X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_objects.c;h=888186c7ec7c94f2b2d079e77e0c3bf64d849638;hb=26b8238659974959780cd49de92595b4b0bdf89f;hp=cd8c26a25db5be7e7dbd6aa0608729b50ba7e9cb;hpb=e1245db78d4de34d5e3236edbfb5b2b0eec638a0;p=fs%2Flustre-release.git diff --git a/lustre/ofd/ofd_objects.c b/lustre/ofd/ofd_objects.c index cd8c26a..888186c 100644 --- a/lustre/ofd/ofd_objects.c +++ b/lustre/ofd/ofd_objects.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) 2009, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2013, Intel Corporation. + * Copyright (c) 2012, 2014 Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -35,24 +31,44 @@ * * lustre/ofd/ofd_objects.c * - * Author: Alex Zhuravlev - * Author: Mikhail Pershin + * This file contains OSD API methods related to OBD Filter Device (OFD) + * object operations. + * + * Author: Alex Zhuravlev + * Author: Mikhail Pershin */ #define DEBUG_SUBSYSTEM S_FILTER #include #include +#include #include "ofd_internal.h" -int ofd_version_get_check(struct ofd_thread_info *info, - struct ofd_object *fo) +/** + * Get object version from disk and check it. + * + * This function checks object version from disk with + * ofd_thread_info::fti_pre_version filled from incoming RPC. This is part of + * VBR (Version-Based Recovery) and ensures that object has the same version + * upon replay as it has during original modification. + * + * \param[in] info execution thread OFD private data + * \param[in] fo OFD object + * + * \retval 0 if version matches + * \retval -EOVERFLOW on version mismatch + */ +static int ofd_version_get_check(struct ofd_thread_info *info, + struct ofd_object *fo) { dt_obj_version_t curr_version; LASSERT(ofd_object_exists(fo)); - LASSERT(info->fti_exp); + + if (info->fti_exp == NULL) + RETURN(0); curr_version = dt_version_get(info->fti_env, ofd_object_child(fo)); if ((__s64)curr_version == -EOPNOTSUPP) @@ -71,6 +87,18 @@ int ofd_version_get_check(struct ofd_thread_info *info, RETURN(0); } +/** + * Get OFD object by FID. + * + * This function finds OFD slice of compound object with the given FID. + * + * \param[in] env execution environment + * \param[in] ofd OFD device + * \param[in] fid FID of the object + * + * \retval pointer to the found ofd_object + * \retval ERR_PTR(errno) in case of error + */ struct ofd_object *ofd_object_find(const struct lu_env *env, struct ofd_device *ofd, const struct lu_fid *fid) @@ -89,63 +117,94 @@ struct ofd_object *ofd_object_find(const struct lu_env *env, RETURN(fo); } -struct ofd_object *ofd_object_find_or_create(const struct lu_env *env, - struct ofd_device *ofd, - const struct lu_fid *fid, - struct lu_attr *attr) +/** + * Get FID of parent MDT object. + * + * This function reads extended attribute XATTR_NAME_FID of OFD object which + * contains the MDT parent object FID and saves it in ofd_object::ofo_pfid. + * + * The filter_fid::ff_parent::f_ver field currently holds + * the OST-object index in the parent MDT-object's layout EA, + * not the actual FID::f_ver of the parent. We therefore access + * it via the macro f_stripe_idx. + * + * \param[in] env execution environment + * \param[in] fo OFD object + * + * \retval 0 if successful + * \retval -ENODATA if there is no such xattr + * \retval negative value on error + */ +int ofd_object_ff_load(const struct lu_env *env, struct ofd_object *fo) { struct ofd_thread_info *info = ofd_info(env); - struct lu_object *fo_obj; - struct dt_object *dto; + struct filter_fid_old *ff = &info->fti_mds_fid_old; + struct lu_buf *buf = &info->fti_buf; + struct lu_fid *pfid = &fo->ofo_pfid; + int rc = 0; - ENTRY; + if (fid_is_sane(pfid)) + return 0; - info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG); + buf->lb_buf = ff; + buf->lb_len = sizeof(*ff); + rc = dt_xattr_get(env, ofd_object_child(fo), buf, XATTR_NAME_FID); + if (rc < 0) + return rc; - dto = dt_find_or_create(env, ofd->ofd_osd, fid, &info->fti_dof, attr); - if (IS_ERR(dto)) - RETURN(ERR_CAST(dto)); + if (rc < sizeof(struct lu_fid)) { + fid_zero(pfid); - fo_obj = lu_object_locate(dto->do_lu.lo_header, - ofd->ofd_dt_dev.dd_lu_dev.ld_type); - RETURN(ofd_obj(fo_obj)); -} - -int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo) -{ - int rc = 0; + return -ENODATA; + } - ENTRY; + pfid->f_seq = le64_to_cpu(ff->ff_parent.f_seq); + pfid->f_oid = le32_to_cpu(ff->ff_parent.f_oid); + pfid->f_stripe_idx = le32_to_cpu(ff->ff_parent.f_stripe_idx); - if (!fo->ofo_ff_exists) { - /* - * This actually means that we don't know whether the object - * has the "fid" EA or not. - */ - rc = dt_xattr_get(env, ofd_object_child(fo), &LU_BUF_NULL, - XATTR_NAME_FID, BYPASS_CAPA); - if (rc >= 0 || rc == -ENODATA) { - /* - * Here we assume that, if the object doesn't have the - * "fid" EA, the caller will add one, unless a fatal - * error (e.g., a memory or disk failure) prevents it - * from doing so. - */ - fo->ofo_ff_exists = 1; - } - if (rc > 0) - rc = 0; - } - RETURN(rc); + return 0; } +/** + * Put OFD object reference. + * + * \param[in] env execution environment + * \param[in] fo OFD object + */ void ofd_object_put(const struct lu_env *env, struct ofd_object *fo) { lu_object_put(env, &fo->ofo_obj.do_lu); } +/** + * Precreate the given number \a nr of objects in the given sequence \a oseq. + * + * This function precreates new OST objects in the given sequence. + * The precreation starts from \a id and creates \a nr objects sequentially. + * + * Notes: + * This function may create fewer objects than requested. + * + * We mark object SUID+SGID to flag it for accepting UID+GID from client on + * first write. Currently the permission bits on the OST are never used, + * so this is OK. + * + * Initialize a/c/m time so any client timestamp will always be newer and + * update the inode. The ctime = 0 case is also handled specially in + * osd_inode_setattr(). See LU-221, LU-1042 for details. + * + * \param[in] env execution environment + * \param[in] ofd OFD device + * \param[in] id object ID to start precreation from + * \param[in] oseq object sequence + * \param[in] nr number of objects to precreate + * \param[in] sync synchronous precreation flag + * + * \retval 0 if successful + * \retval negative value on error + */ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, - obd_id id, struct ofd_seq *oseq, int nr, int sync) + u64 id, struct ofd_seq *oseq, int nr, int sync) { struct ofd_thread_info *info = ofd_info(env); struct ofd_object *fo = NULL; @@ -153,7 +212,7 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, struct thandle *th; struct ofd_object **batch; struct lu_fid *fid = &info->fti_fid; - obd_id tmp; + u64 tmp; int rc; int i; int objects = 0; @@ -179,17 +238,9 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, RETURN(-ENOMEM); info->fti_attr.la_valid = LA_TYPE | LA_MODE; - /* - * We mark object SUID+SGID to flag it for accepting UID+GID from - * client on first write. Currently the permission bits on the OST are - * never used, so this is OK. - */ info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666; info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG); - /* Initialize a/c/m time so any client timestamp will always - * be newer and update the inode. ctime = 0 is also handled - * specially in osd_inode_setattr(). See LU-221, LU-1042 */ info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME; info->fti_attr.la_atime = 0; info->fti_attr.la_mtime = 0; @@ -231,7 +282,7 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, th->th_sync |= sync; - rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp), + rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf, info->fti_off, th); if (rc) GOTO(trans_stop, rc); @@ -253,7 +304,10 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, rc = dt_declare_create(env, next, &info->fti_attr, NULL, &info->fti_dof, th); - if (rc) { + if (rc < 0) { + if (i == 0) + GOTO(trans_stop, rc); + nr = i; break; } @@ -266,11 +320,9 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n", ofd_name(ofd), PFID(fid), nr); - LASSERT(nr > 0); - /* When the LFSCK scanning the whole device to verify the LAST_ID file * consistency, it will load the last_id into RAM firstly, and compare - * the last_id with echo OST-object's ID. If the later one is larger, + * the last_id with each OST-object's ID. If the later one is larger, * then it will regard the LAST_ID file crashed. But during the LFSCK * scanning, the OFD may continue to create new OST-objects. Those new * created OST-objects will have larger IDs than the LFSCK known ones. @@ -295,14 +347,30 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, fo = batch[i]; LASSERT(fo); - if (likely(!ofd_object_exists(fo))) { + /* Only the new created objects need to be recorded. */ + if (ofd->ofd_osd->dd_record_fid_accessed) { + struct lfsck_request *lr = &ofd_info(env)->fti_lr; + + lfsck_pack_rfa(lr, lu_object_fid(&fo->ofo_obj.do_lu), + LE_FID_ACCESSED, + LFSCK_TYPE_LAYOUT); + lfsck_in_notify(env, ofd->ofd_osd, lr, NULL); + } + + if (likely(!ofd_object_exists(fo) && + !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) { next = ofd_object_child(fo); LASSERT(next != NULL); rc = dt_create(env, next, &info->fti_attr, NULL, &info->fti_dof, th); - if (rc) + if (rc < 0) { + if (i == 0) + GOTO(trans_stop, rc); + + rc = 0; break; + } LASSERT(ofd_object_exists(fo)); } ofd_seq_last_oid_set(oseq, id + i); @@ -346,11 +414,22 @@ out: RETURN(objects > 0 ? objects : rc); } -/* - * If the object still has SUID+SGID bits set (see ofd_precreate_object()) then - * we will accept the UID+GID if sent by the client for initializing the +/** + * Fix the OFD object ownership. + * + * If the object still has SUID+SGID bits set, meaning that it was precreated + * by the MDT before it was assigned to any file, (see ofd_precreate_objects()) + * then we will accept the UID+GID if sent by the client for initializing the * ownership of this object. We only allow this to happen once (so clear these * bits) and later only allow setattr. + * + * \param[in] env execution environment + * \param[in] fo OFD object + * \param[in] la object attributes + * \param[in] is_setattr was this function called from setattr or not + * + * \retval 0 if successful + * \retval negative value on error */ int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo, struct lu_attr *la, int is_setattr) @@ -365,7 +444,7 @@ int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo, if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID)) RETURN(0); - rc = dt_attr_get(env, ofd_object_child(fo), ln, BYPASS_CAPA); + rc = dt_attr_get(env, ofd_object_child(fo), ln); if (rc != 0) RETURN(rc); @@ -393,6 +472,22 @@ int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo, RETURN(0); } +/** + * Set OFD object attributes. + * + * This function sets OFD object attributes taken from incoming request. + * It sets not only regular attributes but also XATTR_NAME_FID extended + * attribute if needed. The "fid" xattr allows the object's MDT parent inode + * to be found and verified by LFSCK and other tools in case of inconsistency. + * + * \param[in] env execution environment + * \param[in] fo OFD object + * \param[in] la object attributes + * \param[in] ff filter_fid structure, contains additional attributes + * + * \retval 0 if successful + * \retval negative value on error + */ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo, struct lu_attr *la, struct filter_fid *ff) { @@ -425,7 +520,7 @@ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo, GOTO(unlock, rc); if (ff != NULL) { - rc = ofd_object_ff_check(env, fo); + rc = ofd_object_ff_load(env, fo); if (rc == -ENODATA) ff_needed = 1; else if (rc < 0) @@ -454,25 +549,57 @@ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo, if (rc) GOTO(stop, rc); - rc = dt_attr_set(env, ofd_object_child(fo), la, th, - ofd_object_capa(env, fo)); + rc = dt_attr_set(env, ofd_object_child(fo), la, th); if (rc) GOTO(stop, rc); - if (ff_needed) + if (ff_needed) { rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf, - XATTR_NAME_FID, 0, th, BYPASS_CAPA); + XATTR_NAME_FID, 0, th); + if (rc == 0) { + fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq); + fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid); + /* Currently, the filter_fid::ff_parent::f_ver is not + * the real parent MDT-object's FID::f_ver, instead it + * is the OST-object index in its parent MDT-object's + * layout EA. */ + fo->ofo_pfid.f_stripe_idx = + le32_to_cpu(ff->ff_parent.f_stripe_idx); + } + } + + GOTO(stop, rc); stop: ofd_trans_stop(env, ofd, th, rc); unlock: ofd_write_unlock(env, fo); - RETURN(rc); + + return rc; } +/** + * Truncate/punch OFD object. + * + * This function frees all of the allocated object's space from the \a start + * offset to the \a end offset. For truncate() operations the \a end offset + * is OBD_OBJECT_EOF. The functionality to punch holes in an object via + * fallocate(FALLOC_FL_PUNCH_HOLE) is not yet implemented (see LU-3606). + * + * \param[in] env execution environment + * \param[in] fo OFD object + * \param[in] start start offset to punch from + * \param[in] end end of punch + * \param[in] la object attributes + * \param[in] ff filter_fid structure + * \param[in] oa obdo struct from incoming request + * + * \retval 0 if successful + * \retval negative value on error + */ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo, __u64 start, __u64 end, struct lu_attr *la, - struct filter_fid *ff) + struct filter_fid *ff, struct obdo *oa) { struct ofd_thread_info *info = ofd_info(env); struct ofd_device *ofd = ofd_obj2dev(fo); @@ -487,15 +614,21 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo, /* we support truncate, not punch yet */ LASSERT(end == OBD_OBJECT_EOF); + ofd_write_lock(env, fo); fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid); if (fmd && fmd->fmd_mactime_xid < info->fti_xid) fmd->fmd_mactime_xid = info->fti_xid; ofd_fmd_put(info->fti_exp, fmd); - ofd_write_lock(env, fo); if (!ofd_object_exists(fo)) GOTO(unlock, rc = -ENOENT); + if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) { + rc = ofd_verify_ff(env, fo, oa); + if (rc != 0) + GOTO(unlock, rc); + } + /* VBR: version recovery check */ rc = ofd_version_get_check(info, fo); if (rc) @@ -506,7 +639,7 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo, GOTO(unlock, rc); if (ff != NULL) { - rc = ofd_object_ff_check(env, fo); + rc = ofd_object_ff_load(env, fo); if (rc == -ENODATA) ff_needed = 1; else if (rc < 0) @@ -539,26 +672,53 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo, if (rc) GOTO(stop, rc); - rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th, - ofd_object_capa(env, fo)); + rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th); if (rc) GOTO(stop, rc); - rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo)); + rc = dt_attr_set(env, dob, la, th); if (rc) GOTO(stop, rc); - if (ff_needed) + if (ff_needed) { rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf, - XATTR_NAME_FID, 0, th, BYPASS_CAPA); + XATTR_NAME_FID, 0, th); + if (rc == 0) { + fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq); + fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid); + /* Currently, the filter_fid::ff_parent::f_ver is not + * the real parent MDT-object's FID::f_ver, instead it + * is the OST-object index in its parent MDT-object's + * layout EA. */ + fo->ofo_pfid.f_stripe_idx = + le32_to_cpu(ff->ff_parent.f_stripe_idx); + } + } + + GOTO(stop, rc); stop: ofd_trans_stop(env, ofd, th, rc); unlock: ofd_write_unlock(env, fo); - RETURN(rc); + + return rc; } +/** + * Destroy OFD object. + * + * This function destroys OFD object. If object wasn't used at all (orphan) + * then local transaction is used, which means the transaction data is not + * returned back in reply. + * + * \param[in] env execution environment + * \param[in] fo OFD object + * \param[in] orphan flag to indicate that object is orphaned + * + * \retval 0 if successful + * \retval negative value on error + */ int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo, int orphan) { @@ -576,8 +736,14 @@ int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo, if (IS_ERR(th)) GOTO(unlock, rc = PTR_ERR(th)); - dt_declare_ref_del(env, ofd_object_child(fo), th); - dt_declare_destroy(env, ofd_object_child(fo), th); + rc = dt_declare_ref_del(env, ofd_object_child(fo), th); + if (rc < 0) + GOTO(stop, rc); + + rc = dt_declare_destroy(env, ofd_object_child(fo), th); + if (rc < 0) + GOTO(stop, rc); + if (orphan) rc = dt_trans_start_local(env, ofd->ofd_osd, th); else @@ -596,6 +762,19 @@ unlock: RETURN(rc); } +/** + * Get OFD object attributes. + * + * This function gets OFD object regular attributes. It is used to serve + * incoming request as well as for local OFD purposes. + * + * \param[in] env execution environment + * \param[in] fo OFD object + * \param[in] la object attributes + * + * \retval 0 if successful + * \retval negative value on error + */ int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo, struct lu_attr *la) { @@ -604,10 +783,9 @@ int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo, ENTRY; if (ofd_object_exists(fo)) { - rc = dt_attr_get(env, ofd_object_child(fo), la, - ofd_object_capa(env, fo)); + rc = dt_attr_get(env, ofd_object_child(fo), la); -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) +#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) /* Try to correct for a bug in 2.1.0 (LU-221) that caused * negative timestamps to appear to be in the far future, * due old timestamp being stored on disk as an unsigned value. @@ -620,8 +798,6 @@ int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo, la->la_mtime = 0; if (unlikely(la->la_ctime == LU221_BAD_TIME)) la->la_ctime = 0; -#else -#warning "remove old LU-221/LU-1042 workaround code" #endif } else { rc = -ENOENT;