X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=3d5de6f95cf8c7561dc4181525e621a4f04c97e5;hp=b49e4aeba756d70e7152bedaea6fd989167c154e;hb=2d15edc7be83555a7fa17a666d99b6af4922ab82;hpb=c6a034b746bffc5a11f0c9ec4848d54dea90ca23 diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index b49e4ae..3d5de6f 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -69,6 +69,7 @@ #include #include #include +#include mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_MINMODE] = MDL_MINMODE, @@ -280,15 +281,15 @@ static 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)); @@ -299,7 +300,7 @@ static 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: @@ -383,78 +384,142 @@ static void mdt_pack_size2body(struct mdt_thread_info *info, !(ma->ma_valid & MA_SOM)) return; - b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; - b->size = ma->ma_som->msd_size; - b->blocks = ma->ma_som->msd_blocks; + b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + b->mbo_size = ma->ma_som->msd_size; + b->mbo_blocks = ma->ma_som->msd_blocks; } +#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 + */ +int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody, + struct mdt_object *o, struct lu_nodemap *nodemap) +{ + 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; + + 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); + + b->mbo_atime = attr->la_atime; + b->mbo_mtime = attr->la_mtime; + b->mbo_ctime = attr->la_ctime; + b->mbo_mode = attr->la_mode; + b->mbo_size = attr->la_size; + b->mbo_blocks = attr->la_blocks; + b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID, + NODEMAP_FS_TO_CLIENT, + attr->la_uid); + b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID, + NODEMAP_FS_TO_CLIENT, + attr->la_gid); + b->mbo_flags = attr->la_flags; + b->mbo_nlink = attr->la_nlink; + b->mbo_rdev = attr->la_rdev; + + /* XXX should pack the reply body according to lu_valid */ + b->mbo_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->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 (fid) { + b->mbo_fid1 = *fid; + b->mbo_valid |= OBD_MD_FLID; + CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n", + PFID(fid), b->mbo_nlink, + b->mbo_mode, b->mbo_size); + } - if (info) - mdt_body_reverse_idmap(info, b); + 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) @@ -500,6 +565,12 @@ int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o) 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; } @@ -572,10 +643,26 @@ int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o, rc = mo_xattr_get(info->mti_env, next, buf, name); if (rc > 0) { + +got: if (strcmp(name, XATTR_NAME_LOV) == 0) { - ma->ma_lmm_size = rc; - ma->ma_valid |= MA_LOV; + 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) { @@ -583,6 +670,10 @@ int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o, 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 */ @@ -595,23 +686,7 @@ int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o, rc = mdt_big_xattr_get(info, o, name); if (rc > 0) { info->mti_big_lmm_used = 1; - if (!strcmp(name, XATTR_NAME_LOV)) { - ma->ma_valid |= MA_LOV; - ma->ma_lmm = info->mti_big_lmm; - ma->ma_lmm_size = rc; - } else if (!strcmp(name, XATTR_NAME_LMV)) { - ma->ma_valid |= MA_LMV; - ma->ma_lmv = info->mti_big_lmm; - ma->ma_lmv_size = rc; - } else { - return -EINVAL; - } - - /* 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; } } @@ -671,13 +746,17 @@ 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); @@ -763,38 +842,40 @@ 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); } - if (reqbody->eadatasize > 0) { + if (reqbody->mbo_eadatasize > 0) { buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD); if (buffer->lb_buf == NULL) GOTO(out, rc = -EPROTO); @@ -811,15 +892,15 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, /* 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 | OBD_MD_DEFAULT_MEA))) { + (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->valid & OBD_MD_MEA) + if (reqbody->mbo_valid & OBD_MD_MEA) ma->ma_need |= MA_LMV; - else if (reqbody->valid & OBD_MD_DEFAULT_MEA) + else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) ma->ma_need |= MA_LMV_DEF; } } else { @@ -831,7 +912,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } 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; @@ -851,8 +932,8 @@ 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); @@ -889,30 +970,38 @@ 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); - repbody->eadatasize = ma->ma_lmm_size; - if (S_ISDIR(la->la_mode)) - repbody->valid |= OBD_MD_FLDIREA; - else - repbody->valid |= OBD_MD_FLEASIZE; - mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->valid); + 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->eadatasize = ma->ma_lmv_size; - repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA); + 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->eadatasize = ma->ma_lmv_size; - repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA); + 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", @@ -924,16 +1013,16 @@ 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; @@ -947,62 +1036,37 @@ 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; @@ -1013,7 +1077,7 @@ 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: @@ -1051,7 +1115,7 @@ 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); } @@ -1068,7 +1132,7 @@ static int mdt_getattr(struct tgt_session_info *tsi) 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)); @@ -1084,7 +1148,7 @@ static int mdt_getattr(struct tgt_session_info *tsi) * guess at EA size by just reading it from disk. * Exceptions are readdir and (missing) directory striping */ /* Readlink */ - if (reqbody->valid & OBD_MD_LINKNAME) { + 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; @@ -1094,7 +1158,7 @@ static int mdt_getattr(struct tgt_session_info *tsi) */ } else if (lu_fid_eq(mdt_object_fid(obj), &info->mti_mdt->mdt_md_root_fid) && - (reqbody->valid & OBD_MD_FLDIREA) && + (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 @@ -1106,7 +1170,7 @@ static int mdt_getattr(struct tgt_session_info *tsi) } if (rc < 0) - GOTO(out_shrink, rc); + GOTO(out, rc = err_serious(rc)); req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc); @@ -1116,25 +1180,25 @@ static 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: @@ -1147,35 +1211,6 @@ out: return rc; } -static int mdt_is_subdir(struct tgt_session_info *tsi) -{ - 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; - - LASSERT(o != NULL); - - repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); - - /* - * 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; - - mdt_thread_info_fini(info); - RETURN(rc); -} - static int mdt_swap_layouts(struct tgt_session_info *tsi) { struct mdt_thread_info *info; @@ -1200,28 +1235,28 @@ static int mdt_swap_layouts(struct tgt_session_info *tsi) info = tsi2mdt_info(tsi); if (info->mti_dlm_req != NULL) - ldlm_request_cancel(req, info->mti_dlm_req, 0); + 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); @@ -1274,43 +1309,32 @@ 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); } /* @@ -1328,7 +1352,6 @@ 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; struct mdt_lock_handle *lhp = NULL; @@ -1343,7 +1366,8 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, LASSERT(ergo(is_resent, lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)); - LASSERT(parent != NULL); + if (parent == NULL) + RETURN(-ENOENT); if (info->mti_cross_ref) { /* Only getattr on the child. Parent is on another node. */ @@ -1354,15 +1378,10 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, "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); - } else { + 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); @@ -1409,7 +1428,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (unlikely(reqbody == NULL)) RETURN(err_serious(-EPROTO)); - *child_fid = reqbody->fid2; + *child_fid = reqbody->mbo_fid2; if (unlikely(!fid_is_sane(child_fid))) RETURN(err_serious(-EINVAL)); @@ -1417,7 +1436,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", " "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), - PFID(&reqbody->fid2), ldlm_rep); + PFID(&reqbody->mbo_fid2), ldlm_rep); } mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD); @@ -1437,10 +1456,15 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, } if (lu_name_is_valid(lname)) { - rc = mdt_raw_lookup(info, parent, lname, ldlm_rep); - if (rc != 0) { - if (rc > 0) - rc = 0; + /* Always allow to lookup ".." */ + if (unlikely(lname->ln_namelen == 2 && + lname->ln_name[0] == '.' && + lname->ln_name[1] == '.')) + info->mti_spec.sp_permitted = 1; + + if (info->mti_body->mbo_valid == OBD_MD_FLID) { + rc = mdt_raw_lookup(info, parent, lname, ldlm_rep); + RETURN(rc); } @@ -1457,8 +1481,8 @@ 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); @@ -1487,30 +1511,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, 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); - 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); - } else { -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); @@ -1632,9 +1637,9 @@ static 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)) @@ -1741,17 +1746,17 @@ static 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; @@ -1821,8 +1826,8 @@ 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); @@ -1932,19 +1937,20 @@ 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); } static int mdt_sync(struct tgt_session_info *tsi) @@ -1959,7 +1965,7 @@ static 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); @@ -2001,6 +2007,7 @@ static 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); @@ -2021,9 +2028,11 @@ static 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); @@ -2062,6 +2071,13 @@ static 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)); @@ -2228,12 +2244,13 @@ 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); } /** @@ -2308,10 +2325,11 @@ 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); @@ -2344,6 +2362,41 @@ int mdt_remote_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 (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, const struct lu_fid *fid, struct lustre_handle *lh, ldlm_mode_t mode, @@ -2435,8 +2488,8 @@ static int mdt_object_local_lock(struct mdt_thread_info *info, policy, res_id, dlmflags, info->mti_exp == NULL ? NULL : &info->mti_exp->exp_handle.h_cookie); - if (unlikely(rc)) - RETURN(rc); + if (unlikely(rc != 0)) + GOTO(out_unlock, rc); } /* @@ -2457,15 +2510,15 @@ static int mdt_object_local_lock(struct mdt_thread_info *info, res_id, LDLM_FL_LOCAL_ONLY | dlmflags, info->mti_exp == NULL ? NULL : &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); - } +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 @@ -2687,11 +2740,11 @@ 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); } @@ -2702,17 +2755,17 @@ 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; @@ -2816,6 +2869,7 @@ void mdt_thread_info_init(struct ptlrpc_request *req, 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; @@ -2862,7 +2916,8 @@ 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; @@ -2878,6 +2933,12 @@ static int mdt_tgt_connect(struct tgt_session_info *tsi) 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); @@ -2999,19 +3060,17 @@ static struct mdt_it_flavor { }; static int -mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp, - struct ldlm_lock *new_lock, struct mdt_lock_handle *lh, +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; @@ -3043,6 +3102,8 @@ mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp, */ 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); } @@ -3084,52 +3145,30 @@ mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp, 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, - enum mdt_it_code opcode) + __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]; - /* If the client does not require open lock, it does not need to - * search lock in exp_lock_hash, since the server thread will - * make sure the lock will be released, and the resend request - * can always re-enqueue the lock */ - if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN && - info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) { - /* 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) { - lock_res_and_lock(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); - unlock_res_and_lock(lock); - return; - } - cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); - unlock_res_and_lock(lock); - } + + /* 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 * it. (It's probably an OPEN, for which we don't send a lock. @@ -3144,8 +3183,8 @@ 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, @@ -3162,7 +3201,7 @@ 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, opcode); + 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, @@ -3174,7 +3213,7 @@ 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); @@ -3192,7 +3231,6 @@ 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 mdt_body *reqbody; @@ -3206,9 +3244,9 @@ 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: @@ -3231,7 +3269,7 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, 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, opcode); + 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); @@ -3244,7 +3282,7 @@ 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); @@ -3261,6 +3299,7 @@ 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; @@ -3276,6 +3315,9 @@ 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)); @@ -3283,8 +3325,10 @@ static int mdt_intent_layout(enum mdt_it_code opcode, if (mdt_object_exists(obj) && !mdt_object_remote(obj)) { /* get the length of lsm */ rc = mdt_attr_get_eabuf_size(info, obj); - if (rc < 0) + 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; @@ -3298,11 +3342,14 @@ static int mdt_intent_layout(enum mdt_it_code opcode, 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); @@ -3337,7 +3384,7 @@ static int mdt_intent_reint(enum mdt_it_code opcode, } /* Get lock from request for possible resent case. */ - mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode); + mdt_intent_fixup_resent(info, *lockp, lhc, flags); rc = mdt_reint_internal(info, lhc, opc); @@ -3353,10 +3400,10 @@ 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)) { + if (rc == -MDT_EREMOTE_OPEN || mdt_get_disposition(rep, DISP_OPEN_LOCK)) { LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh)); 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); } @@ -3389,7 +3436,7 @@ static int mdt_intent_reint(enum mdt_it_code opcode, 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); + rc = mdt_intent_lock_replace(info, lockp, lhc, flags); RETURN(rc); } else { lhc->mlh_reg_lh.cookie = 0ull; @@ -3448,17 +3495,16 @@ 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; @@ -3474,33 +3520,33 @@ 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); } @@ -3823,10 +3869,9 @@ 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; @@ -3859,10 +3904,9 @@ 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); @@ -3986,7 +4030,7 @@ 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); @@ -3997,7 +4041,7 @@ 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); @@ -4008,6 +4052,8 @@ 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) @@ -4114,7 +4160,7 @@ 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); @@ -4125,7 +4171,7 @@ 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); @@ -4139,6 +4185,8 @@ 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) @@ -4240,7 +4288,6 @@ 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), @@ -4666,7 +4713,6 @@ 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 */ @@ -4701,8 +4747,7 @@ 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 ? */ @@ -4918,12 +4963,14 @@ 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; @@ -4995,90 +5042,6 @@ 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_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) - mdt_export_stats_init(obd, lexp, localdata); - - /* For phase I, sync for cross-ref operation. */ - lexp->exp_keep_sync = 1; - } - - 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) @@ -5106,25 +5069,26 @@ 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); @@ -5141,12 +5105,13 @@ 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 @@ -5192,23 +5157,120 @@ 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; @@ -5253,34 +5315,15 @@ 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 */ -}; - int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj, struct linkea_data *ldata) { @@ -5312,8 +5355,21 @@ int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj, 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; @@ -5326,6 +5382,7 @@ static int mdt_path_current(struct mdt_thread_info *info, int reclen; struct linkea_data ldata = { 0 }; int rc = 0; + bool first = true; ENTRY; /* temp buffer for path element, the buffer will be finally freed @@ -5335,148 +5392,151 @@ 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 = pli->pli_fids[0]; + *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])) { + 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)) { + if (!mdt_object_exists(mdt_obj)) { mdt_object_put(info->mti_env, mdt_obj); - GOTO(remote_out, rc = -EREMOTE); - } - - 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); - 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) { - struct lmv_mds_md_v1 *lmm1 = &lmm->lmv_md_v1; - - fid_le_to_cpu(tmpfid, &lmm1->lmv_master_fid); - if (!fid_is_sane(tmpfid)) { - mdt_object_put(info->mti_env, mdt_obj); - GOTO(out, rc = -EINVAL); - } - mdt_object_put(info->mti_env, mdt_obj); - pli->pli_fids[pli->pli_fidcount] = *tmpfid; - continue; - } + GOTO(out, rc = -ENOENT); } - if (!mdt_object_exists(mdt_obj)) { + if (mdt_object_remote(mdt_obj)) { mdt_object_put(info->mti_env, mdt_obj); - GOTO(out, rc = -ENOENT); + 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) { @@ -5514,13 +5574,12 @@ 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); @@ -5562,7 +5621,7 @@ 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, @@ -5570,7 +5629,8 @@ 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)); } @@ -5582,7 +5642,8 @@ 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)); } @@ -5896,13 +5957,13 @@ 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); } /** @@ -5932,7 +5993,6 @@ 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") == @@ -5947,11 +6007,7 @@ 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, true, NULL, -#ifndef HAVE_ONLY_PROCFS_SEQ - lvars.module_vars, -#endif LUSTRE_MDT_NAME, &mdt_device_type); if (rc) GOTO(mds_fini, rc);