X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fofd%2Fofd_objects.c;h=7f328eeeb667113a5bfb5cf4f912b7c738b9f8f6;hb=c26a4d72445c7ce02e370cc84932d7cc89416966;hp=8ff3c0de4093e83c3717f7a56dfdf9ed5b23d5c1;hpb=a0a847fee6e30542bffb4e6860499b68cf515998;p=fs%2Flustre-release.git diff --git a/lustre/ofd/ofd_objects.c b/lustre/ofd/ofd_objects.c index 8ff3c0d..7f328ee 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) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2012, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -35,23 +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) @@ -59,17 +76,29 @@ int ofd_version_get_check(struct ofd_thread_info *info, /* VBR: version is checked always because costs nothing */ if (info->fti_pre_version != 0 && info->fti_pre_version != curr_version) { - CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n", + CDEBUG(D_INODE, "Version mismatch %#llx != %#llx\n", info->fti_pre_version, curr_version); - cfs_spin_lock(&info->fti_exp->exp_lock); + spin_lock(&info->fti_exp->exp_lock); info->fti_exp->exp_vbr_failed = 1; - cfs_spin_unlock(&info->fti_exp->exp_lock); + spin_unlock(&info->fti_exp->exp_lock); RETURN (-EOVERFLOW); } info->fti_pre_version = curr_version; 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) @@ -83,93 +112,111 @@ struct ofd_object *ofd_object_find(const struct lu_env *env, if (likely(!IS_ERR(o))) fo = ofd_obj(o); else - fo = (struct ofd_object *)o; /* return error */ + fo = ERR_CAST(o); /* return error */ + 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_ff. + * + * 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; - - ENTRY; - - info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG); - - dto = dt_find_or_create(env, ofd->ofd_osd, fid, &info->fti_dof, attr); - if (IS_ERR(dto)) - RETURN((struct ofd_object *)dto); + struct ofd_thread_info *info = ofd_info(env); + struct filter_fid *ff = &fo->ofo_ff; + struct lu_buf *buf = &info->fti_buf; + int rc = 0; - fo_obj = lu_object_locate(dto->do_lu.lo_header, - ofd->ofd_dt_dev.dd_lu_dev.ld_type); - RETURN(ofd_obj(fo_obj)); -} + if (fid_is_sane(&ff->ff_parent)) + return 0; -int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo) -{ - struct ofd_thread_info *info = ofd_info(env); - int rc = 0; + 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; - ENTRY; + if (unlikely(rc < sizeof(struct lu_fid))) { + fid_zero(&ff->ff_parent); - if (!fo->ofo_ff_exists) { - /* - * This actually means that we don't know whether the object - * has the "fid" EA or not. - */ - info->fti_buf.lb_buf = &info->fti_mds_fid2; - info->fti_buf.lb_len = sizeof(info->fti_mds_fid2); - rc = dt_xattr_get(env, ofd_object_child(fo), &info->fti_buf, - 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 -ENODATA; } - RETURN(rc); -} -void ofd_object_put(const struct lu_env *env, struct ofd_object *fo) -{ - lu_object_put(env, &fo->ofo_obj.do_lu); + filter_fid_le_to_cpu(ff, ff, rc); + + return 0; } +/** + * 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, obd_seq seq, int nr) + u64 id, struct ofd_seq *oseq, int nr, int sync) { struct ofd_thread_info *info = ofd_info(env); struct ofd_object *fo = NULL; struct dt_object *next; struct thandle *th; struct ofd_object **batch; - obd_id tmp; - int rc; - int i; - int objects = 0; - int nr_saved = nr; + struct lu_fid *fid = &info->fti_fid; + u64 tmp; + int rc; + int rc2; + int i; + int objects = 0; + int nr_saved = nr; ENTRY; /* Don't create objects beyond the valid range for this SEQ */ - if (unlikely(fid_seq_is_mdt0(seq) && (id + nr) >= IDIF_MAX_OID)) { - CERROR("%s:"POSTID" hit the IDIF_MAX_OID (1<<48)!\n", - ofd_name(ofd), id, seq); + if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) && + (id + nr) >= IDIF_MAX_OID)) { + CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n", + ofd_name(ofd), id, ostid_seq(&oseq->os_oi)); RETURN(rc = -ENOSPC); - } else if (unlikely(!fid_seq_is_mdt0(seq) && - (id + nr) >= OBIF_MAX_OID)) { - CERROR("%s:"POSTID" hit the OBIF_MAX_OID (1<<32)!\n", - ofd_name(ofd), id, seq); + } else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) && + (id + nr) >= OBIF_MAX_OID)) { + CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n", + ofd_name(ofd), id, ostid_seq(&oseq->os_oi)); RETURN(rc = -ENOSPC); } @@ -178,37 +225,29 @@ 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; info->fti_attr.la_ctime = 0; + LASSERT(id != 0); + /* prepare objects */ + *fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu); for (i = 0; i < nr; i++) { - info->fti_ostid.oi_id = id + i; - info->fti_ostid.oi_seq = seq; - - rc = fid_ostid_unpack(&info->fti_fid, &info->fti_ostid, 0); - if (rc) { + rc = fid_set_id(fid, id + i); + if (rc != 0) { if (i == 0) - GOTO(out, rc = PTR_ERR(fo)); + GOTO(out, rc); nr = i; break; } - fo = ofd_object_find(env, ofd, &info->fti_fid); + fo = ofd_object_find(env, ofd, fid); if (IS_ERR(fo)) { if (i == 0) GOTO(out, rc = PTR_ERR(fo)); @@ -228,8 +267,10 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, if (IS_ERR(th)) GOTO(out, rc = PTR_ERR(th)); - rc = dt_declare_record_write(env, ofd->ofd_lastid_obj[seq], - sizeof(tmp), info->fti_off, th); + th->th_sync |= sync; + + rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf, + info->fti_off, th); if (rc) GOTO(trans_stop, rc); @@ -239,8 +280,9 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, if (unlikely(ofd_object_exists(fo))) { /* object may exist being re-created by write replay */ - CDEBUG(D_INODE, "object "LPD64"/"LPD64" exists: " - DFID"\n", seq, id, PFID(&info->fti_fid)); + CDEBUG(D_INODE, "object %#llx/%#llx exists: " + DFID"\n", ostid_seq(&oseq->os_oi), id, + PFID(lu_object_fid(&fo->ofo_obj.do_lu))); continue; } @@ -249,7 +291,10 @@ 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; } @@ -259,33 +304,90 @@ int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd, if (rc) GOTO(trans_stop, rc); - CDEBUG(D_OTHER, "create new object "DFID"\n", PFID(&info->fti_fid)); + CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n", + ofd_name(ofd), PFID(fid), nr); + + /* 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 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. + * So from the LFSCK view, it needs to re-load the last_id from disk + * file, and if the latest last_id is still smaller than the object's + * ID, then the LAST_ID file is real crashed. + * + * To make above mechanism to work, before OFD pre-create OST-objects, + * it needs to update the LAST_ID file firstly, otherwise, the LFSCK + * may cannot get latest last_id although new OST-object created. */ + if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) { + tmp = cpu_to_le64(id + nr - 1); + dt_write_lock(env, oseq->os_lastid_obj, 0); + rc = dt_record_write(env, oseq->os_lastid_obj, + &info->fti_buf, &info->fti_off, th); + dt_write_unlock(env, oseq->os_lastid_obj); + if (rc != 0) + GOTO(trans_stop, rc); + } for (i = 0; i < nr; i++) { 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_req_local *lrl = &ofd_info(env)->fti_lrl; + + lfsck_pack_rfa(lrl, lu_object_fid(&fo->ofo_obj.do_lu), + LEL_FID_ACCESSED, LFSCK_TYPE_LAYOUT); + lfsck_in_notify_local(env, ofd->ofd_osd, lrl, 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_last_id_set(ofd, id + i, seq); + ofd_seq_last_oid_set(oseq, id + i); } objects = i; - if (objects > 0) { - tmp = cpu_to_le64(ofd_last_id(ofd, seq)); - rc = dt_record_write(env, ofd->ofd_lastid_obj[seq], - &info->fti_buf, &info->fti_off, th); + /* NOT all the wanted objects have been created, + * set the LAST_ID as the real created. */ + if (unlikely(objects < nr)) { + int rc1; + + info->fti_off = 0; + tmp = cpu_to_le64(ofd_seq_last_oid(oseq)); + dt_write_lock(env, oseq->os_lastid_obj, 0); + rc1 = dt_record_write(env, oseq->os_lastid_obj, + &info->fti_buf, &info->fti_off, th); + dt_write_unlock(env, oseq->os_lastid_obj); + if (rc1 != 0) + CERROR("%s: fail to reset the LAST_ID for seq (%#llx" + ") from %llu to %llu\n", ofd_name(ofd), + ostid_seq(&oseq->os_oi), id + nr - 1, + ofd_seq_last_oid(oseq)); } + trans_stop: - ofd_trans_stop(env, ofd, th, rc); + rc2 = ofd_trans_stop(env, ofd, th, rc); + if (rc2) + CERROR("%s: failed to stop transaction: rc = %d\n", + ofd_name(ofd), rc2); + if (!rc) + rc = rc2; out: for (i = 0; i < nr_saved; i++) { fo = batch[i]; @@ -303,13 +405,24 @@ 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 - * ownership of this object. We only allow this to happen once (so clear these - * bits) and later only allow setattr. +/** + * 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/PROJID 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, +int ofd_attr_handle_id(const struct lu_env *env, struct ofd_object *fo, struct lu_attr *la, int is_setattr) { struct ofd_thread_info *info = ofd_info(env); @@ -319,22 +432,29 @@ int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo, ENTRY; - if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID)) + if (!(la->la_valid & LA_UID) && !(la->la_valid & LA_GID) && + !(la->la_valid & LA_PROJID)) 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); LASSERT(ln->la_valid & LA_MODE); + /* + * Only allow setattr to change UID/GID/PROJID, if + * SUID+SGID is not set which means this is not + * initialization of this objects. + */ if (!is_setattr) { if (!(ln->la_mode & S_ISUID)) - la->la_valid &= ~LA_UID; + la->la_valid &= ~(LA_UID | LA_PROJID); if (!(ln->la_mode & S_ISGID)) - la->la_valid &= ~LA_GID; + la->la_valid &= ~(LA_GID | LA_PROJID); } + /* Initialize ownership of this object, clear SUID+SGID bits*/ if ((la->la_valid & LA_UID) && (ln->la_mode & S_ISUID)) mask |= S_ISUID; if ((la->la_valid & LA_GID) && (ln->la_mode & S_ISGID)) @@ -350,6 +470,22 @@ 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) { @@ -357,8 +493,9 @@ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo, struct ofd_device *ofd = ofd_obj2dev(fo); struct thandle *th; struct ofd_mod_data *fmd; - int ff_needed = 0; - int rc; + int ff_needed = 0; + int rc; + int rc2; ENTRY; ofd_write_lock(env, fo); @@ -377,12 +514,12 @@ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo, if (rc) GOTO(unlock, rc); - rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */); + rc = ofd_attr_handle_id(env, fo, la, 1 /* is_setattr */); if (rc != 0) 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) @@ -398,6 +535,11 @@ int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo, GOTO(stop, rc); if (ff_needed) { + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1)) + ff->ff_parent.f_oid = cpu_to_le32(1UL << 31); + else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2)) + le32_add_cpu(&ff->ff_parent.f_oid, -1); + info->fti_buf.lb_buf = ff; info->fti_buf.lb_len = sizeof(*ff); rc = dt_declare_xattr_set(env, ofd_object_child(fo), @@ -411,59 +553,99 @@ 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) { + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID)) + GOTO(stop, rc); + 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) + filter_fid_le_to_cpu(&fo->ofo_ff, ff, sizeof(*ff)); + } + + GOTO(stop, rc); stop: - ofd_trans_stop(env, ofd, th, rc); + rc2 = ofd_trans_stop(env, ofd, th, rc); + if (rc2) + CERROR("%s: failed to stop transaction: rc = %d\n", + ofd_name(ofd), rc2); + if (!rc) + rc = rc2; + 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); struct ofd_mod_data *fmd; struct dt_object *dob = ofd_object_child(fo); struct thandle *th; - int ff_needed = 0; - int rc; + int ff_needed = 0; + int rc; + int rc2; ENTRY; /* 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) GOTO(unlock, rc); - rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */); + rc = ofd_attr_handle_id(env, fo, la, 0 /* !is_setattr */); if (rc != 0) 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) @@ -483,6 +665,11 @@ int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo, GOTO(stop, rc); if (ff_needed) { + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1)) + ff->ff_parent.f_oid = cpu_to_le32(1UL << 31); + else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2)) + le32_add_cpu(&ff->ff_parent.f_oid, -1); + info->fti_buf.lb_buf = ff; info->fti_buf.lb_len = sizeof(*ff); rc = dt_declare_xattr_set(env, ofd_object_child(fo), @@ -496,32 +683,60 @@ 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) { + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID)) + GOTO(stop, rc); + 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) + filter_fid_le_to_cpu(&fo->ofo_ff, ff, sizeof(*ff)); + } + + GOTO(stop, rc); stop: - ofd_trans_stop(env, ofd, th, rc); + rc2 = ofd_trans_stop(env, ofd, th, rc); + if (rc2 != 0) + CERROR("%s: failed to stop transaction: rc = %d\n", + ofd_name(ofd), rc2); + if (!rc) + rc = rc2; unlock: ofd_write_unlock(env, fo); - RETURN(rc); + + return rc; } -int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo, +/** + * 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_destroy(const struct lu_env *env, struct ofd_object *fo, int orphan) { struct ofd_device *ofd = ofd_obj2dev(fo); struct thandle *th; - int rc = 0; + int rc = 0; + int rc2; ENTRY; @@ -533,8 +748,14 @@ 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 @@ -547,12 +768,30 @@ int ofd_object_destroy(const struct lu_env *env, struct ofd_object *fo, dt_ref_del(env, ofd_object_child(fo), th); dt_destroy(env, ofd_object_child(fo), th); stop: - ofd_trans_stop(env, ofd, th, rc); + rc2 = ofd_trans_stop(env, ofd, th, rc); + if (rc2) + CERROR("%s failed to stop transaction: %d\n", + ofd_name(ofd), rc2); + if (!rc) + rc = rc2; unlock: ofd_write_unlock(env, fo); 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) { @@ -561,25 +800,7 @@ 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)); - -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 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. - * This fixes up any bad values stored on disk before - * returning them to the client, and ensures any timestamp - * updates are correct. LU-1042 */ - if (unlikely(la->la_atime == LU221_BAD_TIME)) - la->la_atime = 0; - if (unlikely(la->la_mtime == LU221_BAD_TIME)) - 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 + rc = dt_attr_get(env, ofd_object_child(fo), la); } else { rc = -ENOENT; }