*
* 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
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2015, Intel Corporation.
+ * Copyright (c) 2011, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
static const char dotdot[] = "..";
static struct lu_name lname_dotdot = {
- (char *) dotdot,
- sizeof(dotdot) - 1
+ .ln_name = (char *) dotdot,
+ .ln_namelen = sizeof(dotdot) - 1,
};
static inline int
if (unlikely(mdd_is_dead_obj(mdd_obj)))
RETURN(-ESTALE);
+ if (!mdd_object_exists(mdd_obj))
+ RETURN(-ESTALE);
+
if (mdd_object_remote(mdd_obj)) {
CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n",
mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj)));
- } else if (!mdd_object_exists(mdd_obj)) {
- RETURN(-ESTALE);
}
rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask,
RETURN(rc);
}
+/** Read the link EA into a temp buffer.
+ * Uses the mdd_thread_info::mti_big_buf since it is generally large.
+ * A pointer to the buffer is stored in \a ldata::ld_buf.
+ *
+ * \retval 0 or error
+ */
+static int __mdd_links_read(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct linkea_data *ldata)
+{
+ int rc;
+
+ if (!mdd_object_exists(mdd_obj))
+ return -ENODATA;
+
+ /* First try a small buf */
+ LASSERT(env != NULL);
+ ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
+ PAGE_SIZE);
+ if (ldata->ld_buf->lb_buf == NULL)
+ return -ENOMEM;
+
+ rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
+ if (rc == -ERANGE) {
+ /* Buf was too small, figure out what we need. */
+ lu_buf_free(ldata->ld_buf);
+ rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
+ XATTR_NAME_LINK);
+ if (rc < 0)
+ return rc;
+ ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
+ if (ldata->ld_buf->lb_buf == NULL)
+ return -ENOMEM;
+ rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
+ XATTR_NAME_LINK);
+ }
+ if (rc < 0) {
+ lu_buf_free(ldata->ld_buf);
+ ldata->ld_buf = NULL;
+ return rc;
+ }
+
+ return linkea_init(ldata);
+}
+
+static int mdd_links_read(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct linkea_data *ldata)
+{
+ int rc;
+
+ rc = __mdd_links_read(env, mdd_obj, ldata);
+ if (!rc)
+ rc = linkea_init(ldata);
+
+ return rc;
+}
+
+static int mdd_links_read_with_rec(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct linkea_data *ldata)
+{
+ int rc;
+
+ rc = __mdd_links_read(env, mdd_obj, ldata);
+ if (!rc)
+ rc = linkea_init_with_rec(ldata);
+
+ return rc;
+}
+
/**
* Get parent FID of the directory
*
GOTO(lookup, rc = 0);
ldata.ld_buf = buf;
- rc = mdd_links_read(env, obj, &ldata);
+ rc = mdd_links_read_with_rec(env, obj, &ldata);
if (rc != 0)
GOTO(lookup, rc);
parent = mdd_object_find(env, mdd, pfid);
if (IS_ERR(parent))
GOTO(out, rc = PTR_ERR(parent));
+
+ if (!mdd_object_exists(parent))
+ GOTO(out, rc = -EINVAL);
+
p1 = parent;
}
EXIT;
const struct lu_attr *pattr, struct mdd_object *cobj,
bool check_perm)
{
- struct mdd_thread_info *info = mdd_env_info(env);
- struct lu_buf *xbuf;
int rc = 0;
ENTRY;
if (mdd_is_dead_obj(pobj))
RETURN(-ENOENT);
- /* If the parent is a sub-stripe, check whether it is dead */
- xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key));
- rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV);
- if (unlikely(rc > 0)) {
- struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf;
-
- if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE &&
- le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD)
- RETURN(-ESTALE);
- }
- rc = 0;
-
if (check_perm)
rc = mdd_permission_internal_locked(env, pobj, pattr,
MAY_WRITE | MAY_EXEC,
const struct lu_name *sname)
{
const struct lu_ucred *uc = lu_ucred(env);
- enum changelog_rec_flags crf = 0;
- size_t hdr_size = sizeof(struct llog_changelog_rec) -
- sizeof(struct changelog_rec);
+ enum changelog_rec_flags crf = CLF_EXTRA_FLAGS;
+ enum changelog_rec_extra_flags crfe = CLFE_UIDGID;
if (sname != NULL)
crf |= CLF_RENAME;
if (uc != NULL && uc->uc_jobid[0] != '\0')
crf |= CLF_JOBID;
- return llog_data_len(hdr_size + changelog_rec_offset(crf) +
+ return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
+ changelog_rec_offset(crf, crfe) +
(tname != NULL ? tname->ln_namelen : 0) +
(sname != NULL ? 1 + sname->ln_namelen : 0));
}
return rc;
}
+struct mdd_changelog_gc {
+ struct mdd_device *mcgc_mdd;
+ bool mcgc_found;
+ __u32 mcgc_maxtime;
+ __u64 mcgc_maxindexes;
+ __u32 mcgc_id;
+};
+
+/* return first registered ChangeLog user idle since too long
+ * use ChangeLog's user plain LLOG mtime for this */
+static int mdd_changelog_gc_cb(const struct lu_env *env,
+ struct llog_handle *llh,
+ struct llog_rec_hdr *hdr, void *data)
+{
+ struct llog_changelog_user_rec *rec;
+ struct mdd_changelog_gc *mcgc = (struct mdd_changelog_gc *)data;
+ struct mdd_device *mdd = mcgc->mcgc_mdd;
+ ENTRY;
+
+ if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0)
+ RETURN(-ENXIO);
+
+ rec = container_of(hdr, struct llog_changelog_user_rec,
+ cur_hdr);
+
+ /* find oldest idle user, based on last record update/cancel time (new
+ * behavior), or for old user records, last record index vs current
+ * ChangeLog index. Late users with old record format will be treated
+ * first as we assume they could be idle since longer
+ */
+ if (rec->cur_time != 0) {
+ __u32 time_now = (__u32)get_seconds();
+ __u32 time_out = rec->cur_time +
+ mdd->mdd_changelog_max_idle_time;
+ __u32 idle_time = time_now - rec->cur_time;
+
+ /* treat oldest idle user first, and if no old format user
+ * has been already selected
+ */
+ if (time_after32(time_now, time_out) &&
+ idle_time > mcgc->mcgc_maxtime &&
+ mcgc->mcgc_maxindexes == 0) {
+ mcgc->mcgc_maxtime = idle_time;
+ mcgc->mcgc_id = rec->cur_id;
+ mcgc->mcgc_found = true;
+ }
+ } else {
+ /* old user record with no idle time stamp, so use empirical
+ * method based on its current index/position
+ */
+ __u64 idle_indexes;
+
+ idle_indexes = mdd->mdd_cl.mc_index - rec->cur_endrec;
+
+ /* treat user with the oldest/smallest current index first */
+ if (idle_indexes >= mdd->mdd_changelog_max_idle_indexes &&
+ idle_indexes > mcgc->mcgc_maxindexes) {
+ mcgc->mcgc_maxindexes = idle_indexes;
+ mcgc->mcgc_id = rec->cur_id;
+ mcgc->mcgc_found = true;
+ }
+
+ }
+ RETURN(0);
+}
+
+/* recover space from long-term inactive ChangeLog users */
+static int mdd_chlg_garbage_collect(void *data)
+{
+ struct mdd_device *mdd = (struct mdd_device *)data;
+ struct lu_env *env = NULL;
+ int rc;
+ struct llog_ctxt *ctxt;
+ struct mdd_changelog_gc mcgc = {
+ .mcgc_mdd = mdd,
+ .mcgc_found = false,
+ .mcgc_maxtime = 0,
+ .mcgc_maxindexes = 0,
+ };
+ ENTRY;
+
+ CDEBUG(D_HA, "%s: ChangeLog garbage collect thread start\n",
+ mdd2obd_dev(mdd)->obd_name);
+
+ OBD_ALLOC_PTR(env);
+ if (env == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ rc = lu_env_init(env, LCT_MD_THREAD);
+ if (rc)
+ GOTO(out, rc);
+
+ for (;;) {
+ ctxt = llog_get_context(mdd2obd_dev(mdd),
+ LLOG_CHANGELOG_USER_ORIG_CTXT);
+ if (ctxt == NULL ||
+ (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0)
+ GOTO(out_env, rc = -ENXIO);
+
+ rc = llog_cat_process(env, ctxt->loc_handle,
+ mdd_changelog_gc_cb, &mcgc, 0, 0);
+ if (rc != 0 || mcgc.mcgc_found == false)
+ break;
+ llog_ctxt_put(ctxt);
+
+ CWARN("%s: Force deregister of ChangeLog user cl%d idle more "
+ "than %us\n", mdd2obd_dev(mdd)->obd_name, mcgc.mcgc_id,
+ mcgc.mcgc_maxtime);
+
+ mdd_changelog_user_purge(env, mdd, mcgc.mcgc_id);
+
+ /* try again to search for another candidate */
+ mcgc.mcgc_found = false;
+ mcgc.mcgc_maxtime = 0;
+ mcgc.mcgc_maxindexes = 0;
+ }
+
+out_env:
+ if (ctxt != NULL)
+ llog_ctxt_put(ctxt);
+
+ lu_env_fini(env);
+ GOTO(out, rc);
+out:
+ if (env)
+ OBD_FREE_PTR(env);
+ mdd->mdd_cl.mc_gc_task = NULL;
+ return rc;
+}
+
/** Add a changelog entry \a rec to the changelog llog
* \param mdd
* \param rec
struct llog_ctxt *ctxt;
struct thandle *llog_th;
int rc;
+ bool run_gc_task;
rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
changelog_rec_varsize(&rec->cr));
/* nested journal transaction */
rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
+ /* time to recover some space ?? */
+ spin_lock(&mdd->mdd_cl.mc_lock);
+ if (unlikely(mdd->mdd_changelog_gc && (ktime_get_real_seconds() -
+ mdd->mdd_cl.mc_gc_time > mdd->mdd_changelog_min_gc_interval) &&
+ mdd->mdd_cl.mc_gc_task == NULL &&
+ llog_cat_free_space(ctxt->loc_handle) <=
+ mdd->mdd_changelog_min_free_cat_entries)) {
+ CWARN("%s: low on changelog_catalog free entries, starting "
+ "ChangeLog garbage collection thread\n", obd->obd_name);
+
+ /* indicate further kthread run will occur outside right after
+ * critical section
+ */
+ mdd->mdd_cl.mc_gc_task = (struct task_struct *)(-1);
+ run_gc_task = true;
+ }
+ spin_unlock(&mdd->mdd_cl.mc_lock);
+ if (run_gc_task) {
+ struct task_struct *gc_task;
+
+ gc_task = kthread_run(mdd_chlg_garbage_collect, mdd,
+ "chlg_gc_thread");
+ if (IS_ERR(gc_task)) {
+ CERROR("%s: cannot start ChangeLog garbage collection "
+ "thread: rc = %ld\n", obd->obd_name,
+ PTR_ERR(gc_task));
+ mdd->mdd_cl.mc_gc_task = NULL;
+ } else {
+ CDEBUG(D_HA, "%s: ChangeLog garbage collection thread "
+ "has started with Pid %d\n", obd->obd_name,
+ gc_task->pid);
+ mdd->mdd_cl.mc_gc_task = gc_task;
+ mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
+ }
+ }
out_put:
llog_ctxt_put(ctxt);
if (rc > 0)
const struct lu_fid *spfid,
const struct lu_name *sname)
{
- struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
- size_t extsize = sname->ln_namelen + 1;
+ struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
+ size_t extsize = sname->ln_namelen + 1;
LASSERT(sfid != NULL);
LASSERT(spfid != NULL);
void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
{
- struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
+ struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
if (jobid == NULL || jobid[0] == '\0')
return;
strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
}
+void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
+{
+ struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
+
+ ef->cr_extra_flags = eflags;
+}
+
+void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
+ __u64 uid, __u64 gid)
+{
+ struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
+
+ uidgid->cr_uid = uid;
+ uidgid->cr_gid = gid;
+}
+
/** Store a namespace change changelog record
* If this fails, we must fail the whole transaction; we don't
* want the change to commit without the log entry.
struct llog_changelog_rec *rec;
struct lu_buf *buf;
int reclen;
+ __u64 xflags = CLFE_INVALID;
int rc;
ENTRY;
rec = buf->lb_buf;
crf &= CLF_FLAGMASK;
+ crf |= CLF_EXTRA_FLAGS;
if (uc != NULL && uc->uc_jobid[0] != '\0')
crf |= CLF_JOBID;
else
crf |= CLF_VERSION;
+ xflags |= CLFE_UIDGID;
+
rec->cr.cr_flags = crf;
+
+ if (crf & CLF_EXTRA_FLAGS) {
+ mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
+ if (xflags & CLFE_UIDGID)
+ mdd_changelog_rec_extra_uidgid(&rec->cr,
+ uc->uc_uid, uc->uc_gid);
+ }
+
rec->cr.cr_type = (__u32)type;
rec->cr.cr_pfid = *tpfid;
rec->cr.cr_namelen = tname->ln_namelen;
if (likely(target != NULL)) {
rec->cr.cr_tfid = *mdo2fid(target);
- target->mod_cltime = cfs_time_current_64();
+ target->mod_cltime = ktime_get();
} else {
fid_zero(&rec->cr.cr_tfid);
}
struct linkea_data *ldata)
{
int rc = 0;
- int rc2 = 0;
ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
- return 0;
+ RETURN(0);
LASSERT(oldpfid != NULL || newpfid != NULL);
- if (mdd_obj->mod_flags & DEAD_OBJ) {
- /* Prevent linkea to be updated which is NOT necessary. */
- ldata->ld_reclen = 0;
- /* No more links, don't bother */
+ if (mdd_obj->mod_flags & DEAD_OBJ)
+ /* Unnecessary to update linkEA for dead object. */
RETURN(0);
- }
if (oldpfid != NULL) {
rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
if (rc) {
- if ((check == 1) ||
- (rc != -ENODATA && rc != -ENOENT))
+ if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
RETURN(rc);
+
/* No changes done. */
rc = 0;
}
}
/* If renaming, add the new record */
- if (newpfid != NULL) {
- /* even if the add fails, we still delete the out-of-date
- * old link */
- rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
- first, check);
- }
-
- rc = rc != 0 ? rc : rc2;
+ if (newpfid != NULL)
+ rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
+ first, check);
RETURN(rc);
}
ldata = &mdd_env_info(env)->mti_link_data;
memset(ldata, 0, sizeof(*ldata));
rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
- newpfid, newlname, first, check,
- ldata);
- if (rc != 0)
+ newpfid, newlname, first, check, ldata);
+ if (rc)
GOTO(out, rc);
}
- if (ldata->ld_reclen != 0)
+ if (!(mdd_obj->mod_flags & DEAD_OBJ))
rc = mdd_links_write(env, mdd_obj, ldata, handle);
- EXIT;
+
+ GOTO(out, rc);
+
out:
if (rc != 0) {
- int error = 1;
- if (rc == -EOVERFLOW || rc == -ENOSPC)
- error = 0;
if (newlname == NULL)
- CDEBUG(error ? D_ERROR : D_OTHER,
- "link_ea add failed %d "DFID"\n",
+ CERROR("link_ea add failed %d "DFID"\n",
rc, PFID(mdd_object_fid(mdd_obj)));
else if (oldpfid == NULL)
- CDEBUG(error ? D_ERROR : D_OTHER,
- "link_ea add '%.*s' failed %d "DFID"\n",
- newlname->ln_namelen, newlname->ln_name,
- rc, PFID(mdd_object_fid(mdd_obj)));
+ CERROR("link_ea add '%.*s' failed %d "DFID"\n",
+ newlname->ln_namelen, newlname->ln_name, rc,
+ PFID(mdd_object_fid(mdd_obj)));
else if (newpfid == NULL)
- CDEBUG(error ? D_ERROR : D_OTHER,
- "link_ea del '%.*s' failed %d "DFID"\n",
- oldlname->ln_namelen, oldlname->ln_name,
- rc, PFID(mdd_object_fid(mdd_obj)));
+ CERROR("link_ea del '%.*s' failed %d "DFID"\n",
+ oldlname->ln_namelen, oldlname->ln_name, rc,
+ PFID(mdd_object_fid(mdd_obj)));
else
- CDEBUG(error ? D_ERROR : D_OTHER,
- "link_ea rename '%.*s'->'%.*s' failed %d "
- DFID"\n",
- oldlname->ln_namelen, oldlname->ln_name,
- newlname->ln_namelen, newlname->ln_name,
- rc, PFID(mdd_object_fid(mdd_obj)));
+ CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
+ "\n", oldlname->ln_namelen, oldlname->ln_name,
+ newlname->ln_namelen, newlname->ln_name, rc,
+ PFID(mdd_object_fid(mdd_obj)));
}
if (is_vmalloc_addr(ldata->ld_buf))
}
/** Read the link EA into a temp buffer.
- * Uses the mdd_thread_info::mti_big_buf since it is generally large.
- * A pointer to the buffer is stored in \a ldata::ld_buf.
- *
- * \retval 0 or error
- */
-int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj,
- struct linkea_data *ldata)
-{
- int rc;
-
- if (!mdd_object_exists(mdd_obj))
- return -ENODATA;
-
- /* First try a small buf */
- LASSERT(env != NULL);
- ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf,
- PAGE_CACHE_SIZE);
- if (ldata->ld_buf->lb_buf == NULL)
- return -ENOMEM;
-
- rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
- if (rc == -ERANGE) {
- /* Buf was too small, figure out what we need. */
- lu_buf_free(ldata->ld_buf);
- rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
- XATTR_NAME_LINK);
- if (rc < 0)
- return rc;
- ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
- if (ldata->ld_buf->lb_buf == NULL)
- return -ENOMEM;
- rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
- XATTR_NAME_LINK);
- }
- if (rc < 0) {
- lu_buf_free(ldata->ld_buf);
- ldata->ld_buf = NULL;
- return rc;
- }
-
- return linkea_init(ldata);
-}
-
-/** Read the link EA into a temp buffer.
* Uses the name_buf since it is generally large.
* \retval IS_ERR err
* \retval ptr to \a lu_buf (always \a mti_big_buf)
int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
struct linkea_data *ldata, struct thandle *handle)
{
- const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
- ldata->ld_leh->leh_len);
+ const struct lu_buf *buf;
int rc;
+ if (ldata == NULL || ldata->ld_buf == NULL ||
+ ldata->ld_leh == NULL)
+ return 0;
+
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
return 0;
+again:
+ buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
- if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) &&
- mdd_object_remote(mdd_obj) == 0) {
- struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
- struct thandle *sub_th;
-
- /* XXX: If the linkEA is overflow, then we need to notify the
- * namespace LFSCK to skip "nlink" attribute verification
- * on this object to avoid the "nlink" to be shrinked by
- * wrong. It may be not good an interaction with LFSCK
- * like this. We will consider to replace it with other
- * mechanism in future. LU-5802. */
- lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK,
- LFSCK_TYPE_NAMESPACE);
-
- sub_th = thandle_get_sub_by_dt(env, handle,
- mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom);
- lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
- lr, sub_th);
+ if (unlikely(rc == -ENOSPC)) {
+ rc = linkea_overflow_shrink(ldata);
+ if (likely(rc > 0))
+ goto again;
}
return rc;
}
-int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj,
- struct thandle *handle, struct linkea_data *ldata,
- enum mdd_links_add_overflow overflow)
+static int mdd_declare_links_add(const struct lu_env *env,
+ struct mdd_object *mdd_obj,
+ struct thandle *handle,
+ struct linkea_data *ldata)
{
int rc;
int ea_len;
ea_len = ldata->ld_leh->leh_len;
linkea = ldata->ld_buf->lb_buf;
} else {
- ea_len = DEFAULT_LINKEA_SIZE;
+ ea_len = MAX_LINKEA_SIZE;
linkea = NULL;
}
- /* XXX: max size? */
rc = mdo_declare_xattr_set(env, mdd_obj,
mdd_buf_get_const(env, linkea, ea_len),
XATTR_NAME_LINK, 0, handle);
- if (rc != 0)
- return rc;
-
- if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) {
- struct lfsck_request *lr = &mdd_env_info(env)->mti_lr;
- struct thandle *sub_th;
-
- /* XXX: If the linkEA is overflow, then we need to notify the
- * namespace LFSCK to skip "nlink" attribute verification
- * on this object to avoid the "nlink" to be shrinked by
- * wrong. It may be not good an interaction with LFSCK
- * like this. We will consider to replace it with other
- * mechanism in future. LU-5802. */
- lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE,
- LFSCK_TYPE_NAMESPACE);
-
- sub_th = thandle_get_sub_by_dt(env, handle,
- mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom);
- rc = lfsck_in_notify(env,
- mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom,
- lr, sub_th);
- }
return rc;
}
/* For directory, the linkEA will be removed together
* with the object. */
if (!S_ISDIR(mdd_object_type(c)))
- rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE);
+ rc = mdd_declare_links_add(env, c, handle, NULL);
return rc;
}
struct lu_attr *la,
struct linkea_data *data)
{
+ struct lu_fid tfid = *mdo2fid(c);
int rc;
- rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c),
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
+ tfid.f_oid = cfs_fail_val;
+
+ rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
name->ln_name, handle);
if (rc != 0)
return rc;
if (rc != 0)
return rc;
- if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
- rc = mdo_declare_ref_add(env, c, handle);
- if (rc != 0)
- return rc;
- }
-
la->la_valid = LA_CTIME | LA_MTIME;
rc = mdo_declare_attr_set(env, p, la, handle);
if (rc != 0)
if (rc != 0)
return rc;
- rc = mdd_declare_links_add(env, c, handle, data,
- S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE);
+ rc = mdd_declare_links_add(env, c, handle, data);
if (rc != 0)
return rc;
struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
- struct mdd_device *mdd = mdo2mdd(src_obj);
- struct thandle *handle;
+ struct mdd_device *mdd = mdo2mdd(src_obj);
+ struct thandle *handle;
+ struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2;
struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data;
int rc;
ENTRY;
if (rc != 0)
RETURN(rc);
+ /*
+ * If we are using project inheritance, we only allow hard link
+ * creation in our tree when the project IDs are the same;
+ * otherwise the tree quota mechanism could be circumvented.
+ */
+ if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
+ (tattr->la_projid != cattr->la_projid))
+ RETURN(-EXDEV);
+
handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
LASSERT(ma->ma_attr.la_valid & LA_CTIME);
la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
+ /* Note: even this function will change ldata, but it comes from
+ * thread_info, which is completely temporary and only seen in
+ * this function, so we do not need reset ldata once it fails.*/
+ rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj),
+ lname, 0, 0, ldata);
+ if (rc != 0)
+ GOTO(stop, rc);
+
rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
la, ldata);
if (rc)
GOTO(out_unlock, rc);
}
- if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) {
- rc = mdo_ref_add(env, mdd_sobj, handle);
- if (rc != 0)
- GOTO(out_unlock, rc);
- }
-
- if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) {
- struct lu_fid tfid = *mdo2fid(mdd_sobj);
-
- tfid.f_oid++;
- rc = __mdd_index_insert_only(env, mdd_tobj, &tfid,
- mdd_object_type(mdd_sobj),
- name, handle);
- } else {
- rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj),
- mdd_object_type(mdd_sobj),
- name, handle);
- }
+ *tfid = *mdo2fid(mdd_sobj);
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
+ tfid->f_oid = cfs_fail_val;
+ rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
+ mdd_object_type(mdd_sobj), name, handle);
if (rc != 0) {
mdo_ref_del(env, mdd_sobj, handle);
GOTO(out_unlock, rc);
la->la_valid = LA_CTIME;
rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
- if (rc == 0) {
- rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
- mdo2fid(mdd_tobj), lname, 0, 0,
- ldata);
- if (rc == 0)
- mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
- lname, handle, ldata, 0);
- /* The failure of links_add should not cause the link
- * failure, reset rc here */
- rc = 0;
- }
- EXIT;
+ if (rc == 0)
+ /* Note: The failure of links_add should not cause the
+ * link failure, so do not check return value. */
+ mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj),
+ lname, handle, ldata, 0);
+
+ EXIT;
out_unlock:
- mdd_write_unlock(env, mdd_sobj);
- if (rc == 0)
+ mdd_write_unlock(env, mdd_sobj);
+ if (rc == 0)
rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
mdo2fid(mdd_tobj), NULL, NULL,
lname, NULL, handle);
stop:
- mdd_trans_stop(env, mdd, rc, handle);
-
+ rc = mdd_trans_stop(env, mdd, rc, handle);
if (is_vmalloc_addr(ldata->ld_buf))
/* if we vmalloced a large buffer drop it */
lu_buf_free(ldata->ld_buf);
out_pending:
- return rc;
+ return rc;
}
-static int mdd_mark_dead_object(const struct lu_env *env,
+static int mdd_mark_orphan_object(const struct lu_env *env,
struct mdd_object *obj, struct thandle *handle,
bool declare)
{
struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
int rc;
- if (!declare)
- obj->mod_flags |= DEAD_OBJ;
-
if (!S_ISDIR(mdd_object_type(obj)))
return 0;
attr->la_valid = LA_FLAGS;
- attr->la_flags = LUSTRE_SLAVE_DEAD_FL;
+ attr->la_flags = LUSTRE_ORPHAN_FL;
if (declare)
rc = mdo_declare_attr_set(env, obj, attr, handle);
{
int rc;
- rc = mdd_mark_dead_object(env, obj, handle, true);
+ /* Sigh, we do not know if the unlink object will become orphan in
+ * declare phase, but fortunately the flags here does not matter
+ * in current declare implementation */
+ rc = mdd_mark_orphan_object(env, obj, handle, true);
if (rc != 0)
return rc;
- rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
+ rc = mdo_declare_destroy(env, obj, handle);
if (rc != 0)
return rc;
- rc = mdo_declare_destroy(env, obj, handle);
+ rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle);
if (rc != 0)
return rc;
struct thandle *th)
{
int rc = 0;
- int is_dir = S_ISDIR(ma->ma_attr.la_mode);
- ENTRY;
+ int is_dir = S_ISDIR(ma->ma_attr.la_mode);
+ ENTRY;
- LASSERT(mdd_write_locked(env, obj) != 0);
+ LASSERT(mdd_write_locked(env, obj) != 0);
if (ma->ma_attr.la_nlink == 0 || is_dir) {
- rc = mdd_mark_dead_object(env, obj, th, false);
- if (rc != 0)
- RETURN(rc);
-
- /* add new orphan and the object
- * will be deleted during mdd_close() */
- if (obj->mod_count) {
- rc = __mdd_orphan_add(env, obj, th);
- if (rc == 0)
- CDEBUG(D_HA, "Object "DFID" is inserted into "
- "orphan list, open count = %d\n",
- PFID(mdd_object_fid(obj)),
- obj->mod_count);
- else
- CERROR("Object "DFID" fail to be an orphan, "
- "open count = %d, maybe cause failed "
- "open replay\n",
- PFID(mdd_object_fid(obj)),
- obj->mod_count);
- } else {
+ /* add new orphan and the object
+ * will be deleted during mdd_close() */
+ obj->mod_flags |= DEAD_OBJ;
+ if (obj->mod_count) {
+ rc = __mdd_orphan_add(env, obj, th);
+ if (rc == 0)
+ CDEBUG(D_HA, "Object "DFID" is inserted into "
+ "orphan list, open count = %d\n",
+ PFID(mdd_object_fid(obj)),
+ obj->mod_count);
+ else
+ CERROR("Object "DFID" fail to be an orphan, "
+ "open count = %d, maybe cause failed "
+ "open replay\n",
+ PFID(mdd_object_fid(obj)),
+ obj->mod_count);
+
+ /* mark object as an orphan here, not
+ * before __mdd_orphan_add() as racing
+ * mdd_la_get() may propagate ORPHAN_OBJ
+ * causing the asserition */
+ rc = mdd_mark_orphan_object(env, obj, th, false);
+ } else {
rc = mdo_destroy(env, obj, th);
- }
+ }
} else if (!is_dir) {
/* old files may not have link ea; ignore errors */
mdd_links_del(env, obj, mdo2fid(pobj), lname, th);
if (rc < 0)
RETURN(false);
- ma->ma_valid = MA_HSM;
+ ma->ma_valid |= MA_HSM;
}
if (ma->ma_hsm.mh_flags & HS_EXISTS)
RETURN(true);
struct mdd_object *mdd_cobj = NULL;
struct mdd_device *mdd = mdo2mdd(pobj);
struct thandle *handle;
- int rc, is_dir = 0;
+ int rc, is_dir = 0, cl_flags = 0;
ENTRY;
/* cobj == NULL means only delete name entry */
RETURN(rc);
is_dir = S_ISDIR(cattr->la_mode);
+ /* search for an existing archive.
+ * we should check ahead as the object
+ * can be destroyed in this transaction */
+ if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
+ cl_flags |= CLF_UNLINK_HSM_EXISTS;
}
rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
ma->ma_attr = *cattr;
ma->ma_valid |= MA_INODE;
rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
+ if (rc != 0)
+ GOTO(cleanup, rc);
/* fetch updated nlink */
- if (rc == 0)
- rc = mdd_la_get(env, mdd_cobj, cattr);
+ rc = mdd_la_get(env, mdd_cobj, cattr);
+ /* if object is removed then we can't get its attrs,
+ * use last get */
+ if (rc == -ENOENT) {
+ cattr->la_nlink = 0;
+ rc = 0;
+ }
- /* if object is removed then we can't get its attrs, use last get */
if (cattr->la_nlink == 0) {
ma->ma_attr = *cattr;
ma->ma_valid |= MA_INODE;
}
+
EXIT;
cleanup:
if (likely(mdd_cobj != NULL))
mdd_write_unlock(env, mdd_cobj);
if (rc == 0) {
- int cl_flags = 0;
-
- if (cattr->la_nlink == 0) {
+ if (cattr->la_nlink == 0)
cl_flags |= CLF_UNLINK_LAST;
- /* search for an existing archive */
- if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
- cl_flags |= CLF_UNLINK_HSM_EXISTS;
- }
+ else
+ cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
rc = mdd_changelog_ns_store(env, mdd,
is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
}
stop:
- mdd_trans_stop(env, mdd, rc, handle);
+ rc = mdd_trans_stop(env, mdd, rc, handle);
return rc;
}
RETURN(0);
}
-static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
- struct md_object *cobj, const struct md_op_spec *spec,
- struct md_attr *ma)
+static int mdd_create_data(const struct lu_env *env,
+ struct md_object *pobj,
+ struct md_object *cobj,
+ const struct md_op_spec *spec,
+ struct md_attr *ma)
{
struct mdd_device *mdd = mdo2mdd(cobj);
struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
/* calling ->ah_make_hint() is used to transfer information from parent */
mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
- handle = mdd_trans_create(env, mdd);
- if (IS_ERR(handle))
- GOTO(out_free, rc = PTR_ERR(handle));
+ handle = mdd_trans_create(env, mdd);
+ if (IS_ERR(handle))
+ GOTO(out_free, rc = PTR_ERR(handle));
- /*
- * XXX: Setting the lov ea is not locked but setting the attr is locked?
- * Should this be fixed?
- */
- CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n",
+ /*
+ * XXX: Setting the lov ea is not locked but setting the attr is locked?
+ * Should this be fixed?
+ */
+ CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
spec->sp_cr_flags, spec->no_create);
rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle);
stop:
- mdd_trans_stop(env, mdd, rc, handle);
+ rc = mdd_trans_stop(env, mdd, rc, handle);
+
out_free:
RETURN(rc);
}
static int mdd_declare_object_initialize(const struct lu_env *env,
struct mdd_object *parent,
struct mdd_object *child,
- struct lu_attr *attr,
+ const struct lu_attr *attr,
struct thandle *handle)
{
int rc;
ENTRY;
- /*
- * inode mode has been set in creation time, and it's based on umask,
- * la_mode and acl, don't set here again! (which will go wrong
- * because below function doesn't consider umask).
- * I'd suggest set all object attributes in creation time, see above.
- */
LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
- attr->la_valid &= ~(LA_MODE | LA_TYPE);
- rc = mdo_declare_attr_set(env, child, attr, handle);
- attr->la_valid |= LA_MODE | LA_TYPE;
- if (rc != 0 || !S_ISDIR(attr->la_mode))
- RETURN(rc);
+ if (!S_ISDIR(attr->la_mode))
+ RETURN(0);
rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR,
dot, handle);
int rc;
ENTRY;
- /* EEXIST check */
- if (mdd_is_dead_obj(obj))
- RETURN(-ENOENT);
+ /* EEXIST check */
+ if (mdd_is_dead_obj(obj))
+ RETURN(-ENOENT);
/*
* In some cases this lookup is not needed - we know before if name
check_perm = false;
}
+ if (S_ISDIR(cattr->la_mode) &&
+ unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
+ spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
+ const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
+
+ if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC) &&
+ le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
+ rc = -EINVAL;
+ CERROR("%s: invalid lmv_user_md: magic = %x, "
+ "stripe_offset = %d, stripe_count = %u: "
+ "rc = %d\n", mdd2obd_dev(m)->obd_name,
+ le32_to_cpu(lum->lum_magic),
+ (int)le32_to_cpu(lum->lum_stripe_offset),
+ le32_to_cpu(lum->lum_stripe_count), rc);
+ return rc;
+ }
+ }
+
rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
if (rc != 0)
RETURN(rc);
}
}
+ /* Inherit project ID from parent directory */
+ if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
+ cattr->la_projid = pattr->la_projid;
+ if (S_ISDIR(cattr->la_mode)) {
+ cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
+ cattr->la_valid |= LA_FLAGS;
+ }
+ cattr->la_valid |= LA_PROJID;
+ }
+
rc = mdd_name_check(m, lname);
if (rc < 0)
RETURN(rc);
switch (cattr->la_mode & S_IFMT) {
- case S_IFLNK: {
- unsigned int symlen = strlen(spec->u.sp_symname) + 1;
+ case S_IFLNK: {
+ unsigned int symlen = strlen(spec->u.sp_symname) + 1;
- if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift))
- RETURN(-ENAMETOOLONG);
- else
- RETURN(0);
- }
+ if (symlen > m->mdd_dt_conf.ddp_symlink_max)
+ RETURN(-ENAMETOOLONG);
+ else
+ RETURN(0);
+ }
case S_IFDIR:
case S_IFREG:
case S_IFCHR:
RETURN(rc);
}
-static int mdd_declare_object_create(const struct lu_env *env,
+static int mdd_declare_create_object(const struct lu_env *env,
struct mdd_device *mdd,
struct mdd_object *p, struct mdd_object *c,
struct lu_attr *attr,
struct lu_buf *acl_buf,
struct dt_allocation_hint *hint)
{
+ const struct lu_buf *buf;
int rc;
- rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec,
+ rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
hint);
- if (rc)
- GOTO(out, rc);
+ if (rc)
+ GOTO(out, rc);
#ifdef CONFIG_FS_POSIX_ACL
if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
/* replay case, create LOV EA from client data */
if (spec->no_create ||
(spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
- const struct lu_buf *buf;
-
buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
spec->u.sp_ea.eadatalen);
rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0,
if (rc)
GOTO(out, rc);
}
+
+ if (spec->sp_cr_file_secctx_name != NULL) {
+ buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
+ spec->sp_cr_file_secctx_size);
+ rc = mdo_declare_xattr_set(env, c, buf,
+ spec->sp_cr_file_secctx_name, 0,
+ handle);
+ if (rc < 0)
+ GOTO(out, rc);
+ }
out:
return rc;
}
{
int rc;
- rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec,
+ rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
def_acl_buf, acl_buf, hint);
if (rc)
GOTO(out, rc);
if (rc != 0)
return rc;
- rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE);
+ rc = mdd_declare_links_add(env, c, handle, ldata);
if (rc)
return rc;
/**
* Create a metadata object and initialize it, set acl, xattr.
**/
-static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj,
+static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
struct mdd_object *son, struct lu_attr *attr,
struct md_op_spec *spec, struct lu_buf *acl_buf,
struct lu_buf *def_acl_buf,
struct dt_allocation_hint *hint,
struct thandle *handle)
{
- int rc;
+ const struct lu_buf *buf;
+ int rc;
mdd_write_lock(env, son, MOR_TGT_CHILD);
- rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec,
+ rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
hint);
if (rc)
GOTO(unlock, rc);
if (spec->no_create ||
(S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
S_ISDIR(attr->la_mode)) {
- const struct lu_buf *buf;
-
buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
spec->u.sp_ea.eadatalen);
rc = mdo_xattr_set(env, son, buf,
struct dt_object *dt = mdd_object_child(son);
const char *target_name = spec->u.sp_symname;
int sym_len = strlen(target_name);
- const struct lu_buf *buf;
loff_t pos = 0;
buf = mdd_buf_get_const(env, target_name, sym_len);
GOTO(err_initlized, rc = -EFAULT);
}
+ if (spec->sp_cr_file_secctx_name != NULL) {
+ buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
+ spec->sp_cr_file_secctx_size);
+ rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
+ 0, handle);
+ if (rc < 0)
+ GOTO(err_initlized, rc);
+ }
+
err_initlized:
if (unlikely(rc != 0)) {
int rc2;
if (rc)
GOTO(stop, rc);
stop:
- mdd_trans_stop(env, mdd, rc, handle);
+ rc = mdd_trans_stop(env, mdd, rc, handle);
+
RETURN(rc);
}
-/*
+/**
* Create object and insert it into namespace.
+ *
+ * Two operations have to be performed:
+ *
+ * - an allocation of a new object (->do_create()), and
+ * - an insertion into a parent index (->dio_insert()).
+ *
+ * Due to locking, operation order is not important, when both are
+ * successful, *but* error handling cases are quite different:
+ *
+ * - if insertion is done first, and following object creation fails,
+ * insertion has to be rolled back, but this operation might fail
+ * also leaving us with dangling index entry.
+ *
+ * - if creation is done first, is has to be undone if insertion fails,
+ * leaving us with leaked space, which is not good but not fatal.
+ *
+ * It seems that creation-first is simplest solution, but it is sub-optimal
+ * in the frequent
+ *
+ * $ mkdir foo
+ * $ mkdir foo
+ *
+ * case, because second mkdir is bound to create object, only to
+ * destroy it immediately.
+ *
+ * To avoid this follow local file systems that do double lookup:
+ *
+ * 0. lookup -> -EEXIST (mdd_create_sanity_check())
+ * 1. create (mdd_create_object_internal())
+ * 2. insert (__mdd_index_insert(), lookup again)
+ *
+ * \param[in] pobj parent object
+ * \param[in] lname name of child being created
+ * \param[in,out] child child object being created
+ * \param[in] spec additional create parameters
+ * \param[in] ma attributes for new child object
+ *
+ * \retval 0 on success
+ * \retval negative errno on failure
*/
static int mdd_create(const struct lu_env *env, struct md_object *pobj,
const struct lu_name *lname, struct md_object *child,
- struct md_op_spec *spec, struct md_attr* ma)
+ struct md_op_spec *spec, struct md_attr *ma)
{
struct mdd_thread_info *info = mdd_env_info(env);
struct lu_attr *la = &info->mti_la_for_fix;
int rc2;
ENTRY;
- /*
- * Two operations have to be performed:
- *
- * - an allocation of a new object (->do_create()), and
- *
- * - an insertion into a parent index (->dio_insert()).
- *
- * Due to locking, operation order is not important, when both are
- * successful, *but* error handling cases are quite different:
- *
- * - if insertion is done first, and following object creation fails,
- * insertion has to be rolled back, but this operation might fail
- * also leaving us with dangling index entry.
- *
- * - if creation is done first, is has to be undone if insertion
- * fails, leaving us with leaked space, which is neither good, nor
- * fatal.
- *
- * It seems that creation-first is simplest solution, but it is
- * sub-optimal in the frequent
- *
- * $ mkdir foo
- * $ mkdir foo
- *
- * case, because second mkdir is bound to create object, only to
- * destroy it immediately.
- *
- * To avoid this follow local file systems that do double lookup:
- *
- * 0. lookup -> -EEXIST (mdd_create_sanity_check())
- *
- * 1. create (mdd_object_create_internal())
- *
- * 2. insert (__mdd_index_insert(), lookup again)
- */
-
rc = mdd_la_get(env, mdd_pobj, pattr);
if (rc != 0)
RETURN(rc);
if (rc)
RETURN(rc);
- if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
+ if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
GOTO(out_free, rc = -EINPROGRESS);
handle = mdd_trans_create(env, mdd);
if (IS_ERR(handle))
GOTO(out_free, rc = PTR_ERR(handle));
- acl_buf.lb_buf = info->mti_xattr_buf;
- acl_buf.lb_len = sizeof(info->mti_xattr_buf);
+ lu_buf_check_and_alloc(&info->mti_xattr_buf,
+ mdd->mdd_dt_conf.ddp_max_ea_size);
+ acl_buf = info->mti_xattr_buf;
def_acl_buf.lb_buf = info->mti_key;
def_acl_buf.lb_len = sizeof(info->mti_key);
rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
handle, spec, ldata, &def_acl_buf, &acl_buf,
hint);
- if (rc)
- GOTO(out_stop, rc);
+ if (rc)
+ GOTO(out_stop, rc);
- rc = mdd_trans_start(env, mdd, handle);
- if (rc)
- GOTO(out_stop, rc);
+ rc = mdd_trans_start(env, mdd, handle);
+ if (rc)
+ GOTO(out_stop, rc);
- rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf,
+ rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
&def_acl_buf, hint, handle);
if (rc != 0)
GOTO(out_stop, rc);
if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
mdd_write_lock(env, son, MOR_TGT_CHILD);
+ son->mod_flags |= VOLATILE_OBJ;
rc = __mdd_orphan_add(env, son, handle);
GOTO(out_volatile, rc);
} else {
EXIT;
err_insert:
if (rc != 0) {
- int rc2;
-
if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
rc2 = __mdd_orphan_del(env, son, handle);
else
* before mdd_rename and enable MDS_PERM_BYPASS. */
LASSERT(sobj);
+ /*
+ * If we are using project inheritance, we only allow renames
+ * into our tree when the project IDs are the same; otherwise
+ * tree quota mechanism would be circumvented.
+ */
+ if (((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
+ tpattr->la_projid != cattr->la_projid) ||
+ ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
+ (pattr->la_projid != tpattr->la_projid)))
+ RETURN(-EXDEV);
+
rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
if (rc)
RETURN(rc);
struct mdd_object *mdd_tpobj,
struct mdd_object *mdd_sobj,
struct mdd_object *mdd_tobj,
- const struct lu_name *tname,
const struct lu_name *sname,
+ const struct lu_name *tname,
struct md_attr *ma,
struct linkea_data *ldata,
struct thandle *handle)
if (rc)
return rc;
- rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata,
- S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE);
+ rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
if (rc)
return rc;
rc = mdd_la_get(env, mdd_tobj, tattr);
if (rc)
GOTO(out_pending, rc);
+ /* search for an existing archive.
+ * we should check ahead as the object
+ * can be destroyed in this transaction */
+ if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
+ cl_flags |= CLF_RENAME_LAST_EXISTS;
}
rc = mdd_la_get(env, mdd_tpobj, tpattr);
GOTO(out_pending, rc = PTR_ERR(handle));
memset(ldata, 0, sizeof(*ldata));
- mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname,
- mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata);
+ rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
+ lsname, mdd_object_fid(mdd_tpobj), ltname,
+ 1, 0, ldata);
+ if (rc)
+ GOTO(stop, rc);
+
rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
mdd_tobj, lsname, ltname, ma, ldata, handle);
if (rc)
/* fetch updated nlink */
rc = mdd_la_get(env, mdd_tobj, tattr);
- if (rc != 0) {
+ if (rc == -ENOENT) {
+ /* the object got removed, let's
+ * return the latest known attributes */
+ tattr->la_nlink = 0;
+ rc = 0;
+ } else if (rc != 0) {
CERROR("%s: Failed to get nlink for tobj "
DFID": rc = %d\n",
mdd2obd_dev(mdd)->obd_name,
ma->ma_attr = *tattr;
ma->ma_valid |= MA_INODE;
- if (tattr->la_nlink == 0) {
+ if (tattr->la_nlink == 0)
cl_flags |= CLF_RENAME_LAST;
- if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
- cl_flags |= CLF_RENAME_LAST_EXISTS;
- }
+ else
+ cl_flags &= ~CLF_RENAME_LAST_EXISTS;
}
la->la_valid = LA_CTIME | LA_MTIME;
ltname, lsname, handle);
stop:
- mdd_trans_stop(env, mdd, rc, handle);
+ rc = mdd_trans_stop(env, mdd, rc, handle);
out_pending:
mdd_object_put(env, mdd_sobj);
**/
static int mdd_linkea_update_child_internal(const struct lu_env *env,
struct mdd_object *parent,
+ struct mdd_object *newparent,
struct mdd_object *child,
const char *name, int namelen,
struct thandle *handle,
&lname, &fid);
if (strncmp(lname.ln_name, name, namelen) != 0 ||
- lu_fid_eq(&fid, mdd_object_fid(parent))) {
+ !lu_fid_eq(&fid, mdd_object_fid(parent))) {
ldata.ld_lee = (struct link_ea_entry *)
((char *)ldata.ld_lee +
ldata.ld_reclen);
CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n",
mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)),
lname.ln_namelen, lname.ln_name,
- PFID(mdd_object_fid(parent)));
+ PFID(mdd_object_fid(newparent)));
/* update to the new parent fid */
linkea_entry_pack(ldata.ld_lee, &lname,
- mdd_object_fid(parent));
+ mdd_object_fid(newparent));
if (declare)
- rc = mdd_declare_links_add(env, child, handle, &ldata,
- MLAO_IGNORE);
+ rc = mdd_declare_links_add(env, child, handle, &ldata);
else
rc = mdd_links_write(env, child, &ldata, handle);
break;
static int mdd_linkea_declare_update_child(const struct lu_env *env,
struct mdd_object *parent,
+ struct mdd_object *newparent,
struct mdd_object *child,
const char *name, int namelen,
struct thandle *handle)
{
- return mdd_linkea_update_child_internal(env, parent, child, name,
+ return mdd_linkea_update_child_internal(env, parent, newparent,
+ child, name,
namelen, handle, true);
}
static int mdd_linkea_update_child(const struct lu_env *env,
struct mdd_object *parent,
+ struct mdd_object *newparent,
struct mdd_object *child,
const char *name, int namelen,
struct thandle *handle)
{
- return mdd_linkea_update_child_internal(env, parent, child, name,
+ return mdd_linkea_update_child_internal(env, parent, newparent,
+ child, name,
namelen, handle, false);
}
ENTRY;
LASSERT(ldata->ld_buf != NULL);
+ LASSERT(ldata->ld_leh != NULL);
-again:
/* If it is mulitple links file, we need update the name entry for
* all parent */
- LASSERT(ldata->ld_leh != NULL);
ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj);
CWARN("%s: cannot find obj "DFID": rc = %ld\n",
mdd2obd_dev(mdd)->obd_name, PFID(&fid),
PTR_ERR(pobj));
- linkea_del_buf(ldata, &lname);
- goto again;
+ continue;
}
if (!mdd_object_exists(pobj)) {
CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
mdd2obd_dev(mdd)->obd_name, PFID(&fid));
- linkea_del_buf(ldata, &lname);
- mdd_object_put(env, pobj);
- goto again;
+ goto next_put;
}
if (pobj == mdd_pobj &&
CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n",
mdd2obd_dev(mdd)->obd_name, child_name->ln_name,
PFID(&fid));
- linkea_del_buf(ldata, &lname);
- mdd_object_put(env, pobj);
- goto again;
+ goto next_put;
}
CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n",
/* lnamelen is too big(> NAME_MAX + 16),
* something wrong about this linkea, let's
* skip it */
- linkea_del_buf(ldata, &lname);
- mdd_object_put(env, pobj);
- goto again;
+ CWARN("%s: the name %.*s is too long under "
+ DFID"\n", mdd2obd_dev(mdd)->obd_name,
+ lname.ln_namelen, lname.ln_name,
+ PFID(&fid));
+ goto next_put;
}
/* Note: lname might be without \0 at the end, see
* it might be packed into the RPC buffer. */
rc = mdd_lookup(env, &pobj->mod_obj, &lname,
&info->mti_fid, NULL);
- if (rc < 0 ||
- !lu_fid_eq(&info->mti_fid,
- mdd_object_fid(mdd_sobj))) {
- /* skip invalid linkea entry */
- linkea_del_buf(ldata, &lname);
- mdd_object_put(env, pobj);
- goto again;
- }
+ if (rc < 0 || !lu_fid_eq(&info->mti_fid,
+ mdd_object_fid(mdd_sobj)))
+ GOTO(next_put, rc == -ENOENT ? 0 : rc);
rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle);
if (rc != 0)
int list_xsize;
struct lu_buf list_xbuf;
int rc;
- int rc1;
/* retrieve xattr list from the old object */
list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL);
xname = list_xbuf.lb_buf;
while (rem > 0) {
xlen = strnlen(xname, rem - 1) + 1;
- if (strcmp(XATTR_NAME_LINK, xname) == 0 ||
- strcmp(XATTR_NAME_LMA, xname) == 0 ||
+ if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
strcmp(XATTR_NAME_LMV, xname) == 0)
goto next;
if (rc != 0)
GOTO(stop_trans, rc);
+again:
rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle);
if (rc == -EEXIST)
GOTO(stop_trans, rc = 0);
+ if (unlikely(rc == -ENOSPC &&
+ strcmp(xname, XATTR_NAME_LINK) == 0)) {
+ rc = linkea_overflow_shrink(
+ (struct linkea_data *)(xbuf.lb_buf));
+ if (likely(rc > 0)) {
+ xbuf.lb_len = rc;
+ goto again;
+ }
+ }
+
if (rc != 0)
GOTO(stop_trans, rc);
stop_trans:
- rc1 = mdd_trans_stop(env, mdd, rc, handle);
- if (rc == 0)
- rc = rc1;
+ rc = mdd_trans_stop(env, mdd, rc, handle);
if (rc != 0)
GOTO(out, rc);
next:
int rc;
int mgr_easize;
- rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la,
+ rc = mdd_declare_create_object_internal(env, mdd_pobj, mdd_tobj, la,
handle, spec, NULL);
if (rc != 0)
return rc;
buf, 0, handle);
if (rc != 0)
return rc;
- } else if (S_ISDIR(la->la_mode)) {
- rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata,
- MLAO_IGNORE);
+ } else if (S_ISDIR(la->la_mode) && ldata != NULL) {
+ rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata);
if (rc != 0)
return rc;
}
struct md_op_spec *spec = &info->mti_spec;
struct lu_buf lmm_buf = { NULL };
struct lu_buf link_buf = { NULL };
- const struct lu_buf *buf;
+ struct lu_buf mgr_buf;
struct thandle *handle;
struct lmv_mds_md_v1 *mgr_ea;
struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix);
spec->sp_cr_lookup = 0;
spec->sp_feat = &dt_directory_features;
if (S_ISLNK(la->la_mode)) {
+ const struct lu_buf *buf;
+
buf = lu_buf_check_and_alloc(
&mdd_env_info(env)->mti_big_buf,
la->la_size + 1);
spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
}
} else if (S_ISDIR(la->la_mode)) {
- rc = mdd_links_read(env, mdd_sobj, ldata);
- if (rc < 0 && rc != -ENODATA)
+ rc = mdd_links_read_with_rec(env, mdd_sobj, ldata);
+ if (rc == -ENODATA) {
+ /* ignore the non-linkEA error */
+ ldata = NULL;
+ rc = 0;
+ }
+ if (rc < 0)
RETURN(rc);
}
- mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf;
+ mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
+ lu_buf_check_and_alloc(&info->mti_xattr_buf, mgr_easize);
+ mgr_buf.lb_buf = info->mti_xattr_buf.lb_buf;
+ mgr_buf.lb_len = mgr_easize;
+ mgr_ea = mgr_buf.lb_buf;
memset(mgr_ea, 0, sizeof(*mgr_ea));
mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
mgr_ea->lmv_stripe_count = cpu_to_le32(2);
* the last step of migration, so we set th_local = 1 to avoid
* update last rcvd for this transaction */
handle->th_local = 1;
- rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj,
- spec, la,
- (union lmv_mds_md *)info->mti_xattr_buf,
- ldata, handle);
+ rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec,
+ la, mgr_buf.lb_buf, ldata, handle);
if (rc != 0)
GOTO(stop_trans, rc);
if (rc != 0)
GOTO(stop_trans, rc);
+ /* don't set nlink from the original object */
+ la->la_valid &= ~LA_NLINK;
+
/* create the target object */
- rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
+ rc = mdd_create_object(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL,
hint, handle);
if (rc != 0)
GOTO(stop_trans, rc);
- if (S_ISDIR(la->la_mode)) {
+ if (S_ISDIR(la->la_mode) && ldata != NULL) {
rc = mdd_links_write(env, mdd_tobj, ldata, handle);
if (rc != 0)
GOTO(stop_trans, rc);
/* Set MIGRATE EA on the source inode, so once the migration needs
* to be re-done during failover, the re-do process can locate the
* target object which is already being created. */
- mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1);
- buf = mdd_buf_get_const(env, mgr_ea, mgr_easize);
- rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle);
+ rc = mdo_xattr_set(env, mdd_sobj, &mgr_buf, XATTR_NAME_LMV, 0, handle);
if (rc != 0)
GOTO(stop_trans, rc);
la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL;
rc = mdo_attr_set(env, mdd_sobj, la_flag, handle);
stop_trans:
- if (handle != NULL) {
- int rc1;
-
- rc1 = mdd_trans_stop(env, mdd, rc, handle);
- if (rc == 0)
- rc = rc1;
- }
+ if (handle != NULL)
+ rc = mdd_trans_stop(env, mdd, rc, handle);
out_free:
if (lmm_buf.lb_buf != NULL)
OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len);
struct thandle *handle;
struct dt_it *it;
const struct dt_it_ops *iops;
- int rc;
int result;
struct lu_dirent *ent;
+ int rc;
ENTRY;
OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1);
struct mdd_object *child;
char *name = mdd_env_info(env)->mti_key;
int len;
- int recsize;
int is_dir;
bool target_exist = false;
- int rc1;
len = iops->key_size(env, it);
if (len == 0)
}
fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
- recsize = le16_to_cpu(ent->lde_reclen);
/* Insert new fid with target name into target dir */
if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') ||
GOTO(out_put, rc);
}
- rc = mdd_linkea_declare_update_child(env, mdd_tobj,
+ rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj,
child, name,
strlen(name),
handle);
GOTO(out_put, rc);
}
- rc = mdd_linkea_update_child(env, mdd_tobj, child, name,
+ rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj,
+ child, name,
strlen(name), handle);
out_put:
mdd_write_unlock(env, child);
mdd_object_put(env, child);
- rc1 = mdd_trans_stop(env, mdd, rc, handle);
- if (rc == 0)
- rc = rc1;
-
+ rc = mdd_trans_stop(env, mdd, rc, handle);
if (rc != 0)
GOTO(out, rc);
next:
if (rc != 0)
return rc;
- rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE);
+ rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL);
if (rc != 0)
return rc;
if (rc != 0)
GOTO(stop_trans, rc);
- linkea_add_buf(ldata, lname, mdd_object_fid(mdd_pobj));
- rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle,
- ldata, 1);
- if (rc != 0)
- GOTO(stop_trans, rc);
-
mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD);
- /* Increase mod_count to add the source object to the orphan list,
- * so if other clients still send RPC to the old object, then these
- * objects can help the request to find the new object, see
- * mdt_reint_open() */
- mdd_sobj->mod_count++;
- rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle);
- mdd_sobj->mod_count--;
+ mdd_sobj->mod_flags |= DEAD_OBJ;
+ rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false);
+ if (rc != 0)
+ GOTO(out_unlock, rc);
+
+ rc = __mdd_orphan_add(env, mdd_sobj, handle);
if (rc != 0)
GOTO(out_unlock, rc);
mdd_write_unlock(env, mdd_sobj);
stop_trans:
- mdd_trans_stop(env, mdd, rc, handle);
+ rc = mdd_trans_stop(env, mdd, rc, handle);
RETURN(rc);
}
/* If there are still links locally, then the file will not be
* migrated. */
LASSERT(ldata->ld_leh != NULL);
+
+ /* If the linkEA is overflow, then means there are some unknown name
+ * entries under unknown parents, that will prevent the migration. */
+ if (unlikely(ldata->ld_leh->leh_overflow_time))
+ RETURN(1);
+
ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1);
for (count = 0; count < ldata->ld_leh->leh_reccount; count++) {
struct lu_name lname;
struct mdd_object *mdd_tobj = md2mdd_obj(tobj);
struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr);
struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
+ bool created = false;
int rc;
ENTRY;
* the file is being opened by someone else right now */
mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD);
if (mdd_sobj->mod_count > 0) {
- CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n",
+ CDEBUG(D_OTHER,
+ "%s: "DFID"%s is already opened count %d: rc = %d\n",
mdd2obd_dev(mdd)->obd_name,
PFID(mdd_object_fid(mdd_sobj)), lname->ln_name,
mdd_sobj->mod_count, -EBUSY);
lname, so_attr);
if (rc != 0)
GOTO(put, rc);
+ created = true;
}
LASSERT(mdd_object_exists(mdd_tobj));
ma);
if (rc != 0)
GOTO(put, rc);
+
+ /* newly created target was not locked, don't cache its attributes */
+ if (created)
+ mdd_invalidate(env, tobj);
put:
RETURN(rc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff