X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=c3e8cf1445a09509080926f191c50326ab4d64d4;hb=1bf196aaf5adb9a1c67886c0cd6a780ec6838040;hp=b19f7c7b28df3a75723820e89e5fe984aaa9a5de;hpb=0e638c2fb9dd45323ccff0d9fb43f378faf6127f;p=fs%2Flustre-release.git diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index b19f7c7..c3e8cf1 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2010, 2014, Intel Corporation. + * Copyright (c) 2010, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -49,27 +45,31 @@ #define DEBUG_SUBSYSTEM S_MDS #include -/* - * struct OBD_{ALLOC,FREE}*() - */ -#include +#include + +#include +#include +#include #include -/* struct ptlrpc_request */ +#include +#include #include -/* struct obd_export */ -#include -/* struct obd_device */ -#include -/* lu2dt_dev() */ -#include +#include #include -#include -#include "mdt_internal.h" -#include #include #include -#include -#include +#include +#include +#include + +#include + +#include "mdt_internal.h" + + +static unsigned int max_mod_rpcs_per_client = 8; +module_param(max_mod_rpcs_per_client, uint, 0644); +MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client"); mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_MINMODE] = MDL_MINMODE, @@ -82,15 +82,15 @@ mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_GROUP] = MDL_GROUP }; -ldlm_mode_t mdt_dlm_lock_modes[] = { - [MDL_MINMODE] = LCK_MINMODE, - [MDL_EX] = LCK_EX, - [MDL_PW] = LCK_PW, - [MDL_PR] = LCK_PR, - [MDL_CW] = LCK_CW, - [MDL_CR] = LCK_CR, - [MDL_NL] = LCK_NL, - [MDL_GROUP] = LCK_GROUP +enum ldlm_mode mdt_dlm_lock_modes[] = { + [MDL_MINMODE] = LCK_MINMODE, + [MDL_EX] = LCK_EX, + [MDL_PW] = LCK_PW, + [MDL_PR] = LCK_PR, + [MDL_CW] = LCK_CW, + [MDL_CR] = LCK_CR, + [MDL_NL] = LCK_NL, + [MDL_GROUP] = LCK_GROUP }; static struct mdt_device *mdt_dev(struct lu_device *d); @@ -128,47 +128,48 @@ static struct lu_kmem_descr mdt_caches[] = { } }; -int mdt_get_disposition(struct ldlm_reply *rep, int flag) +__u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag) { - if (!rep) - return 0; - return (rep->lock_policy_res1 & flag); + if (!rep) + return 0; + return rep->lock_policy_res1 & op_flag; } void mdt_clear_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, __u64 op_flag) { if (info) { - info->mti_opdata &= ~flag; - tgt_opdata_clear(info->mti_env, flag); + info->mti_opdata &= ~op_flag; + tgt_opdata_clear(info->mti_env, op_flag); } if (rep) - rep->lock_policy_res1 &= ~flag; + rep->lock_policy_res1 &= ~op_flag; } void mdt_set_disposition(struct mdt_thread_info *info, - struct ldlm_reply *rep, int flag) + struct ldlm_reply *rep, __u64 op_flag) { if (info) { - info->mti_opdata |= flag; - tgt_opdata_set(info->mti_env, flag); + info->mti_opdata |= op_flag; + tgt_opdata_set(info->mti_env, op_flag); } if (rep) - rep->lock_policy_res1 |= flag; + rep->lock_policy_res1 |= op_flag; } -void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm) +void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm) { - lh->mlh_pdo_hash = 0; - lh->mlh_reg_mode = lm; + lh->mlh_pdo_hash = 0; + lh->mlh_reg_mode = lm; lh->mlh_rreg_mode = lm; - lh->mlh_type = MDT_REG_LOCK; + lh->mlh_type = MDT_REG_LOCK; } -void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode, +void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode, const struct lu_name *lname) { lh->mlh_reg_mode = lock_mode; + lh->mlh_pdo_mode = LCK_MINMODE; lh->mlh_rreg_mode = lock_mode; lh->mlh_type = MDT_PDO_LOCK; @@ -264,12 +265,98 @@ static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o EXIT; } -static int mdt_getstatus(struct tgt_session_info *tsi) +static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset, + struct lu_fid *fid) +{ + struct mdt_device *mdt = info->mti_mdt; + struct lu_name *lname = &info->mti_name; + char *name = NULL; + struct mdt_object *parent; + u32 mode; + int rc = 0; + + LASSERT(!info->mti_cross_ref); + + OBD_ALLOC(name, NAME_MAX + 1); + if (name == NULL) + return -ENOMEM; + lname->ln_name = name; + + /* + * We may want to allow this to mount a completely separate + * fileset from the MDT in the future, but keeping it to + * ROOT/ only for now avoid potential security issues. + */ + *fid = mdt->mdt_md_root_fid; + + while (rc == 0 && fileset != NULL && *fileset != '\0') { + const char *s1 = fileset; + const char *s2; + + while (*++s1 == '/') + ; + s2 = s1; + while (*s2 != '/' && *s2 != '\0') + s2++; + + if (s2 == s1) + break; + + fileset = s2; + + lname->ln_namelen = s2 - s1; + if (lname->ln_namelen > NAME_MAX) { + rc = -EINVAL; + break; + } + + /* reject .. as a path component */ + if (lname->ln_namelen == 2 && + strncmp(s1, "..", 2) == 0) { + rc = -EINVAL; + break; + } + + strncpy(name, s1, lname->ln_namelen); + name[lname->ln_namelen] = '\0'; + + parent = mdt_object_find(info->mti_env, mdt, fid); + if (IS_ERR(parent)) { + rc = PTR_ERR(parent); + break; + } + /* Only got the fid of this obj by name */ + fid_zero(fid); + rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, + fid, &info->mti_spec); + mdt_object_put(info->mti_env, parent); + } + if (!rc) { + parent = mdt_object_find(info->mti_env, mdt, fid); + if (IS_ERR(parent)) + rc = PTR_ERR(parent); + else { + mode = lu_object_attr(&parent->mot_obj); + mdt_object_put(info->mti_env, parent); + if (!S_ISDIR(mode)) + rc = -ENOTDIR; + } + } + + OBD_FREE(name, NAME_MAX + 1); + + return rc; +} + +static int mdt_get_root(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); struct mdt_device *mdt = info->mti_mdt; struct mdt_body *repbody; + char *fileset = NULL, *buffer = NULL; int rc; + struct obd_export *exp = info->mti_exp; + char *nodemap_fileset; ENTRY; @@ -277,34 +364,50 @@ static int mdt_getstatus(struct tgt_session_info *tsi) if (rc) GOTO(out, rc = err_serious(rc)); - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK)) GOTO(out, rc = err_serious(-ENOMEM)); repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - repbody->mbo_fid1 = mdt->mdt_md_root_fid; - repbody->mbo_valid |= OBD_MD_FLID; + if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) { + fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME); + if (fileset == NULL) + GOTO(out, rc = err_serious(-EFAULT)); + } - 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; + nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap); + if (nodemap_fileset && nodemap_fileset[0]) { + CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset); + if (fileset) { + /* consider fileset from client as a sub-fileset + * of the nodemap one */ + OBD_ALLOC(buffer, PATH_MAX + 1); + if (buffer == NULL) + GOTO(out, rc = err_serious(-ENOMEM)); + if (snprintf(buffer, PATH_MAX + 1, "%s/%s", + nodemap_fileset, fileset) >= PATH_MAX + 1) + GOTO(out, rc = err_serious(-EINVAL)); + fileset = buffer; + } else { + /* enforce fileset as specified in the nodemap */ + fileset = nodemap_fileset; + } + } + + if (fileset) { + CDEBUG(D_INFO, "Getting fileset %s\n", fileset); + rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1); + if (rc < 0) + GOTO(out, rc = err_serious(rc)); + } else { + repbody->mbo_fid1 = mdt->mdt_md_root_fid; + } + repbody->mbo_valid |= OBD_MD_FLID; - root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1); - if (IS_ERR(root)) - GOTO(out, rc = PTR_ERR(root)); - - capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1); - LASSERT(capa); - capa->lc_opc = CAPA_OPC_MDS_DEFAULT; - rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa, - 0); - mdt_object_put(info->mti_env, root); - if (rc == 0) - repbody->mbo_valid |= OBD_MD_FLMDSCAPA; - } EXIT; out: mdt_thread_info_fini(info); + if (buffer) + OBD_FREE(buffer, PATH_MAX+1); return rc; } @@ -344,15 +447,15 @@ static int mdt_statfs(struct tgt_session_info *tsi) rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs); if (rc) GOTO(out, rc); - spin_lock(&info->mti_mdt->mdt_osfs_lock); + spin_lock(&info->mti_mdt->mdt_lock); info->mti_mdt->mdt_osfs = *osfs; info->mti_mdt->mdt_osfs_age = cfs_time_current_64(); - spin_unlock(&info->mti_mdt->mdt_osfs_lock); + spin_unlock(&info->mti_mdt->mdt_lock); } else { /** use cached statfs data */ - spin_lock(&info->mti_mdt->mdt_osfs_lock); + spin_lock(&info->mti_mdt->mdt_lock); *osfs = info->mti_mdt->mdt_osfs; - spin_unlock(&info->mti_mdt->mdt_osfs_lock); + spin_unlock(&info->mti_mdt->mdt_lock); } if (rc == 0) @@ -362,33 +465,6 @@ out: RETURN(rc); } -/** - * Pack SOM attributes into the reply. - * Call under a DLM UPDATE lock. - */ -static void mdt_pack_size2body(struct mdt_thread_info *info, - struct mdt_object *mo) -{ - 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; - - 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. @@ -448,9 +524,9 @@ int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody, 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; - struct obd_export *exp = info->mti_exp; - struct lu_nodemap *nodemap = exp->exp_target_data.ted_nodemap; + struct md_attr *ma = &info->mti_attr; + struct obd_export *exp = info->mti_exp; + struct lu_nodemap *nodemap = NULL; LASSERT(ma->ma_valid & MA_INODE); @@ -474,6 +550,11 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, b->mbo_nlink = attr->la_nlink; b->mbo_valid |= OBD_MD_FLNLINK; } + if (attr->la_valid & (LA_UID|LA_GID)) { + nodemap = nodemap_get_from_exp(exp); + if (IS_ERR(nodemap)) + goto out; + } if (attr->la_valid & LA_UID) { b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID, NODEMAP_FS_TO_CLIENT, @@ -486,6 +567,7 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, 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; @@ -495,13 +577,10 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, if (fid != NULL) { b->mbo_fid1 = *fid; b->mbo_valid |= OBD_MD_FLID; - CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n", + CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\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; @@ -535,6 +614,10 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, 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); + +out: + if (!IS_ERR_OR_NULL(nodemap)) + nodemap_putref(nodemap); } static inline int mdt_body_has_lov(const struct lu_attr *la, @@ -814,19 +897,6 @@ int mdt_attr_get_complex(struct mdt_thread_info *info, 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_HSM && S_ISREG(mode)) { buf->lb_buf = info->mti_xattr_buf; buf->lb_len = sizeof(info->mti_xattr_buf); @@ -857,7 +927,7 @@ int mdt_attr_get_complex(struct mdt_thread_info *info, #endif out: ma->ma_need = need; - CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n", + CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n", rc, ma->ma_valid, ma->ma_lmm); RETURN(rc); } @@ -875,7 +945,6 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, 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; @@ -941,22 +1010,22 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, 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)) { - CERROR("%s: getattr error for "DFID": rc = %d\n", + CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR, + "%s: getattr error for "DFID": rc = %d\n", mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(o)), rc); RETURN(rc); } /* 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->mbo_t_state = MS_RESTORE; + if (ma->ma_valid & MA_HSM) { repbody->mbo_valid |= OBD_MD_TSTATE; + if ((ma->ma_hsm.mh_flags & HS_RELEASED) && + mdt_hsm_restore_is_running(info, mdt_object_fid(o))) + repbody->mbo_t_state = MS_RESTORE; } is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid); @@ -1032,7 +1101,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, PFID(mdt_object_fid(o)), rc); rc = -EFAULT; } else { - int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc); + int print_limit = min_t(int, PAGE_SIZE - 128, rc); if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO)) rc -= 2; @@ -1060,49 +1129,24 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } 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); + CDEBUG(D_INODE, "changing the max MD size to %u\n", + repbody->mbo_max_mdsize); } - 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->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->mbo_valid & OBD_MD_FLACL)) + if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) && + (reqbody->mbo_valid & OBD_MD_FLACL)) { + struct lu_nodemap *nodemap = nodemap_get_from_exp(exp); + if (IS_ERR(nodemap)) + RETURN(PTR_ERR(nodemap)); + rc = mdt_pack_acl2body(info, repbody, o, nodemap); + nodemap_putref(nodemap); + } #endif - 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; - - capa = req_capsule_server_get(pill, &RMF_CAPA1); - LASSERT(capa); - capa->lc_opc = CAPA_OPC_MDS_DEFAULT; - rc = mo_capa_get(env, next, capa, 0); - if (rc) - RETURN(rc); - repbody->mbo_valid |= OBD_MD_FLMDSCAPA; - } - out: if (rc == 0) mdt_counter_incr(req, LPROC_MDT_GETATTR); @@ -1110,38 +1154,6 @@ out: RETURN(rc); } -static int mdt_renew_capa(struct mdt_thread_info *info) -{ - struct mdt_object *obj = info->mti_object; - struct mdt_body *body; - struct lustre_capa *capa, *c; - int rc; - ENTRY; - - /* if object doesn't exist, or server has disabled capability, - * return directly, client will find body->valid OBD_MD_FLOSSCAPA - * flag not set. - */ - if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa || - !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA)) - RETURN(0); - - body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - LASSERT(body != NULL); - - c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1); - LASSERT(c); - - capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2); - LASSERT(capa); - - *capa = *c; - rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1); - if (rc == 0) - body->mbo_valid |= OBD_MD_FLOSSCAPA; - RETURN(rc); -} - static int mdt_getattr(struct tgt_session_info *tsi) { struct mdt_thread_info *info = tsi2mdt_info(tsi); @@ -1154,15 +1166,6 @@ static int mdt_getattr(struct tgt_session_info *tsi) reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY); LASSERT(reqbody); - - if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) { - rc = req_capsule_server_pack(pill); - if (unlikely(rc)) - RETURN(err_serious(rc)); - rc = mdt_renew_capa(info); - GOTO(out_shrink, rc); - } - LASSERT(obj != NULL); LASSERT(lu_object_assert_exists(&obj->mot_obj)); @@ -1206,23 +1209,13 @@ static int mdt_getattr(struct tgt_session_info *tsi) repbody->mbo_eadatasize = 0; repbody->mbo_aclsize = 0; - if (reqbody->mbo_valid & OBD_MD_FLRMTPERM) - rc = mdt_init_ucred(info, reqbody); - else - rc = mdt_check_ucred(info); + rc = mdt_check_ucred(info); if (unlikely(rc)) GOTO(out_shrink, rc); 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->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: mdt_client_compatibility(info); @@ -1234,6 +1227,28 @@ out: return rc; } +/** + * 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) +{ + unsigned int o1_lov_created = o1->mot_lov_created; + + mutex_lock(&o1->mot_lov_mutex); + mutex_lock(&o2->mot_lov_mutex); + + o1->mot_lov_created = o2->mot_lov_created; + o2->mot_lov_created = o1_lov_created; + + mutex_unlock(&o2->mot_lov_mutex); + mutex_unlock(&o1->mot_lov_mutex); +} + static int mdt_swap_layouts(struct tgt_session_info *tsi) { struct mdt_thread_info *info; @@ -1260,16 +1275,6 @@ static int mdt_swap_layouts(struct tgt_session_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->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->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->mbo_fid2); @@ -1308,18 +1313,22 @@ static int mdt_swap_layouts(struct tgt_session_info *tsi) mdt_lock_reg_init(lh2, LCK_EX); rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT | - MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); + MDS_INODELOCK_XATTR); if (rc < 0) GOTO(put, rc); rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT | - MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK); + MDS_INODELOCK_XATTR); if (rc < 0) GOTO(unlock1, rc); 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: @@ -1361,7 +1370,7 @@ static int mdt_raw_lookup(struct mdt_thread_info *info, } /* - * UPDATE lock should be taken against parent, and be release before exit; + * UPDATE lock should be taken against parent, and be released before exit; * child_bits lock should be taken against child, and be returned back: * (1)normal request should release the child lock; * (2)intent request will grant the lock to client. @@ -1416,8 +1425,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, MDS_INODELOCK_LAYOUT); child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE; - rc = mdt_object_lock(info, child, lhc, child_bits, - MDT_LOCAL_LOCK); + rc = mdt_object_lock(info, child, lhc, child_bits); if (rc < 0) RETURN(rc); } @@ -1426,12 +1434,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (!mdt_object_exists(child)) { LU_OBJECT_DEBUG(D_INFO, info->mti_env, &child->mot_obj, - "remote object doesn't exist.\n"); + "remote object doesn't exist."); 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); @@ -1467,7 +1474,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, 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"); + "Parent doesn't exist!"); RETURN(-ESTALE); } @@ -1495,12 +1502,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (S_ISDIR(lu_object_attr(&parent->mot_obj))) { 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); - if (unlikely(rc != 0)) - RETURN(rc); - } + rc = mdt_object_lock(info, parent, lhp, + MDS_INODELOCK_UPDATE); + if (unlikely(rc != 0)) + RETURN(rc); + } /* step 2: lookup child's fid by name */ fid_zero(child_fid); @@ -1535,22 +1541,22 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (unlikely(IS_ERR(child))) GOTO(out_parent, rc = PTR_ERR(child)); + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2); + if (!mdt_object_exists(child)) { + LU_OBJECT_DEBUG(D_INODE, info->mti_env, + &child->mot_obj, + "Object doesn't exist!"); + GOTO(out_child, rc = -ENOENT); + } + 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, - &child->mot_obj, - "Object doesn't exist!\n"); - GOTO(out_child, rc = -ENOENT); - } - if (!(child_bits & MDS_INODELOCK_UPDATE) && mdt_object_exists(child) && !mdt_object_remote(child)) { struct md_attr *ma = &info->mti_attr; @@ -1589,12 +1595,12 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, child_bits |= MDS_INODELOCK_LAYOUT; /* try layout lock, it may fail to be granted due to * contention at LOOKUP or UPDATE */ - if (!mdt_object_lock_try(info, child, lhc, child_bits, - MDT_CROSS_LOCK)) { + if (!mdt_object_lock_try(info, child, lhc, + child_bits)) { child_bits &= ~MDS_INODELOCK_LAYOUT; LASSERT(child_bits != 0); rc = mdt_object_lock(info, child, lhc, - child_bits, MDT_CROSS_LOCK); + child_bits); } else { ma_need |= MA_LOV; } @@ -1603,22 +1609,15 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, * 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); + rc = mdt_object_lock(info, child, lhc, child_bits); } if (unlikely(rc != 0)) GOTO(out_child, rc); } 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); rc = mdt_getattr_internal(info, child, ma_need); if (unlikely(rc != 0)) { mdt_object_unlock(info, child, lhc, 1); @@ -1630,8 +1629,6 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, "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); @@ -1664,7 +1661,7 @@ static int mdt_getattr_name(struct tgt_session_info *tsi) repbody->mbo_eadatasize = 0; repbody->mbo_aclsize = 0; - rc = mdt_init_ucred(info, reqbody); + rc = mdt_init_ucred_intent_getattr(info, reqbody); if (unlikely(rc)) GOTO(out_shrink, rc); @@ -1685,7 +1682,7 @@ out_shrink: } static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, - void *karg, void *uarg); + void *karg, void __user *uarg); static int mdt_set_info(struct tgt_session_info *tsi) { @@ -1739,6 +1736,9 @@ static int mdt_set_info(struct tgt_session_info *tsi) rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export, vallen, val, NULL); + } else if (KEY_IS(KEY_EVICT_BY_NID)) { + if (vallen > 0) + obd_export_evict_by_nid(req->rq_export->exp_obd, val); } else { RETURN(-EINVAL); } @@ -1771,7 +1771,7 @@ static int mdt_readpage(struct tgt_session_info *tsi) */ rdpg->rp_hash = reqbody->mbo_size; if (rdpg->rp_hash != reqbody->mbo_size) { - CERROR("Invalid hash: "LPX64" != "LPX64"\n", + CERROR("Invalid hash: %#llx != %#llx\n", rdpg->rp_hash, reqbody->mbo_size); RETURN(-EFAULT); } @@ -1781,14 +1781,14 @@ static int mdt_readpage(struct tgt_session_info *tsi) rdpg->rp_attrs |= LUDA_64BITHASH; 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; + rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> + PAGE_SHIFT; OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]); if (rdpg->rp_pages == NULL) RETURN(-ENOMEM); for (i = 0; i < rdpg->rp_npages; ++i) { - rdpg->rp_pages[i] = alloc_page(GFP_IOFS); + rdpg->rp_pages[i] = alloc_page(GFP_NOFS); if (rdpg->rp_pages[i] == NULL) GOTO(free_rdpg, rc = -ENOMEM); } @@ -1815,6 +1815,26 @@ free_rdpg: return rc; } +static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op) +{ + struct lu_ucred *uc = mdt_ucred_check(info); + struct lu_attr *attr = &info->mti_attr.ma_attr; + + if (uc == NULL) + return -EINVAL; + + if (op != REINT_SETATTR) { + if ((attr->la_valid & LA_UID) && (attr->la_uid != -1)) + attr->la_uid = uc->uc_fsuid; + /* for S_ISGID, inherit gid from his parent, such work will be + * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */ + if ((attr->la_valid & LA_GID) && (attr->la_gid != -1)) + attr->la_gid = uc->uc_fsgid; + } + + return 0; +} + static int mdt_reint_internal(struct mdt_thread_info *info, struct mdt_lock_handle *lhc, __u32 op) @@ -1868,7 +1888,10 @@ static int mdt_reint_internal(struct mdt_thread_info *info, if (rc != 0) GOTO(out_ucred, rc = err_serious(rc)); - if (mdt_check_resent(info, mdt_reconstruct, lhc)) { + rc = mdt_check_resent(info, mdt_reconstruct, lhc); + if (rc < 0) { + GOTO(out_ucred, rc); + } else if (rc == 1) { 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); @@ -2030,7 +2053,7 @@ 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; + struct lu_nodemap *nodemap; ENTRY; oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); @@ -2041,59 +2064,30 @@ static int mdt_quotactl(struct tgt_session_info *tsi) if (rc) RETURN(err_serious(rc)); + nodemap = nodemap_get_from_exp(exp); + if (IS_ERR(nodemap)) + RETURN(PTR_ERR(nodemap)); + switch (oqctl->qc_cmd) { - case Q_QUOTACHECK: - case LUSTRE_Q_INVALIDATE: - case LUSTRE_Q_FINVALIDATE: - case Q_QUOTAON: - case Q_QUOTAOFF: - case Q_INITQUOTA: - /* deprecated, not used any more */ - RETURN(-EOPNOTSUPP); /* master quotactl */ case Q_SETINFO: case Q_SETQUOTA: if (!nodemap_can_setquota(nodemap)) - RETURN(-EPERM); + GOTO(out_nodemap, rc = -EPERM); case Q_GETINFO: case Q_GETQUOTA: if (qmt == NULL) - RETURN(-EOPNOTSUPP); + GOTO(out_nodemap, rc = -EOPNOTSUPP); /* slave quotactl */ case Q_GETOINFO: case Q_GETOQUOTA: break; default: CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd); - RETURN(-EFAULT); + GOTO(out_nodemap, rc = -EFAULT); } - /* map uid/gid for remote client */ id = oqctl->qc_id; - if (exp_connect_rmtclient(exp)) { - struct lustre_idmap_table *idmap; - - idmap = exp->exp_mdt_data.med_idmap; - - if (unlikely(oqctl->qc_cmd != Q_GETQUOTA && - oqctl->qc_cmd != Q_GETINFO)) - RETURN(-EPERM); - - if (oqctl->qc_type == USRQUOTA) - id = lustre_idmap_lookup_uid(NULL, idmap, 0, - oqctl->qc_id); - else if (oqctl->qc_type == GRPQUOTA) - id = lustre_idmap_lookup_gid(NULL, idmap, 0, - oqctl->qc_id); - else - RETURN(-EINVAL); - - if (id == CFS_IDMAP_NOTFOUND) { - CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id); - RETURN(-EACCES); - } - } - if (oqctl->qc_type == USRQUOTA) id = nodemap_map_id(nodemap, NODEMAP_UID, NODEMAP_CLIENT_TO_FS, id); @@ -2103,7 +2097,7 @@ static int mdt_quotactl(struct tgt_session_info *tsi) repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL); if (repoqc == NULL) - RETURN(err_serious(-EFAULT)); + GOTO(out_nodemap, rc = err_serious(-EFAULT)); if (oqctl->qc_id != id) swap(oqctl->qc_id, id); @@ -2127,14 +2121,20 @@ static int mdt_quotactl(struct tgt_session_info *tsi) default: CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd); - RETURN(-EFAULT); + GOTO(out_nodemap, rc = -EFAULT); } if (oqctl->qc_id != id) swap(oqctl->qc_id, id); *repoqc = *oqctl; - RETURN(rc); + + EXIT; + +out_nodemap: + nodemap_putref(nodemap); + + return rc; } /** clone llog ctxt from child (mdd) @@ -2186,21 +2186,9 @@ static int mdt_llog_ctxt_unclone(const struct lu_env *env, */ static int mdt_sec_ctx_handle(struct tgt_session_info *tsi) { - int rc; - - rc = mdt_handle_idmap(tsi); - if (unlikely(rc)) { - struct ptlrpc_request *req = tgt_ses_req(tsi); - __u32 opc; - - opc = lustre_msg_get_opc(req->rq_reqmsg); - if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT) - sptlrpc_svc_ctx_invalidate(req); - } - CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val); - return rc; + return 0; } /* @@ -2269,11 +2257,14 @@ static void mdt_device_commit_async(const struct lu_env *env, { struct dt_device *dt = mdt->mdt_bottom; int rc; + ENTRY; 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); + atomic_inc(&mdt->mdt_async_commit_count); + EXIT; } /** @@ -2329,17 +2320,22 @@ int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, if (flag == LDLM_CB_CANCELING) RETURN(0); + lock_res_and_lock(lock); if (lock->l_blocking_ast != mdt_blocking_ast) { unlock_res_and_lock(lock); RETURN(0); } - if (mdt_cos_is_enabled(mdt) && - lock->l_req_mode & (LCK_PW | LCK_EX) && - lock->l_blocking_lock != NULL && - lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) { - mdt_set_lock_sync(lock); - } + if (lock->l_req_mode & (LCK_PW | LCK_EX) && + lock->l_blocking_lock != NULL) { + if (mdt_cos_is_enabled(mdt) && + lock->l_client_cookie != + lock->l_blocking_lock->l_client_cookie) + mdt_set_lock_sync(lock); + else if (mdt_slc_is_enabled(mdt) && + ldlm_is_cos_incompat(lock->l_blocking_lock)) + mdt_set_lock_sync(lock); + } rc = ldlm_blocking_ast_nocheck(lock); /* There is no lock conflict if l_blocking_lock == NULL, @@ -2360,29 +2356,77 @@ int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, RETURN(rc); } -/* Used for cross-MDT lock */ +/* + * Blocking AST for cross-MDT lock + * + * Discard lock from uncommitted_slc_locks and cancel it. + * + * \param lock the lock which blocks a request or cancelling lock + * \param desc unused + * \param data unused + * \param flag indicates whether this cancelling or blocking callback + * \retval 0 on success + * \retval negative number on error + */ int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, void *data, int flag) { - struct lustre_handle lockh; - int rc; + int rc = 0; + ENTRY; switch (flag) { - case LDLM_CB_BLOCKING: + case LDLM_CB_BLOCKING: { + struct lustre_handle lockh; + ldlm_lock2handle(lock, &lockh); - rc = ldlm_cli_cancel(&lockh, LCF_ASYNC); + rc = ldlm_cli_cancel(&lockh, + ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC); if (rc < 0) { CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc); RETURN(rc); } break; - case LDLM_CB_CANCELING: + } + case LDLM_CB_CANCELING: { + struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd; + struct mdt_device *mdt = + mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev); + LDLM_DEBUG(lock, "Revoke remote lock\n"); + + /* discard slc lock here so that it can be cleaned anytime, + * especially for cleanup_resource() */ + tgt_discard_slc_lock(&mdt->mdt_lut, lock); + + /* once we cache lock, l_ast_data is set to mdt_object */ + if (lock->l_ast_data != NULL) { + struct mdt_object *mo = lock->l_ast_data; + struct lu_env env; + + rc = lu_env_init(&env, LCT_MD_THREAD); + if (unlikely(rc != 0)) { + CWARN("%s: lu_env initialization failed, object" + "%p "DFID" is leaked!\n", + obd->obd_name, mo, + PFID(mdt_object_fid(mo))); + RETURN(rc); + } + + if (lock->l_policy_data.l_inodebits.bits & + (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) { + rc = mo_invalidate(&env, mdt_object_child(mo)); + mo->mot_cache_attr = 0; + } + mdt_object_put(&env, mo); + lu_env_fini(&env); + } break; + } default: LBUG(); } - RETURN(0); + + RETURN(rc); } int mdt_check_resent_lock(struct mdt_thread_info *info, @@ -2400,7 +2444,7 @@ int mdt_check_resent_lock(struct mdt_thread_info *info, /* 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", + LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx", lhc->mlh_reg_lh.cookie); RETURN(-ESTALE); } @@ -2420,21 +2464,19 @@ int mdt_check_resent_lock(struct mdt_thread_info *info, 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, - __u64 ibits) +int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o, + const struct lu_fid *fid, struct lustre_handle *lh, + enum ldlm_mode mode, __u64 ibits, bool nonblock, + bool cache) { struct ldlm_enqueue_info *einfo = &mti->mti_einfo; - ldlm_policy_data_t *policy = &mti->mti_policy; + union ldlm_policy_data *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); - fid_build_reg_res_name(fid, res_id); memset(einfo, 0, sizeof(*einfo)); @@ -2444,32 +2486,54 @@ int mdt_remote_object_lock(struct mdt_thread_info *mti, einfo->ei_cb_cp = ldlm_completion_ast; einfo->ei_enq_slave = 0; einfo->ei_res_id = res_id; + if (nonblock) + einfo->ei_nonblock = 1; + if (cache) { + /* + * if we cache lock, couple lock with mdt_object, so that object + * can be easily found in lock ASTs. + */ + mdt_object_get(mti->mti_env, o); + einfo->ei_cbdata = o; + } memset(policy, 0, sizeof(*policy)); policy->l_inodebits.bits = ibits; rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo, policy); + if (rc < 0 && cache) { + mdt_object_put(mti->mti_env, o); + einfo->ei_cbdata = NULL; + } RETURN(rc); } 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) + bool nonblock, bool cos_incompat) { - struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace; - ldlm_policy_data_t *policy = &info->mti_policy; - struct ldlm_res_id *res_id = &info->mti_res_id; - __u64 dlmflags; - int rc; - ENTRY; + struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace; + union ldlm_policy_data *policy = &info->mti_policy; + struct ldlm_res_id *res_id = &info->mti_res_id; + __u64 dlmflags = 0; + int rc; + ENTRY; LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh)); LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh)); LASSERT(lh->mlh_reg_mode != LCK_MINMODE); LASSERT(lh->mlh_type != MDT_NUL_LOCK); + if (cos_incompat) { + LASSERT(lh->mlh_reg_mode == LCK_PW || + lh->mlh_reg_mode == LCK_EX); + dlmflags |= LDLM_FL_COS_INCOMPAT; + } else if (mdt_cos_is_enabled(info->mti_mdt)) { + dlmflags |= LDLM_FL_COS_ENABLED; + } + /* Only enqueue LOOKUP lock for remote object */ if (mdt_object_remote(o)) LASSERT(ibits == MDS_INODELOCK_LOOKUP); @@ -2489,7 +2553,7 @@ static int mdt_object_local_lock(struct mdt_thread_info *info, memset(policy, 0, sizeof(*policy)); fid_build_reg_res_name(mdt_object_fid(o), res_id); - dlmflags = LDLM_FL_ATOMIC_CB; + dlmflags |= LDLM_FL_ATOMIC_CB; if (nonblock) dlmflags |= LDLM_FL_BLOCK_NOWAIT; @@ -2546,26 +2610,33 @@ out_unlock: 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) + struct mdt_lock_handle *lh, __u64 ibits, bool nonblock, + bool cos_incompat) { + struct mdt_lock_handle *local_lh = NULL; 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); + if (!mdt_object_remote(o)) { + rc = mdt_object_local_lock(info, o, lh, ibits, nonblock, + cos_incompat); + RETURN(rc); } /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */ ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR); + + /* Only enqueue LOOKUP lock for remote object */ + if (ibits & MDS_INODELOCK_LOOKUP) { + rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP, + nonblock, cos_incompat); + if (rc != ELDLM_OK) + RETURN(rc); + + local_lh = lh; + } + 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 @@ -2583,36 +2654,55 @@ mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o, 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) + MDS_INODELOCK_UPDATE, nonblock, + false); + if (rc != ELDLM_OK) { + if (local_lh != NULL) + mdt_object_unlock(info, o, local_lh, rc); 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) + struct mdt_lock_handle *lh, __u64 ibits) { - return mdt_object_lock_internal(info, o, lh, ibits, false, locality); + return mdt_object_lock_internal(info, o, lh, ibits, false, false); +} + +int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o, + struct mdt_lock_handle *lh, __u64 ibits, + bool cos_incompat) +{ + LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX); + return mdt_object_lock_internal(info, o, lh, ibits, false, + cos_incompat); } int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o, - struct mdt_lock_handle *lh, __u64 ibits, int locality) + struct mdt_lock_handle *lh, __u64 ibits) +{ + struct mdt_lock_handle tmp = *lh; + int rc; + + rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false); + if (rc == 0) + *lh = tmp; + + return rc == 0; +} + +int mdt_reint_object_lock_try(struct mdt_thread_info *info, + struct mdt_object *o, struct mdt_lock_handle *lh, + __u64 ibits, bool cos_incompat) { struct mdt_lock_handle tmp = *lh; int rc; - rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality); + LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX); + rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat); if (rc == 0) *lh = tmp; @@ -2632,38 +2722,40 @@ int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o, * \param mode lock mode * \param decref force immediate lock releasing */ -static -void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, - ldlm_mode_t mode, int decref) +static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, + enum ldlm_mode mode, int decref) { - ENTRY; + ENTRY; - if (lustre_handle_is_used(h)) { - if (decref || !info->mti_has_trans || - !(mode & (LCK_PW | LCK_EX))){ - mdt_fid_unlock(h, mode); - } else { - struct mdt_device *mdt = info->mti_mdt; - struct ldlm_lock *lock = ldlm_handle2lock(h); - struct ptlrpc_request *req = mdt_info_req(info); - int no_ack = 0; + if (lustre_handle_is_used(h)) { + if (decref || !info->mti_has_trans || + !(mode & (LCK_PW | LCK_EX))) { + mdt_fid_unlock(h, mode); + } else { + struct mdt_device *mdt = info->mti_mdt; + struct ldlm_lock *lock = ldlm_handle2lock(h); + struct ptlrpc_request *req = mdt_info_req(info); + int cos; + + cos = (mdt_cos_is_enabled(mdt) || + mdt_slc_is_enabled(mdt)); + + LASSERTF(lock != NULL, "no lock for cookie %#llx\n", + h->cookie); - LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n", - h->cookie); /* 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, + " transno = %lld\n", req, req->rq_reply_state, req->rq_transno); - if (mdt_cos_is_enabled(mdt)) { - no_ack = 1; + if (cos) { ldlm_lock_downgrade(lock, LCK_COS); mode = LCK_COS; } - ptlrpc_save_lock(req, h, mode, no_ack); + ptlrpc_save_lock(req, h, mode, cos); } else { - ldlm_lock_decref(h, mode); + mdt_fid_unlock(h, mode); } if (mdt_is_lock_sync(lock)) { CDEBUG(D_HA, "found sync-lock," @@ -2680,10 +2772,53 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, } /** + * Save cross-MDT lock in uncommitted_slc_locks + * + * Keep the lock referenced until transaction commit happens or release the lock + * immediately depending on input parameters. + * + * \param info thead info object + * \param h lock handle + * \param mode lock mode + * \param decref force immediate lock releasing + */ +static void mdt_save_remote_lock(struct mdt_thread_info *info, + struct mdt_object *o, struct lustre_handle *h, + enum ldlm_mode mode, int decref) +{ + ENTRY; + + if (lustre_handle_is_used(h)) { + struct ldlm_lock *lock = ldlm_handle2lock(h); + + if (o != NULL && + (lock->l_policy_data.l_inodebits.bits & + (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE))) + mo_invalidate(info->mti_env, mdt_object_child(o)); + + if (decref || !info->mti_has_trans || + !(mode & (LCK_PW | LCK_EX))) { + ldlm_lock_decref_and_cancel(h, mode); + LDLM_LOCK_PUT(lock); + } else { + struct ptlrpc_request *req = mdt_info_req(info); + + LASSERT(req != NULL); + tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock, + req->rq_transno); + ldlm_lock_decref(h, mode); + } + h->cookie = 0ull; + } + + EXIT; +} + +/** * Unlock mdt object. * * Immeditely release the regular lock and the PDO lock or save the - * lock in reqeuest and keep them referenced until client ACK or + * lock in request and keep them referenced until client ACK or * transaction commit. * * \param info thread info object @@ -2692,17 +2827,16 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, * \param decref force immediate lock releasing */ void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o, - struct mdt_lock_handle *lh, int decref) + struct mdt_lock_handle *lh, int decref) { - ENTRY; - - mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref); - mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref); + ENTRY; - if (lustre_handle_is_used(&lh->mlh_rreg_lh)) - ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode); + mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref); + mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref); + mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, + decref); - EXIT; + EXIT; } struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info, @@ -2716,8 +2850,7 @@ struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info, if (!IS_ERR(o)) { int rc; - rc = mdt_object_lock(info, o, lh, ibits, - MDT_LOCAL_LOCK); + rc = mdt_object_lock(info, o, lh, ibits); if (rc != 0) { mdt_object_put(info->mti_env, o); o = ERR_PTR(rc); @@ -2771,27 +2904,11 @@ static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags) RETURN(-EINVAL); } - /* - * Do not get size or any capa fields before we check that request - * contains capa actually. There are some requests which do not, for - * instance MDS_IS_SUBDIR. - */ - 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->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)) { + 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; + rc = -ENOENT; } else { info->mti_object = obj; rc = 0; @@ -2827,17 +2944,6 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) RETURN(rc); } -static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m) -{ - struct md_device *next = m->mdt_child; - - return next->md_ops->mdo_init_capa_ctxt(env, next, - m->mdt_lut.lut_mds_capa, - m->mdt_capa_timeout, - m->mdt_capa_alg, - m->mdt_capa_keys); -} - void mdt_lock_handle_init(struct mdt_lock_handle *lh) { lh->mlh_type = MDT_NUL_LOCK; @@ -2893,6 +2999,7 @@ 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.sp_migrate_close = 0; info->mti_spec.u.sp_ea.eadata = NULL; info->mti_spec.u.sp_ea.eadatalen = 0; @@ -2920,28 +3027,14 @@ void mdt_thread_info_fini(struct mdt_thread_info *info) struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi) { struct mdt_thread_info *mti; - struct lustre_capa *lc; mti = mdt_th_info(tsi->tsi_env); LASSERT(mti != NULL); mdt_thread_info_init(tgt_ses_req(tsi), mti); if (tsi->tsi_corpus != NULL) { - struct req_capsule *pill = tsi->tsi_pill; - mti->mti_object = mdt_obj(tsi->tsi_corpus); lu_object_get(tsi->tsi_corpus); - - /* - * XXX: must be part of tgt_mdt_body_unpack but moved here - * due to mdt_set_capainfo(). - */ - 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->mbo_fid1, - lc); - } } mti->mti_body = tsi->tsi_mdt_body; mti->mti_dlm_req = tsi->tsi_dlm_req; @@ -2951,28 +3044,14 @@ struct mdt_thread_info *tsi2mdt_info(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); + tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) { + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC)); + } - rc = mdt_init_idmap(tsi); - if (rc != 0) - GOTO(err, rc); - RETURN(0); -err: - obd_disconnect(class_export_get(req->rq_export)); - return rc; + return tgt_connect(tsi); } enum mdt_it_code { @@ -3086,7 +3165,7 @@ static int mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp, struct mdt_lock_handle *lh, - __u64 flags) + __u64 flags, int result) { struct ptlrpc_request *req = mdt_info_req(info); struct ldlm_lock *lock = *lockp; @@ -3100,8 +3179,19 @@ mdt_intent_lock_replace(struct mdt_thread_info *info, RETURN(0); } - LASSERTF(new_lock != NULL, - "lockh "LPX64"\n", lh->mlh_reg_lh.cookie); + if (new_lock == NULL && (flags & LDLM_FL_RESENT)) { + /* 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 %#llx\n", + lh->mlh_reg_lh.cookie); + lh->mlh_reg_lh.cookie = 0; + RETURN(-ESTALE); + } + + LASSERTF(new_lock != NULL, + "lockh %#llx flags %#llx : rc = %d\n", + lh->mlh_reg_lh.cookie, flags, result); /* * If we've already given this lock to a client once, then we should @@ -3187,17 +3277,17 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, 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, + DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx", 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. - */ - if (req_xid_is_last(req)) - 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. + */ + if (req_can_reconstruct(req, NULL)) + return; /* * This remote handle isn't enqueued, so we never received or processed @@ -3206,7 +3296,7 @@ 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, + DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx", dlmreq->lock_handle[0].cookie); } @@ -3217,7 +3307,8 @@ static int mdt_intent_getxattr(enum mdt_it_code opcode, { struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT]; struct ldlm_reply *ldlm_rep = NULL; - int rc, grc; + int rc; + ENTRY; /* * Initialize lhc->mlh_reg_lh either from a previously granted lock @@ -3228,24 +3319,35 @@ static int mdt_intent_getxattr(enum mdt_it_code opcode, 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, - MDS_INODELOCK_XATTR, - MDT_LOCAL_LOCK); + MDS_INODELOCK_XATTR); if (rc) return rc; } - grc = mdt_getxattr(info); - - rc = mdt_intent_lock_replace(info, lockp, lhc, flags); + rc = mdt_getxattr(info); if (mdt_info_req(info)->rq_repmsg != NULL) ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); - if (ldlm_rep == NULL) + + if (ldlm_rep == NULL || + OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) { + mdt_object_unlock(info, info->mti_object, lhc, 1); RETURN(err_serious(-EFAULT)); + } + + ldlm_rep->lock_policy_res2 = clear_serious(rc); - ldlm_rep->lock_policy_res2 = grc; + /* This is left for interop instead of adding a new interop flag. + * LU-7433 */ +#if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0) + if (ldlm_rep->lock_policy_res2) { + mdt_object_unlock(info, info->mti_object, lhc, 1); + RETURN(ELDLM_LOCK_ABORTED); + } +#endif - return rc; + rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc); + RETURN(rc); } static int mdt_intent_getattr(enum mdt_it_code opcode, @@ -3284,9 +3386,9 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, GOTO(out_shrink, rc = -EINVAL); } - rc = mdt_init_ucred(info, reqbody); - if (rc) - GOTO(out_shrink, rc); + rc = mdt_init_ucred_intent_getattr(info, reqbody); + if (rc) + GOTO(out_shrink, rc); ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD); @@ -3305,7 +3407,7 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED); } - rc = mdt_intent_lock_replace(info, lockp, lhc, flags); + rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc); EXIT; out_ucred: mdt_exit_ucred(info); @@ -3326,6 +3428,7 @@ static int mdt_intent_layout(enum mdt_it_code opcode, struct layout_intent *layout; struct lu_fid *fid; struct mdt_object *obj = NULL; + int layout_size = 0; int rc = 0; ENTRY; @@ -3335,6 +3438,16 @@ static int mdt_intent_layout(enum mdt_it_code opcode, RETURN(-EINVAL); } + layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT); + if (layout == NULL) + RETURN(-EPROTO); + + if (layout->li_opc != LAYOUT_INTENT_ACCESS) { + CERROR("%s: Unsupported layout intent opc %d\n", + mdt_obd_name(info->mti_mdt), layout->li_opc); + RETURN(-EINVAL); + } + fid = &info->mti_tmp_fid2; fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name); @@ -3343,40 +3456,33 @@ static int mdt_intent_layout(enum mdt_it_code opcode, obj = mdt_object_find(info->mti_env, info->mti_mdt, fid); if (IS_ERR(obj)) - RETURN(PTR_ERR(obj)); + GOTO(out, rc = PTR_ERR(obj)); 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) { - mdt_object_put(info->mti_env, obj); - RETURN(rc); - } + layout_size = mdt_attr_get_eabuf_size(info, obj); + if (layout_size < 0) + GOTO(out_obj, rc = layout_size); - if (rc > info->mti_mdt->mdt_max_mdsize) - info->mti_mdt->mdt_max_mdsize = rc; + if (layout_size > info->mti_mdt->mdt_max_mdsize) + info->mti_mdt->mdt_max_mdsize = layout_size; } - 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, rc); + req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, + layout_size); rc = req_capsule_server_pack(info->mti_pill); - if (rc != 0) - RETURN(-EINVAL); + GOTO(out_obj, rc); - if (lustre_handle_is_used(&lhc->mlh_reg_lh)) - rc = mdt_intent_lock_replace(info, lockp, lhc, flags); +out_obj: + mdt_object_put(info->mti_env, obj); - layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT); - LASSERT(layout != NULL); - if (layout->li_opc == LAYOUT_INTENT_ACCESS) - /* return to normal/resent ldlm handling */ - RETURN(rc); + if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh)) + rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc); + +out: + lhc->mlh_reg_lh.cookie = 0; - CERROR("%s: Unsupported layout intent (%d)\n", - mdt_obd_name(info->mti_mdt), layout->li_opc); - RETURN(-EINVAL); + return rc; } static int mdt_intent_reint(enum mdt_it_code opcode, @@ -3426,15 +3532,16 @@ static int mdt_intent_reint(enum mdt_it_code opcode, 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, lhc, flags); + rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc); RETURN(rc); } rep->lock_policy_res2 = clear_serious(rc); if (rep->lock_policy_res2 == -ENOENT && - mdt_get_disposition(rep, DISP_LOOKUP_NEG)) - rep->lock_policy_res2 = 0; + mdt_get_disposition(rep, DISP_LOOKUP_NEG) && + !mdt_get_disposition(rep, DISP_OPEN_CREATE)) + rep->lock_policy_res2 = 0; lhc->mlh_reg_lh.cookie = 0ull; if (rc == -ENOTCONN || rc == -ENODEV || @@ -3454,54 +3561,55 @@ static int mdt_intent_reint(enum mdt_it_code opcode, RETURN(ELDLM_LOCK_ABORTED); } -static int mdt_intent_code(long itcode) +static int mdt_intent_code(enum ldlm_intent_flags itcode) { - int rc; + int rc; - switch(itcode) { - case IT_OPEN: - rc = MDT_IT_OPEN; - break; - case IT_OPEN|IT_CREAT: - rc = MDT_IT_OCREAT; - break; - case IT_CREAT: - rc = MDT_IT_CREATE; - break; - case IT_READDIR: - rc = MDT_IT_READDIR; - break; - case IT_GETATTR: - rc = MDT_IT_GETATTR; - break; - case IT_LOOKUP: - rc = MDT_IT_LOOKUP; - break; - case IT_UNLINK: - rc = MDT_IT_UNLINK; - break; - case IT_TRUNC: - rc = MDT_IT_TRUNC; - break; - case IT_GETXATTR: - rc = MDT_IT_GETXATTR; - break; - case IT_LAYOUT: - rc = MDT_IT_LAYOUT; - break; + switch (itcode) { + case IT_OPEN: + rc = MDT_IT_OPEN; + break; + case IT_OPEN|IT_CREAT: + rc = MDT_IT_OCREAT; + break; + case IT_CREAT: + rc = MDT_IT_CREATE; + break; + case IT_READDIR: + rc = MDT_IT_READDIR; + break; + case IT_GETATTR: + rc = MDT_IT_GETATTR; + break; + case IT_LOOKUP: + rc = MDT_IT_LOOKUP; + break; + case IT_UNLINK: + rc = MDT_IT_UNLINK; + break; + case IT_TRUNC: + rc = MDT_IT_TRUNC; + break; + case IT_GETXATTR: + rc = MDT_IT_GETXATTR; + break; + case IT_LAYOUT: + rc = MDT_IT_LAYOUT; + break; case IT_QUOTA_DQACQ: case IT_QUOTA_CONN: rc = MDT_IT_QUOTA; break; - default: - CERROR("Unknown intent opcode: %ld\n", itcode); - rc = -EINVAL; - break; - } - return rc; + default: + CERROR("Unknown intent opcode: 0x%08x\n", itcode); + rc = -EINVAL; + break; + } + return rc; } -static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, +static int mdt_intent_opc(enum ldlm_intent_flags itopc, + struct mdt_thread_info *info, struct ldlm_lock **lockp, __u64 flags) { struct req_capsule *pill = info->mti_pill; @@ -3544,6 +3652,8 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, if (flv->it_act != NULL) { struct ldlm_reply *rep; + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10); + /* execute policy */ rc = flv->it_act(opc, info, lockp, flags); @@ -3560,8 +3670,8 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, } static int mdt_intent_policy(struct ldlm_namespace *ns, - struct ldlm_lock **lockp, void *req_cookie, - ldlm_mode_t mode, __u64 flags, void *data) + struct ldlm_lock **lockp, void *req_cookie, + enum ldlm_mode mode, __u64 flags, void *data) { struct tgt_session_info *tsi; struct mdt_thread_info *info; @@ -3664,20 +3774,26 @@ static int mdt_register_lwp_callback(void *data) LASSERT(mdt_seq_site(mdt)->ss_node_id != 0); - if (!likely(fld->lsf_new)) - RETURN(0); - rc = lu_env_init(&env, LCT_MD_THREAD); - if (rc) { + if (rc < 0) { 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); + /* Allocate new sequence now to avoid creating local transaction + * in the normal transaction process */ + rc = seq_server_check_and_alloc_super(&env, + mdt_seq_site(mdt)->ss_server_seq); + if (rc < 0) GOTO(out, rc); + + if (fld->lsf_new) { + 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); @@ -3798,7 +3914,7 @@ static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt) GOTO(out_seq_fini, rc); if (ss->ss_node_id != 0) - /* register controler export through lwp */ + /* register controller export through lwp */ rc = mdt_register_seq_exp(mdt); EXIT; @@ -4040,7 +4156,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt, lcfg = lustre_cfg_new(LCFG_ATTACH, bufs); if (lcfg == NULL) - GOTO(free_bufs, rc = -ENOMEM); + GOTO(put_profile, rc = -ENOMEM); rc = class_attach(lcfg); if (rc) @@ -4100,6 +4216,8 @@ class_detach: class_detach(obd, lcfg); lcfg_cleanup: lustre_cfg_free(lcfg); +put_profile: + class_put_profile(lprof); free_bufs: OBD_FREE_PTR(bufs); cleanup_mem: @@ -4170,7 +4288,7 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt, lcfg = lustre_cfg_new(LCFG_ATTACH, bufs); if (lcfg == NULL) - GOTO(cleanup_mem, rc = -ENOMEM); + GOTO(put_profile, rc = -ENOMEM); rc = class_attach(lcfg); if (rc) @@ -4232,6 +4350,8 @@ class_detach: class_detach(obd, lcfg); lcfg_cleanup: lustre_cfg_free(lcfg); +put_profile: + class_put_profile(lprof); cleanup_mem: if (bufs) OBD_FREE_PTR(bufs); @@ -4285,7 +4405,7 @@ TGT_RPC_HANDLER(MDS_FIRST_OPC, HABEO_REFERO, MDS_SET_INFO, mdt_set_info, &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION), TGT_MDT_HDL(0, MDS_GET_INFO, mdt_get_info), -TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GETSTATUS, mdt_getstatus), +TGT_MDT_HDL(0 | HABEO_REFERO, MDS_GET_ROOT, mdt_get_root), TGT_MDT_HDL(HABEO_CORPUS, MDS_GETATTR, mdt_getattr), TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME, mdt_getattr_name), @@ -4293,8 +4413,6 @@ 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(0, MDS_QUOTACTL, mdt_quotactl), @@ -4395,9 +4513,8 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) 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)); + ping_evictor_stop(); if (m->mdt_opts.mo_coordinator) mdt_hsm_cdt_stop(m); @@ -4406,6 +4523,11 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT); mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); + + if (m->mdt_namespace != NULL) + ldlm_namespace_free_prior(m->mdt_namespace, NULL, + d->ld_obd->obd_force); + obd_exports_barrier(obd); obd_zombie_barrier(); @@ -4416,11 +4538,15 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) upcall_cache_cleanup(m->mdt_identity_cache); m->mdt_identity_cache = NULL; - if (m->mdt_namespace != NULL) { - ldlm_namespace_free(m->mdt_namespace, NULL, - d->ld_obd->obd_force); - d->ld_obd->obd_namespace = m->mdt_namespace = NULL; - } + if (m->mdt_namespace != NULL) { + ldlm_namespace_free_post(m->mdt_namespace); + d->ld_obd->obd_namespace = m->mdt_namespace = NULL; + } + + if (m->mdt_md_root != NULL) { + mdt_object_put(env, m->mdt_md_root); + m->mdt_md_root = NULL; + } mdt_quota_fini(env, m); @@ -4429,10 +4555,6 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) mdt_seq_fini(env, m); mdt_fld_fini(env, m); - next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL); - cfs_timer_disarm(&m->mdt_ck_timer); - mdt_ck_thread_stop(m); - /* * Finish the stack */ @@ -4485,9 +4607,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, LASSERT(obd != NULL); m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */ - - m->mdt_som_conf = 0; - + m->mdt_opts.mo_evict_tgt_nids = 1; m->mdt_opts.mo_cos = MDT_COS_DEFAULT; /* default is coordinator off, it is started through conf_param @@ -4504,21 +4624,23 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, LASSERT(num); node_id = simple_strtol(num, NULL, 10); obd->u.obt.obt_magic = OBT_MAGIC; + if (lsi->lsi_lmd != NULL && + lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK) + m->mdt_skip_lfsck = 1; } - 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.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); + spin_lock_init(&m->mdt_lock); m->mdt_osfs_age = cfs_time_shift_64(-1000); m->mdt_enable_remote_dir = 0; m->mdt_enable_remote_dir_gid = 0; + atomic_set(&m->mdt_mds_mds_conns, 0); + atomic_set(&m->mdt_async_commit_count, 0); + m->mdt_lu_dev.ld_ops = &mdt_lu_ops; m->mdt_lu_dev.ld_obd = obd; /* Set this lu_device to obd for error handling purposes. */ @@ -4580,8 +4702,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, /* set obd_namespace for compatibility with old code */ obd->obd_namespace = m->mdt_namespace; - cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m); - rc = mdt_hsm_cdt_init(m); if (rc != 0) { CERROR("%s: error initializing coordinator, rc %d\n", @@ -4589,15 +4709,11 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, GOTO(err_free_ns, rc); } - rc = mdt_ck_thread_start(m); - if (rc) - GOTO(err_free_hsm, rc); - rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice, OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_MDS_ALL_REPLY_NET); if (rc) - GOTO(err_capa, rc); + GOTO(err_free_hsm, rc); rc = mdt_fs_setup(env, m, obd, lsi); if (rc) @@ -4659,8 +4775,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, * when the whole stack is complete and ready * to serve the requests */ - mdt_init_capa_ctxt(env, m); - /* Reduce the initial timeout on an MDS because it doesn't need such * a long timeout as an OST does. Adaptive timeouts will adjust this * value appropriately. */ @@ -4678,9 +4792,6 @@ err_fs_cleanup: mdt_fs_cleanup(env, m); err_tgt: tgt_fini(env, &m->mdt_lut); -err_capa: - cfs_timer_disarm(&m->mdt_ck_timer); - mdt_ck_thread_stop(m); err_free_hsm: mdt_hsm_cdt_fini(m); err_free_ns: @@ -4705,6 +4816,7 @@ static struct cfg_interop_param mdt_interop_param[] = { { "mdt.group_upcall", NULL }, { "mdt.quota_type", NULL }, { "mdd.quota_type", NULL }, + { "mdt.som", NULL }, { "mdt.rootsquash", "mdt.root_squash" }, { "mdt.nosquash_nid", "mdt.nosquash_nids" }, { NULL } @@ -4801,7 +4913,7 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env, lu_object_init(o, h, d); lu_object_add_top(h, o); o->lo_ops = &mdt_obj_ops; - mutex_init(&mo->mot_ioepoch_mutex); + spin_lock_init(&mo->mot_write_lock); mutex_init(&mo->mot_lov_mutex); init_rwsem(&mo->mot_open_sem); RETURN(o); @@ -4852,13 +4964,15 @@ 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(%s %s, writecount=%d)", + mdto, mdto->mot_lov_created ? "lov_created" : "", + mdto->mot_cache_attr ? "cache_attr" : "", + mdto->mot_write_count); } static int mdt_prepare(const struct lu_env *env, @@ -4899,6 +5013,7 @@ static int mdt_prepare(const struct lu_env *env, } LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state)); + target_recovery_init(&mdt->mdt_lut, tgt_request_handle); set_bit(MDT_FL_CFGLOG, &mdt->mdt_state); LASSERT(obd->obd_no_conn); @@ -4955,6 +5070,10 @@ static int mdt_obd_set_info_async(const struct lu_env *env, * connect flags from the obd_connect_data::ocd_connect_flags field of the * reply. \see mdt_connect(). * + * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a + * full struct obd_connect_data. So care must be taken when accessing fields + * that are not present in struct obd_connect_data_v1. See LU-16. + * * \param exp the obd_export associated with this client/target pair * \param mdt the target device for the connection * \param data stores data for this connect request @@ -4970,12 +5089,16 @@ static int mdt_connect_internal(struct obd_export *exp, LASSERT(data != NULL); data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED; + + if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2) + data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2; + data->ocd_ibits_known &= MDS_INODELOCK_FULL; 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", + "very old or an invalid client: flags %#llx\n", mdt_obd_name(mdt), exp->exp_client_uuid.uuid, data->ocd_connect_flags); return -EBADE; @@ -4987,14 +5110,11 @@ 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); if (data->ocd_brw_size == 0) { - CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64 + CERROR("%s: cli %s/%p ocd_connect_flags: %#llx" " ocd_version: %x ocd_grant: %d " "ocd_index: %u ocd_brw_size is " "unexpectedly zero, network data " @@ -5027,15 +5147,6 @@ 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); @@ -5048,6 +5159,18 @@ static int mdt_connect_internal(struct obd_export *exp, data->ocd_max_easize = mdt->mdt_max_ea_size; + /* NB: Disregard the rule against updating + * exp_connect_data.ocd_connect_flags in this case, since + * tgt_client_new() needs to know if this is client supports + * multiple modify RPCs, and it is safe to expose this flag before + * connection processing completes. */ + if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) { + data->ocd_maxmodrpcs = max_mod_rpcs_per_client; + spin_lock(&exp->exp_lock); + *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS; + spin_unlock(&exp->exp_lock); + } + return 0; } @@ -5137,7 +5260,7 @@ 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 */ @@ -5158,6 +5281,18 @@ static int mdt_export_cleanup(struct obd_export *exp) RETURN(rc); } +static inline void mdt_enable_slc(struct mdt_device *mdt) +{ + if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL) + mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL; +} + +static inline void mdt_disable_slc(struct mdt_device *mdt) +{ + if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL) + mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL; +} + static int mdt_obd_disconnect(struct obd_export *exp) { int rc; @@ -5166,12 +5301,20 @@ static int mdt_obd_disconnect(struct obd_export *exp) LASSERT(exp); class_export_get(exp); - nodemap_del_member(exp); + if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) && + !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) { + struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev); + + if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns)) + mdt_disable_slc(mdt); + } + rc = server_disconnect_export(exp); if (rc != 0) CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc); rc = mdt_export_cleanup(exp); + nodemap_del_member(exp); class_export_put(exp); RETURN(rc); } @@ -5191,21 +5334,30 @@ static int mdt_obd_connect(const struct lu_env *env, ENTRY; LASSERT(env != NULL); + LASSERT(data != NULL); + if (!exp || !obd || !cluuid) RETURN(-EINVAL); mdt = mdt_dev(obd->obd_lu_dev); + if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) && + !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) { + atomic_inc(&mdt->mdt_mds_mds_conns); + mdt_enable_slc(mdt); + } + /* * 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)) { + if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && + !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) && + !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) { rc = obd_get_info(env, mdt->mdt_child_exp, sizeof(KEY_OSP_CONNECTED), - KEY_OSP_CONNECTED, NULL, NULL, NULL); + KEY_OSP_CONNECTED, NULL, NULL); if (rc) RETURN(-EAGAIN); set_bit(MDT_FL_SYNCED, &mdt->mdt_state); @@ -5218,6 +5370,10 @@ static int mdt_obd_connect(const struct lu_env *env, lexp = class_conn2export(&conn); LASSERT(lexp != NULL); + rc = nodemap_add_member(*client_nid, lexp); + if (rc != 0 && rc != -EEXIST) + GOTO(out, rc); + rc = mdt_connect_internal(lexp, mdt, data); if (rc == 0) { struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd; @@ -5225,25 +5381,23 @@ static int mdt_obd_connect(const struct lu_env *env, 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; - + if (rc == 0) 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); + nodemap_del_member(lexp); *exp = NULL; } else { *exp = lexp; + /* Because we do not want this export to be evicted by pinger, + * let's not add this export to the timed chain list. */ + if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) { + spin_lock(&lexp->exp_obd->obd_dev_lock); + list_del_init(&lexp->exp_obd_chain_timed); + spin_unlock(&lexp->exp_obd->obd_dev_lock); + } } RETURN(rc); @@ -5262,12 +5416,15 @@ static int mdt_obd_reconnect(const struct lu_env *env, if (exp == NULL || obd == NULL || cluuid == NULL) RETURN(-EINVAL); + rc = nodemap_add_member(*client_nid, exp); + if (rc != 0 && rc != -EEXIST) + RETURN(rc); + 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); - } + if (rc == 0) + mdt_export_stats_init(obd, exp, localdata); + else + nodemap_del_member(exp); RETURN(rc); } @@ -5281,8 +5438,6 @@ static int mdt_init_export(struct obd_export *exp) INIT_LIST_HEAD(&med->med_open_head); spin_lock_init(&med->med_open_lock); - mutex_init(&med->med_idmap_mutex); - med->med_idmap = NULL; spin_lock(&exp->exp_lock); exp->exp_connecting = 1; spin_unlock(&exp->exp_lock); @@ -5314,9 +5469,6 @@ static int mdt_destroy_export(struct obd_export *exp) { ENTRY; - if (exp_connect_rmtclient(exp)) - mdt_cleanup_idmap(&exp->exp_mdt_data); - target_destroy_export(exp); /* destroy can be called from failed obd_setup, so * checking uuid is safer than obd_self_export */ @@ -5371,6 +5523,7 @@ int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj, * \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 + * \param[in] root_fid Root FID of current path should reach * * \retval 0 Lookup successful, path information stored in fp * \retval -EAGAIN Lookup failed, usually because object is being moved @@ -5378,7 +5531,8 @@ int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj, */ static int mdt_path_current(struct mdt_thread_info *info, struct mdt_object *obj, - struct getinfo_fid2path *fp) + struct getinfo_fid2path *fp, + struct lu_fid *root_fid) { struct mdt_device *mdt = info->mti_mdt; struct mdt_object *mdt_obj; @@ -5401,16 +5555,18 @@ static int mdt_path_current(struct mdt_thread_info *info, RETURN(-ENOMEM); ldata.ld_buf = buf; - ptr = fp->gf_path + fp->gf_pathlen - 1; + ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1; *ptr = 0; --ptr; *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, &fp->gf_fid)) { + while (!lu_fid_eq(root_fid, &fp->gf_fid)) { struct lu_buf lmv_buf; + if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) && + lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) + GOTO(out, rc = -ENOENT); + mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid); if (IS_ERR(mdt_obj)) GOTO(out, rc = PTR_ERR(mdt_obj)); @@ -5471,7 +5627,7 @@ static int mdt_path_current(struct mdt_thread_info *info, /* Pack the name in the end of the buffer */ ptr -= tmpname->ln_namelen; - if (ptr - 1 <= fp->gf_path) + if (ptr - 1 <= fp->gf_u.gf_path) GOTO(out, rc = -EOVERFLOW); strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen); *(--ptr) = '/'; @@ -5486,7 +5642,8 @@ static int mdt_path_current(struct mdt_thread_info *info, remote_out: ptr++; /* skip leading / */ - memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr); + memmove(fp->gf_u.gf_path, ptr, + fp->gf_u.gf_path + fp->gf_pathlen - ptr); out: RETURN(rc); @@ -5508,7 +5665,7 @@ out: * \retval negative errno if there was a problem */ static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj, - struct getinfo_fid2path *fp) + struct getinfo_fid2path *fp, struct lu_fid *root_fid) { struct mdt_device *mdt = info->mti_mdt; int tries = 3; @@ -5518,14 +5675,17 @@ static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj, if (fp->gf_pathlen < 3) RETURN(-EOVERFLOW); - if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) { - fp->gf_path[0] = '\0'; + if (root_fid == NULL) + root_fid = &mdt->mdt_md_root_fid; + + if (lu_fid_eq(root_fid, mdt_object_fid(obj))) { + fp->gf_u.gf_path[0] = '\0'; RETURN(0); } /* Retry multiple times in case file is being moved */ while (tries-- && rc == -EAGAIN) - rc = mdt_path_current(info, obj, fp); + rc = mdt_path_current(info, obj, fp, root_fid); RETURN(rc); } @@ -5547,6 +5707,7 @@ static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj, * \retval negative errno if there was a problem */ static int mdt_fid2path(struct mdt_thread_info *info, + struct lu_fid *root_fid, struct getinfo_fid2path *fp) { struct mdt_device *mdt = info->mti_mdt; @@ -5554,24 +5715,25 @@ static int mdt_fid2path(struct mdt_thread_info *info, int rc; ENTRY; - CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n", + CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n", PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno); if (!fid_is_sane(&fp->gf_fid)) RETURN(-EINVAL); if (!fid_is_namespace_visible(&fp->gf_fid)) { - CWARN("%s: "DFID" is invalid, sequence should be " - ">= "LPX64"\n", mdt_obd_name(mdt), - PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL); + CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx" + ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt), + PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL); RETURN(-EINVAL); } 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)) @@ -5588,20 +5750,22 @@ static int mdt_fid2path(struct mdt_thread_info *info, RETURN(rc); } - rc = mdt_path(info, obj, fp); + rc = mdt_path(info, obj, fp, root_fid); - 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); + CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n", + PFID(&fp->gf_fid), fp->gf_u.gf_path, + fp->gf_recno, fp->gf_linkno); mdt_object_put(info->mti_env, obj); RETURN(rc); } -static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, +static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen, void *val, int vallen) { struct getinfo_fid2path *fpout, *fpin; + struct lu_fid *root_fid = NULL; int rc = 0; fpin = key + cfs_size_round(sizeof(KEY_FID2PATH)); @@ -5614,7 +5778,18 @@ static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, if (fpout->gf_pathlen != vallen - sizeof(*fpin)) RETURN(-EINVAL); - rc = mdt_fid2path(info, fpout); + if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) + + sizeof(struct lu_fid)) { + /* client sent its root FID, which is normally fileset FID */ + root_fid = fpin->gf_u.gf_root_fid; + if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) + lustre_swab_lu_fid(root_fid); + + if (root_fid != NULL && !fid_is_sane(root_fid)) + RETURN(-EINVAL); + } + + rc = mdt_fid2path(info, root_fid, fpout); RETURN(rc); } @@ -5659,7 +5834,7 @@ int mdt_get_info(struct tgt_session_info *tsi) if (KEY_IS(KEY_FID2PATH)) { struct mdt_thread_info *info = tsi2mdt_info(tsi); - rc = mdt_rpc_fid2path(info, key, valout, *vallen); + rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen); mdt_thread_info_fini(info); } else { rc = -EINVAL; @@ -5740,7 +5915,7 @@ static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg) /* ioctls on obd dev */ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, - void *karg, void *uarg) + void *karg, void __user *uarg) { struct lu_env env; struct obd_device *obd = exp->exp_obd; @@ -5754,15 +5929,18 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, if (rc) RETURN(rc); - switch (cmd) { - case OBD_IOC_SYNC: - rc = mdt_device_sync(&env, mdt); - break; - case OBD_IOC_SET_READONLY: - rc = dt->dd_ops->dt_ro(&env, dt); - break; + switch (cmd) { + case OBD_IOC_SYNC: + rc = mdt_device_sync(&env, mdt); + break; + case OBD_IOC_SET_READONLY: + rc = dt_sync(&env, dt); + if (rc == 0) + rc = dt_ro(&env, dt); + break; case OBD_IOC_ABORT_RECOVERY: CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt)); + obd->obd_abort_recovery = 1; target_stop_recovery_thread(obd); rc = 0; break; @@ -5802,6 +5980,14 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop); break; } + case OBD_IOC_QUERY_LFSCK: { + struct md_device *next = mdt->mdt_child; + struct obd_ioctl_data *data = karg; + + rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, + data->ioc_inlbuf1); + break; + } case OBD_IOC_GET_OBJ_VERSION: { struct mdt_thread_info *mti; mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key); @@ -5835,18 +6021,21 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) { struct lu_device *ld = md2lu_dev(mdt->mdt_child); - struct lfsck_start_param lsp; int rc; ENTRY; - 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); + if (!mdt->mdt_skip_lfsck) { + struct lfsck_start_param lsp; + + 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); @@ -6000,7 +6189,7 @@ static struct lu_device_type mdt_device_type = { .ldt_ctx_tags = LCT_MD_THREAD }; -static int __init mdt_mod_init(void) +static int __init mdt_init(void) { int rc; @@ -6029,15 +6218,17 @@ mds_fini: return rc; } -static void __exit mdt_mod_exit(void) +static void __exit mdt_exit(void) { class_unregister_type(LUSTRE_MDT_NAME); mds_mod_exit(); lu_kmem_fini(mdt_caches); } -MODULE_AUTHOR("Sun Microsystems, Inc. "); +MODULE_AUTHOR("OpenSFS, Inc. "); MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit); +module_init(mdt_init); +module_exit(mdt_exit);