt->mdt_qmt_dev; + struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap; ENTRY; oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); @@ -1898,9 +2032,11 @@ int mdt_quotactl(struct tgt_session_info *tsi) /* deprecated, not used any more */ RETURN(-EOPNOTSUPP); /* master quotactl */ - case Q_GETINFO: case Q_SETINFO: case Q_SETQUOTA: + if (!nodemap_can_setquota(nodemap)) + RETURN(-EPERM); + case Q_GETINFO: case Q_GETQUOTA: if (qmt == NULL) RETURN(-EOPNOTSUPP); @@ -1939,6 +2075,13 @@ int mdt_quotactl(struct tgt_session_info *tsi) } } + if (oqctl->qc_type == USRQUOTA) + id = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_CLIENT_TO_FS, id); + else if (oqctl->qc_type == GRPQUOTA) + id = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_CLIENT_TO_FS, id); + repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL); if (repoqc == NULL) RETURN(err_serious(-EFAULT)); @@ -2022,7 +2165,7 @@ static int mdt_llog_ctxt_unclone(const struct lu_env *env, /* * sec context handlers */ -int mdt_sec_ctx_handle(struct tgt_session_info *tsi) +static int mdt_sec_ctx_handle(struct tgt_session_info *tsi) { int rc; @@ -2044,7 +2187,7 @@ int mdt_sec_ctx_handle(struct tgt_session_info *tsi) /* * quota request handlers */ -int mdt_quota_dqacq(struct tgt_session_info *tsi) +static int mdt_quota_dqacq(struct tgt_session_info *tsi) { struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp); struct lu_device *qmt = mdt->mdt_qmt_dev; @@ -2105,12 +2248,13 @@ struct mdt_object *mdt_object_find(const struct lu_env *env, static void mdt_device_commit_async(const struct lu_env *env, struct mdt_device *mdt) { - struct dt_device *dt = mdt->mdt_bottom; - int rc; + struct dt_device *dt = mdt->mdt_bottom; + int rc; - rc = dt->dd_ops->dt_commit_async(env, dt); - if (unlikely(rc != 0)) - CWARN("async commit start failed with rc = %d", rc); + rc = dt->dd_ops->dt_commit_async(env, dt); + if (unlikely(rc != 0)) + CWARN("%s: async commit start failed: rc = %d\n", + mdt_obd_name(mdt), rc); } /** @@ -2185,10 +2329,11 @@ int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) { struct lu_env env; - rc = lu_env_init(&env, LCT_LOCAL); - if (unlikely(rc != 0)) - CWARN("lu_env initialization failed with rc = %d," - "cannot start asynchronous commit\n", rc); + rc = lu_env_init(&env, LCT_LOCAL); + if (unlikely(rc != 0)) + CWARN("%s: lu_env initialization failed, cannot " + "start asynchronous commit: rc = %d\n", + obd->obd_name, rc); else mdt_device_commit_async(&env, mdt); lu_env_fini(&env); @@ -2196,8 +2341,9 @@ int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, RETURN(rc); } -int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, - void *data, int flag) +/* Used for cross-MDT lock */ +int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, + void *data, int flag) { struct lustre_handle lockh; int rc; @@ -2220,24 +2366,65 @@ int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, RETURN(0); } +int mdt_check_resent_lock(struct mdt_thread_info *info, + struct mdt_object *mo, + struct mdt_lock_handle *lhc) +{ + /* the lock might already be gotten in ldlm_handle_enqueue() */ + if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) { + struct ptlrpc_request *req = mdt_info_req(info); + struct ldlm_lock *lock; + + lock = ldlm_handle2lock(&lhc->mlh_reg_lh); + LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT); + if (lock == NULL) { + /* Lock is pinned by ldlm_handle_enqueue0() as it is + * a resend case, however, it could be already destroyed + * due to client eviction or a raced cancel RPC. */ + LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n", + lhc->mlh_reg_lh.cookie); + RETURN(-ESTALE); + } + + if (!fid_res_name_eq(mdt_object_fid(mo), + &lock->l_resource->lr_name)) { + CWARN("%s: Although resent, but still not " + "get child lock:"DFID"\n", + info->mti_exp->exp_obd->obd_name, + PFID(mdt_object_fid(mo))); + LDLM_LOCK_PUT(lock); + RETURN(-EPROTO); + } + LDLM_LOCK_PUT(lock); + return 0; + } + return 1; +} + int mdt_remote_object_lock(struct mdt_thread_info *mti, - struct mdt_object *o, struct lustre_handle *lh, - ldlm_mode_t mode, __u64 ibits) + struct mdt_object *o, const struct lu_fid *fid, + struct lustre_handle *lh, ldlm_mode_t mode, + __u64 ibits) { struct ldlm_enqueue_info *einfo = &mti->mti_einfo; ldlm_policy_data_t *policy = &mti->mti_policy; + struct ldlm_res_id *res_id = &mti->mti_res_id; int rc = 0; ENTRY; LASSERT(mdt_object_remote(o)); - LASSERT(ibits & MDS_INODELOCK_UPDATE); + LASSERT(ibits == MDS_INODELOCK_UPDATE); + + fid_build_reg_res_name(fid, res_id); memset(einfo, 0, sizeof(*einfo)); einfo->ei_type = LDLM_IBITS; einfo->ei_mode = mode; - einfo->ei_cb_bl = mdt_md_blocking_ast; + einfo->ei_cb_bl = mdt_remote_blocking_ast; einfo->ei_cb_cp = ldlm_completion_ast; + einfo->ei_enq_slave = 0; + einfo->ei_res_id = res_id; memset(policy, 0, sizeof(*policy)); policy->l_inodebits.bits = ibits; @@ -2247,9 +2434,10 @@ int mdt_remote_object_lock(struct mdt_thread_info *mti, RETURN(rc); } -static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, - struct mdt_lock_handle *lh, __u64 ibits, - bool nonblock, int locality) +static int mdt_object_local_lock(struct mdt_thread_info *info, + struct mdt_object *o, + struct mdt_lock_handle *lh, __u64 ibits, + bool nonblock, int locality) { struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace; ldlm_policy_data_t *policy = &info->mti_policy; @@ -2263,21 +2451,9 @@ static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, LASSERT(lh->mlh_reg_mode != LCK_MINMODE); LASSERT(lh->mlh_type != MDT_NUL_LOCK); - if (mdt_object_remote(o)) { - if (locality == MDT_CROSS_LOCK) { - ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM); - ibits |= MDS_INODELOCK_LOOKUP; - } else { - LASSERTF(!(ibits & - (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM)), - "%s: wrong bit "LPX64" for remote obj "DFID"\n", - mdt_obd_name(info->mti_mdt), ibits, - PFID(mdt_object_fid(o))); - LASSERT(ibits & MDS_INODELOCK_LOOKUP); - } - /* No PDO lock on remote object */ - LASSERT(lh->mlh_type != MDT_PDO_LOCK); - } + /* Only enqueue LOOKUP lock for remote object */ + if (mdt_object_remote(o)) + LASSERT(ibits == MDS_INODELOCK_LOOKUP); if (lh->mlh_type == MDT_PDO_LOCK) { /* check for exists after object is locked */ @@ -2312,11 +2488,12 @@ static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, * want it slowed down due to possible cancels. */ policy->l_inodebits.bits = MDS_INODELOCK_UPDATE; - rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, - policy, res_id, dlmflags, - &info->mti_exp->exp_handle.h_cookie); - if (unlikely(rc)) - RETURN(rc); + rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, + policy, res_id, dlmflags, + info->mti_exp == NULL ? NULL : + &info->mti_exp->exp_handle.h_cookie); + if (unlikely(rc != 0)) + GOTO(out_unlock, rc); } /* @@ -2333,24 +2510,81 @@ static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o, * going to be sent to client. If it is - mdt_intent_policy() path will * fix it up and turn FL_LOCAL flag off. */ - rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy, - res_id, LDLM_FL_LOCAL_ONLY | dlmflags, - &info->mti_exp->exp_handle.h_cookie); - if (rc) - mdt_object_unlock(info, o, lh, 1); - else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) && - lh->mlh_pdo_hash != 0 && - (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) { - OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15); - } + rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy, + res_id, LDLM_FL_LOCAL_ONLY | dlmflags, + info->mti_exp == NULL ? NULL : + &info->mti_exp->exp_handle.h_cookie); +out_unlock: + if (rc != 0) + mdt_object_unlock(info, o, lh, 1); + else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) && + lh->mlh_pdo_hash != 0 && + (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15); - RETURN(rc); + RETURN(rc); +} + +static int +mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o, + struct mdt_lock_handle *lh, __u64 ibits, + bool nonblock, int locality) +{ + int rc; + ENTRY; + + if (!mdt_object_remote(o)) + return mdt_object_local_lock(info, o, lh, ibits, nonblock, + locality); + + if (locality == MDT_LOCAL_LOCK) { + CERROR("%s: try to get local lock for remote object" + DFID".\n", mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o))); + RETURN(-EPROTO); + } + + /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */ + ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT | + MDS_INODELOCK_XATTR); + if (ibits & MDS_INODELOCK_UPDATE) { + /* Sigh, PDO needs to enqueue 2 locks right now, but + * enqueue RPC can only request 1 lock, to avoid extra + * RPC, so it will instead enqueue EX lock for remote + * object anyway XXX*/ + if (lh->mlh_type == MDT_PDO_LOCK && + lh->mlh_pdo_hash != 0) { + CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to" + "EX lock.\n", mdt_obd_name(info->mti_mdt), + PFID(mdt_object_fid(o))); + lh->mlh_pdo_hash = 0; + lh->mlh_rreg_mode = LCK_EX; + lh->mlh_type = MDT_REG_LOCK; + } + rc = mdt_remote_object_lock(info, o, mdt_object_fid(o), + &lh->mlh_rreg_lh, + lh->mlh_rreg_mode, + MDS_INODELOCK_UPDATE); + if (rc != ELDLM_OK) + RETURN(rc); + } + + /* Only enqueue LOOKUP lock for remote object */ + if (ibits & MDS_INODELOCK_LOOKUP) { + rc = mdt_object_local_lock(info, o, lh, + MDS_INODELOCK_LOOKUP, + nonblock, locality); + if (rc != ELDLM_OK) + RETURN(rc); + } + + RETURN(0); } int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle *lh, __u64 ibits, int locality) { - return mdt_object_lock0(info, o, lh, ibits, false, locality); + return mdt_object_lock_internal(info, o, lh, ibits, false, locality); } int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o, @@ -2359,7 +2593,7 @@ int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle tmp = *lh; int rc; - rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality); + rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality); if (rc == 0) *lh = tmp; @@ -2397,15 +2631,21 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n", h->cookie); - CDEBUG(D_HA, "request = %p reply state = %p" - " transno = "LPD64"\n", - req, req->rq_reply_state, req->rq_transno); - if (mdt_cos_is_enabled(mdt)) { - no_ack = 1; - ldlm_lock_downgrade(lock, LCK_COS); - mode = LCK_COS; - } - ptlrpc_save_lock(req, h, mode, no_ack); + /* there is no request if mdt_object_unlock() is called + * from mdt_export_cleanup()->mdt_add_dirty_flag() */ + if (likely(req != NULL)) { + CDEBUG(D_HA, "request = %p reply state = %p" + " transno = "LPD64"\n", req, + req->rq_reply_state, req->rq_transno); + if (mdt_cos_is_enabled(mdt)) { + no_ack = 1; + ldlm_lock_downgrade(lock, LCK_COS); + mode = LCK_COS; + } + ptlrpc_save_lock(req, h, mode, no_ack); + } else { + ldlm_lock_decref(h, mode); + } if (mdt_is_lock_sync(lock)) { CDEBUG(D_HA, "found sync-lock," " async commit started\n"); @@ -2504,11 +2744,11 @@ static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags) if (body == NULL) RETURN(-EFAULT); - if (!(body->valid & OBD_MD_FLID)) - RETURN(0); + if (!(body->mbo_valid & OBD_MD_FLID)) + RETURN(0); - if (!fid_is_sane(&body->fid1)) { - CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1)); + if (!fid_is_sane(&body->mbo_fid1)) { + CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1)); RETURN(-EINVAL); } @@ -2519,17 +2759,17 @@ static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags) */ if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) && req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT)) - mdt_set_capainfo(info, 0, &body->fid1, - req_capsule_client_get(pill, &RMF_CAPA1)); - - obj = mdt_object_find(env, info->mti_mdt, &body->fid1); - if (!IS_ERR(obj)) { - if ((flags & HABEO_CORPUS) && - !mdt_object_exists(obj)) { - mdt_object_put(env, obj); - /* for capability renew ENOENT will be handled in - * mdt_renew_capa */ - if (body->valid & OBD_MD_FLOSSCAPA) + mdt_set_capainfo(info, 0, &body->mbo_fid1, + req_capsule_client_get(pill, &RMF_CAPA1)); + + obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1); + if (!IS_ERR(obj)) { + if ((flags & HABEO_CORPUS) && + !mdt_object_exists(obj)) { + mdt_object_put(env, obj); + /* for capability renew ENOENT will be handled in + * mdt_renew_capa */ + if (body->mbo_valid & OBD_MD_FLOSSCAPA) rc = 0; else rc = -ENOENT; @@ -2558,7 +2798,7 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) /* Pack reply. */ if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, - info->mti_body->eadatasize); + DEF_REP_MD_SIZE); if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0); @@ -2620,7 +2860,6 @@ void mdt_thread_info_init(struct ptlrpc_request *req, info->mti_mdt = NULL; info->mti_env = req->rq_svc_thread->t_env; info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg); - info->mti_mos = NULL; memset(&info->mti_attr, 0, sizeof(info->mti_attr)); info->mti_big_buf = LU_BUF_NULL; @@ -2632,9 +2871,12 @@ void mdt_thread_info_init(struct ptlrpc_request *req, info->mti_opdata = 0; info->mti_big_lmm_used = 0; - /* To not check for split by default. */ info->mti_spec.no_create = 0; info->mti_spec.sp_rm_entry = 0; + info->mti_spec.sp_permitted = 0; + + info->mti_spec.u.sp_ea.eadata = NULL; + info->mti_spec.u.sp_ea.eadatalen = 0; } void mdt_thread_info_fini(struct mdt_thread_info *info) @@ -2678,7 +2920,8 @@ struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi) if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) && req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) { lc = req_capsule_client_get(pill, &RMF_CAPA1); - mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc); + mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1, + lc); } } mti->mti_body = tsi->tsi_mdt_body; @@ -2687,13 +2930,19 @@ struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi) return mti; } -int mdt_tgt_connect(struct tgt_session_info *tsi) +static int mdt_tgt_connect(struct tgt_session_info *tsi) { struct ptlrpc_request *req = tgt_ses_req(tsi); int rc; ENTRY; + if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) && + cfs_fail_val == + tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) + schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, + msecs_to_jiffies(3 * MSEC_PER_SEC)); + rc = tgt_connect(tsi); if (rc != 0) RETURN(rc); @@ -2814,21 +3063,18 @@ static struct mdt_it_flavor { } }; -int mdt_intent_lock_replace(struct mdt_thread_info *info, - struct ldlm_lock **lockp, - struct ldlm_lock *new_lock, - struct mdt_lock_handle *lh, - __u64 flags) +static int +mdt_intent_lock_replace(struct mdt_thread_info *info, + struct ldlm_lock **lockp, + struct mdt_lock_handle *lh, + __u64 flags) { struct ptlrpc_request *req = mdt_info_req(info); struct ldlm_lock *lock = *lockp; + struct ldlm_lock *new_lock; - /* - * Get new lock only for cases when possible resent did not find any - * lock. - */ - if (new_lock == NULL) - new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0); + /* If possible resent found a lock, @lh is set to its handle */ + new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0); if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) { lh->mlh_reg_lh.cookie = 0; @@ -2860,6 +3106,8 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info, */ LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT); + + LDLM_LOCK_RELEASE(new_lock); lh->mlh_reg_lh.cookie = 0; RETURN(ELDLM_LOCK_REPLACED); } @@ -2900,43 +3148,30 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info, } static void mdt_intent_fixup_resent(struct mdt_thread_info *info, - struct ldlm_lock *new_lock, - struct ldlm_lock **old_lock, - struct mdt_lock_handle *lh) + struct ldlm_lock *new_lock, + struct mdt_lock_handle *lh, + __u64 flags) { struct ptlrpc_request *req = mdt_info_req(info); - struct obd_export *exp = req->rq_export; - struct lustre_handle remote_hdl; struct ldlm_request *dlmreq; - struct ldlm_lock *lock; if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)) return; dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ); - remote_hdl = dlmreq->lock_handle[0]; - - /* In the function below, .hs_keycmp resolves to - * ldlm_export_lock_keycmp() */ - /* coverity[overrun-buffer-val] */ - lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl); - if (lock) { - if (lock != new_lock) { - lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie; - lh->mlh_reg_mode = lock->l_granted_mode; - - LDLM_DEBUG(lock, "Restoring lock cookie"); - DEBUG_REQ(D_DLMTRACE, req, - "restoring lock cookie "LPX64, - lh->mlh_reg_lh.cookie); - if (old_lock) - *old_lock = LDLM_LOCK_GET(lock); - cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); - return; - } - cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); - } + /* Check if this is a resend case (MSG_RESENT is set on RPC) and a + * lock was found by ldlm_handle_enqueue(); if so @lh must be + * initialized. */ + if (flags & LDLM_FL_RESENT) { + lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie; + lh->mlh_reg_mode = new_lock->l_granted_mode; + + LDLM_DEBUG(new_lock, "Restoring lock cookie"); + DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64, + lh->mlh_reg_lh.cookie); + return; + } /* * If the xid matches, then we know this is a resent request, and allow @@ -2952,8 +3187,8 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, */ lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT); - DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64, - remote_hdl.cookie); + DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64, + dlmreq->lock_handle[0].cookie); } static int mdt_intent_getxattr(enum mdt_it_code opcode, @@ -2970,7 +3205,7 @@ static int mdt_intent_getxattr(enum mdt_it_code opcode, * (for the resend case) or a new lock. Below we will use it to * replace the original lock. */ - mdt_intent_fixup_resent(info, *lockp, NULL, lhc); + mdt_intent_fixup_resent(info, *lockp, lhc, flags); if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) { mdt_lock_reg_init(lhc, (*lockp)->l_req_mode); rc = mdt_object_lock(info, info->mti_object, lhc, @@ -2982,7 +3217,7 @@ static int mdt_intent_getxattr(enum mdt_it_code opcode, grc = mdt_getxattr(info); - rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); + rc = mdt_intent_lock_replace(info, lockp, lhc, flags); if (mdt_info_req(info)->rq_repmsg != NULL) ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); @@ -3000,10 +3235,8 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, __u64 flags) { struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT]; - struct ldlm_lock *new_lock = NULL; __u64 child_bits; struct ldlm_reply *ldlm_rep; - struct ptlrpc_request *req; struct mdt_body *reqbody; struct mdt_body *repbody; int rc, rc2; @@ -3015,9 +3248,9 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(repbody); - info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF); - repbody->eadatasize = 0; - repbody->aclsize = 0; + info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF); + repbody->mbo_eadatasize = 0; + repbody->mbo_aclsize = 0; switch (opcode) { case MDT_IT_LOOKUP: @@ -3036,12 +3269,11 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, if (rc) GOTO(out_shrink, rc); - req = info->mti_pill->rc_req; ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD); - /* Get lock from request for possible resent case. */ - mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc); + /* Get lock from request for possible resent case. */ + mdt_intent_fixup_resent(info, *lockp, lhc, flags); rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep); ldlm_rep->lock_policy_res2 = clear_serious(rc); @@ -3054,7 +3286,7 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED); } - rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags); + rc = mdt_intent_lock_replace(info, lockp, lhc, flags); EXIT; out_ucred: mdt_exit_ucred(info); @@ -3071,11 +3303,11 @@ static int mdt_intent_layout(enum mdt_it_code opcode, struct ldlm_lock **lockp, __u64 flags) { + struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT]; struct layout_intent *layout; struct lu_fid *fid; struct mdt_object *obj = NULL; - struct md_object *child = NULL; - int rc; + int rc = 0; ENTRY; if (opcode != MDT_IT_LAYOUT) { @@ -3087,16 +3319,20 @@ static int mdt_intent_layout(enum mdt_it_code opcode, fid = &info->mti_tmp_fid2; fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name); + /* Get lock from request for possible resent case. */ + mdt_intent_fixup_resent(info, *lockp, lhc, flags); + obj = mdt_object_find(info->mti_env, info->mti_mdt, fid); if (IS_ERR(obj)) RETURN(PTR_ERR(obj)); if (mdt_object_exists(obj) && !mdt_object_remote(obj)) { - child = mdt_object_child(obj); - /* get the length of lsm */ - rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL, - XATTR_NAME_LOV); + rc = mdt_attr_get_eabuf_size(info, obj); + if (rc < 0) { + mdt_object_put(info->mti_env, obj); + RETURN(rc); + } if (rc > info->mti_mdt->mdt_max_mdsize) info->mti_mdt->mdt_max_mdsize = rc; @@ -3105,17 +3341,19 @@ static int mdt_intent_layout(enum mdt_it_code opcode, mdt_object_put(info->mti_env, obj); (*lockp)->l_lvb_type = LVB_T_LAYOUT; - req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, - ldlm_lvbo_size(*lockp)); + req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc); rc = req_capsule_server_pack(info->mti_pill); if (rc != 0) RETURN(-EINVAL); + if (lustre_handle_is_used(&lhc->mlh_reg_lh)) + rc = mdt_intent_lock_replace(info, lockp, lhc, flags); + layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT); LASSERT(layout != NULL); if (layout->li_opc == LAYOUT_INTENT_ACCESS) - /* return to normal ldlm handling */ - RETURN(0); + /* return to normal/resent ldlm handling */ + RETURN(rc); CERROR("%s: Unsupported layout intent (%d)\n", mdt_obd_name(info->mti_mdt), layout->li_opc); @@ -3149,8 +3387,8 @@ static int mdt_intent_reint(enum mdt_it_code opcode, RETURN(err_serious(-EPROTO)); } - /* Get lock from request for possible resent case. */ - mdt_intent_fixup_resent(info, *lockp, NULL, lhc); + /* Get lock from request for possible resent case. */ + mdt_intent_fixup_resent(info, *lockp, lhc, flags); rc = mdt_reint_internal(info, lhc, opc); @@ -3166,10 +3404,10 @@ static int mdt_intent_reint(enum mdt_it_code opcode, /* the open lock or the lock for cross-ref object should be * returned to the client */ - if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) { - LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh)); + if (lustre_handle_is_used(&lhc->mlh_reg_lh) && + (rc == 0 || rc == -MDT_EREMOTE_OPEN)) { rep->lock_policy_res2 = 0; - rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); + rc = mdt_intent_lock_replace(info, lockp, lhc, flags); RETURN(rc); } @@ -3179,6 +3417,7 @@ static int mdt_intent_reint(enum mdt_it_code opcode, mdt_get_disposition(rep, DISP_LOOKUP_NEG)) rep->lock_policy_res2 = 0; + lhc->mlh_reg_lh.cookie = 0ull; if (rc == -ENOTCONN || rc == -ENODEV || rc == -EOVERFLOW) { /**< if VBR failure then return error */ /* @@ -3187,28 +3426,13 @@ static int mdt_intent_reint(enum mdt_it_code opcode, * will detect this, then disconnect, reconnect the import * immediately, instead of impacting the following the rpc. */ - lhc->mlh_reg_lh.cookie = 0ull; RETURN(rc); - } else { - /* - * For other cases, the error will be returned by intent. - * and client will retrieve the result from intent. - */ - /* - * FIXME: when open lock is finished, that should be - * checked here. - */ - if (lustre_handle_is_used(&lhc->mlh_reg_lh)) { - LASSERTF(rc == 0, "Error occurred but lock handle " - "is still in use, rc = %d\n", rc); - rep->lock_policy_res2 = 0; - rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); - RETURN(rc); - } else { - lhc->mlh_reg_lh.cookie = 0ull; - RETURN(ELDLM_LOCK_ABORTED); - } } + /* + * For other cases, the error will be returned by intent, and client + * will retrieve the result from intent. + */ + RETURN(ELDLM_LOCK_ABORTED); } static int mdt_intent_code(long itcode) @@ -3261,17 +3485,16 @@ static int mdt_intent_code(long itcode) static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, struct ldlm_lock **lockp, __u64 flags) { - struct req_capsule *pill; - struct mdt_it_flavor *flv; - int opc; - int rc; - ENTRY; - - opc = mdt_intent_code(itopc); - if (opc < 0) - RETURN(-EINVAL); + struct req_capsule *pill = info->mti_pill; + struct ptlrpc_request *req = mdt_info_req(info); + struct mdt_it_flavor *flv; + int opc; + int rc; + ENTRY; - pill = info->mti_pill; + opc = mdt_intent_code(itopc); + if (opc < 0) + RETURN(-EINVAL); if (opc == MDT_IT_QUOTA) { struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev; @@ -3287,33 +3510,33 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, RETURN(rc); } - flv = &mdt_it_flavor[opc]; - if (flv->it_fmt != NULL) - req_capsule_extend(pill, flv->it_fmt); + flv = &mdt_it_flavor[opc]; + if (flv->it_fmt != NULL) + req_capsule_extend(pill, flv->it_fmt); - rc = mdt_unpack_req_pack_rep(info, flv->it_flags); - if (rc == 0) { - struct ptlrpc_request *req = mdt_info_req(info); - if (flv->it_flags & MUTABOR && - exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY) - RETURN(-EROFS); - } - if (rc == 0 && flv->it_act != NULL) { + rc = mdt_unpack_req_pack_rep(info, flv->it_flags); + if (rc < 0) + RETURN(rc); + + if (flv->it_flags & MUTABOR && + exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY) + RETURN(-EROFS); + + if (flv->it_act != NULL) { struct ldlm_reply *rep; /* execute policy */ rc = flv->it_act(opc, info, lockp, flags); /* Check whether the reply has been packed successfully. */ - if (mdt_info_req(info)->rq_repmsg != NULL) { + if (req->rq_repmsg != NULL) { rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); rep->lock_policy_res2 = ptlrpc_status_hton(rep->lock_policy_res2); } - } else { - rc = -EPROTO; } + RETURN(rc); } @@ -3370,97 +3593,199 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, RETURN(rc); } -static int mdt_seq_fini(const struct lu_env *env, - struct mdt_device *m) +static void mdt_deregister_seq_exp(struct mdt_device *mdt) +{ + struct seq_server_site *ss = mdt_seq_site(mdt); + + if (ss->ss_node_id == 0) + return; + + if (ss->ss_client_seq != NULL) { + lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp); + ss->ss_client_seq->lcs_exp = NULL; + } + + if (ss->ss_server_fld != NULL) { + lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp); + ss->ss_server_fld->lsf_control_exp = NULL; + } +} + +static void mdt_seq_fini_cli(struct mdt_device *mdt) { - return seq_site_fini(env, mdt_seq_site(m)); + struct seq_server_site *ss = mdt_seq_site(mdt); + + if (ss == NULL) + return; + + if (ss->ss_server_seq != NULL) + seq_server_set_cli(NULL, ss->ss_server_seq, NULL); } -static int mdt_seq_init(const struct lu_env *env, - const char *uuid, - struct mdt_device *m) +static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt) { - struct seq_server_site *ss; - char *prefix; - int rc; + mdt_seq_fini_cli(mdt); + mdt_deregister_seq_exp(mdt); + + return seq_site_fini(env, mdt_seq_site(mdt)); +} + +/** + * It will retrieve its FLDB entries from MDT0, and it only happens + * when upgrading existent FS to 2.6 or when local FLDB is corrupted, + * and it needs to refresh FLDB from the MDT0. + **/ +static int mdt_register_lwp_callback(void *data) +{ + struct lu_env env; + struct mdt_device *mdt = data; + struct lu_server_fld *fld = mdt_seq_site(mdt)->ss_server_fld; + int rc; ENTRY; - ss = mdt_seq_site(m); + LASSERT(mdt_seq_site(mdt)->ss_node_id != 0); - /* - * This is sequence-controller node. Init seq-controller server on local - * MDT. - */ - if (ss->ss_node_id == 0) { - LASSERT(ss->ss_control_seq == NULL); + if (!likely(fld->lsf_new)) + RETURN(0); - OBD_ALLOC_PTR(ss->ss_control_seq); - if (ss->ss_control_seq == NULL) - RETURN(-ENOMEM); + rc = lu_env_init(&env, LCT_MD_THREAD); + if (rc) { + CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc); + RETURN(rc); + } + + rc = fld_update_from_controller(&env, fld); + if (rc != 0) { + CERROR("%s: cannot update controller: rc = %d\n", + mdt_obd_name(mdt), rc); + GOTO(out, rc); + } +out: + lu_env_fini(&env); + RETURN(rc); +} - rc = seq_server_init(ss->ss_control_seq, - m->mdt_bottom, uuid, - LUSTRE_SEQ_CONTROLLER, - ss, - env); +static int mdt_register_seq_exp(struct mdt_device *mdt) +{ + struct seq_server_site *ss = mdt_seq_site(mdt); + char *lwp_name = NULL; + int rc; - if (rc) - GOTO(out_seq_fini, rc); + if (ss->ss_node_id == 0) + return 0; - OBD_ALLOC_PTR(ss->ss_client_seq); - if (ss->ss_client_seq == NULL) - GOTO(out_seq_fini, rc = -ENOMEM); + OBD_ALLOC(lwp_name, MAX_OBD_NAME); + if (lwp_name == NULL) + GOTO(out_free, rc = -ENOMEM); - OBD_ALLOC(prefix, MAX_OBD_NAME + 5); - if (prefix == NULL) { - OBD_FREE_PTR(ss->ss_client_seq); - GOTO(out_seq_fini, rc = -ENOMEM); - } + rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0); + if (rc != 0) + GOTO(out_free, rc); - snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", - uuid); + rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp, + NULL, NULL); + if (rc != 0) + GOTO(out_free, rc); - /* - * Init seq-controller client after seq-controller server is - * ready. Pass ss->ss_control_seq to it for direct talking. - */ - rc = seq_client_init(ss->ss_client_seq, NULL, - LUSTRE_SEQ_METADATA, prefix, - ss->ss_control_seq); - OBD_FREE(prefix, MAX_OBD_NAME + 5); + rc = lustre_register_lwp_item(lwp_name, + &ss->ss_server_fld->lsf_control_exp, + mdt_register_lwp_callback, mdt); + if (rc != 0) { + lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp); + ss->ss_client_seq->lcs_exp = NULL; + GOTO(out_free, rc); + } +out_free: + if (lwp_name != NULL) + OBD_FREE(lwp_name, MAX_OBD_NAME); + + return rc; +} + +/* + * Init client sequence manager which is used by local MDS to talk to sequence + * controller on remote node. + */ +static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt) +{ + struct seq_server_site *ss = mdt_seq_site(mdt); + int rc; + char *prefix; + ENTRY; + + /* check if this is adding the first MDC and controller is not yet + * initialized. */ + OBD_ALLOC_PTR(ss->ss_client_seq); + if (ss->ss_client_seq == NULL) + RETURN(-ENOMEM); + + OBD_ALLOC(prefix, MAX_OBD_NAME + 5); + if (prefix == NULL) { + OBD_FREE_PTR(ss->ss_client_seq); + ss->ss_client_seq = NULL; + RETURN(-ENOMEM); + } + + /* Note: seq_client_fini will be called in seq_site_fini */ + snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt)); + rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA, + prefix, ss->ss_node_id == 0 ? ss->ss_control_seq : + NULL); + OBD_FREE(prefix, MAX_OBD_NAME + 5); + if (rc != 0) { + OBD_FREE_PTR(ss->ss_client_seq); + ss->ss_client_seq = NULL; + RETURN(rc); + } + + rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq); + + RETURN(rc); +} + +static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt) +{ + struct seq_server_site *ss; + int rc; + ENTRY; + ss = mdt_seq_site(mdt); + /* init sequence controller server(MDT0) */ + if (ss->ss_node_id == 0) { + OBD_ALLOC_PTR(ss->ss_control_seq); + if (ss->ss_control_seq == NULL) + RETURN(-ENOMEM); + + rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom, + mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER, + ss); if (rc) GOTO(out_seq_fini, rc); } - /* Init seq-server on local MDT */ - LASSERT(ss->ss_server_seq == NULL); - + /* Init normal sequence server */ OBD_ALLOC_PTR(ss->ss_server_seq); if (ss->ss_server_seq == NULL) GOTO(out_seq_fini, rc = -ENOMEM); - rc = seq_server_init(ss->ss_server_seq, - m->mdt_bottom, uuid, - LUSTRE_SEQ_SERVER, - ss, - env); + rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom, + mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss); if (rc) - GOTO(out_seq_fini, rc = -ENOMEM); + GOTO(out_seq_fini, rc); - /* Assign seq-controller client to local seq-server. */ - if (ss->ss_node_id == 0) { - LASSERT(ss->ss_client_seq != NULL); + /* init seq client for seq server to talk to seq controller(MDT0) */ + rc = mdt_seq_init_cli(env, mdt); + if (rc != 0) + GOTO(out_seq_fini, rc); - rc = seq_server_set_cli(ss->ss_server_seq, - ss->ss_client_seq, - env); - } + if (ss->ss_node_id != 0) + /* register controler export through lwp */ + rc = mdt_register_seq_exp(mdt); EXIT; out_seq_fini: if (rc) - mdt_seq_fini(env, m); + mdt_seq_fini(env, mdt); return rc; } @@ -3498,7 +3823,7 @@ static int mdt_fld_init(const struct lu_env *env, RETURN(rc = -ENOMEM); rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid, - ss->ss_node_id, LU_SEQ_RANGE_MDT); + LU_SEQ_RANGE_MDT); if (rc) { OBD_FREE_PTR(ss->ss_server_fld); ss->ss_server_fld = NULL; @@ -3511,7 +3836,6 @@ static int mdt_fld_init(const struct lu_env *env, static void mdt_stack_pre_fini(const struct lu_env *env, struct mdt_device *m, struct lu_device *top) { - struct obd_device *obd; struct lustre_cfg_bufs *bufs; struct lustre_cfg *lcfg; struct mdt_thread_info *info; @@ -3526,7 +3850,6 @@ static void mdt_stack_pre_fini(const struct lu_env *env, LASSERT(m->mdt_child_exp); LASSERT(m->mdt_child_exp->exp_obd); - obd = m->mdt_child_exp->exp_obd; /* process cleanup, pass mdt obd name to get obd umount flags */ /* XXX: this is needed because all layers are referenced by @@ -3536,10 +3859,9 @@ static void mdt_stack_pre_fini(const struct lu_env *env, lustre_cfg_bufs_reset(bufs, mdt_obd_name(m)); lustre_cfg_bufs_set_string(bufs, 1, NULL); lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs); - if (!lcfg) { - CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m)); - return; - } + if (lcfg == NULL) + RETURN_EXIT; + top->ld_ops->ldo_process_config(env, top, lcfg); lustre_cfg_free(lcfg); EXIT; @@ -3572,10 +3894,9 @@ static void mdt_stack_fini(const struct lu_env *env, strcat(flags, "A"); lustre_cfg_bufs_set_string(bufs, 1, flags); lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs); - if (!lcfg) { - CERROR("Cannot alloc lcfg!\n"); - return; - } + if (lcfg == NULL) + RETURN_EXIT; + LASSERT(top); top->ld_ops->ldo_process_config(env, top, lcfg); lustre_cfg_free(lcfg); @@ -3681,7 +4002,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, strcpy(name, dev); p = strstr(name, "-MDT"); if (p == NULL) - GOTO(cleanup_mem, rc = -ENOMEM); + GOTO(free_bufs, rc = -ENOMEM); p[3] = 'D'; snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name); @@ -3690,7 +4011,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, if (lprof == NULL || lprof->lp_dt == NULL) { CERROR("can't find the profile: %s\n", lustre_cfg_string(cfg, 0)); - GOTO(cleanup_mem, rc = -EINVAL); + GOTO(free_bufs, rc = -EINVAL); } lustre_cfg_bufs_reset(bufs, name); @@ -3699,7 +4020,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt); lcfg = lustre_cfg_new(LCFG_ATTACH, bufs); - if (!lcfg) + if (lcfg == NULL) GOTO(free_bufs, rc = -ENOMEM); rc = class_attach(lcfg); @@ -3710,7 +4031,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, if (!obd) { CERROR("Can not find obd %s (%s in config)\n", MDD_OBD_NAME, lustre_cfg_string(cfg, 0)); - GOTO(class_detach, rc = -EINVAL); + GOTO(lcfg_cleanup, rc = -EINVAL); } lustre_cfg_free(lcfg); @@ -3721,6 +4042,8 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt); lcfg = lustre_cfg_new(LCFG_SETUP, bufs); + if (lcfg == NULL) + GOTO(class_detach, rc = -ENOMEM); rc = class_setup(obd, lcfg); if (rc) @@ -3729,7 +4052,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, /* connect to MDD we just setup */ rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp); if (rc) - RETURN(rc); + GOTO(class_detach, rc); site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site; LASSERT(site); @@ -3738,16 +4061,14 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, site->ls_top_dev = &mdt->mdt_lu_dev; mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev); - /* now connect to bottom OSD */ snprintf(name, MAX_OBD_NAME, "%s-osd", dev); rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp); if (rc) - RETURN(rc); + GOTO(class_detach, rc); mdt->mdt_bottom = lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev); - rc = lu_env_refill((struct lu_env *)env); if (rc != 0) CERROR("Failure to refill session: '%d'\n", rc); @@ -3829,7 +4150,7 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt); lcfg = lustre_cfg_new(LCFG_ATTACH, bufs); - if (!lcfg) + if (lcfg == NULL) GOTO(cleanup_mem, rc = -ENOMEM); rc = class_attach(lcfg); @@ -3840,7 +4161,7 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, if (!obd) { CERROR("Can not find obd %s (%s in config)\n", qmtname, lustre_cfg_string(cfg, 0)); - GOTO(class_detach, rc = -EINVAL); + GOTO(lcfg_cleanup, rc = -EINVAL); } lustre_cfg_free(lcfg); @@ -3854,6 +4175,8 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name); lcfg = lustre_cfg_new(LCFG_SETUP, bufs); + if (lcfg == NULL) + GOTO(class_detach, rc = -ENOMEM); rc = class_setup(obd, lcfg); if (rc) @@ -3921,7 +4244,7 @@ static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt) /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper * for now. This will be removed along with converting rest of MDT code * to use tgt_session_info */ -int mdt_tgt_getxattr(struct tgt_session_info *tsi) +static int mdt_tgt_getxattr(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); int rc; @@ -3951,11 +4274,8 @@ TGT_MDT_HDL(HABEO_CORPUS, MDS_GETXATTR, mdt_tgt_getxattr), TGT_MDT_HDL(0 | HABEO_REFERO, MDS_STATFS, mdt_statfs), TGT_MDT_HDL(0 | MUTABOR, MDS_REINT, mdt_reint), TGT_MDT_HDL(HABEO_CORPUS, MDS_CLOSE, mdt_close), -TGT_MDT_HDL(HABEO_CORPUS, MDS_DONE_WRITING, - mdt_done_writing), TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE, mdt_readpage), TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC, mdt_sync), -TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR, mdt_is_subdir), TGT_MDT_HDL(0, MDS_QUOTACTL, mdt_quotactl), TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS, mdt_hsm_progress), @@ -3970,8 +4290,9 @@ TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET, TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action), TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST, mdt_hsm_request), -TGT_MDT_HDL(HABEO_CORPUS|HABEO_REFERO | MUTABOR, MDS_SWAP_LAYOUTS, - mdt_swap_layouts) +TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR, + MDS_SWAP_LAYOUTS, + mdt_swap_layouts), }; static struct tgt_handler mdt_sec_ctx_ops[] = { @@ -4006,8 +4327,8 @@ static struct tgt_opc_slice mdt_common_slice[] = { .tos_hs = mdt_sec_ctx_ops }, { - .tos_opc_start = UPDATE_OBJ, - .tos_opc_end = UPDATE_LAST_OPC, + .tos_opc_start = OUT_UPDATE_FIRST_OPC, + .tos_opc_end = OUT_UPDATE_LAST_OPC, .tos_hs = tgt_out_handlers }, { @@ -4030,6 +4351,11 @@ static struct tgt_opc_slice mdt_common_slice[] = { .tos_opc_end = LLOG_LAST_OPC, .tos_hs = tgt_llog_handlers }, + { + .tos_opc_start = LFSCK_FIRST_OPC, + .tos_opc_end = LFSCK_LAST_OPC, + .tos_hs = tgt_lfsck_handlers + }, { .tos_hs = NULL @@ -4038,15 +4364,18 @@ static struct tgt_opc_slice mdt_common_slice[] = { static void mdt_fini(const struct lu_env *env, struct mdt_device *m) { - struct md_device *next = m->mdt_child; - struct lu_device *d = &m->mdt_lu_dev; - struct obd_device *obd = mdt2obd_dev(m); - ENTRY; - - target_recovery_fini(obd); + struct md_device *next = m->mdt_child; + struct lu_device *d = &m->mdt_lu_dev; + struct obd_device *obd = mdt2obd_dev(m); + struct lfsck_stop stop; + ENTRY; - ping_evictor_stop(); + stop.ls_status = LS_PAUSED; + stop.ls_flags = 0; + next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop); + target_recovery_fini(obd); + ping_evictor_stop(); mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child)); if (m->mdt_opts.mo_coordinator) @@ -4074,14 +4403,7 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) mdt_quota_fini(env, m); - cfs_free_nidlist(&m->mdt_nosquash_nids); - if (m->mdt_nosquash_str) { - OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen); - m->mdt_nosquash_str = NULL; - m->mdt_nosquash_strlen = 0; - } - - next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL); + cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids); mdt_seq_fini(env, m); mdt_fld_fini(env, m); @@ -4095,38 +4417,14 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) */ mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); - LASSERT(cfs_atomic_read(&d->ld_ref) == 0); + LASSERT(atomic_read(&d->ld_ref) == 0); - server_put_mount(mdt_obd_name(m), NULL); + server_put_mount(mdt_obd_name(m), true); EXIT; } -static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial) -{ - struct mdt_device *m = mdt_dev(obd->obd_lu_dev); - struct sptlrpc_rule_set tmp_rset; - int rc; - - sptlrpc_rule_set_init(&tmp_rset); - rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial); - if (rc) { - CERROR("mdt %s: failed get sptlrpc rules: %d\n", - mdt_obd_name(m), rc); - return rc; - } - - sptlrpc_target_update_exp_flavor(obd, &tmp_rset); - - write_lock(&m->mdt_lut.lut_sptlrpc_lock); - sptlrpc_rule_set_free(&m->mdt_lut.lut_sptlrpc_rset); - m->mdt_lut.lut_sptlrpc_rset = tmp_rset; - write_unlock(&m->mdt_lut.lut_sptlrpc_lock); - - return 0; -} - -int mdt_postrecov(const struct lu_env *, struct mdt_device *); +static int mdt_postrecov(const struct lu_env *, struct mdt_device *); static int mdt_init0(const struct lu_env *env, struct mdt_device *m, struct lu_device_type *ldt, struct lustre_cfg *cfg) @@ -4167,8 +4465,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */ - m->mdt_som_conf = 0; - m->mdt_opts.mo_cos = MDT_COS_DEFAULT; /* default is coordinator off, it is started through conf_param @@ -4187,16 +4483,13 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, obd->u.obt.obt_magic = OBT_MAGIC; } - spin_lock_init(&m->mdt_ioepoch_lock); m->mdt_capa_timeout = CAPA_TIMEOUT; m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1; m->mdt_ck_timeout = CAPA_KEY_TIMEOUT; - m->mdt_squash_uid = 0; - m->mdt_squash_gid = 0; - CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids); - m->mdt_nosquash_str = NULL; - m->mdt_nosquash_strlen = 0; - init_rwsem(&m->mdt_squash_sem); + m->mdt_squash.rsi_uid = 0; + m->mdt_squash.rsi_gid = 0; + INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids); + init_rwsem(&m->mdt_squash.rsi_sem); spin_lock_init(&m->mdt_osfs_lock); m->mdt_osfs_age = cfs_time_shift_64(-1000); m->mdt_enable_remote_dir = 0; @@ -4243,7 +4536,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, if (rc) GOTO(err_fini_stack, rc); - rc = mdt_seq_init(env, mdt_obd_name(m), m); + rc = mdt_seq_init(env, m); if (rc) GOTO(err_fini_fld, rc); @@ -4286,7 +4579,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, if (rc) GOTO(err_tgt, rc); - mdt_adapt_sptlrpc_conf(obd, 1); + tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1); next = m->mdt_child; rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0, @@ -4377,7 +4670,7 @@ err_fini_stack: mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); err_lmi: if (lmi) - server_put_mount(dev, lmi->lmi_mnt); + server_put_mount(dev, true); return(rc); } @@ -4405,7 +4698,6 @@ static int mdt_process_config(const struct lu_env *env, switch (cfg->lcfg_command) { case LCFG_PARAM: { - struct lprocfs_static_vars lvars; struct obd_device *obd = d->ld_obd; /* For interoperability */ @@ -4440,8 +4732,7 @@ static int mdt_process_config(const struct lu_env *env, } } - lprocfs_mdt_init_vars(&lvars); - rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, + rc = class_process_proc_param(PARAM_MDT, obd->obd_vars, cfg, obd); if (rc > 0 || rc == -ENOSYS) { /* is it an HSM var ? */ @@ -4468,25 +4759,25 @@ static int mdt_process_config(const struct lu_env *env, } static struct lu_object *mdt_object_alloc(const struct lu_env *env, - const struct lu_object_header *hdr, - struct lu_device *d) + const struct lu_object_header *hdr, + struct lu_device *d) { - struct mdt_object *mo; + struct mdt_object *mo; - ENTRY; + ENTRY; - OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO); - if (mo != NULL) { - struct lu_object *o; - struct lu_object_header *h; + OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS); + if (mo != NULL) { + struct lu_object *o; + struct lu_object_header *h; o = &mo->mot_obj; - h = &mo->mot_header; - lu_object_header_init(h); - lu_object_init(o, h, d); - lu_object_add_top(h, o); - o->lo_ops = &mdt_obj_ops; - mutex_init(&mo->mot_ioepoch_mutex); + h = &mo->mot_header; + lu_object_header_init(h); + lu_object_init(o, h, d); + lu_object_add_top(h, o); + o->lo_ops = &mdt_obj_ops; + spin_lock_init(&mo->mot_write_lock); mutex_init(&mo->mot_lov_mutex); init_rwsem(&mo->mot_open_sem); RETURN(o); @@ -4537,13 +4828,13 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o) } static int mdt_object_print(const struct lu_env *env, void *cookie, - lu_printer_t p, const struct lu_object *o) + lu_printer_t p, const struct lu_object *o) { - struct mdt_object *mdto = mdt_obj((struct lu_object *)o); - return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" " - "flags="LPX64", epochcount=%d, writecount=%d)", - mdto, mdto->mot_ioepoch, mdto->mot_flags, - mdto->mot_ioepoch_count, mdto->mot_writecount); + struct mdt_object *mdto = mdt_obj((struct lu_object *)o); + + return (*p)(env, cookie, + LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)", + mdto, mdto->mot_flags, mdto->mot_write_count); } static int mdt_prepare(const struct lu_env *env, @@ -4553,7 +4844,6 @@ static int mdt_prepare(const struct lu_env *env, struct mdt_device *mdt = mdt_dev(cdev); struct lu_device *next = &mdt->mdt_child->md_lu_dev; struct obd_device *obd = cdev->ld_obd; - struct lfsck_start_param lsp; int rc; ENTRY; @@ -4572,16 +4862,10 @@ static int mdt_prepare(const struct lu_env *env, if (rc) RETURN(rc); - lsp.lsp_start = NULL; - lsp.lsp_namespace = mdt->mdt_namespace; - rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child, - OBD_IOC_START_LFSCK, - 0, &lsp); - if (rc != 0) { - CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", - mdt_obd_name(mdt), rc); - rc = 0; - } + rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace); + /* The LFSCK instance is registered just now, so it must be there when + * register the namespace to such instance. */ + LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc); if (mdt->mdt_seq_site.ss_node_id == 0) { rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child, @@ -4622,18 +4906,16 @@ static int mdt_obd_set_info_async(const struct lu_env *env, __u32 vallen, void *val, struct ptlrpc_request_set *set) { - struct obd_device *obd = exp->exp_obd; - int rc; - ENTRY; + int rc; - LASSERT(obd); + ENTRY; - if (KEY_IS(KEY_SPTLRPC_CONF)) { - rc = mdt_adapt_sptlrpc_conf(obd, 0); - RETURN(rc); - } + if (KEY_IS(KEY_SPTLRPC_CONF)) { + rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0); + RETURN(rc); + } - RETURN(0); + RETURN(0); } /** @@ -4666,12 +4948,14 @@ static int mdt_connect_internal(struct obd_export *exp, data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED; data->ocd_ibits_known &= MDS_INODELOCK_FULL; - /* If no known bits (which should not happen, probably, - as everybody should support LOOKUP and UPDATE bits at least) - revert to compat mode with plain locks. */ - if (!data->ocd_ibits_known && - data->ocd_connect_flags & OBD_CONNECT_IBITS) - data->ocd_connect_flags &= ~OBD_CONNECT_IBITS; + if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) && + !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) { + CWARN("%s: client %s does not support ibits lock, either " + "very old or an invalid client: flags "LPX64"\n", + mdt_obd_name(mdt), exp->exp_client_uuid.uuid, + data->ocd_connect_flags); + return -EBADE; + } if (!mdt->mdt_opts.mo_acl) data->ocd_connect_flags &= ~OBD_CONNECT_ACL; @@ -4679,9 +4963,6 @@ static int mdt_connect_internal(struct obd_export *exp, if (!mdt->mdt_opts.mo_user_xattr) data->ocd_connect_flags &= ~OBD_CONNECT_XATTR; - if (!mdt->mdt_som_conf) - data->ocd_connect_flags &= ~OBD_CONNECT_SOM; - if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) { data->ocd_brw_size = min(data->ocd_brw_size, (__u32)MD_MAX_BRW_SIZE); @@ -4719,15 +5000,6 @@ static int mdt_connect_internal(struct obd_export *exp, return -EBADE; } - if (mdt->mdt_som_conf && - !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT | - OBD_CONNECT_MDS_MDS | - OBD_CONNECT_SOM))) { - CWARN("%s: MDS has SOM enabled, but client does not support " - "it\n", mdt_obd_name(mdt)); - return -EBADE; - } - if (OCD_HAS_FLAG(data, PINGLESS)) { if (ptlrpc_pinger_suppress_pings()) { spin_lock(&exp->exp_obd->obd_dev_lock); @@ -4743,85 +5015,6 @@ static int mdt_connect_internal(struct obd_export *exp, return 0; } -/* mds_connect copy */ -static int mdt_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) -{ - struct obd_export *lexp; - struct lustre_handle conn = { 0 }; - struct mdt_device *mdt; - int rc; - ENTRY; - - LASSERT(env != NULL); - if (!exp || !obd || !cluuid) - RETURN(-EINVAL); - - mdt = mdt_dev(obd->obd_lu_dev); - - /* - * first, check whether the stack is ready to handle requests - * XXX: probably not very appropriate method is used now - * at some point we should find a better one - */ - if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL && - !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) { - rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd); - if (rc) - RETURN(-EAGAIN); - set_bit(MDT_FL_SYNCED, &mdt->mdt_state); - } - - rc = class_connect(&conn, obd, cluuid); - if (rc) - RETURN(rc); - - lexp = class_conn2export(&conn); - LASSERT(lexp != NULL); - - rc = mdt_connect_internal(lexp, mdt, data); - if (rc == 0) { - struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd; - - LASSERT(lcd); - memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); - rc = tgt_client_new(env, lexp); - if (rc == 0) - mdt_export_stats_init(obd, lexp, localdata); - } - - if (rc != 0) { - class_disconnect(lexp); - *exp = NULL; - } else { - *exp = lexp; - } - - RETURN(rc); -} - -static int mdt_obd_reconnect(const struct lu_env *env, - struct obd_export *exp, struct obd_device *obd, - struct obd_uuid *cluuid, - struct obd_connect_data *data, - void *localdata) -{ - int rc; - ENTRY; - - if (exp == NULL || obd == NULL || cluuid == NULL) - RETURN(-EINVAL); - - rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); - if (rc == 0) - mdt_export_stats_init(obd, exp, localdata); - - RETURN(rc); -} - static int mdt_ctxt_add_dirty_flag(struct lu_env *env, struct mdt_thread_info *info, struct mdt_file_data *mfd) @@ -4849,25 +5042,26 @@ static int mdt_ctxt_add_dirty_flag(struct lu_env *env, static int mdt_export_cleanup(struct obd_export *exp) { - struct mdt_export_data *med = &exp->exp_mdt_data; - struct obd_device *obd = exp->exp_obd; - struct mdt_device *mdt; - struct mdt_thread_info *info; - struct lu_env env; - CFS_LIST_HEAD(closing_list); - struct mdt_file_data *mfd, *n; - int rc = 0; - ENTRY; + struct list_head closing_list; + struct mdt_export_data *med = &exp->exp_mdt_data; + struct obd_device *obd = exp->exp_obd; + struct mdt_device *mdt; + struct mdt_thread_info *info; + struct lu_env env; + struct mdt_file_data *mfd, *n; + int rc = 0; + ENTRY; + INIT_LIST_HEAD(&closing_list); spin_lock(&med->med_open_lock); - while (!cfs_list_empty(&med->med_open_head)) { - cfs_list_t *tmp = med->med_open_head.next; - mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list); + while (!list_empty(&med->med_open_head)) { + struct list_head *tmp = med->med_open_head.next; + mfd = list_entry(tmp, struct mdt_file_data, mfd_list); /* Remove mfd handle so it can't be found again. * We are consuming the mfd_list reference here. */ class_handle_unhash(&mfd->mfd_handle); - cfs_list_move_tail(&mfd->mfd_list, &closing_list); + list_move_tail(&mfd->mfd_list, &closing_list); } spin_unlock(&med->med_open_lock); mdt = mdt_dev(obd->obd_lu_dev); @@ -4884,12 +5078,13 @@ static int mdt_export_cleanup(struct obd_export *exp) info->mti_mdt = mdt; info->mti_exp = exp; - if (!cfs_list_empty(&closing_list)) { - struct md_attr *ma = &info->mti_attr; + if (!list_empty(&closing_list)) { + struct md_attr *ma = &info->mti_attr; - /* Close any open files (which may also cause orphan unlinking). */ - cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { - cfs_list_del_init(&mfd->mfd_list); + /* Close any open files (which may also cause orphan + * unlinking). */ + list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { + list_del_init(&mfd->mfd_list); ma->ma_need = ma->ma_valid = 0; /* This file is being closed due to an eviction, it @@ -4906,7 +5101,7 @@ static int mdt_export_cleanup(struct obd_export *exp) * archive request into a noop if it's not actually * dirty. */ - if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC)) + if (mfd->mfd_mode & FMODE_WRITE) rc = mdt_ctxt_add_dirty_flag(&env, info, mfd); /* Don't unlink orphan on failover umount, LU-184 */ @@ -4935,23 +5130,120 @@ static int mdt_obd_disconnect(struct obd_export *exp) LASSERT(exp); class_export_get(exp); - rc = server_disconnect_export(exp); - if (rc != 0) - CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc); + nodemap_del_member(exp); + rc = server_disconnect_export(exp); + if (rc != 0) + CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc); - rc = mdt_export_cleanup(exp); - class_export_put(exp); - RETURN(rc); + rc = mdt_export_cleanup(exp); + class_export_put(exp); + RETURN(rc); +} + +/* mds_connect copy */ +static int mdt_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) +{ + struct obd_export *lexp; + struct lustre_handle conn = { 0 }; + struct mdt_device *mdt; + int rc; + lnet_nid_t *client_nid = localdata; + ENTRY; + + LASSERT(env != NULL); + if (!exp || !obd || !cluuid) + RETURN(-EINVAL); + + mdt = mdt_dev(obd->obd_lu_dev); + + /* + * first, check whether the stack is ready to handle requests + * XXX: probably not very appropriate method is used now + * at some point we should find a better one + */ + if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL && + !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) { + rc = obd_get_info(env, mdt->mdt_child_exp, + sizeof(KEY_OSP_CONNECTED), + KEY_OSP_CONNECTED, NULL, NULL, NULL); + if (rc) + RETURN(-EAGAIN); + set_bit(MDT_FL_SYNCED, &mdt->mdt_state); + } + + rc = class_connect(&conn, obd, cluuid); + if (rc) + RETURN(rc); + + lexp = class_conn2export(&conn); + LASSERT(lexp != NULL); + + rc = mdt_connect_internal(lexp, mdt, data); + if (rc == 0) { + struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd; + + LASSERT(lcd); + memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); + rc = tgt_client_new(env, lexp); + if (rc == 0) { + rc = nodemap_add_member(*client_nid, lexp); + if (rc != 0 && rc != -EEXIST) + goto out; + + mdt_export_stats_init(obd, lexp, localdata); + } + + /* For phase I, sync for cross-ref operation. */ + spin_lock(&lexp->exp_lock); + lexp->exp_keep_sync = 1; + spin_unlock(&lexp->exp_lock); + } +out: + if (rc != 0) { + class_disconnect(lexp); + *exp = NULL; + } else { + *exp = lexp; + } + + RETURN(rc); +} + +static int mdt_obd_reconnect(const struct lu_env *env, + struct obd_export *exp, struct obd_device *obd, + struct obd_uuid *cluuid, + struct obd_connect_data *data, + void *localdata) +{ + lnet_nid_t *client_nid = localdata; + int rc; + ENTRY; + + if (exp == NULL || obd == NULL || cluuid == NULL) + RETURN(-EINVAL); + + rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); + if (rc == 0) { + rc = nodemap_add_member(*client_nid, exp); + if (rc == 0 || rc == -EEXIST) + mdt_export_stats_init(obd, exp, localdata); + } + + RETURN(rc); } /* FIXME: Can we avoid using these two interfaces? */ static int mdt_init_export(struct obd_export *exp) { - struct mdt_export_data *med = &exp->exp_mdt_data; - int rc; - ENTRY; + struct mdt_export_data *med = &exp->exp_mdt_data; + int rc; + ENTRY; - CFS_INIT_LIST_HEAD(&med->med_open_head); + INIT_LIST_HEAD(&med->med_open_head); spin_lock_init(&med->med_open_lock); mutex_init(&med->med_idmap_mutex); med->med_idmap = NULL; @@ -4996,36 +5288,17 @@ static int mdt_destroy_export(struct obd_export *exp) &exp->exp_client_uuid))) RETURN(0); - ldlm_destroy_export(exp); - tgt_client_free(exp); + ldlm_destroy_export(exp); + tgt_client_free(exp); - LASSERT(cfs_list_empty(&exp->exp_outstanding_replies)); - LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head)); + LASSERT(list_empty(&exp->exp_outstanding_replies)); + LASSERT(list_empty(&exp->exp_mdt_data.med_open_head)); - RETURN(0); + RETURN(0); } -/** The maximum depth that fid2path() will search. - * This is limited only because we want to store the fids for - * historical path lookup purposes. - */ -#define MAX_PATH_DEPTH 100 - -/** mdt_path() lookup structure. */ -struct path_lookup_info { - __u64 pli_recno; /**< history point */ - __u64 pli_currec; /**< current record */ - struct lu_fid pli_fid; - struct lu_fid pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */ - struct mdt_object *pli_mdt_obj; - char *pli_path; /**< full path */ - int pli_pathlen; - int pli_linkno; /**< which hardlink to follow */ - int pli_fidcount; /**< number of \a pli_fids */ -}; - -static int mdt_links_read(struct mdt_thread_info *info, - struct mdt_object *mdt_obj, struct linkea_data *ldata) +int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj, + struct linkea_data *ldata) { int rc; @@ -5055,8 +5328,21 @@ static int mdt_links_read(struct mdt_thread_info *info, return linkea_init(ldata); } +/** + * Given an MDT object, try to look up the full path to the object. + * Part of the MDT layer implementation of lfs fid2path. + * + * \param[in] info Per-thread common data shared by MDT level handlers. + * \param[in] obj Object to do path lookup of + * \param[in,out] fp User-provided struct to store path information + * + * \retval 0 Lookup successful, path information stored in fp + * \retval -EAGAIN Lookup failed, usually because object is being moved + * \retval negative errno if there was a problem + */ static int mdt_path_current(struct mdt_thread_info *info, - struct path_lookup_info *pli) + struct mdt_object *obj, + struct getinfo_fid2path *fp) { struct mdt_device *mdt = info->mti_mdt; struct mdt_object *mdt_obj; @@ -5067,8 +5353,9 @@ static int mdt_path_current(struct mdt_thread_info *info, struct lu_buf *buf = &info->mti_big_buf; char *ptr; int reclen; - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; int rc = 0; + bool first = true; ENTRY; /* temp buffer for path element, the buffer will be finally freed @@ -5078,121 +5365,151 @@ static int mdt_path_current(struct mdt_thread_info *info, RETURN(-ENOMEM); ldata.ld_buf = buf; - ptr = pli->pli_path + pli->pli_pathlen - 1; + ptr = fp->gf_path + fp->gf_pathlen - 1; *ptr = 0; --ptr; - pli->pli_fidcount = 0; - pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj); + *tmpfid = fp->gf_fid = *mdt_object_fid(obj); /* root FID only exists on MDT0, and fid2path should also ends at MDT0, * so checking root_fid can only happen on MDT0. */ - while (!lu_fid_eq(&mdt->mdt_md_root_fid, - &pli->pli_fids[pli->pli_fidcount])) { - mdt_obj = mdt_object_find(info->mti_env, mdt, - &pli->pli_fids[pli->pli_fidcount]); + while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) { + struct lu_buf lmv_buf; + + mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid); if (IS_ERR(mdt_obj)) GOTO(out, rc = PTR_ERR(mdt_obj)); - if (mdt_object_remote(mdt_obj)) { - mdt_object_put(info->mti_env, mdt_obj); - GOTO(remote_out, rc = -EREMOTE); - } + if (!mdt_object_exists(mdt_obj)) { mdt_object_put(info->mti_env, mdt_obj); GOTO(out, rc = -ENOENT); } + if (mdt_object_remote(mdt_obj)) { + mdt_object_put(info->mti_env, mdt_obj); + GOTO(remote_out, rc = -EREMOTE); + } + rc = mdt_links_read(info, mdt_obj, &ldata); - mdt_object_put(info->mti_env, mdt_obj); - if (rc != 0) + if (rc != 0) { + mdt_object_put(info->mti_env, mdt_obj); GOTO(out, rc); + } leh = buf->lb_buf; lee = (struct link_ea_entry *)(leh + 1); /* link #0 */ linkea_entry_unpack(lee, &reclen, tmpname, tmpfid); /* If set, use link #linkno for path lookup, otherwise use link #0. Only do this for the final path element. */ - if (pli->pli_fidcount == 0 && - pli->pli_linkno < leh->leh_reccount) { + if (first && fp->gf_linkno < leh->leh_reccount) { int count; - for (count = 0; count < pli->pli_linkno; count++) { + for (count = 0; count < fp->gf_linkno; count++) { lee = (struct link_ea_entry *) ((char *)lee + reclen); linkea_entry_unpack(lee, &reclen, tmpname, tmpfid); } - if (pli->pli_linkno < leh->leh_reccount - 1) + if (fp->gf_linkno < leh->leh_reccount - 1) /* indicate to user there are more links */ - pli->pli_linkno++; + fp->gf_linkno++; + } + + lmv_buf.lb_buf = info->mti_xattr_buf; + lmv_buf.lb_len = sizeof(info->mti_xattr_buf); + /* Check if it is slave stripes */ + rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj), + &lmv_buf, XATTR_NAME_LMV); + mdt_object_put(info->mti_env, mdt_obj); + if (rc > 0) { + union lmv_mds_md *lmm = lmv_buf.lb_buf; + + /* For slave stripes, get its master */ + if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) { + fp->gf_fid = *tmpfid; + continue; + } + } else if (rc < 0 && rc != -ENODATA) { + GOTO(out, rc); } + rc = 0; + /* Pack the name in the end of the buffer */ ptr -= tmpname->ln_namelen; - if (ptr - 1 <= pli->pli_path) + if (ptr - 1 <= fp->gf_path) GOTO(out, rc = -EOVERFLOW); strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen); *(--ptr) = '/'; - /* Store the parent fid for historic lookup */ - if (++pli->pli_fidcount >= MAX_PATH_DEPTH) - GOTO(out, rc = -EOVERFLOW); - pli->pli_fids[pli->pli_fidcount] = *tmpfid; + /* keep the last resolved fid to the client, so the + * client will build the left path on another MDT for + * remote object */ + fp->gf_fid = *tmpfid; + + first = false; } remote_out: ptr++; /* skip leading / */ - memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr); + memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr); - EXIT; out: - return rc; + RETURN(rc); } -/* Returns the full path to this fid, as of changelog record recno. */ +/** + * Given an MDT object, use mdt_path_current to get the path. + * Essentially a wrapper to retry mdt_path_current a set number of times + * if -EAGAIN is returned (usually because an object is being moved). + * + * Part of the MDT layer implementation of lfs fid2path. + * + * \param[in] info Per-thread common data shared by mdt level handlers. + * \param[in] obj Object to do path lookup of + * \param[in,out] fp User-provided struct for arguments and to store path + * information + * + * \retval 0 Lookup successful, path information stored in fp + * \retval negative errno if there was a problem + */ static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj, - char *path, int pathlen, __u64 *recno, int *linkno, - struct lu_fid *fid) + struct getinfo_fid2path *fp) { struct mdt_device *mdt = info->mti_mdt; - struct path_lookup_info *pli; int tries = 3; int rc = -EAGAIN; ENTRY; - if (pathlen < 3) + if (fp->gf_pathlen < 3) RETURN(-EOVERFLOW); if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) { - path[0] = '\0'; + fp->gf_path[0] = '\0'; RETURN(0); } - OBD_ALLOC_PTR(pli); - if (pli == NULL) - RETURN(-ENOMEM); - - pli->pli_mdt_obj = obj; - pli->pli_recno = *recno; - pli->pli_path = path; - pli->pli_pathlen = pathlen; - pli->pli_linkno = *linkno; - /* Retry multiple times in case file is being moved */ while (tries-- && rc == -EAGAIN) - rc = mdt_path_current(info, pli); - - /* return the last resolved fids to the client, so the client will - * build the left path on another MDT for remote object */ - *fid = pli->pli_fids[pli->pli_fidcount]; - - *recno = pli->pli_currec; - /* Return next link index to caller */ - *linkno = pli->pli_linkno; - - OBD_FREE_PTR(pli); + rc = mdt_path_current(info, obj, fp); RETURN(rc); } +/** + * Get the full path of the provided FID, as of changelog record recno. + * + * This checks sanity and looks up object for user provided FID + * before calling the actual path lookup code. + * + * Part of the MDT layer implementation of lfs fid2path. + * + * \param[in] info Per-thread common data shared by mdt level handlers. + * \param[in,out] fp User-provided struct for arguments and to store path + * information + * + * \retval 0 Lookup successful, path information and recno stored in fp + * \retval -ENOENT, object does not exist + * \retval negative errno if there was a problem + */ static int mdt_fid2path(struct mdt_thread_info *info, struct getinfo_fid2path *fp) { @@ -5215,10 +5532,11 @@ static int mdt_fid2path(struct mdt_thread_info *info, } obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid); - if (obj == NULL || IS_ERR(obj)) { - CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid), - PTR_ERR(obj)); - RETURN(-EINVAL); + if (IS_ERR(obj)) { + rc = PTR_ERR(obj); + CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n", + PFID(&fp->gf_fid), rc); + RETURN(rc); } if (mdt_object_remote(obj)) @@ -5230,13 +5548,12 @@ static int mdt_fid2path(struct mdt_thread_info *info, if (rc < 0) { mdt_object_put(info->mti_env, obj); - CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n", + CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n", PFID(&fp->gf_fid), rc); RETURN(rc); } - rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno, - &fp->gf_linkno, &fp->gf_fid); + rc = mdt_path(info, obj, fp); CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n", PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno); @@ -5278,7 +5595,7 @@ int mdt_get_info(struct tgt_session_info *tsi) key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY); if (key == NULL) { - CDEBUG(D_IOCTL, "No GETINFO key"); + CDEBUG(D_IOCTL, "No GETINFO key\n"); RETURN(err_serious(-EFAULT)); } keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY, @@ -5286,7 +5603,8 @@ int mdt_get_info(struct tgt_session_info *tsi) vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN); if (vallen == NULL) { - CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer"); + CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n", + tgt_name(tsi->tsi_tgt)); RETURN(err_serious(-EFAULT)); } @@ -5298,7 +5616,8 @@ int mdt_get_info(struct tgt_session_info *tsi) valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL); if (valout == NULL) { - CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer"); + CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n", + tgt_name(tsi->tsi_tgt)); RETURN(err_serious(-EFAULT)); } @@ -5428,14 +5747,24 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, } lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1); - lsp.lsp_namespace = mdt->mdt_namespace; + lsp.lsp_index_valid = 0; rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp); break; } case OBD_IOC_STOP_LFSCK: { - struct md_device *next = mdt->mdt_child; + struct md_device *next = mdt->mdt_child; + struct obd_ioctl_data *data = karg; + struct lfsck_stop stop; + + stop.ls_status = LS_STOPPED; + /* Old lfsck utils may pass NULL @stop. */ + if (data->ioc_inlbuf1 == NULL) + stop.ls_flags = 0; + else + stop.ls_flags = + ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags; - rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL); + rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop); break; } case OBD_IOC_GET_OBJ_VERSION: { @@ -5449,6 +5778,15 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, rc = mdt_ioc_version_get(mti, karg); break; } + case OBD_IOC_CATLOGLIST: { + struct mdt_thread_info *mti; + + mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key); + lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID); + rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg, + &mti->mti_tmp_fid1); + break; + } default: rc = -EOPNOTSUPP; CERROR("%s: Not supported cmd = %d, rc = %d\n", @@ -5459,17 +5797,27 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, RETURN(rc); } -int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) +static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) { - struct lu_device *ld = md2lu_dev(mdt->mdt_child); - int rc; - ENTRY; + struct lu_device *ld = md2lu_dev(mdt->mdt_child); + struct lfsck_start_param lsp; + int rc; + ENTRY; - rc = ld->ld_ops->ldo_recovery_complete(env, ld); - RETURN(rc); + lsp.lsp_start = NULL; + lsp.lsp_index_valid = 0; + rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child, + OBD_IOC_START_LFSCK, + 0, &lsp); + if (rc != 0 && rc != -EALREADY) + CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n", + mdt_obd_name(mdt), rc); + + rc = ld->ld_ops->ldo_recovery_complete(env, ld); + RETURN(rc); } -int mdt_obd_postrecov(struct obd_device *obd) +static int mdt_obd_postrecov(struct obd_device *obd) { struct lu_env env; int rc; @@ -5583,13 +5931,13 @@ void mdt_enable_cos(struct mdt_device *mdt, int val) mdt->mdt_opts.mo_cos = !!val; rc = lu_env_init(&env, LCT_LOCAL); - if (unlikely(rc != 0)) { - CWARN("lu_env initialization failed with rc = %d," - "cannot sync\n", rc); - return; - } - mdt_device_sync(&env, mdt); - lu_env_fini(&env); + if (unlikely(rc != 0)) { + CWARN("%s: lu_env initialization failed, cannot " + "sync: rc = %d\n", mdt_obd_name(mdt), rc); + return; + } + mdt_device_sync(&env, mdt); + lu_env_fini(&env); } /** @@ -5619,7 +5967,6 @@ static struct lu_device_type mdt_device_type = { static int __init mdt_mod_init(void) { - struct lprocfs_static_vars lvars; int rc; CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") == @@ -5634,10 +5981,8 @@ static int __init mdt_mod_init(void) if (rc) GOTO(lu_fini, rc); - lprocfs_mdt_init_vars(&lvars); - rc = class_register_type(&mdt_obd_device_ops, NULL, - lvars.module_vars, LUSTRE_MDT_NAME, - &mdt_device_type); + rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL, + LUSTRE_MDT_NAME, &mdt_device_type); if (rc) GOTO(mds_fini, rc); lu_fini: