4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/mdd/mdd_dir.c
33 * Lustre Metadata Server (mdd) routines
35 * Author: Wang Di <wangdi@intel.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <obd_class.h>
41 #include <obd_support.h>
42 #include <lustre_mds.h>
43 #include <lustre_fid.h>
44 #include <lustre_lmv.h>
45 #include <lustre_idmap.h>
47 #include "mdd_internal.h"
49 static const char dot[] = ".";
50 static const char dotdot[] = "..";
52 static struct lu_name lname_dotdot = {
53 .ln_name = (char *) dotdot,
54 .ln_namelen = sizeof(dotdot) - 1,
58 mdd_name_check(const struct lu_env *env, struct mdd_device *m,
59 const struct lu_name *ln)
61 struct mdd_thread_info *info = mdd_env_info(env);
62 bool enc = info->mdi_pattr.la_valid & LA_FLAGS &&
63 info->mdi_pattr.la_flags & LUSTRE_ENCRYPT_FL;
65 if (!lu_name_is_valid(ln))
67 else if (!enc && ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)
73 /* Get FID from name and parent */
75 __mdd_lookup(const struct lu_env *env, struct md_object *pobj,
76 const struct lu_attr *pattr, const struct lu_name *lname,
77 struct lu_fid *fid, unsigned int may_mask)
79 const char *name = lname->ln_name;
80 const struct dt_key *key = (const struct dt_key *)name;
81 struct mdd_object *mdd_obj = md2mdd_obj(pobj);
82 struct dt_object *dir = mdd_object_child(mdd_obj);
87 if (unlikely(mdd_is_dead_obj(mdd_obj)))
90 if (!mdd_object_exists(mdd_obj))
93 if (mdd_object_remote(mdd_obj)) {
94 CDEBUG(D_INFO, "%s: Object "DFID" located on remote server\n",
95 mdd_obj_dev_name(mdd_obj),
96 PFID(mdd_object_fid(mdd_obj)));
99 rc = mdd_permission_internal_locked(env, mdd_obj, pattr, may_mask,
104 if (likely(dt_try_as_dir(env, dir, true)))
105 rc = dt_lookup(env, dir, (struct dt_rec *)fid, key);
112 int mdd_lookup(const struct lu_env *env,
113 struct md_object *pobj, const struct lu_name *lname,
114 struct lu_fid *fid, struct md_op_spec *spec)
116 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
120 rc = mdd_la_get(env, md2mdd_obj(pobj), pattr);
124 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
125 (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC);
129 /** Read the link EA into a temp buffer.
130 * Uses the mdd_thread_info::mdi_link_buf since it is generally large.
131 * A pointer to the buffer is stored in \a ldata::ld_buf.
135 static int __mdd_links_read(const struct lu_env *env,
136 struct mdd_object *mdd_obj,
137 struct linkea_data *ldata)
141 if (!mdd_object_exists(mdd_obj))
144 /* First try a small buf */
145 LASSERT(env != NULL);
146 ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_link_buf,
148 if (ldata->ld_buf->lb_buf == NULL)
151 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK);
153 /* Buf was too small, figure out what we need. */
154 lu_buf_free(ldata->ld_buf);
155 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
159 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
160 if (ldata->ld_buf->lb_buf == NULL)
162 rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf,
166 lu_buf_free(ldata->ld_buf);
167 ldata->ld_buf = NULL;
171 return linkea_init(ldata);
174 int mdd_links_read(const struct lu_env *env,
175 struct mdd_object *mdd_obj,
176 struct linkea_data *ldata)
180 rc = __mdd_links_read(env, mdd_obj, ldata);
182 rc = linkea_init(ldata);
187 static int mdd_links_read_with_rec(const struct lu_env *env,
188 struct mdd_object *mdd_obj,
189 struct linkea_data *ldata)
193 rc = __mdd_links_read(env, mdd_obj, ldata);
195 rc = linkea_init_with_rec(ldata);
201 * Get parent FID of the directory
203 * Read parent FID from linkEA, if that fails, then do lookup
204 * dotdot to get the parent FID.
206 * \param[in] env execution environment
207 * \param[in] obj object from which to find the parent FID
208 * \param[in] attr attribute of the object
209 * \param[out] fid fid to get the parent FID
211 * \retval 0 if getting the parent FID succeeds.
212 * \retval negative errno if getting the parent FID fails.
214 static inline int mdd_parent_fid(const struct lu_env *env,
215 struct mdd_object *obj,
216 const struct lu_attr *attr,
219 struct mdd_thread_info *info = mdd_env_info(env);
220 struct linkea_data ldata = { NULL };
221 struct lu_buf *buf = &info->mdi_link_buf;
222 struct lu_name lname;
227 LASSERTF(S_ISDIR(mdd_object_type(obj)),
228 "%s: FID "DFID" is not a directory type = %o\n",
229 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)),
230 mdd_object_type(obj));
232 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
233 if (buf->lb_buf == NULL)
234 GOTO(lookup, rc = 0);
237 rc = mdd_links_read_with_rec(env, obj, &ldata);
241 /* the obj is not locked, don't cache attributes */
242 mdd_invalidate(env, &obj->mod_obj);
244 LASSERT(ldata.ld_leh != NULL);
245 /* Directory should only have 1 parent */
246 if (ldata.ld_leh->leh_reccount > 1)
249 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
251 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid);
252 if (likely(fid_is_sane(fid)))
255 rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0);
260 * For root fid use special function, which does not compare version component
261 * of fid. Version component is different for root fids on all MDTs.
263 int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid)
265 return fid_seq(&mdd->mdd_root_fid) == fid_seq(fid) &&
266 fid_oid(&mdd->mdd_root_fid) == fid_oid(fid);
270 * return 1: if \a tfid is the fid of the ancestor of \a mo;
272 * otherwise: values < 0, errors.
274 static int mdd_is_parent(const struct lu_env *env,
275 struct mdd_device *mdd,
276 struct mdd_object *mo,
277 const struct lu_attr *attr,
278 const struct lu_fid *tfid)
280 struct mdd_object *mp;
284 LASSERT(!lu_fid_eq(mdd_object_fid(mo), tfid));
285 pfid = &mdd_env_info(env)->mdi_fid;
287 if (mdd_is_root(mdd, mdd_object_fid(mo)))
290 if (mdd_is_root(mdd, tfid))
293 rc = mdd_parent_fid(env, mo, attr, pfid);
298 if (lu_fid_eq(pfid, tfid))
301 if (mdd_is_root(mdd, pfid))
304 mp = mdd_object_find(env, mdd, pfid);
308 if (!mdd_object_exists(mp)) {
309 mdd_object_put(env, mp);
313 rc = mdd_parent_fid(env, mp, attr, pfid);
314 mdd_object_put(env, mp);
323 * No permission check is needed.
325 * returns 1: if fid is ancestor of @mo;
326 * returns 0: if fid is not an ancestor of @mo;
327 * returns < 0: if error
329 int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
330 const struct lu_fid *fid)
332 struct mdd_device *mdd = mdo2mdd(mo);
333 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
337 if (!mdd_object_exists(md2mdd_obj(mo)))
340 if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo))))
343 rc = mdd_la_get(env, md2mdd_obj(mo), attr);
347 rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid);
352 * Check that @dir contains no entries except (possibly) dot and dotdot.
357 * -ENOTDIR not a directory object
358 * -ENOTEMPTY not empty
362 int mdd_dir_is_empty(const struct lu_env *env, struct mdd_object *dir)
365 struct dt_object *obj;
366 const struct dt_it_ops *iops;
370 obj = mdd_object_child(dir);
371 if (!dt_try_as_dir(env, obj, true))
374 iops = &obj->do_index_ops->dio_it;
375 it = iops->init(env, obj, LUDA_64BITHASH);
377 result = iops->get(env, it, (const struct dt_key *)"");
380 for (result = 0, i = 0; result == 0 && i < 3; ++i)
381 result = iops->next(env, it);
384 else if (result == 1)
386 } else if (result == 0)
388 * Huh? Index contains no zero key?
395 result = PTR_ERR(it);
396 /* -ENODEV means no valid stripe */
397 if (result == -ENODEV)
404 * Determine if the target object can be hard linked, and right now it only
405 * checks if the link count reach the maximum limit. Note: for ldiskfs, the
406 * directory nlink count might exceed the maximum link count(see
407 * osd_object_ref_add), so it only check nlink for non-directories.
409 * \param[in] env thread environment
410 * \param[in] obj object being linked to
411 * \param[in] la attributes of \a obj
413 * \retval 0 if \a obj can be hard linked
414 * \retval negative error if \a obj is a directory or has too
417 static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj,
418 const struct lu_attr *la)
420 struct mdd_device *m = mdd_obj2mdd_dev(obj);
425 /* Subdir count limitation can be broken through
426 * (see osd_object_ref_add), so only check non-directory here. */
427 if (!S_ISDIR(la->la_mode) &&
428 la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink)
435 * Check whether it may create the cobj under the pobj.
437 * \param[in] env execution environment
438 * \param[in] pobj the parent directory
439 * \param[in] pattr the attribute of the parent directory
440 * \param[in] cobj the child to be created
441 * \param[in] check_perm if check WRITE|EXEC permission for parent
443 * \retval = 0 create the child under this dir is allowed
444 * \retval negative errno create the child under this dir is
447 int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj,
448 const struct lu_attr *pattr, struct mdd_object *cobj,
454 if (cobj && mdd_object_exists(cobj))
457 if (mdd_is_dead_obj(pobj))
461 rc = mdd_permission_internal_locked(env, pobj, pattr,
462 MAY_WRITE | MAY_EXEC,
468 * Check whether can unlink from the pobj in the case of "cobj == NULL".
470 int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj,
471 const struct lu_attr *pattr, const struct lu_attr *attr)
476 if (mdd_is_dead_obj(pobj))
479 if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
482 rc = mdd_permission_internal_locked(env, pobj, pattr,
483 MAY_WRITE | MAY_EXEC,
488 if (pattr->la_flags & LUSTRE_APPEND_FL)
495 * pobj == NULL is remote ops case, under such case, pobj's
496 * VTX feature has been checked already, no need check again.
498 static inline int mdd_is_sticky(const struct lu_env *env,
499 struct mdd_object *pobj,
500 const struct lu_attr *pattr,
501 struct mdd_object *cobj,
502 const struct lu_attr *cattr)
504 struct lu_ucred *uc = lu_ucred_assert(env);
507 LASSERT(pattr != NULL);
508 if (!(pattr->la_mode & S_ISVTX) ||
509 (pattr->la_uid == uc->uc_fsuid))
513 LASSERT(cattr != NULL);
514 if (cattr->la_uid == uc->uc_fsuid)
517 return !cap_raised(uc->uc_cap, CAP_FOWNER);
520 static int mdd_may_delete_entry(const struct lu_env *env,
521 struct mdd_object *pobj,
522 const struct lu_attr *pattr,
527 LASSERT(pobj != NULL);
528 if (!mdd_object_exists(pobj))
531 if (mdd_is_dead_obj(pobj))
536 rc = mdd_permission_internal_locked(env, pobj, pattr,
537 MAY_WRITE | MAY_EXEC,
543 if (pattr->la_flags & LUSTRE_APPEND_FL)
550 * Check whether it may delete the cobj from the pobj.
553 int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj,
554 const struct lu_attr *tpattr, struct mdd_object *tobj,
555 const struct lu_attr *tattr, const struct lu_attr *cattr,
556 int check_perm, int check_empty)
562 LASSERT(tpattr != NULL);
563 rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm);
571 if (!mdd_object_exists(tobj))
574 if (mdd_is_dead_obj(tobj))
577 if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr))
580 if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))
583 /* additional check the rename case */
585 if (S_ISDIR(cattr->la_mode)) {
586 if (!S_ISDIR(tattr->la_mode))
589 if (mdd_is_root(mdo2mdd(&tobj->mod_obj),
590 mdd_object_fid(tobj)))
592 } else if (S_ISDIR(tattr->la_mode))
596 if (S_ISDIR(tattr->la_mode) && check_empty)
597 rc = mdd_dir_is_empty(env, tobj);
603 * Check whether it can create the link file(linked to @src_obj) under
604 * the target directory(@tgt_obj), and src_obj has been locked by
607 * \param[in] env execution environment
608 * \param[in] tgt_obj the target directory
609 * \param[in] tattr attributes of target directory
610 * \param[in] lname the link name
611 * \param[in] src_obj source object for link
612 * \param[in] cattr attributes for source object
614 * \retval = 0 it is allowed to create the link file under tgt_obj
615 * \retval negative error not allowed to create the link file
617 static int mdd_link_sanity_check(const struct lu_env *env,
618 struct mdd_object *tgt_obj,
619 const struct lu_attr *tattr,
620 const struct lu_name *lname,
621 struct mdd_object *src_obj,
622 const struct lu_attr *cattr)
624 struct mdd_device *m = mdd_obj2mdd_dev(src_obj);
628 if (!mdd_object_exists(src_obj))
631 if (mdd_is_dead_obj(src_obj))
634 /* Local ops, no lookup before link, check filename length here. */
635 rc = mdd_name_check(env, m, lname);
639 if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL))
642 if (S_ISDIR(mdd_object_type(src_obj)))
645 LASSERT(src_obj != tgt_obj);
646 rc = mdd_may_create(env, tgt_obj, tattr, NULL, true);
650 rc = __mdd_may_link(env, src_obj, cattr);
655 static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
656 const char *name, struct thandle *handle)
658 struct dt_object *next = mdd_object_child(pobj);
662 if (dt_try_as_dir(env, next, true))
663 rc = dt_delete(env, next, (struct dt_key *)name, handle);
670 static int __mdd_index_insert_only(const struct lu_env *env,
671 struct mdd_object *pobj,
672 const struct lu_fid *lf, __u32 type,
673 const char *name, struct thandle *handle)
675 struct dt_object *next = mdd_object_child(pobj);
679 if (dt_try_as_dir(env, next, true)) {
680 struct dt_insert_rec *rec = &mdd_env_info(env)->mdi_dt_rec;
683 rec->rec_type = type;
684 rc = dt_insert(env, next, (const struct dt_rec *)rec,
685 (const struct dt_key *)name, handle);
692 /* insert named index, add reference if isdir */
693 static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
694 const struct lu_fid *lf, __u32 type,
695 const char *name, struct thandle *handle)
700 rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle);
701 if (rc == 0 && S_ISDIR(type)) {
702 mdd_write_lock(env, pobj, DT_TGT_PARENT);
703 mdo_ref_add(env, pobj, handle);
704 mdd_write_unlock(env, pobj);
710 /* delete named index, drop reference if isdir */
711 static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
712 const char *name, int is_dir,
713 struct thandle *handle)
718 rc = __mdd_index_delete_only(env, pobj, name, handle);
719 if (rc == 0 && is_dir) {
720 mdd_write_lock(env, pobj, DT_TGT_PARENT);
721 mdo_ref_del(env, pobj, handle);
722 mdd_write_unlock(env, pobj);
728 static int mdd_llog_record_calc_size(const struct lu_env *env,
729 const struct lu_name *tname,
730 const struct lu_name *sname)
732 const struct lu_ucred *uc = lu_ucred(env);
733 enum changelog_rec_flags clf_flags = CLF_EXTRA_FLAGS;
734 enum changelog_rec_extra_flags crfe = CLFE_UIDGID | CLFE_NID;
737 clf_flags |= CLF_RENAME;
739 if (uc != NULL && uc->uc_jobid[0] != '\0')
740 clf_flags |= CLF_JOBID;
742 return llog_data_len(LLOG_CHANGELOG_HDR_SZ +
743 changelog_rec_offset(clf_flags, crfe) +
744 (tname != NULL ? tname->ln_namelen : 0) +
745 (sname != NULL ? 1 + sname->ln_namelen : 0));
748 int mdd_declare_changelog_store(const struct lu_env *env,
749 struct mdd_device *mdd,
750 enum changelog_rec_type type,
751 const struct lu_name *tname,
752 const struct lu_name *sname,
753 struct thandle *handle)
755 struct obd_device *obd = mdd2obd_dev(mdd);
756 struct llog_ctxt *ctxt;
757 struct llog_rec_hdr rec_hdr;
758 struct thandle *llog_th;
761 if (!mdd_changelog_enabled(env, mdd, type))
764 rec_hdr.lrh_len = mdd_llog_record_calc_size(env, tname, sname);
765 rec_hdr.lrh_type = CHANGELOG_REC;
767 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
771 llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj);
773 GOTO(out_put, rc = PTR_ERR(llog_th));
775 rc = llog_declare_add(env, ctxt->loc_handle, &rec_hdr, llog_th);
783 int mdd_changelog_write_rec(const struct lu_env *env,
784 struct llog_handle *loghandle,
785 struct llog_rec_hdr *r,
786 struct llog_cookie *cookie,
787 int idx, struct thandle *th)
791 if (r->lrh_type == CHANGELOG_REC) {
792 struct mdd_device *mdd;
793 struct llog_changelog_rec *rec;
795 mdd = lu2mdd_dev(loghandle->lgh_ctxt->loc_obd->obd_lu_dev);
796 rec = container_of(r, struct llog_changelog_rec, cr_hdr);
798 spin_lock(&mdd->mdd_cl.mc_lock);
799 rec->cr.cr_index = mdd->mdd_cl.mc_index + 1;
800 spin_unlock(&mdd->mdd_cl.mc_lock);
802 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
806 * if current llog is full, we will generate a new
807 * llog, and since it's actually not an error, let's
808 * avoid increasing index so that userspace apps
809 * should not see a gap in the changelog sequence
811 if (!(rc == -ENOSPC && llog_is_full(loghandle))) {
812 spin_lock(&mdd->mdd_cl.mc_lock);
813 ++mdd->mdd_cl.mc_index;
814 spin_unlock(&mdd->mdd_cl.mc_lock);
817 rc = llog_osd_ops.lop_write_rec(env, loghandle, r,
825 * Checks that changelog consumes safe amount of space comparing
828 * \param env - current lu_env
829 * \param mdd - current MDD device
830 * \param lgh - changelog catalog llog handle
831 * \param estimate - get exact llog size or estimate it.
835 bool mdd_changelog_is_space_safe(const struct lu_env *env,
836 struct mdd_device *mdd,
837 struct llog_handle *lgh,
840 struct obd_statfs sfs;
841 unsigned long long free_space_limit;
842 unsigned long long llog_size;
845 rc = dt_statfs(env, mdd->mdd_bottom, &sfs);
847 /* check is ignored if OSD is not healthy for any reason */
851 * if changelog consumes more than 1/4 of available space then start
854 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_ENOSPC))
855 free_space_limit = cfs_fail_val;
857 free_space_limit = (sfs.os_bfree * sfs.os_bsize) >> 2;
859 /* if \estimate parameter is used then calculate llog size from
860 * number of used catalog entries and plain llog maximum size.
861 * Plain llog maximum size if set as 1/64 of FS free space limited
862 * by 128MB as maximum and 2MB as minimum, see llog_cat_new_log()
863 * Estimation helps to avoid full llog processing to get exact size
864 * by llog_cat_size().
867 /* use 1/64 of FS size but keep it between 2MB and 128MB */
868 llog_size = clamp_t(unsigned long long,
869 (sfs.os_blocks * sfs.os_bsize) >> 6,
872 * llog_cat_free_space() gives free slots, we need occupied,
873 * so subtruct free from total slots minus one for header
875 llog_size *= LLOG_HDR_BITMAP_SIZE(lgh->lgh_hdr) - 1 -
876 llog_cat_free_space(lgh);
878 /* get exact llog size */
879 llog_size = llog_cat_size(env, lgh);
881 CDEBUG(D_HA, "%s:%s changelog size is %lluMB, space limit is %lluMB\n",
882 mdd2obd_dev(mdd)->obd_name, estimate ? " estimated" : "",
883 llog_size >> 20, free_space_limit >> 20);
885 if (llog_size > free_space_limit) {
886 CWARN("%s: changelog uses %lluMB with %lluMB space limit\n",
887 mdd2obd_dev(mdd)->obd_name, llog_size >> 20,
888 free_space_limit >> 20);
896 * Checks if there is enough space in changelog itself and in FS and force
897 * emergency changelog cleanup if needed. It will purge users one by one
898 * from the oldest one while emergency conditions are true.
900 * \param env - current lu_env
901 * \param mdd - current MDD device
902 * \param lgh - changelog catalog llog handle
904 * \retval true if emergency cleanup is needed for changelog
906 static bool mdd_changelog_emrg_cleanup(const struct lu_env *env,
907 struct mdd_device *mdd,
908 struct llog_handle *lgh)
910 unsigned long free_entries = llog_cat_free_space(lgh);
912 /* free space GC is disabled or is in progress already */
913 if (!mdd->mdd_changelog_free_space_gc || mdd->mdd_changelog_emrg_gc)
916 if (free_entries <= mdd->mdd_changelog_min_free_cat_entries) {
917 CWARN("%s: changelog has only %lu free catalog entries\n",
918 mdd2obd_dev(mdd)->obd_name, free_entries);
919 mdd->mdd_changelog_emrg_gc = true;
923 if (!mdd_changelog_is_space_safe(env, mdd, lgh, true)) {
924 mdd->mdd_changelog_emrg_gc = true;
931 bool mdd_changelog_need_gc(const struct lu_env *env, struct mdd_device *mdd,
932 struct llog_handle *lgh)
934 struct mdd_changelog *mc = &mdd->mdd_cl;
936 return mdd_changelog_emrg_cleanup(env, mdd, lgh) ||
937 mdd_changelog_is_too_idle(mdd, mc->mc_minrec, mc->mc_mintime) ||
938 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD);
941 /** Add a changelog entry \a rec to the changelog llog
944 * \param handle - currently ignored since llogs start their own transaction;
945 * this will hopefully be fixed in llog rewrite
948 int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd,
949 struct llog_changelog_rec *rec, struct thandle *th)
951 struct obd_device *obd = mdd2obd_dev(mdd);
952 struct llog_ctxt *ctxt;
953 struct thandle *llog_th;
957 rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) +
958 changelog_rec_varsize(&rec->cr));
960 /* llog_lvfs_write_rec sets the llog tail len */
961 rec->cr_hdr.lrh_type = CHANGELOG_REC;
962 rec->cr.cr_time = cl_time();
964 ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT);
968 llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj);
970 GOTO(out_put, rc = PTR_ERR(llog_th));
972 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_CHANGELOG_REORDER, cfs_fail_val);
973 /* nested journal transaction */
974 rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th);
976 /* time to recover some space ?? */
977 if (likely(!mdd->mdd_changelog_gc ||
978 mdd->mdd_cl.mc_gc_task != MDD_CHLG_GC_NONE ||
979 mdd->mdd_changelog_min_gc_interval >=
980 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time))
981 /* save a spin_lock trip */
984 if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_CHANGELOG_IDX_PUMP)) {
985 spin_lock(&mdd->mdd_cl.mc_lock);
986 mdd->mdd_cl.mc_index += cfs_fail_val;
987 spin_unlock(&mdd->mdd_cl.mc_lock);
990 need_gc = mdd_changelog_need_gc(env, mdd, ctxt->loc_handle);
991 spin_lock(&mdd->mdd_cl.mc_lock);
992 if (likely(mdd->mdd_changelog_gc &&
993 mdd->mdd_cl.mc_gc_task == MDD_CHLG_GC_NONE &&
994 ktime_get_real_seconds() - mdd->mdd_cl.mc_gc_time >
995 mdd->mdd_changelog_min_gc_interval)) {
996 if (unlikely(need_gc)) {
997 CWARN("%s:%s starting changelog garbage collection\n",
999 OBD_FAIL_CHECK(OBD_FAIL_FORCE_GC_THREAD) ?
1001 /* indicate further kthread run will occur outside
1002 * right after current journal transaction filling has
1005 mdd->mdd_cl.mc_gc_task = MDD_CHLG_GC_NEED;
1007 /* next check in mdd_changelog_min_gc_interval anyway
1009 mdd->mdd_cl.mc_gc_time = ktime_get_real_seconds();
1011 spin_unlock(&mdd->mdd_cl.mc_lock);
1013 llog_ctxt_put(ctxt);
1019 static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec,
1020 const struct lu_fid *sfid,
1021 const struct lu_fid *spfid,
1022 const struct lu_name *sname)
1024 struct changelog_ext_rename *rnm = changelog_rec_rename(rec);
1027 LASSERT(sfid != NULL);
1028 LASSERT(spfid != NULL);
1029 LASSERT(sname != NULL);
1031 extsize = sname->ln_namelen + 1;
1033 rnm->cr_sfid = *sfid;
1034 rnm->cr_spfid = *spfid;
1036 changelog_rec_name(rec)[rec->cr_namelen] = '\0';
1037 strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize);
1038 rec->cr_namelen += extsize;
1041 void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid)
1043 struct changelog_ext_jobid *jid = changelog_rec_jobid(rec);
1045 if (jobid == NULL || jobid[0] == '\0')
1048 strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid));
1051 void mdd_changelog_rec_ext_extra_flags(struct changelog_rec *rec, __u64 eflags)
1053 struct changelog_ext_extra_flags *ef = changelog_rec_extra_flags(rec);
1055 ef->cr_extra_flags = eflags;
1058 void mdd_changelog_rec_extra_uidgid(struct changelog_rec *rec,
1059 __u64 uid, __u64 gid)
1061 struct changelog_ext_uidgid *uidgid = changelog_rec_uidgid(rec);
1063 uidgid->cr_uid = uid;
1064 uidgid->cr_gid = gid;
1067 void mdd_changelog_rec_extra_nid(struct changelog_rec *rec,
1070 struct changelog_ext_nid *clnid = changelog_rec_nid(rec);
1072 clnid->cr_nid = nid;
1075 void mdd_changelog_rec_extra_omode(struct changelog_rec *rec, u32 flags)
1077 struct changelog_ext_openmode *omd = changelog_rec_openmode(rec);
1079 omd->cr_openflags = flags;
1082 void mdd_changelog_rec_extra_xattr(struct changelog_rec *rec,
1083 const char *xattr_name)
1085 struct changelog_ext_xattr *xattr = changelog_rec_xattr(rec);
1087 strlcpy(xattr->cr_xattr, xattr_name, sizeof(xattr->cr_xattr));
1090 /** Store a namespace change changelog record
1091 * If this fails, we must fail the whole transaction; we don't
1092 * want the change to commit without the log entry.
1093 * \param target - mdd_object of change
1094 * \param tpfid - target parent dir/object fid
1095 * \param sfid - source object fid
1096 * \param spfid - source parent fid
1097 * \param tname - target name string
1098 * \param sname - source name string
1099 * \param handle - transaction handle
1101 int mdd_changelog_ns_store(const struct lu_env *env,
1102 struct mdd_device *mdd,
1103 enum changelog_rec_type type,
1104 enum changelog_rec_flags clf_flags,
1105 struct mdd_object *target,
1106 const struct lu_fid *tpfid,
1107 const struct lu_fid *sfid,
1108 const struct lu_fid *spfid,
1109 const struct lu_name *tname,
1110 const struct lu_name *sname,
1111 struct thandle *handle)
1113 const struct lu_ucred *uc = lu_ucred(env);
1114 struct llog_changelog_rec *rec;
1117 __u64 xflags = CLFE_INVALID;
1121 if (!mdd_changelog_enabled(env, mdd, type))
1124 LASSERT(tpfid != NULL);
1125 LASSERT(tname != NULL);
1126 LASSERT(handle != NULL);
1128 reclen = mdd_llog_record_calc_size(env, tname, sname);
1129 buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mdi_chlg_buf, reclen);
1130 if (buf->lb_buf == NULL)
1134 clf_flags &= CLF_FLAGMASK;
1135 clf_flags |= CLF_EXTRA_FLAGS;
1138 if (uc->uc_jobid[0] != '\0')
1139 clf_flags |= CLF_JOBID;
1140 xflags |= CLFE_UIDGID;
1145 clf_flags |= CLF_RENAME;
1147 clf_flags |= CLF_VERSION;
1149 rec->cr.cr_flags = clf_flags;
1151 if (clf_flags & CLF_EXTRA_FLAGS) {
1152 mdd_changelog_rec_ext_extra_flags(&rec->cr, xflags);
1153 if (xflags & CLFE_UIDGID)
1154 mdd_changelog_rec_extra_uidgid(&rec->cr,
1155 uc->uc_uid, uc->uc_gid);
1156 if (xflags & CLFE_NID)
1157 mdd_changelog_rec_extra_nid(&rec->cr, uc->uc_nid);
1160 rec->cr.cr_type = (__u32)type;
1161 rec->cr.cr_pfid = *tpfid;
1162 rec->cr.cr_namelen = tname->ln_namelen;
1163 memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen);
1165 if (clf_flags & CLF_RENAME)
1166 mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname);
1168 if (clf_flags & CLF_JOBID)
1169 mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid);
1171 if (likely(target != NULL)) {
1172 rec->cr.cr_tfid = *mdd_object_fid(target);
1173 target->mod_cltime = ktime_get();
1175 fid_zero(&rec->cr.cr_tfid);
1178 rc = mdd_changelog_store(env, mdd, rec, handle);
1180 CERROR("%s: cannot store changelog record: type = %d, "
1181 "name = '%s', t = "DFID", p = "DFID": rc = %d\n",
1182 mdd2obd_dev(mdd)->obd_name, type, tname->ln_name,
1183 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc);
1190 static int __mdd_links_add(const struct lu_env *env,
1191 struct mdd_object *mdd_obj,
1192 struct linkea_data *ldata,
1193 const struct lu_name *lname,
1194 const struct lu_fid *pfid,
1195 int first, int check)
1197 /* cattr is set in mdd_link */
1198 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1201 if (ldata->ld_leh == NULL) {
1202 rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata);
1206 rc = linkea_data_new(ldata,
1207 &mdd_env_info(env)->mdi_link_buf);
1214 rc = linkea_links_find(ldata, lname, pfid);
1215 if (rc && rc != -ENOENT)
1221 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) {
1222 struct lu_fid *tfid = &mdd_env_info(env)->mdi_fid2;
1226 linkea_add_buf(ldata, lname, tfid, false);
1229 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2))
1230 linkea_add_buf(ldata, lname, pfid, false);
1232 /* For encrypted file, we want to limit number of hard links to what
1233 * linkEA can contain. So ask to return error in case of overflow.
1234 * Currently linkEA stores 4KiB of links, that is 14 NAME_MAX links,
1235 * or 119 16-byte names.
1237 return linkea_add_buf(ldata, lname, pfid,
1238 cattr->la_valid & LA_FLAGS &&
1239 cattr->la_flags & LUSTRE_ENCRYPT_FL);
1242 static int __mdd_links_del(const struct lu_env *env,
1243 struct mdd_object *mdd_obj,
1244 struct linkea_data *ldata,
1245 const struct lu_name *lname,
1246 const struct lu_fid *pfid)
1248 /* cattr is set in mdd_link */
1249 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1252 if (ldata->ld_leh == NULL) {
1253 rc = mdd_links_read(env, mdd_obj, ldata);
1258 rc = linkea_links_find(ldata, lname, pfid);
1262 linkea_del_buf(ldata, lname,
1263 cattr->la_valid & LA_FLAGS &&
1264 cattr->la_flags & LUSTRE_ENCRYPT_FL);
1268 static int mdd_linkea_prepare(const struct lu_env *env,
1269 struct mdd_object *mdd_obj,
1270 const struct lu_fid *oldpfid,
1271 const struct lu_name *oldlname,
1272 const struct lu_fid *newpfid,
1273 const struct lu_name *newlname,
1274 int first, int check,
1275 struct linkea_data *ldata)
1280 if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF))
1283 LASSERT(oldpfid != NULL || newpfid != NULL);
1285 if (mdd_obj->mod_flags & DEAD_OBJ)
1286 /* Unnecessary to update linkEA for dead object. */
1289 if (oldpfid != NULL) {
1290 rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid);
1292 if ((check == 1) || (rc != -ENODATA && rc != -ENOENT))
1295 /* No changes done. */
1300 /* If renaming, add the new record */
1301 if (newpfid != NULL)
1302 rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid,
1308 int mdd_links_rename(const struct lu_env *env,
1309 struct mdd_object *mdd_obj,
1310 const struct lu_fid *oldpfid,
1311 const struct lu_name *oldlname,
1312 const struct lu_fid *newpfid,
1313 const struct lu_name *newlname,
1314 struct thandle *handle,
1315 struct linkea_data *ldata,
1316 int first, int check)
1321 if (ldata == NULL) {
1322 ldata = &mdd_env_info(env)->mdi_link_data;
1323 memset(ldata, 0, sizeof(*ldata));
1324 rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname,
1325 newpfid, newlname, first, check, ldata);
1330 if (!(mdd_obj->mod_flags & DEAD_OBJ))
1331 rc = mdd_links_write(env, mdd_obj, ldata, handle);
1337 if (newlname == NULL)
1338 CERROR("link_ea add failed %d "DFID"\n",
1339 rc, PFID(mdd_object_fid(mdd_obj)));
1340 else if (oldpfid == NULL)
1341 CERROR("link_ea add '%.*s' failed %d "DFID"\n",
1342 newlname->ln_namelen, newlname->ln_name, rc,
1343 PFID(mdd_object_fid(mdd_obj)));
1344 else if (newpfid == NULL)
1345 CERROR("link_ea del '%.*s' failed %d "DFID"\n",
1346 oldlname->ln_namelen, oldlname->ln_name, rc,
1347 PFID(mdd_object_fid(mdd_obj)));
1349 CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID
1350 "\n", oldlname->ln_namelen, oldlname->ln_name,
1351 newlname->ln_namelen, newlname->ln_name, rc,
1352 PFID(mdd_object_fid(mdd_obj)));
1355 if (is_vmalloc_addr(ldata->ld_buf))
1356 /* if we vmalloced a large buffer drop it */
1357 lu_buf_free(ldata->ld_buf);
1362 static inline int mdd_links_add(const struct lu_env *env,
1363 struct mdd_object *mdd_obj,
1364 const struct lu_fid *pfid,
1365 const struct lu_name *lname,
1366 struct thandle *handle,
1367 struct linkea_data *ldata, int first)
1369 return mdd_links_rename(env, mdd_obj, NULL, NULL,
1370 pfid, lname, handle, ldata, first, 0);
1373 static inline int mdd_links_del(const struct lu_env *env,
1374 struct mdd_object *mdd_obj,
1375 const struct lu_fid *pfid,
1376 const struct lu_name *lname,
1377 struct thandle *handle)
1379 return mdd_links_rename(env, mdd_obj, pfid, lname,
1380 NULL, NULL, handle, NULL, 0, 0);
1383 /** Read the link EA into a temp buffer.
1384 * Uses the name_buf since it is generally large.
1385 * \retval IS_ERR err
1386 * \retval ptr to \a lu_buf (always \a mdi_link_buf)
1388 struct lu_buf *mdd_links_get(const struct lu_env *env,
1389 struct mdd_object *mdd_obj)
1391 struct linkea_data ldata = { NULL };
1394 rc = mdd_links_read(env, mdd_obj, &ldata);
1395 return rc ? ERR_PTR(rc) : ldata.ld_buf;
1398 int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj,
1399 struct linkea_data *ldata, struct thandle *handle)
1401 const struct lu_buf *buf;
1404 if (ldata == NULL || ldata->ld_buf == NULL ||
1405 ldata->ld_leh == NULL)
1408 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA))
1412 buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf,
1413 ldata->ld_leh->leh_len);
1414 rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle);
1415 if (unlikely(rc == -ENOSPC)) {
1416 rc = linkea_overflow_shrink(ldata);
1424 static int mdd_declare_links_add(const struct lu_env *env,
1425 struct mdd_object *mdd_obj,
1426 struct thandle *handle,
1427 struct linkea_data *ldata)
1433 if (ldata != NULL && ldata->ld_leh != NULL) {
1434 ea_len = ldata->ld_leh->leh_len;
1435 linkea = ldata->ld_buf->lb_buf;
1437 ea_len = MAX_LINKEA_SIZE;
1441 rc = mdo_declare_xattr_set(env, mdd_obj,
1442 mdd_buf_get_const(env, linkea, ea_len),
1443 XATTR_NAME_LINK, 0, handle);
1448 static inline int mdd_declare_links_del(const struct lu_env *env,
1449 struct mdd_object *c,
1450 struct thandle *handle)
1454 /* For directory, the linkEA will be removed together
1455 * with the object. */
1456 if (!S_ISDIR(mdd_object_type(c)))
1457 rc = mdd_declare_links_add(env, c, handle, NULL);
1462 static int mdd_declare_link(const struct lu_env *env,
1463 struct mdd_device *mdd,
1464 struct mdd_object *p,
1465 struct mdd_object *c,
1466 const struct lu_name *name,
1467 struct thandle *handle,
1469 struct linkea_data *data)
1471 struct lu_fid tfid = *mdd_object_fid(c);
1474 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1475 tfid.f_oid = cfs_fail_val;
1477 rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c),
1478 name->ln_name, handle);
1482 rc = mdo_declare_ref_add(env, c, handle);
1486 la->la_valid = LA_CTIME | LA_MTIME;
1487 rc = mdo_declare_attr_set(env, p, la, handle);
1491 la->la_valid = LA_CTIME;
1492 rc = mdo_declare_attr_set(env, c, la, handle);
1496 rc = mdd_declare_links_add(env, c, handle, data);
1500 rc = mdd_declare_changelog_store(env, mdd, CL_HARDLINK, name, NULL,
1506 static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj,
1507 struct md_object *src_obj, const struct lu_name *lname,
1510 const char *name = lname->ln_name;
1511 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
1512 struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj);
1513 struct mdd_object *mdd_sobj = md2mdd_obj(src_obj);
1514 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1515 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
1516 struct mdd_device *mdd = mdo2mdd(src_obj);
1517 struct thandle *handle;
1518 struct lu_fid *tfid = &mdd_env_info(env)->mdi_fid2;
1519 struct linkea_data *ldata = &mdd_env_info(env)->mdi_link_data;
1523 rc = mdd_la_get(env, mdd_sobj, cattr);
1527 rc = mdd_la_get(env, mdd_tobj, tattr);
1532 * If we are using project inheritance, we only allow hard link
1533 * creation in our tree when the project IDs are the same;
1534 * otherwise the tree quota mechanism could be circumvented.
1536 if ((tattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
1537 (tattr->la_projid != cattr->la_projid))
1540 handle = mdd_trans_create(env, mdd);
1542 GOTO(out_pending, rc = PTR_ERR(handle));
1544 memset(ldata, 0, sizeof(*ldata));
1546 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1547 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1549 /* Note: even this function will change ldata, but it comes from
1550 * thread_info, which is completely temporary and only seen in
1551 * this function, so we do not need reset ldata once it fails.*/
1552 rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL,
1553 mdd_object_fid(mdd_tobj), lname, 0, 0, ldata);
1557 rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle,
1562 rc = mdd_trans_start(env, mdd, handle);
1566 mdd_write_lock(env, mdd_sobj, DT_TGT_CHILD);
1567 rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj,
1570 GOTO(out_unlock, rc);
1572 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) {
1573 rc = mdo_ref_add(env, mdd_sobj, handle);
1575 GOTO(out_unlock, rc);
1578 *tfid = *mdd_object_fid(mdd_sobj);
1579 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3))
1580 tfid->f_oid = cfs_fail_val;
1582 rc = __mdd_index_insert_only(env, mdd_tobj, tfid,
1583 mdd_object_type(mdd_sobj), name, handle);
1585 mdo_ref_del(env, mdd_sobj, handle);
1586 GOTO(out_unlock, rc);
1589 la->la_valid = LA_CTIME | LA_MTIME;
1590 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
1592 GOTO(out_unlock, rc);
1594 la->la_valid = LA_CTIME;
1595 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
1597 /* Note: The failure of links_add should not cause the
1598 * link failure, so do not check return value. */
1599 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_tobj),
1600 lname, handle, ldata, 0);
1604 mdd_write_unlock(env, mdd_sobj);
1606 rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
1607 mdd_object_fid(mdd_tobj), NULL,
1608 NULL, lname, NULL, handle);
1610 rc = mdd_trans_stop(env, mdd, rc, handle);
1611 if (is_vmalloc_addr(ldata->ld_buf))
1612 /* if we vmalloced a large buffer drop it */
1613 lu_buf_free(ldata->ld_buf);
1618 static int mdd_mark_orphan_object(const struct lu_env *env,
1619 struct mdd_object *obj, struct thandle *handle,
1622 struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start);
1625 attr->la_valid = LA_FLAGS;
1626 attr->la_flags = LUSTRE_ORPHAN_FL;
1629 rc = mdo_declare_attr_set(env, obj, attr, handle);
1631 rc = mdo_attr_set(env, obj, attr, handle);
1636 static int mdd_declare_finish_unlink(const struct lu_env *env,
1637 struct mdd_object *obj,
1638 struct thandle *handle)
1642 /* Sigh, we do not know if the unlink object will become orphan in
1643 * declare phase, but fortunately the flags here does not matter
1644 * in current declare implementation */
1645 rc = mdd_mark_orphan_object(env, obj, handle, true);
1649 rc = mdo_declare_destroy(env, obj, handle);
1653 rc = mdd_orphan_declare_insert(env, obj, mdd_object_type(obj), handle);
1657 return mdd_declare_links_del(env, obj, handle);
1660 /* caller should take a lock before calling */
1661 int mdd_finish_unlink(const struct lu_env *env,
1662 struct mdd_object *obj, struct md_attr *ma,
1663 struct mdd_object *pobj,
1664 const struct lu_name *lname,
1668 int is_dir = S_ISDIR(ma->ma_attr.la_mode);
1671 LASSERT(mdd_write_locked(env, obj) != 0);
1673 if (ma->ma_attr.la_nlink == 0 || is_dir) {
1674 /* add new orphan and the object
1675 * will be deleted during mdd_close() */
1676 obj->mod_flags |= DEAD_OBJ;
1677 if (obj->mod_count) {
1678 rc = mdd_orphan_insert(env, obj, th);
1680 CDEBUG(D_HA, "Object "DFID" is inserted into "
1681 "orphan list, open count = %d\n",
1682 PFID(mdd_object_fid(obj)),
1685 CERROR("Object "DFID" fail to be an orphan, "
1686 "open count = %d, maybe cause failed "
1688 PFID(mdd_object_fid(obj)),
1691 /* mark object as an orphan here, not
1692 * before mdd_orphan_insert() as racing
1693 * mdd_la_get() may propagate ORPHAN_OBJ
1694 * causing the asserition */
1695 rc = mdd_mark_orphan_object(env, obj, th, false);
1697 rc = mdo_destroy(env, obj, th);
1699 } else if (!is_dir) {
1700 /* old files may not have link ea; ignore errors */
1701 mdd_links_del(env, obj, mdd_object_fid(pobj), lname, th);
1709 * has mdd_write_lock on cobj already, but not on pobj yet
1711 int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj,
1712 const struct lu_attr *pattr,
1713 struct mdd_object *cobj,
1714 const struct lu_attr *cattr)
1719 rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1);
1724 static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd,
1725 struct mdd_object *p, struct mdd_object *c,
1726 const struct lu_name *name, struct md_attr *ma,
1727 struct thandle *handle, int no_name, int is_dir)
1729 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
1732 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1733 if (likely(no_name == 0)) {
1734 rc = mdo_declare_index_delete(env, p, name->ln_name,
1741 rc = mdo_declare_ref_del(env, p, handle);
1747 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1748 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1749 la->la_valid = LA_CTIME | LA_MTIME;
1750 rc = mdo_declare_attr_set(env, p, la, handle);
1755 rc = mdo_declare_ref_del(env, c, handle);
1759 rc = mdo_declare_ref_del(env, c, handle);
1763 la->la_valid = LA_CTIME;
1764 rc = mdo_declare_attr_set(env, c, la, handle);
1768 rc = mdd_declare_finish_unlink(env, c, handle);
1772 /* FIXME: need changelog for remove entry */
1773 rc = mdd_declare_changelog_store(env, mdd, CL_UNLINK, name,
1781 * test if a file has an HSM archive
1782 * if HSM attributes are not found in ma update them from
1785 static bool mdd_hsm_archive_exists(const struct lu_env *env,
1786 struct mdd_object *obj,
1791 if (!(ma->ma_valid & MA_HSM)) {
1792 /* no HSM MD provided, read xattr */
1793 struct lu_buf *hsm_buf;
1794 const size_t buflen = sizeof(struct hsm_attrs);
1797 hsm_buf = mdd_buf_get(env, NULL, 0);
1798 lu_buf_alloc(hsm_buf, buflen);
1799 rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM);
1800 rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm);
1801 lu_buf_free(hsm_buf);
1805 ma->ma_valid |= MA_HSM;
1807 if (ma->ma_hsm.mh_flags & HS_EXISTS)
1813 * Delete name entry and the object.
1814 * Note: no_name == 1 means it only destory the object, i.e. name_entry
1815 * does not exist for this object, and it could only happen during resending
1816 * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname
1817 * is also needed in this case(needed by changelog), so we have to add another
1818 * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II,
1819 * the ENOENT failure should be able to be fixed by redo mechanism.
1821 static int mdd_unlink(const struct lu_env *env, struct md_object *pobj,
1822 struct md_object *cobj, const struct lu_name *lname,
1823 struct md_attr *ma, int no_name)
1825 char *name = (char *)lname->ln_name;
1826 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
1827 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
1828 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
1829 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
1830 struct mdd_object *mdd_cobj = NULL;
1831 struct mdd_device *mdd = mdo2mdd(pobj);
1832 struct thandle *handle;
1833 int rc, is_dir = 0, cl_flags = 0;
1836 /* let shutdown to start */
1837 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 1);
1839 /* cobj == NULL means only delete name entry */
1840 if (likely(cobj != NULL)) {
1841 mdd_cobj = md2mdd_obj(cobj);
1842 if (mdd_object_exists(mdd_cobj) == 0)
1846 rc = mdd_la_get(env, mdd_pobj, pattr);
1850 if (likely(mdd_cobj != NULL)) {
1852 rc = mdd_la_get(env, mdd_cobj, cattr);
1856 is_dir = S_ISDIR(cattr->la_mode);
1857 /* search for an existing archive.
1858 * we should check ahead as the object
1859 * can be destroyed in this transaction */
1860 if (mdd_hsm_archive_exists(env, mdd_cobj, ma))
1861 cl_flags |= CLF_UNLINK_HSM_EXISTS;
1864 rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr);
1868 handle = mdd_trans_create(env, mdd);
1870 RETURN(PTR_ERR(handle));
1872 rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj,
1873 lname, ma, handle, no_name, is_dir);
1877 rc = mdd_trans_start(env, mdd, handle);
1881 if (likely(mdd_cobj != NULL))
1882 mdd_write_lock(env, mdd_cobj, DT_TGT_CHILD);
1884 if (lname->ln_name[lname->ln_namelen] != '\0') {
1885 /* lname->ln_name is not necessarily NUL terminated */
1886 name = kmalloc(lname->ln_namelen + 1, GFP_NOFS);
1888 GOTO(cleanup, rc = -ENOMEM);
1890 memcpy(name, lname->ln_name, lname->ln_namelen);
1891 name[lname->ln_namelen] = '\0';
1894 if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) {
1895 rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle);
1900 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) ||
1901 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY))
1902 GOTO(cleanup, rc = 0);
1904 if (likely(mdd_cobj != NULL)) {
1905 rc = mdo_ref_del(env, mdd_cobj, handle);
1907 __mdd_index_insert_only(env, mdd_pobj,
1908 mdd_object_fid(mdd_cobj),
1909 mdd_object_type(mdd_cobj),
1916 mdo_ref_del(env, mdd_cobj, handle);
1918 /* fetch updated nlink */
1919 rc = mdd_la_get(env, mdd_cobj, cattr);
1924 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
1925 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
1927 la->la_valid = LA_CTIME | LA_MTIME;
1928 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
1932 /* Enough for only unlink the entry */
1933 if (unlikely(mdd_cobj == NULL))
1936 if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) {
1937 /* update ctime of an unlinked file only if it is still
1938 * opened or a link still exists */
1939 la->la_valid = LA_CTIME;
1940 rc = mdd_update_time(env, mdd_cobj, cattr, la, handle);
1945 /* XXX: this transfer to ma will be removed with LOD/OSP */
1946 ma->ma_attr = *cattr;
1947 ma->ma_valid |= MA_INODE;
1948 rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle);
1952 /* fetch updated nlink */
1953 rc = mdd_la_get(env, mdd_cobj, cattr);
1954 /* if object is removed then we can't get its attrs,
1956 if (rc == -ENOENT) {
1957 cattr->la_nlink = 0;
1961 if (cattr->la_nlink == 0) {
1962 ma->ma_attr = *cattr;
1963 ma->ma_valid |= MA_INODE;
1968 if (name != lname->ln_name)
1971 if (likely(mdd_cobj != NULL))
1972 mdd_write_unlock(env, mdd_cobj);
1975 if (cattr->la_nlink == 0)
1976 cl_flags |= CLF_UNLINK_LAST;
1978 cl_flags &= ~CLF_UNLINK_HSM_EXISTS;
1980 rc = mdd_changelog_ns_store(env, mdd,
1981 is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
1982 mdd_cobj, mdd_object_fid(mdd_pobj), NULL, NULL,
1983 lname, NULL, handle);
1987 rc = mdd_trans_stop(env, mdd, rc, handle);
1993 * The permission has been checked when obj created, no need check again.
1995 static int mdd_cd_sanity_check(const struct lu_env *env,
1996 struct mdd_object *obj)
2001 if (!obj || mdd_is_dead_obj(obj))
2007 static int mdd_create_data(const struct lu_env *env, struct md_object *pobj,
2008 struct md_object *cobj,
2009 const struct md_op_spec *spec, struct md_attr *ma)
2011 struct mdd_device *mdd = mdo2mdd(cobj);
2012 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2013 struct mdd_object *son = md2mdd_obj(cobj);
2014 struct thandle *handle;
2015 const struct lu_buf *buf;
2016 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
2017 struct dt_allocation_hint *hint = &mdd_env_info(env)->mdi_hint;
2021 rc = mdd_cd_sanity_check(env, son);
2025 if (!md_should_create(spec->sp_cr_flags))
2029 * there are following use cases for this function:
2030 * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE
2031 * striping can be specified or not
2034 rc = mdd_la_get(env, son, attr);
2038 /* calling ->ah_make_hint() is used to transfer information from parent */
2039 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2041 handle = mdd_trans_create(env, mdd);
2043 GOTO(out_free, rc = PTR_ERR(handle));
2046 * XXX: Setting the lov ea is not locked but setting the attr is locked?
2047 * Should this be fixed?
2049 CDEBUG(D_OTHER, "ea %p/%u, cr_flags %#llo, no_create %u\n",
2050 spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen,
2051 spec->sp_cr_flags, spec->no_create);
2053 if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) {
2054 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2055 spec->u.sp_ea.eadatalen);
2060 rc = dt_declare_xattr_set(env, mdd_object_child(son), buf,
2061 XATTR_NAME_LOV, 0, handle);
2065 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
2070 rc = mdd_trans_start(env, mdd, handle);
2074 rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV,
2080 rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle,
2084 rc = mdd_trans_stop(env, mdd, rc, handle);
2090 static int mdd_declare_object_initialize(const struct lu_env *env,
2091 struct mdd_object *parent,
2092 struct mdd_object *child,
2093 const struct lu_attr *attr,
2094 struct thandle *handle)
2099 LASSERT(attr->la_valid & (LA_MODE | LA_TYPE));
2100 if (!S_ISDIR(attr->la_mode))
2103 rc = mdo_declare_index_insert(env, child, mdd_object_fid(child),
2104 S_IFDIR, dot, handle);
2108 rc = mdo_declare_ref_add(env, child, handle);
2112 rc = mdo_declare_index_insert(env, child, mdd_object_fid(parent),
2113 S_IFDIR, dotdot, handle);
2118 static int mdd_object_initialize(const struct lu_env *env,
2119 const struct lu_fid *pfid,
2120 struct mdd_object *child,
2121 struct lu_attr *attr,
2122 struct thandle *handle)
2127 if (S_ISDIR(attr->la_mode)) {
2128 /* Add "." and ".." for newly created dir */
2129 mdo_ref_add(env, child, handle);
2130 rc = __mdd_index_insert_only(env, child, mdd_object_fid(child),
2131 S_IFDIR, dot, handle);
2133 rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR,
2136 mdo_ref_del(env, child, handle);
2143 * This function checks whether it can create a file/dir under the
2144 * directory(@pobj). The directory(@pobj) is not being locked by
2147 * \param[in] env execution environment
2148 * \param[in] pobj the directory to create files
2149 * \param[in] pattr the attributes of the directory
2150 * \param[in] lname the name of the created file/dir
2151 * \param[in] cattr the attributes of the file/dir
2152 * \param[in] spec create specification
2154 * \retval = 0 it is allowed to create file/dir under
2156 * \retval negative error not allowed to create file/dir
2157 * under the directory
2159 static int mdd_create_sanity_check(const struct lu_env *env,
2160 struct md_object *pobj,
2161 const struct lu_attr *pattr,
2162 const struct lu_name *lname,
2163 struct lu_attr *cattr,
2164 struct md_op_spec *spec)
2166 struct mdd_thread_info *info = mdd_env_info(env);
2167 struct lu_fid *fid = &info->mdi_fid;
2168 struct mdd_object *obj = md2mdd_obj(pobj);
2169 struct mdd_device *m = mdo2mdd(pobj);
2170 bool check_perm = true;
2175 if (mdd_is_dead_obj(obj))
2179 * In some cases this lookup is not needed - we know before if name
2180 * exists or not because MDT performs lookup for it.
2181 * name length check is done in lookup.
2183 if (spec->sp_cr_lookup) {
2185 * Check if the name already exist, though it will be checked in
2186 * _index_insert also, for avoiding rolling back if exists
2189 rc = __mdd_lookup(env, pobj, pattr, lname, fid,
2190 MAY_WRITE | MAY_EXEC);
2192 RETURN(rc ? : -EEXIST);
2194 /* Permission is already being checked in mdd_lookup */
2198 if (S_ISDIR(cattr->la_mode) &&
2199 unlikely(spec != NULL && spec->sp_cr_flags & MDS_OPEN_HAS_EA) &&
2200 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen > 0) {
2201 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
2203 if (!lmv_user_magic_supported(le32_to_cpu(lum->lum_magic)) &&
2204 le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC_V0) {
2206 CERROR("%s: invalid lmv_user_md: magic=%x hash=%x stripe_offset=%d stripe_count=%u: rc = %d\n",
2207 mdd2obd_dev(m)->obd_name,
2208 le32_to_cpu(lum->lum_magic),
2209 le32_to_cpu(lum->lum_hash_type),
2210 (int)le32_to_cpu(lum->lum_stripe_offset),
2211 le32_to_cpu(lum->lum_stripe_count), rc);
2217 rc = mdd_may_create(env, obj, pattr, NULL, check_perm);
2222 if (pattr->la_mode & S_ISGID) {
2223 struct lu_ucred *uc = lu_ucred(env);
2225 cattr->la_gid = pattr->la_gid;
2227 /* Directories are special, and always inherit S_ISGID */
2228 if (S_ISDIR(cattr->la_mode)) {
2229 cattr->la_mode |= S_ISGID;
2230 cattr->la_valid |= LA_MODE;
2231 } else if ((cattr->la_mode & (S_ISGID | S_IXGRP))
2232 == (S_ISGID | S_IXGRP) &&
2233 !lustre_in_group_p(uc,
2234 (cattr->la_valid & LA_GID) ?
2235 cattr->la_gid : pattr->la_gid) &&
2236 !cap_raised(uc->uc_cap, CAP_FSETID)) {
2237 cattr->la_mode &= ~S_ISGID;
2238 cattr->la_valid |= LA_MODE;
2242 /* Inherit project ID from parent directory */
2243 if (pattr->la_flags & LUSTRE_PROJINHERIT_FL) {
2244 cattr->la_projid = pattr->la_projid;
2245 if (S_ISDIR(cattr->la_mode)) {
2246 cattr->la_flags |= LUSTRE_PROJINHERIT_FL;
2247 cattr->la_valid |= LA_FLAGS;
2249 cattr->la_valid |= LA_PROJID;
2252 rc = mdd_name_check(env, m, lname);
2256 switch (cattr->la_mode & S_IFMT) {
2258 unsigned int symlen = spec->u.sp_symname.ln_namelen + 1;
2260 if (symlen > m->mdd_dt_conf.ddp_symlink_max)
2261 RETURN(-ENAMETOOLONG);
2280 static int mdd_declare_create_object(const struct lu_env *env,
2281 struct mdd_device *mdd,
2282 struct mdd_object *p, struct mdd_object *c,
2283 struct lu_attr *attr,
2284 struct thandle *handle,
2285 const struct md_op_spec *spec,
2286 struct lu_buf *def_acl_buf,
2287 struct lu_buf *acl_buf,
2288 struct lu_buf *hsm_buf,
2289 struct dt_allocation_hint *hint)
2291 const struct lu_buf *buf;
2294 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2295 /* ldiskfs OSD needs this information for credit allocation */
2297 hint->dah_acl_len = def_acl_buf->lb_len;
2299 rc = mdd_declare_create_object_internal(env, p, c, attr, handle, spec,
2304 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2305 if (def_acl_buf && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) {
2306 /* if dir, then can inherit default ACl */
2307 rc = mdo_declare_xattr_set(env, c, def_acl_buf,
2308 XATTR_NAME_ACL_DEFAULT,
2314 if (acl_buf && acl_buf->lb_len > 0) {
2315 rc = mdo_declare_attr_set(env, c, attr, handle);
2319 rc = mdo_declare_xattr_set(env, c, acl_buf,
2320 XATTR_NAME_ACL_ACCESS, 0, handle);
2325 rc = mdd_declare_object_initialize(env, p, c, attr, handle);
2329 /* replay case, create LOV EA from client data */
2330 if ((!(spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) && spec->no_create) ||
2331 (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) {
2332 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2333 spec->u.sp_ea.eadatalen);
2334 rc = mdo_declare_xattr_set(env, c, buf,
2335 S_ISDIR(attr->la_mode) ?
2336 XATTR_NAME_LMV : XATTR_NAME_LOV,
2337 LU_XATTR_CREATE, handle);
2341 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2342 rc = mdo_declare_xattr_set(env, c, hsm_buf,
2350 if (S_ISLNK(attr->la_mode)) {
2351 const char *target_name = spec->u.sp_symname.ln_name;
2352 int sym_len = spec->u.sp_symname.ln_namelen;
2353 const struct lu_buf *buf;
2355 buf = mdd_buf_get_const(env, target_name, sym_len);
2356 rc = dt_declare_record_write(env, mdd_object_child(c),
2362 if (spec->sp_cr_file_secctx_name != NULL) {
2363 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2364 spec->sp_cr_file_secctx_size);
2365 rc = mdo_declare_xattr_set(env, c, buf,
2366 spec->sp_cr_file_secctx_name, 0,
2372 if (spec->sp_cr_file_encctx != NULL) {
2373 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2374 spec->sp_cr_file_encctx_size);
2375 rc = mdo_declare_xattr_set(env, c, buf,
2376 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2385 static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd,
2386 struct mdd_object *p, struct mdd_object *c,
2387 const struct lu_name *name,
2388 struct lu_attr *attr,
2389 struct thandle *handle,
2390 const struct md_op_spec *spec,
2391 struct linkea_data *ldata,
2392 struct lu_buf *def_acl_buf,
2393 struct lu_buf *acl_buf,
2394 struct lu_buf *hsm_buf,
2395 struct dt_allocation_hint *hint)
2399 rc = mdd_declare_create_object(env, mdd, p, c, attr, handle, spec,
2400 def_acl_buf, acl_buf, hsm_buf, hint);
2404 if (S_ISDIR(attr->la_mode)) {
2405 rc = mdo_declare_ref_add(env, p, handle);
2410 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2411 rc = mdd_orphan_declare_insert(env, c, attr->la_mode, handle);
2415 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
2416 enum changelog_rec_type type;
2418 rc = mdo_declare_index_insert(env, p, mdd_object_fid(c),
2419 attr->la_mode, name->ln_name,
2424 rc = mdd_declare_links_add(env, c, handle, ldata);
2429 la->la_valid = LA_CTIME | LA_MTIME;
2430 rc = mdo_declare_attr_set(env, p, la, handle);
2434 type = S_ISDIR(attr->la_mode) ? CL_MKDIR :
2435 S_ISREG(attr->la_mode) ? CL_CREATE :
2436 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD;
2438 rc = mdd_declare_changelog_store(env, mdd, type, name, NULL,
2447 static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj,
2448 struct lu_attr *la, struct lu_buf *def_acl_buf,
2449 struct lu_buf *acl_buf)
2455 if (S_ISLNK(la->la_mode)) {
2456 acl_buf->lb_len = 0;
2457 def_acl_buf->lb_len = 0;
2461 mdd_read_lock(env, pobj, DT_TGT_PARENT);
2462 rc = mdo_xattr_get(env, pobj, def_acl_buf,
2463 XATTR_NAME_ACL_DEFAULT);
2464 mdd_read_unlock(env, pobj);
2466 /* ACL buffer size is not enough, need realloc */
2467 if (rc > acl_buf->lb_len)
2470 /* If there are default ACL, fix mode/ACL by default ACL */
2471 def_acl_buf->lb_len = rc;
2472 memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc);
2473 acl_buf->lb_len = rc;
2474 rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode);
2477 } else if (rc == -ENODATA || rc == -EOPNOTSUPP) {
2478 /* If there are no default ACL, fix mode by mask */
2479 struct lu_ucred *uc = lu_ucred(env);
2481 /* The create triggered by MDT internal events, such as
2482 * LFSCK reset, will not contain valid "uc". */
2483 if (unlikely(uc != NULL))
2484 la->la_mode &= ~uc->uc_umask;
2486 acl_buf->lb_len = 0;
2487 def_acl_buf->lb_len = 0;
2494 * Create a metadata object and initialize it, set acl, xattr.
2496 static int mdd_create_object(const struct lu_env *env, struct mdd_object *pobj,
2497 struct mdd_object *son, struct lu_attr *attr,
2498 struct md_op_spec *spec, struct lu_buf *acl_buf,
2499 struct lu_buf *def_acl_buf,
2500 struct lu_buf *hsm_buf,
2501 struct dt_allocation_hint *hint,
2502 struct thandle *handle, bool initsecctx)
2504 const struct lu_buf *buf;
2507 mdd_write_lock(env, son, DT_TGT_CHILD);
2508 rc = mdd_create_object_internal(env, NULL, son, attr, handle, spec,
2513 /* Note: In DNE phase I, for striped dir, though sub-stripes will be
2514 * created in declare phase, they also needs to be added to master
2515 * object as sub-directory entry. So it has to initialize the master
2516 * object, then set dir striped EA.(in mdo_xattr_set) */
2517 rc = mdd_object_initialize(env, mdd_object_fid(pobj), son, attr,
2520 GOTO(err_destroy, rc);
2523 * in case of replay we just set LOVEA provided by the client
2524 * XXX: I think it would be interesting to try "old" way where
2525 * MDT calls this xattr_set(LOV) in a different transaction.
2526 * probably this way we code can be made better.
2529 /* During creation, there are only a few cases we need do xattr_set to
2531 * 1. regular file: see comments above.
2532 * 2. dir: inherit default striping or pool settings from parent.
2533 * 3. create striped directory with provided stripeEA.
2534 * 4. create striped directory because inherit default layout from the
2537 if (spec->no_create ||
2538 (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) ||
2539 S_ISDIR(attr->la_mode)) {
2540 buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata,
2541 spec->u.sp_ea.eadatalen);
2542 rc = mdo_xattr_set(env, son, buf,
2543 S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV :
2545 LU_XATTR_CREATE, handle);
2547 GOTO(err_destroy, rc);
2550 if (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_PCC) {
2553 memset(&mh, 0, sizeof(mh));
2554 mh.mh_flags = HS_EXISTS | HS_ARCHIVED | HS_RELEASED;
2555 mh.mh_arch_id = spec->sp_archive_id;
2556 lustre_hsm2buf(hsm_buf->lb_buf, &mh);
2557 rc = mdo_xattr_set(env, son, hsm_buf, XATTR_NAME_HSM,
2560 GOTO(err_destroy, rc);
2563 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
2564 if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 &&
2565 S_ISDIR(attr->la_mode)) {
2566 /* set default acl */
2567 rc = mdo_xattr_set(env, son, def_acl_buf,
2568 XATTR_NAME_ACL_DEFAULT, 0,
2571 GOTO(err_destroy, rc);
2573 /* set its own acl */
2574 if (acl_buf != NULL && acl_buf->lb_len > 0) {
2575 rc = mdo_xattr_set(env, son, acl_buf,
2576 XATTR_NAME_ACL_ACCESS,
2579 GOTO(err_destroy, rc);
2583 if (S_ISLNK(attr->la_mode)) {
2584 struct dt_object *dt = mdd_object_child(son);
2585 const char *target_name = spec->u.sp_symname.ln_name;
2586 int sym_len = spec->u.sp_symname.ln_namelen;
2589 buf = mdd_buf_get_const(env, target_name, sym_len);
2590 rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle);
2594 GOTO(err_initlized, rc = -EFAULT);
2597 if (initsecctx && spec->sp_cr_file_secctx_name != NULL) {
2598 buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx,
2599 spec->sp_cr_file_secctx_size);
2600 rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name,
2603 GOTO(err_initlized, rc);
2606 if (spec->sp_cr_file_encctx != NULL) {
2607 buf = mdd_buf_get_const(env, spec->sp_cr_file_encctx,
2608 spec->sp_cr_file_encctx_size);
2609 rc = mdo_xattr_set(env, son, buf,
2610 LL_XATTR_NAME_ENCRYPTION_CONTEXT, 0,
2613 GOTO(err_initlized, rc);
2617 if (unlikely(rc != 0)) {
2619 if (S_ISDIR(attr->la_mode)) {
2620 /* Drop the reference, no need to delete "."/"..",
2621 * because the object to be destroyed directly. */
2622 rc2 = mdo_ref_del(env, son, handle);
2626 rc2 = mdo_ref_del(env, son, handle);
2630 mdo_destroy(env, son, handle);
2633 mdd_write_unlock(env, son);
2637 static int mdd_index_delete(const struct lu_env *env,
2638 struct mdd_object *mdd_pobj,
2639 struct lu_attr *cattr,
2640 const struct lu_name *lname)
2642 struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj);
2643 struct thandle *handle;
2647 handle = mdd_trans_create(env, mdd);
2649 RETURN(PTR_ERR(handle));
2651 rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name,
2656 if (S_ISDIR(cattr->la_mode)) {
2657 rc = mdo_declare_ref_del(env, mdd_pobj, handle);
2662 /* Since this will only be used in the error handler path,
2663 * Let's set the thandle to be local and not mess the transno */
2664 handle->th_local = 1;
2665 rc = mdd_trans_start(env, mdd, handle);
2669 rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name,
2670 S_ISDIR(cattr->la_mode), handle);
2674 rc = mdd_trans_stop(env, mdd, rc, handle);
2680 * Create object and insert it into namespace.
2682 * Two operations have to be performed:
2684 * - an allocation of a new object (->do_create()), and
2685 * - an insertion into a parent index (->dio_insert()).
2687 * Due to locking, operation order is not important, when both are
2688 * successful, *but* error handling cases are quite different:
2690 * - if insertion is done first, and following object creation fails,
2691 * insertion has to be rolled back, but this operation might fail
2692 * also leaving us with dangling index entry.
2694 * - if creation is done first, is has to be undone if insertion fails,
2695 * leaving us with leaked space, which is not good but not fatal.
2697 * It seems that creation-first is simplest solution, but it is sub-optimal
2703 * case, because second mkdir is bound to create object, only to
2704 * destroy it immediately.
2706 * To avoid this follow local file systems that do double lookup:
2708 * 0. lookup -> -EEXIST (mdd_create_sanity_check())
2709 * 1. create (mdd_create_object_internal())
2710 * 2. insert (__mdd_index_insert(), lookup again)
2712 * \param[in] pobj parent object
2713 * \param[in] lname name of child being created
2714 * \param[in,out] child child object being created
2715 * \param[in] spec additional create parameters
2716 * \param[in] ma attributes for new child object
2718 * \retval 0 on success
2719 * \retval negative errno on failure
2721 int mdd_create(const struct lu_env *env, struct md_object *pobj,
2722 const struct lu_name *lname, struct md_object *child,
2723 struct md_op_spec *spec, struct md_attr *ma)
2725 struct mdd_thread_info *info = mdd_env_info(env);
2726 struct lu_attr *la = &info->mdi_la_for_fix;
2727 struct mdd_object *mdd_pobj = md2mdd_obj(pobj);
2728 struct mdd_object *son = md2mdd_obj(child);
2729 struct mdd_device *mdd = mdo2mdd(pobj);
2730 struct lu_attr *attr = &ma->ma_attr;
2731 struct thandle *handle;
2732 struct lu_attr *pattr = &info->mdi_pattr;
2733 struct lu_buf acl_buf;
2734 struct lu_buf def_acl_buf;
2735 struct lu_buf hsm_buf;
2736 struct linkea_data *ldata = &info->mdi_link_data;
2737 const char *name = lname->ln_name;
2738 struct dt_allocation_hint *hint = &mdd_env_info(env)->mdi_hint;
2739 int acl_size = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
2744 rc = mdd_la_get(env, mdd_pobj, pattr);
2748 /* Sanity checks before big job. */
2749 rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec);
2753 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET))
2754 GOTO(out_free, rc = -EINPROGRESS);
2756 handle = mdd_trans_create(env, mdd);
2758 GOTO(out_free, rc = PTR_ERR(handle));
2761 acl_buf = *lu_buf_check_and_alloc(&info->mdi_xattr_buf, acl_size);
2762 if (!acl_buf.lb_buf)
2763 GOTO(out_stop, rc = -ENOMEM);
2765 def_acl_buf = *lu_buf_check_and_alloc(&info->mdi_big_buf, acl_size);
2766 if (!def_acl_buf.lb_buf)
2767 GOTO(out_stop, rc = -ENOMEM);
2769 rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf);
2770 if (unlikely(rc == -ERANGE &&
2771 acl_size == LUSTRE_POSIX_ACL_MAX_SIZE_OLD)) {
2772 /* use maximum-sized xattr buffer for too-big default ACL */
2773 acl_size = min_t(unsigned int, mdd->mdd_dt_conf.ddp_max_ea_size,
2775 goto use_bigger_buffer;
2780 if (S_ISDIR(attr->la_mode)) {
2781 struct lmv_user_md *lmu = spec->u.sp_ea.eadata;
2784 * migrate may create 1-stripe directory, so lod_ah_init()
2785 * doesn't adjust stripe count from lmu.
2787 if (lmu && lmu->lum_stripe_count == cpu_to_le32(1))
2788 lmu->lum_stripe_count = 0;
2791 mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint);
2793 memset(ldata, 0, sizeof(*ldata));
2794 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
2795 struct lu_fid tfid = *mdd_object_fid(mdd_pobj);
2798 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2799 &tfid, lname, 1, 0, ldata);
2801 rc = mdd_linkea_prepare(env, son, NULL, NULL,
2802 mdd_object_fid(mdd_pobj),
2803 lname, 1, 0, ldata);
2806 if (spec->sp_cr_flags & MDS_OPEN_PCC) {
2807 LASSERT(spec->sp_cr_flags & MDS_OPEN_HAS_EA);
2809 memset(&hsm_buf, 0, sizeof(hsm_buf));
2810 lu_buf_alloc(&hsm_buf, sizeof(struct hsm_attrs));
2811 if (hsm_buf.lb_buf == NULL)
2812 GOTO(out_stop, rc = -ENOMEM);
2815 rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr,
2816 handle, spec, ldata, &def_acl_buf, &acl_buf,
2821 rc = mdd_trans_start(env, mdd, handle);
2825 rc = mdd_create_object(env, mdd_pobj, son, attr, spec, &acl_buf,
2826 &def_acl_buf, &hsm_buf, hint, handle, true);
2830 if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) {
2831 mdd_write_lock(env, son, DT_TGT_CHILD);
2832 son->mod_flags |= VOLATILE_OBJ;
2833 rc = mdd_orphan_insert(env, son, handle);
2834 GOTO(out_volatile, rc);
2836 rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(son),
2837 attr->la_mode, name, handle);
2839 GOTO(err_created, rc);
2841 mdd_links_add(env, son, mdd_object_fid(mdd_pobj), lname,
2844 /* update parent directory mtime/ctime */
2846 la->la_valid = LA_CTIME | LA_MTIME;
2847 rc = mdd_update_time(env, mdd_pobj, pattr, la, handle);
2849 GOTO(err_insert, rc);
2855 if (spec->sp_cr_flags & MDS_OPEN_VOLATILE)
2856 rc2 = mdd_orphan_delete(env, son, handle);
2858 rc2 = __mdd_index_delete(env, mdd_pobj, name,
2859 S_ISDIR(attr->la_mode),
2865 mdd_write_lock(env, son, DT_TGT_CHILD);
2866 if (S_ISDIR(attr->la_mode)) {
2867 /* Drop the reference, no need to delete "."/"..",
2868 * because the object is to be destroyed directly. */
2869 rc2 = mdo_ref_del(env, son, handle);
2871 mdd_write_unlock(env, son);
2876 /* For volatile files drop one link immediately, since there is
2877 * no filename in the namespace, and save any error returned. */
2878 rc2 = mdo_ref_del(env, son, handle);
2880 mdd_write_unlock(env, son);
2881 if (unlikely(rc == 0))
2886 /* Don't destroy the volatile object on success */
2887 if (likely(rc != 0))
2888 mdo_destroy(env, son, handle);
2889 mdd_write_unlock(env, son);
2892 if (rc == 0 && fid_is_namespace_visible(mdd_object_fid(son)) &&
2893 likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0))
2894 rc = mdd_changelog_ns_store(env, mdd,
2895 S_ISDIR(attr->la_mode) ? CL_MKDIR :
2896 S_ISREG(attr->la_mode) ? CL_CREATE :
2897 S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
2898 0, son, mdd_object_fid(mdd_pobj), NULL, NULL,
2899 lname, NULL, handle);
2901 rc2 = mdd_trans_stop(env, mdd, rc, handle);
2903 /* If creation fails, it is most likely due to the remote update
2904 * failure, because local transaction will mostly succeed at
2905 * this stage. There is no easy way to rollback all of previous
2906 * updates, so let's remove the object from namespace, and
2907 * LFSCK should handle the orphan object. */
2908 if (rc2 < 0 && !mdd_object_remote(mdd_pobj))
2909 mdd_index_delete(env, mdd_pobj, attr, lname);
2913 if (is_vmalloc_addr(ldata->ld_buf))
2914 /* if we vmalloced a large buffer drop it */
2915 lu_buf_free(ldata->ld_buf);
2917 if (spec->sp_cr_flags & MDS_OPEN_PCC)
2918 lu_buf_free(&hsm_buf);
2920 /* The child object shouldn't be cached anymore */
2922 set_bit(LU_OBJECT_HEARD_BANSHEE,
2923 &child->mo_lu.lo_header->loh_flags);
2927 /* has not mdd_write{read}_lock on any obj yet. */
2928 static int mdd_rename_sanity_check(const struct lu_env *env,
2929 struct mdd_object *src_pobj,
2930 const struct lu_attr *pattr,
2931 struct mdd_object *tgt_pobj,
2932 const struct lu_attr *tpattr,
2933 struct mdd_object *sobj,
2934 const struct lu_attr *cattr,
2935 struct mdd_object *tobj,
2936 const struct lu_attr *tattr)
2941 /* XXX: when get here, sobj must NOT be NULL,
2942 * the other case has been processed in cld_rename
2943 * before mdd_rename and enable MDS_PERM_BYPASS. */
2947 * If we are using project inheritance, we only allow renames
2948 * into our tree when the project IDs are the same; otherwise
2949 * tree quota mechanism would be circumvented.
2951 if ((((tpattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2952 tpattr->la_projid != cattr->la_projid) ||
2953 ((pattr->la_flags & LUSTRE_PROJINHERIT_FL) &&
2954 (pattr->la_projid != tpattr->la_projid))) &&
2955 S_ISDIR(cattr->la_mode))
2958 /* we prevent an encrypted file from being renamed
2959 * into an unencrypted dir
2961 if ((pattr->la_valid & LA_FLAGS &&
2962 pattr->la_flags & LUSTRE_ENCRYPT_FL) &&
2963 !(tpattr->la_valid & LA_FLAGS &&
2964 tpattr->la_flags & LUSTRE_ENCRYPT_FL))
2967 rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0);
2971 /* XXX: when get here, "tobj == NULL" means tobj must
2972 * NOT exist (neither on remote MDS, such case has been
2973 * processed in cld_rename before mdd_rename and enable
2975 * So check may_create, but not check may_unlink. */
2977 rc = mdd_may_create(env, tgt_pobj, tpattr, NULL,
2978 (src_pobj != tgt_pobj));
2980 rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr,
2981 (src_pobj != tgt_pobj), 1);
2983 if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode))
2984 rc = __mdd_may_link(env, tgt_pobj, tpattr);
2989 static int mdd_declare_rename(const struct lu_env *env,
2990 struct mdd_device *mdd,
2991 struct mdd_object *mdd_spobj,
2992 struct mdd_object *mdd_tpobj,
2993 struct mdd_object *mdd_sobj,
2994 struct mdd_object *mdd_tobj,
2995 const struct lu_name *sname,
2996 const struct lu_name *tname,
2998 struct linkea_data *ldata, bool change_projid,
2999 struct thandle *handle)
3001 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
3004 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3005 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3011 /* name from source dir */
3012 rc = mdo_declare_index_delete(env, mdd_spobj, sname->ln_name, handle);
3016 /* .. from source child */
3017 if (S_ISDIR(mdd_object_type(mdd_sobj))) {
3018 /* source child can be directory, count by source dir's nlink */
3019 rc = mdo_declare_ref_del(env, mdd_spobj, handle);
3022 if (mdd_spobj != mdd_tpobj) {
3023 rc = mdo_declare_index_delete(env, mdd_sobj, dotdot,
3028 rc = mdo_declare_index_insert(env, mdd_sobj,
3029 mdd_object_fid(mdd_tpobj),
3030 S_IFDIR, dotdot, handle);
3035 /* new target child can be directory,
3036 * counted by target dir's nlink */
3037 rc = mdo_declare_ref_add(env, mdd_tpobj, handle);
3042 la->la_valid = LA_CTIME | LA_MTIME;
3043 rc = mdo_declare_attr_set(env, mdd_spobj, la, handle);
3047 rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle);
3051 la->la_valid = LA_CTIME;
3053 la->la_valid |= LA_PROJID;
3054 rc = mdo_declare_attr_set(env, mdd_sobj, la, handle);
3058 rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata);
3063 rc = mdo_declare_index_insert(env, mdd_tpobj, mdd_object_fid(mdd_sobj),
3064 mdd_object_type(mdd_sobj),
3065 tname->ln_name, handle);
3069 if (mdd_tobj && mdd_object_exists(mdd_tobj)) {
3070 /* delete target child in target parent directory */
3071 rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name,
3076 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
3080 if (S_ISDIR(mdd_object_type(mdd_tobj))) {
3081 /* target child can be directory,
3082 * delete "." reference in target child directory */
3083 rc = mdo_declare_ref_del(env, mdd_tobj, handle);
3087 /* delete ".." reference in target parent directory */
3088 rc = mdo_declare_ref_del(env, mdd_tpobj, handle);
3093 la->la_valid = LA_CTIME;
3094 rc = mdo_declare_attr_set(env, mdd_tobj, la, handle);
3098 rc = mdd_declare_finish_unlink(env, mdd_tobj, handle);
3103 rc = mdd_declare_changelog_store(env, mdd, CL_RENAME, tname, sname,
3111 static int mdd_migrate_object(const struct lu_env *env,
3112 struct mdd_object *spobj,
3113 struct mdd_object *tpobj,
3114 struct mdd_object *sobj,
3115 struct mdd_object *tobj,
3116 const struct lu_name *sname,
3117 const struct lu_name *tname,
3118 struct md_op_spec *spec,
3119 struct md_attr *ma);
3121 /* src object can be remote that is why we use only fid and type of object */
3122 static int mdd_rename(const struct lu_env *env,
3123 struct md_object *src_pobj, struct md_object *tgt_pobj,
3124 const struct lu_fid *lf, const struct lu_name *lsname,
3125 struct md_object *tobj, const struct lu_name *ltname,
3128 const char *sname = lsname->ln_name;
3129 const char *tname = ltname->ln_name;
3130 struct lu_attr *la = &mdd_env_info(env)->mdi_la_for_fix;
3131 struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */
3132 struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj);
3133 struct mdd_device *mdd = mdo2mdd(src_pobj);
3134 struct mdd_object *mdd_sobj = NULL; /* source object */
3135 struct mdd_object *mdd_tobj = NULL;
3136 struct lu_attr *cattr = MDD_ENV_VAR(env, cattr);
3137 struct lu_attr *pattr = MDD_ENV_VAR(env, pattr);
3138 struct lu_attr *tattr = MDD_ENV_VAR(env, tattr);
3139 struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr);
3140 struct thandle *handle;
3141 struct linkea_data *ldata = &mdd_env_info(env)->mdi_link_data;
3142 const struct lu_fid *tpobj_fid = mdd_object_fid(mdd_tpobj);
3143 const struct lu_fid *spobj_fid = mdd_object_fid(mdd_spobj);
3146 bool tobj_locked = 0;
3147 bool change_projid = false;
3148 unsigned cl_flags = 0;
3152 /* let unlink to complete and commit */
3153 CFS_FAIL_TIMEOUT(OBD_FAIL_TGT_REPLY_DATA_RACE, 2 + cfs_fail_val);
3156 mdd_tobj = md2mdd_obj(tobj);
3158 mdd_sobj = mdd_object_find(env, mdd, lf);
3159 if (IS_ERR(mdd_sobj))
3160 RETURN(PTR_ERR(mdd_sobj));
3162 rc = mdd_la_get(env, mdd_sobj, cattr);
3164 GOTO(out_pending, rc);
3166 /* if rename is cross MDTs, migrate symlink if it doesn't have other
3167 * hard links, and target doesn't exist.
3169 if (mdd_object_remote(mdd_sobj) && S_ISLNK(cattr->la_mode) &&
3170 cattr->la_nlink == 1 && !tobj) {
3171 struct md_op_spec *spec = &mdd_env_info(env)->mdi_spec;
3172 struct lu_device *ld = &mdd->mdd_md_dev.md_lu_dev;
3175 rc = ld->ld_ops->ldo_fid_alloc(env, ld, &tfid, &tgt_pobj->mo_lu,
3178 GOTO(out_pending, rc);
3180 mdd_tobj = mdd_object_find(env, mdd, &tfid);
3181 if (IS_ERR(mdd_tobj))
3182 GOTO(out_pending, rc = PTR_ERR(mdd_tobj));
3184 memset(spec, 0, sizeof(*spec));
3185 rc = mdd_migrate_object(env, mdd_spobj, mdd_tpobj, mdd_sobj,
3186 mdd_tobj, lsname, ltname, spec, ma);
3187 mdd_object_put(env, mdd_tobj);
3188 GOTO(out_pending, rc);
3191 rc = mdd_la_get(env, mdd_spobj, pattr);
3193 GOTO(out_pending, rc);
3196 rc = mdd_la_get(env, mdd_tobj, tattr);
3198 GOTO(out_pending, rc);
3199 /* search for an existing archive.
3200 * we should check ahead as the object
3201 * can be destroyed in this transaction */
3202 if (mdd_hsm_archive_exists(env, mdd_tobj, ma))
3203 cl_flags |= CLF_RENAME_LAST_EXISTS;
3206 rc = mdd_la_get(env, mdd_tpobj, tpattr);
3208 GOTO(out_pending, rc);
3210 rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr,
3211 mdd_sobj, cattr, mdd_tobj, tattr);
3213 GOTO(out_pending, rc);
3215 rc = mdd_name_check(env, mdd, ltname);
3217 GOTO(out_pending, rc);
3219 handle = mdd_trans_create(env, mdd);
3221 GOTO(out_pending, rc = PTR_ERR(handle));
3223 memset(ldata, 0, sizeof(*ldata));
3224 rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3225 lsname, mdd_object_fid(mdd_tpobj), ltname,
3230 if (tpattr->la_projid != cattr->la_projid &&
3231 tpattr->la_flags & LUSTRE_PROJINHERIT_FL)
3232 change_projid = true;
3234 rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj,
3235 mdd_tobj, lsname, ltname, ma, ldata,
3236 change_projid, handle);
3240 rc = mdd_trans_start(env, mdd, handle);
3244 is_dir = S_ISDIR(cattr->la_mode);
3246 /* Remove source name from source directory */
3247 rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle);
3251 /* "mv dir1 dir2" needs "dir1/.." link update */
3252 if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) {
3253 rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3255 GOTO(fixup_spobj2, rc);
3257 rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR,
3260 GOTO(fixup_spobj, rc);
3263 if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) {
3264 rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3266 /* tname might been renamed to something else */
3267 GOTO(fixup_spobj, rc);
3270 /* Insert new fid with target name into target dir */
3271 rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode,
3274 GOTO(fixup_tpobj, rc);
3276 LASSERT(ma->ma_attr.la_valid & LA_CTIME);
3277 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
3279 /* XXX: mdd_sobj must be local one if it is NOT NULL. */
3280 la->la_valid = LA_CTIME;
3281 if (change_projid) {
3282 /* mdd_update_time honors other valid flags except TIME ones */
3283 la->la_valid |= LA_PROJID;
3284 la->la_projid = tpattr->la_projid;
3286 rc = mdd_update_time(env, mdd_sobj, cattr, la, handle);
3288 GOTO(fixup_tpobj, rc);
3290 /* Update the linkEA for the source object */
3291 mdd_write_lock(env, mdd_sobj, DT_SRC_CHILD);
3292 rc = mdd_links_rename(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3293 lsname, mdd_object_fid(mdd_tpobj), ltname,
3294 handle, ldata, 0, 0);
3296 /* Old files might not have EA entry */
3297 mdd_links_add(env, mdd_sobj, mdd_object_fid(mdd_spobj),
3298 lsname, handle, NULL, 0);
3299 mdd_write_unlock(env, mdd_sobj);
3300 /* We don't fail the transaction if the link ea can't be
3301 updated -- fid2path will use alternate lookup method. */
3304 /* Remove old target object
3305 * For tobj is remote case cmm layer has processed
3306 * and set tobj to NULL then. So when tobj is NOT NULL,
3307 * it must be local one.
3309 if (tobj && mdd_object_exists(mdd_tobj)) {
3310 mdd_write_lock(env, mdd_tobj, DT_TGT_CHILD);
3312 if (mdd_is_dead_obj(mdd_tobj)) {
3313 /* shld not be dead, something is wrong */
3314 CERROR("tobj is dead, something is wrong\n");
3318 mdo_ref_del(env, mdd_tobj, handle);
3320 /* Remove dot reference. */
3321 if (S_ISDIR(tattr->la_mode))
3322 mdo_ref_del(env, mdd_tobj, handle);
3325 /* fetch updated nlink */
3326 rc = mdd_la_get(env, mdd_tobj, tattr);
3328 CERROR("%s: Failed to get nlink for tobj "
3330 mdd2obd_dev(mdd)->obd_name,
3331 PFID(tpobj_fid), rc);
3332 GOTO(fixup_tpobj, rc);
3335 la->la_valid = LA_CTIME;
3336 rc = mdd_update_time(env, mdd_tobj, tattr, la, handle);
3338 CERROR("%s: Failed to set ctime for tobj "
3340 mdd2obd_dev(mdd)->obd_name,
3341 PFID(tpobj_fid), rc);
3342 GOTO(fixup_tpobj, rc);
3345 /* XXX: this transfer to ma will be removed with LOD/OSP */
3346 ma->ma_attr = *tattr;
3347 ma->ma_valid |= MA_INODE;
3348 rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname,
3351 CERROR("%s: Failed to unlink tobj "
3353 mdd2obd_dev(mdd)->obd_name,
3354 PFID(tpobj_fid), rc);
3355 GOTO(fixup_tpobj, rc);
3358 /* fetch updated nlink */
3359 rc = mdd_la_get(env, mdd_tobj, tattr);
3360 if (rc == -ENOENT) {
3361 /* the object got removed, let's
3362 * return the latest known attributes */
3363 tattr->la_nlink = 0;
3365 } else if (rc != 0) {
3366 CERROR("%s: Failed to get nlink for tobj "
3368 mdd2obd_dev(mdd)->obd_name,
3369 PFID(tpobj_fid), rc);
3370 GOTO(fixup_tpobj, rc);
3372 /* XXX: this transfer to ma will be removed with LOD/OSP */
3373 ma->ma_attr = *tattr;
3374 ma->ma_valid |= MA_INODE;
3376 if (tattr->la_nlink == 0)
3377 cl_flags |= CLF_RENAME_LAST;
3379 cl_flags &= ~CLF_RENAME_LAST_EXISTS;
3382 la->la_valid = LA_CTIME | LA_MTIME;
3383 rc = mdd_update_time(env, mdd_spobj, pattr, la, handle);
3385 GOTO(fixup_tpobj, rc);
3387 if (mdd_spobj != mdd_tpobj) {
3388 la->la_valid = LA_CTIME | LA_MTIME;
3389 rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle);
3391 GOTO(fixup_tpobj, rc);
3398 rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle);
3400 CWARN("tp obj fix error %d\n",rc2);
3402 if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
3403 !mdd_is_dead_obj(mdd_tobj)) {
3405 mdo_ref_add(env, mdd_tobj, handle);
3407 mdo_ref_add(env, mdd_tobj, handle);
3410 rc2 = __mdd_index_insert(env, mdd_tpobj,
3411 mdd_object_fid(mdd_tobj),
3412 mdd_object_type(mdd_tobj),
3415 CWARN("tp obj fix error: rc = %d\n", rc2);
3420 if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) {
3421 rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle);
3423 CWARN("%s: sp obj dotdot delete error: rc = %d\n",
3424 mdd2obd_dev(mdd)->obd_name, rc2);
3427 rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR,
3430 CWARN("%s: sp obj dotdot insert error: rc = %d\n",
3431 mdd2obd_dev(mdd)->obd_name, rc2);
3436 rc2 = __mdd_index_insert(env, mdd_spobj, lf,
3437 mdd_object_type(mdd_sobj), sname,
3440 CWARN("sp obj fix error: rc = %d\n", rc2);
3445 mdd_write_unlock(env, mdd_tobj);
3448 rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags,
3449 mdd_tobj, tpobj_fid, lf, spobj_fid,
3450 ltname, lsname, handle);
3453 rc = mdd_trans_stop(env, mdd, rc, handle);
3456 mdd_object_put(env, mdd_sobj);
3461 * Check whether we should migrate the file/dir
3463 * < 0 permission check failed or other error.
3464 * = 0 the file can be migrated.
3466 static int mdd_migrate_sanity_check(const struct lu_env *env,
3467 struct mdd_device *mdd,
3468 struct mdd_object *spobj,
3469 struct mdd_object *tpobj,
3470 struct mdd_object *sobj,
3471 struct mdd_object *tobj,
3472 const struct lu_attr *spattr,
3473 const struct lu_attr *tpattr,
3474 const struct lu_attr *attr)
3480 if (!mdd_object_remote(sobj)) {
3481 mdd_read_lock(env, sobj, DT_SRC_CHILD);
3482 if (sobj->mod_count > 0) {
3483 CDEBUG(D_INFO, "%s: "DFID" is opened, count %d\n",
3484 mdd_obj_dev_name(sobj),
3485 PFID(mdd_object_fid(sobj)),
3487 mdd_read_unlock(env, sobj);
3490 mdd_read_unlock(env, sobj);
3493 if (mdd_object_exists(tobj))
3496 rc = mdd_may_delete(env, spobj, spattr, sobj, attr, NULL, 1, 0);
3500 rc = mdd_may_create(env, tpobj, tpattr, NULL, true);
3505 struct mdd_xattr_entry {
3506 struct list_head mxe_linkage;
3508 struct lu_buf mxe_buf;
3512 struct lu_buf mx_namebuf;
3513 struct list_head mx_list;
3516 static inline void mdd_xattrs_init(struct mdd_xattrs *xattrs)
3518 INIT_LIST_HEAD(&xattrs->mx_list);
3519 xattrs->mx_namebuf.lb_buf = NULL;
3520 xattrs->mx_namebuf.lb_len = 0;
3523 static inline void mdd_xattrs_fini(struct mdd_xattrs *xattrs)
3525 struct mdd_xattr_entry *entry;
3526 struct mdd_xattr_entry *tmp;
3528 list_for_each_entry_safe(entry, tmp, &xattrs->mx_list, mxe_linkage) {
3529 lu_buf_free(&entry->mxe_buf);
3530 list_del(&entry->mxe_linkage);
3531 OBD_FREE_PTR(entry);
3534 lu_buf_free(&xattrs->mx_namebuf);
3537 /* read xattrs into buf, but skip LMA, LMV, LINKEA if 'skip_linkea' is
3538 * set, and DMV if 'skip_dmv" is set.
3540 static int mdd_xattrs_migrate_prep(const struct lu_env *env,
3541 struct mdd_xattrs *xattrs,
3542 struct mdd_object *sobj,
3546 struct lu_attr *attr = MDD_ENV_VAR(env, cattr);
3547 struct mdd_xattr_entry *entry;
3548 bool needencxattr = false;
3549 bool encxattrfound = false;
3559 list_xsize = mdo_xattr_list(env, sobj, &LU_BUF_NULL);
3560 if (list_xsize == -ENODATA)
3566 if (attr->la_valid & LA_FLAGS &&
3567 attr->la_flags & LUSTRE_ENCRYPT_FL) {
3568 needencxattr = true;
3570 strlen(LL_XATTR_NAME_ENCRYPTION_CONTEXT) + 1;
3573 lu_buf_alloc(&xattrs->mx_namebuf, list_xsize);
3574 if (xattrs->mx_namebuf.lb_buf == NULL)
3577 rc = mdo_xattr_list(env, sobj, &xattrs->mx_namebuf);
3583 xname = xattrs->mx_namebuf.lb_buf;
3585 for (; rem > 0; xname += xlen, rem -= xlen) {
3587 strcmp(xname, LL_XATTR_NAME_ENCRYPTION_CONTEXT) == 0)
3588 encxattrfound = true;
3589 xlen = strnlen(xname, rem - 1) + 1;
3590 if (strcmp(XATTR_NAME_LMA, xname) == 0 ||
3591 strcmp(XATTR_NAME_LMV, xname) == 0)
3595 strcmp(XATTR_NAME_LINK, xname) == 0)
3599 strcmp(XATTR_NAME_DEFAULT_LMV, xname) == 0)
3602 xsize = mdo_xattr_get(env, sobj, &LU_BUF_NULL, xname);
3603 if (xsize == -ENODATA)
3606 GOTO(fini, rc = xsize);
3608 OBD_ALLOC_PTR(entry);
3610 GOTO(fini, rc = -ENOMEM);
3612 lu_buf_alloc(&entry->mxe_buf, xsize);
3613 if (!entry->mxe_buf.lb_buf) {
3614 OBD_FREE_PTR(entry);
3615 GOTO(fini, rc = -ENOMEM);
3618 rc = mdo_xattr_get(env, sobj, &entry->mxe_buf, xname);
3620 lu_buf_free(&entry->mxe_buf);
3621 OBD_FREE_PTR(entry);
3627 entry->mxe_name = xname;
3628 list_add_tail(&entry->mxe_linkage, &xattrs->mx_list);
3631 if (needencxattr && !encxattrfound) {
3632 xlen = strlen(LL_XATTR_NAME_ENCRYPTION_CONTEXT) + 1;
3633 strncpy(xname, LL_XATTR_NAME_ENCRYPTION_CONTEXT, xlen);
3640 mdd_xattrs_fini(xattrs);
3644 typedef int (*mdd_xattr_cb)(const struct lu_env *env,
3645 struct mdd_object *obj,
3646 const struct lu_buf *buf,
3648 int fl, struct thandle *handle);
3650 static int mdd_foreach_xattr(const struct lu_env *env,
3651 struct mdd_object *tobj,
3652 struct mdd_xattrs *xattrs,
3653 struct thandle *handle,
3656 struct mdd_xattr_entry *entry;
3659 list_for_each_entry(entry, &xattrs->mx_list, mxe_linkage) {
3660 rc = cb(env, tobj, &entry->mxe_buf, entry->mxe_name, 0, handle);
3668 typedef int (*mdd_linkea_cb)(const struct lu_env *env,
3669 struct mdd_object *sobj,
3670 struct mdd_object *tobj,
3671 const struct lu_name *sname,
3672 const struct lu_fid *sfid,
3673 const struct lu_name *lname,
3674 const struct lu_fid *fid,
3676 struct thandle *handle);
3678 static int mdd_declare_update_link(const struct lu_env *env,
3679 struct mdd_object *sobj,
3680 struct mdd_object *tobj,
3681 const struct lu_name *tname,
3682 const struct lu_fid *tpfid,
3683 const struct lu_name *lname,
3684 const struct lu_fid *fid,
3686 struct thandle *handle)
3688 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3689 struct mdd_object *pobj;
3693 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3694 !strcmp(tname->ln_name, lname->ln_name))
3697 pobj = mdd_object_find(env, mdd, fid);
3699 return PTR_ERR(pobj);
3702 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
3704 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
3705 mdd_object_type(sobj),
3706 lname->ln_name, handle);
3707 mdd_object_put(env, pobj);
3711 rc = mdo_declare_ref_add(env, tobj, handle);
3715 rc = mdo_declare_ref_del(env, sobj, handle);
3719 static int mdd_update_link(const struct lu_env *env,
3720 struct mdd_object *sobj,
3721 struct mdd_object *tobj,
3722 const struct lu_name *tname,
3723 const struct lu_fid *tpfid,
3724 const struct lu_name *lname,
3725 const struct lu_fid *fid,
3727 struct thandle *handle)
3729 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3730 struct mdd_object *pobj;
3736 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3737 !memcmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3740 CDEBUG(D_INFO, "update "DFID"/"DNAME":"DFID"\n",
3741 PFID(fid), PNAME(lname), PFID(mdd_object_fid(tobj)));
3743 pobj = mdd_object_find(env, mdd, fid);
3745 CWARN("%s: cannot find obj "DFID": %ld\n",
3746 mdd2obd_dev(mdd)->obd_name, PFID(fid), PTR_ERR(pobj));
3747 RETURN(PTR_ERR(pobj));
3750 if (!mdd_object_exists(pobj)) {
3751 CDEBUG(D_INFO, DFID" doesn't exist\n", PFID(fid));
3752 mdd_object_put(env, pobj);
3756 mdd_write_lock(env, pobj, DT_TGT_PARENT);
3757 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
3759 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(tobj),
3760 mdd_object_type(sobj),
3761 lname->ln_name, handle);
3762 mdd_write_unlock(env, pobj);
3763 mdd_object_put(env, pobj);
3767 mdd_write_lock(env, tobj, DT_TGT_CHILD);
3768 rc = mdo_ref_add(env, tobj, handle);
3769 mdd_write_unlock(env, tobj);
3773 mdd_write_lock(env, sobj, DT_SRC_CHILD);
3774 rc = mdo_ref_del(env, sobj, handle);
3775 mdd_write_unlock(env, sobj);
3780 static inline int mdd_fld_lookup(const struct lu_env *env,
3781 struct mdd_device *mdd,
3782 const struct lu_fid *fid,
3785 struct lu_seq_range *range = &mdd_env_info(env)->mdi_range;
3786 struct seq_server_site *ss;
3789 ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site;
3791 range->lsr_flags = LU_SEQ_RANGE_MDT;
3792 rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range);
3796 *mdt_index = range->lsr_index;
3801 static int mdd_is_link_on_source_mdt(const struct lu_env *env,
3802 struct mdd_object *sobj,
3803 struct mdd_object *tobj,
3804 const struct lu_name *tname,
3805 const struct lu_fid *tpfid,
3806 const struct lu_name *lname,
3807 const struct lu_fid *fid,
3809 struct thandle *handle)
3811 struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj);
3812 __u32 source_mdt_index = *(__u32 *)opaque;
3813 __u32 link_mdt_index;
3819 if (lu_fid_eq(tpfid, fid) && tname->ln_namelen == lname->ln_namelen &&
3820 !memcmp(tname->ln_name, lname->ln_name, lname->ln_namelen))
3823 rc = mdd_fld_lookup(env, mdd, fid, &link_mdt_index);
3827 RETURN(link_mdt_index == source_mdt_index);
3830 static int mdd_iterate_linkea(const struct lu_env *env,
3831 struct mdd_object *sobj,
3832 struct mdd_object *tobj,
3833 const struct lu_name *tname,
3834 const struct lu_fid *tpfid,
3835 struct linkea_data *ldata,
3837 struct thandle *handle,
3840 struct mdd_thread_info *info = mdd_env_info(env);
3841 char *filename = info->mdi_name;
3842 struct lu_name lname;
3849 for (linkea_first_entry(ldata); ldata->ld_lee && !rc;
3850 linkea_next_entry(ldata)) {
3851 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
3854 /* Note: lname might miss \0 at the end */
3855 snprintf(filename, sizeof(info->mdi_name), "%.*s",
3856 lname.ln_namelen, lname.ln_name);
3857 lname.ln_name = filename;
3859 CDEBUG(D_INFO, DFID"/"DNAME"\n", PFID(&fid), PNAME(&lname));
3861 rc = cb(env, sobj, tobj, tname, tpfid, &lname, &fid, opaque,
3869 * Prepare linkea, and check whether file needs migrate: if source still has
3870 * link on source MDT, no need to migrate, just update namespace on source and
3873 * \retval 0 do migrate
3874 * \retval 1 don't migrate
3875 * \retval -errno on failure
3877 static int mdd_migrate_linkea_prepare(const struct lu_env *env,
3878 struct mdd_device *mdd,
3879 struct mdd_object *spobj,
3880 struct mdd_object *tpobj,
3881 struct mdd_object *sobj,
3882 const struct lu_name *sname,
3883 const struct lu_name *tname,
3884 const struct lu_attr *attr,
3885 struct linkea_data *ldata)
3887 __u32 source_mdt_index;
3892 memset(ldata, 0, sizeof(*ldata));
3893 rc = mdd_linkea_prepare(env, sobj, mdd_object_fid(spobj), sname,
3894 mdd_object_fid(tpobj), tname, 1, 0, ldata);
3899 * Then it will check if the file should be migrated. If the file has
3900 * mulitple links, we only need migrate the file if all of its entries
3901 * has been migrated to the remote MDT.
3903 if (S_ISDIR(attr->la_mode) || attr->la_nlink < 2)
3906 /* If there are still links locally, don't migrate this file */
3907 LASSERT(ldata->ld_leh != NULL);
3910 * If linkEA is overflow, it means there are some unknown name entries
3911 * under unknown parents, which will prevent the migration.
3913 if (unlikely(ldata->ld_leh->leh_overflow_time))
3916 rc = mdd_fld_lookup(env, mdd, mdd_object_fid(sobj), &source_mdt_index);
3920 rc = mdd_iterate_linkea(env, sobj, NULL, tname, mdd_object_fid(tpobj),
3921 ldata, &source_mdt_index, NULL,
3922 mdd_is_link_on_source_mdt);
3926 static int mdd_declare_migrate_update(const struct lu_env *env,
3927 struct mdd_object *spobj,
3928 struct mdd_object *tpobj,
3929 struct mdd_object *obj,
3930 const struct lu_name *sname,
3931 const struct lu_name *tname,
3932 struct lu_attr *attr,
3933 struct lu_attr *spattr,
3934 struct lu_attr *tpattr,
3935 struct linkea_data *ldata,
3937 struct thandle *handle)
3939 struct mdd_thread_info *info = mdd_env_info(env);
3940 struct lu_attr *la = &info->mdi_la_for_fix;
3943 rc = mdo_declare_index_delete(env, spobj, sname->ln_name, handle);
3947 if (S_ISDIR(attr->la_mode)) {
3948 rc = mdo_declare_ref_del(env, spobj, handle);
3953 rc = mdo_declare_index_insert(env, tpobj, mdd_object_fid(obj),
3954 attr->la_mode & S_IFMT,
3955 tname->ln_name, handle);
3959 rc = mdd_declare_links_add(env, obj, handle, ldata);
3963 if (S_ISDIR(attr->la_mode)) {
3964 rc = mdo_declare_ref_add(env, tpobj, handle);
3969 la->la_valid = LA_CTIME | LA_MTIME;
3970 rc = mdo_declare_attr_set(env, spobj, la, handle);
3974 if (tpobj != spobj) {
3975 rc = mdo_declare_attr_set(env, tpobj, la, handle);
3983 static int mdd_declare_migrate_create(const struct lu_env *env,
3984 struct mdd_object *spobj,
3985 struct mdd_object *tpobj,
3986 struct mdd_object *sobj,
3987 struct mdd_object *tobj,
3988 const struct lu_name *sname,
3989 const struct lu_name *tname,
3990 struct lu_attr *spattr,
3991 struct lu_attr *tpattr,
3992 struct lu_attr *attr,
3993 struct lu_buf *sbuf,
3994 struct linkea_data *ldata,
3995 struct mdd_xattrs *xattrs,
3997 struct md_op_spec *spec,
3998 struct dt_allocation_hint *hint,
3999 struct thandle *handle)
4001 struct mdd_thread_info *info = mdd_env_info(env);
4002 struct md_layout_change *mlc = &info->mdi_mlc;
4003 struct lmv_mds_md_v1 *lmv = sbuf->lb_buf;
4008 if (S_ISDIR(attr->la_mode)) {
4009 struct lmv_user_md *lum = spec->u.sp_ea.eadata;
4011 mlc->mlc_opc = MD_LAYOUT_DETACH;
4012 rc = mdo_declare_layout_change(env, sobj, mlc, handle);
4016 lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_MIGRATION);
4017 } else if (S_ISLNK(attr->la_mode)) {
4018 spec->u.sp_symname.ln_name = sbuf->lb_buf;
4019 /* don't count NUL */
4020 spec->u.sp_symname.ln_namelen = sbuf->lb_len - 1;
4021 } else if (S_ISREG(attr->la_mode)) {
4022 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
4023 spec->sp_cr_flags &= ~MDS_OPEN_HAS_EA;
4026 mdd_object_make_hint(env, tpobj, tobj, attr, spec, hint);
4028 rc = mdd_declare_create(env, mdo2mdd(&tpobj->mod_obj), tpobj, tobj,
4029 tname, attr, handle, spec, ldata, NULL, NULL,
4035 * tobj mode will be used in mdo_declare_layout_change(), but it's not
4036 * createb yet, copy from sobj.
4038 tobj->mod_obj.mo_lu.lo_header->loh_attr &= ~S_IFMT;
4039 tobj->mod_obj.mo_lu.lo_header->loh_attr |=
4040 sobj->mod_obj.mo_lu.lo_header->loh_attr & S_IFMT;
4042 if (S_ISDIR(attr->la_mode)) {
4044 /* if sobj is not striped, fake a 1-stripe LMV */
4045 LASSERT(sizeof(info->mdi_key) >
4046 lmv_mds_md_size(1, LMV_MAGIC_V1));
4047 lmv = (typeof(lmv))info->mdi_key;
4048 memset(lmv, 0, sizeof(*lmv));
4049 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
4050 lmv->lmv_stripe_count = cpu_to_le32(1);
4051 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
4052 fid_le_to_cpu(&lmv->lmv_stripe_fids[0],
4053 mdd_object_fid(sobj));
4054 mlc->mlc_buf.lb_buf = lmv;
4055 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
4057 mlc->mlc_buf = *sbuf;
4059 mlc->mlc_opc = MD_LAYOUT_ATTACH;
4060 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4065 rc = mdd_foreach_xattr(env, tobj, xattrs, handle,
4066 mdo_declare_xattr_set);
4070 if (S_ISREG(attr->la_mode)) {
4071 struct lu_buf fid_buf;
4073 handle->th_complex = 1;
4075 /* target may be remote, update PFID via sobj. */
4076 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
4077 fid_buf.lb_len = sizeof(struct lu_fid);
4078 rc = mdo_declare_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID,
4083 rc = mdo_declare_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4088 if (!S_ISDIR(attr->la_mode)) {
4089 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
4090 mdd_object_fid(tpobj), ldata, NULL,
4091 handle, mdd_declare_update_link);
4096 if (!S_ISDIR(attr->la_mode) || lmv) {
4097 rc = mdo_declare_ref_del(env, sobj, handle);
4101 if (S_ISDIR(attr->la_mode)) {
4102 rc = mdo_declare_ref_del(env, sobj, handle);
4107 rc = mdo_declare_destroy(env, sobj, handle);
4112 rc = mdd_declare_migrate_update(env, spobj, tpobj, tobj, sname, tname,
4113 attr, spattr, tpattr, ldata, ma,
4119 * migrate dirent from \a spobj to \a tpobj.
4121 static int mdd_migrate_update(const struct lu_env *env,
4122 struct mdd_object *spobj,
4123 struct mdd_object *tpobj,
4124 struct mdd_object *obj,
4125 const struct lu_name *sname,
4126 const struct lu_name *tname,
4127 struct lu_attr *attr,
4128 struct lu_attr *spattr,
4129 struct lu_attr *tpattr,
4130 struct linkea_data *ldata,
4132 struct thandle *handle)
4134 struct mdd_thread_info *info = mdd_env_info(env);
4135 struct lu_attr *la = &info->mdi_la_for_fix;
4140 CDEBUG(D_INFO, "update "DFID" from "DFID"/%s to "DFID"/%s\n",
4141 PFID(mdd_object_fid(obj)), PFID(mdd_object_fid(spobj)),
4142 sname->ln_name, PFID(mdd_object_fid(tpobj)), tname->ln_name);
4144 rc = __mdd_index_delete(env, spobj, sname->ln_name,
4145 S_ISDIR(attr->la_mode), handle);
4149 rc = __mdd_index_insert(env, tpobj, mdd_object_fid(obj),
4150 attr->la_mode & S_IFMT,
4151 tname->ln_name, handle);
4155 rc = mdd_links_write(env, obj, ldata, handle);
4159 la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
4160 la->la_valid = LA_CTIME | LA_MTIME;
4161 mdd_write_lock(env, spobj, DT_SRC_PARENT);
4162 rc = mdd_update_time(env, spobj, spattr, la, handle);
4163 mdd_write_unlock(env, spobj);
4167 if (tpobj != spobj) {
4168 la->la_valid = LA_CTIME | LA_MTIME;
4169 mdd_write_lock(env, tpobj, DT_TGT_PARENT);
4170 rc = mdd_update_time(env, tpobj, tpattr, la, handle);
4171 mdd_write_unlock(env, tpobj);
4180 * Migrate file/dir to target MDT.
4182 * Create target according to \a spec, and then migrate xattrs, if it's
4183 * directory, migrate source stripes to target.
4185 * \param[in] env execution environment
4186 * \param[in] spobj source parent object
4187 * \param[in] tpobj target parent object
4188 * \param[in] sobj source object
4189 * \param[in] tobj target object
4190 * \param[in] lname file name
4191 * \param[in] spattr source parent attributes
4192 * \param[in] tpattr target parent attributes
4193 * \param[in] attr source attributes
4194 * \param[in] sbuf source LMV buf
4195 * \param[in] spec migrate create spec
4196 * \param[in] hint target creation hint
4197 * \param[in] handle tranasction handle
4199 * \retval 0 on success
4200 * \retval -errno on failure
4202 static int mdd_migrate_create(const struct lu_env *env,
4203 struct mdd_object *spobj,
4204 struct mdd_object *tpobj,
4205 struct mdd_object *sobj,
4206 struct mdd_object *tobj,
4207 const struct lu_name *sname,
4208 const struct lu_name *tname,
4209 struct lu_attr *spattr,
4210 struct lu_attr *tpattr,
4211 struct lu_attr *attr,
4212 const struct lu_buf *sbuf,
4213 struct linkea_data *ldata,
4214 struct mdd_xattrs *xattrs,
4216 struct md_op_spec *spec,
4217 struct dt_allocation_hint *hint,
4218 struct thandle *handle)
4225 * migrate sobj stripes to tobj if it's directory:
4226 * 1. detach stripes from sobj.
4227 * 2. attach stripes to tobj, see mdd_declare_migrate_mdt().
4228 * 3. create stripes for tobj, see lod_xattr_set_lmv().
4230 if (S_ISDIR(attr->la_mode)) {
4231 struct mdd_thread_info *info = mdd_env_info(env);
4232 struct md_layout_change *mlc = &info->mdi_mlc;
4234 mlc->mlc_opc = MD_LAYOUT_DETACH;
4236 mdd_write_lock(env, sobj, DT_SRC_PARENT);
4237 rc = mdo_layout_change(env, sobj, mlc, handle);
4238 mdd_write_unlock(env, sobj);
4243 /* don't set nlink from sobj */
4244 attr->la_valid &= ~LA_NLINK;
4246 rc = mdd_create_object(env, tpobj, tobj, attr, spec, NULL, NULL, NULL,
4247 hint, handle, false);
4251 mdd_write_lock(env, tobj, DT_TGT_CHILD);
4252 rc = mdd_foreach_xattr(env, tobj, xattrs, handle, mdo_xattr_set);
4253 mdd_write_unlock(env, tobj);
4257 /* for regular file, update OST objects XATTR_NAME_FID */
4258 if (S_ISREG(attr->la_mode)) {
4259 struct lu_buf fid_buf;
4261 /* target may be remote, update PFID via sobj. */
4262 fid_buf.lb_buf = (void *)mdd_object_fid(tobj);
4263 fid_buf.lb_len = sizeof(struct lu_fid);
4264 rc = mdo_xattr_set(env, sobj, &fid_buf, XATTR_NAME_FID, 0,
4269 /* delete LOV to avoid deleting OST objs when destroying sobj */
4270 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4271 rc = mdo_xattr_del(env, sobj, XATTR_NAME_LOV, handle);
4272 mdd_write_unlock(env, sobj);
4273 /* O_DELAY_CREATE file may not have LOV, ignore -ENODATA */
4274 if (rc && rc != -ENODATA)
4279 /* update links FID */
4280 if (!S_ISDIR(attr->la_mode)) {
4281 rc = mdd_iterate_linkea(env, sobj, tobj, tname,
4282 mdd_object_fid(tpobj), ldata,
4283 NULL, handle, mdd_update_link);
4288 /* don't destroy sobj if it's plain directory */
4289 if (!S_ISDIR(attr->la_mode) || sbuf->lb_buf) {
4290 mdd_write_lock(env, sobj, DT_SRC_CHILD);
4291 rc = mdo_ref_del(env, sobj, handle);
4293 if (S_ISDIR(attr->la_mode))
4294 rc = mdo_ref_del(env, sobj, handle);
4296 rc = mdo_destroy(env, sobj, handle);
4298 mdd_write_unlock(env, sobj);
4303 rc = mdd_migrate_update(env, spobj, tpobj, tobj, sname, tname, attr,
4304 spattr, tpattr, ldata, ma, handle);
4309 /* NB: if user issued different migrate command, we can't adjust it silently
4310 * here, because this command will decide target MDT in subdir migration in
4313 static int mdd_migrate_cmd_check(struct mdd_device *mdd,
4314 const struct lmv_mds_md_v1 *lmv,
4315 const struct lmv_user_md_v1 *lum,
4316 const struct lu_name *lname)
4318 __u32 lum_stripe_count = lum->lum_stripe_count;
4319 __u32 lum_hash_type = lum->lum_hash_type &
4320 cpu_to_le32(LMV_HASH_TYPE_MASK);
4321 __u32 lmv_hash_type = lmv->lmv_hash_type &
4322 cpu_to_le32(LMV_HASH_TYPE_MASK);
4324 if (!lmv_is_sane(lmv))
4327 /* if stripe_count unspecified, set to 1 */
4328 if (!lum_stripe_count)
4329 lum_stripe_count = cpu_to_le32(1);
4331 /* TODO: check specific MDTs */
4332 if (lum_stripe_count != lmv->lmv_migrate_offset ||
4333 lum->lum_stripe_offset != lmv->lmv_master_mdt_index ||
4334 (lum_hash_type && lum_hash_type != lmv_hash_type)) {
4335 CERROR("%s: '"DNAME"' migration was interrupted, run 'lfs migrate -m %d -c %d -H %s "DNAME"' to finish migration.\n",
4336 mdd2obd_dev(mdd)->obd_name, PNAME(lname),
4337 le32_to_cpu(lmv->lmv_master_mdt_index),
4338 le32_to_cpu(lmv->lmv_migrate_offset),
4339 mdt_hash_name[le32_to_cpu(lmv_hash_type)],
4348 * Internal function to migrate directory or file between MDTs.
4350 * migrate source to target in following steps:
4351 * 1. create target, append source stripes after target's if it's directory,
4352 * migrate xattrs and update fid of source links.
4353 * 2. update namespace: migrate dirent from source parent to target parent,
4354 * update file linkea, and destroy source if it's not needed any more.
4356 * \param[in] env execution environment
4357 * \param[in] spobj source parent object
4358 * \param[in] tpobj target parent object
4359 * \param[in] sobj source object
4360 * \param[in] tobj target object
4361 * \param[in] sname source file name
4362 * \param[in] tname target file name
4363 * \param[in] spec target creation spec
4364 * \param[in] ma used to update \a pobj mtime and ctime
4366 * \retval 0 on success
4367 * \retval -errno on failure
4369 static int mdd_migrate_object(const struct lu_env *env,
4370 struct mdd_object *spobj,
4371 struct mdd_object *tpobj,
4372 struct mdd_object *sobj,
4373 struct mdd_object *tobj,
4374 const struct lu_name *sname,
4375 const struct lu_name *tname,
4376 struct md_op_spec *spec,
4379 struct mdd_thread_info *info = mdd_env_info(env);
4380 struct mdd_device *mdd = mdo2mdd(&spobj->mod_obj);
4381 struct lu_attr *spattr = &info->mdi_pattr;
4382 struct lu_attr *tpattr = &info->mdi_tpattr;
4383 struct lu_attr *attr = &info->mdi_cattr;
4384 struct linkea_data *ldata = &info->mdi_link_data;
4385 struct dt_allocation_hint *hint = &info->mdi_hint;
4386 struct lu_buf sbuf = { NULL };
4387 struct mdd_xattrs xattrs;
4388 struct lmv_mds_md_v1 *lmv;
4389 struct thandle *handle;
4394 rc = mdd_la_get(env, sobj, attr);
4398 rc = mdd_la_get(env, spobj, spattr);
4402 rc = mdd_la_get(env, tpobj, tpattr);
4406 rc = mdd_migrate_sanity_check(env, mdd, spobj, tpobj, sobj, tobj,
4407 spattr, tpattr, attr);
4411 mdd_xattrs_init(&xattrs);
4413 if (S_ISDIR(attr->la_mode) && !spec->sp_migrate_nsonly) {
4414 struct lmv_user_md_v1 *lum = spec->u.sp_ea.eadata;
4418 /* if user use default value '0' for stripe_count, we need to
4419 * adjust it to '1' to create a 1-stripe directory.
4421 if (lum->lum_stripe_count == 0)
4422 lum->lum_stripe_count = cpu_to_le32(1);
4424 rc = mdd_stripe_get(env, sobj, &sbuf, XATTR_NAME_LMV);
4425 if (rc && rc != -ENODATA)
4430 if (!lmv_is_sane(lmv))
4431 GOTO(out, rc = -EBADF);
4432 if (lmv_is_migrating(lmv)) {
4433 rc = mdd_migrate_cmd_check(mdd, lmv, lum,
4438 } else if (!S_ISDIR(attr->la_mode)) {
4440 GOTO(out, rc = -EALREADY);
4442 /* update namespace only if @sobj is on MDT where @tpobj is. */
4443 if (!mdd_object_remote(tpobj) && !mdd_object_remote(sobj))
4444 spec->sp_migrate_nsonly = true;
4446 if (S_ISLNK(attr->la_mode)) {
4447 lu_buf_check_and_alloc(&sbuf, attr->la_size + 1);
4449 GOTO(out, rc = -ENOMEM);
4451 rc = mdd_readlink(env, &sobj->mod_obj, &sbuf);
4454 CERROR("%s: "DFID" readlink failed: rc = %d\n",
4455 mdd2obd_dev(mdd)->obd_name,
4456 PFID(mdd_object_fid(sobj)), rc);
4462 /* linkea needs update upon FID or parent stripe change */
4463 rc = mdd_migrate_linkea_prepare(env, mdd, spobj, tpobj, sobj, sname,
4464 tname, attr, ldata);
4466 /* update namespace only if @sobj has link on its MDT. */
4467 spec->sp_migrate_nsonly = true;
4471 /* migrate inode will migrate xattrs, prepare xattrs early to avoid
4472 * RPCs inside transaction.
4474 if (!spec->sp_migrate_nsonly) {
4475 rc = mdd_xattrs_migrate_prep(env, &xattrs, sobj, true, true);
4480 handle = mdd_trans_create(env, mdd);
4482 GOTO(out, rc = PTR_ERR(handle));
4484 if (spec->sp_migrate_nsonly)
4485 rc = mdd_declare_migrate_update(env, spobj, tpobj, sobj, sname,
4486 tname, attr, spattr, tpattr,
4489 rc = mdd_declare_migrate_create(env, spobj, tpobj, sobj, tobj,
4490 sname, tname, spattr, tpattr,
4491 attr, &sbuf, ldata, &xattrs, ma,
4492 spec, hint, handle);
4496 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, tname, sname,
4501 rc = mdd_trans_start(env, mdd, handle);
4505 if (spec->sp_migrate_nsonly)
4506 rc = mdd_migrate_update(env, spobj, tpobj, sobj, sname, tname,
4507 attr, spattr, tpattr, ldata, ma,
4510 rc = mdd_migrate_create(env, spobj, tpobj, sobj, tobj, sname,
4511 tname, spattr, tpattr, attr, &sbuf,
4512 ldata, &xattrs, ma, spec, hint, handle);
4516 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0,
4517 spec->sp_migrate_nsonly ? sobj : tobj,
4518 mdd_object_fid(spobj), mdd_object_fid(sobj),
4519 mdd_object_fid(tpobj), tname, sname,
4526 rc = mdd_trans_stop(env, mdd, rc, handle);
4528 mdd_xattrs_fini(&xattrs);
4535 * Migrate directory or file between MDTs.
4537 * \param[in] env execution environment
4538 * \param[in] md_pobj parent master object
4539 * \param[in] md_sobj source object
4540 * \param[in] lname file name
4541 * \param[in] md_tobj target object
4542 * \param[in] spec target creation spec
4543 * \param[in] ma used to update \a pobj mtime and ctime
4545 * \retval 0 on success
4546 * \retval -errno on failure
4548 static int mdd_migrate(const struct lu_env *env, struct md_object *md_pobj,
4549 struct md_object *md_sobj, const struct lu_name *lname,
4550 struct md_object *md_tobj, struct md_op_spec *spec,
4553 struct mdd_thread_info *info = mdd_env_info(env);
4554 struct mdd_device *mdd = mdo2mdd(md_pobj);
4555 struct mdd_object *pobj = md2mdd_obj(md_pobj);
4556 struct mdd_object *sobj = md2mdd_obj(md_sobj);
4557 struct mdd_object *tobj = md2mdd_obj(md_tobj);
4558 struct mdd_object *spobj = NULL;
4559 struct mdd_object *tpobj = NULL;
4560 struct lu_buf pbuf = { NULL };
4561 struct lu_fid *fid = &info->mdi_fid2;
4562 struct lmv_mds_md_v1 *lmv;
4567 /* locate source and target stripe on pobj, which are the real parent */
4568 rc = mdd_stripe_get(env, pobj, &pbuf, XATTR_NAME_LMV);
4569 if (rc < 0 && rc != -ENODATA)
4576 if (!lmv_is_sane(lmv))
4577 GOTO(out, rc = -EBADF);
4579 /* locate target parent stripe */
4580 /* fail check here to make sure top dir migration succeed. */
4581 if (lmv_is_migrating(lmv) &&
4582 OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_ENTRIES, 0))
4583 GOTO(out, rc = -EIO);
4585 index = lmv_name_to_stripe_index(lmv, lname->ln_name,
4588 GOTO(out, rc = index);
4590 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4591 tpobj = mdd_object_find(env, mdd, fid);
4593 GOTO(out, rc = PTR_ERR(tpobj));
4595 /* locate source parent stripe */
4596 if (lmv_is_layout_changing(lmv)) {
4597 index = lmv_name_to_stripe_index_old(lmv,
4601 GOTO(out, rc = index);
4603 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[index]);
4604 spobj = mdd_object_find(env, mdd, fid);
4606 GOTO(out, rc = PTR_ERR(spobj));
4608 /* parent stripe unchanged */
4609 if (spobj == tpobj) {
4610 if (!lmv_is_restriping(lmv))
4611 GOTO(out, rc = -EINVAL);
4612 GOTO(out, rc = -EALREADY);
4616 mdd_object_get(spobj);
4621 mdd_object_get(tpobj);
4622 mdd_object_get(spobj);
4625 rc = mdd_migrate_object(env, spobj, tpobj, sobj, tobj, lname, lname,
4630 if (!IS_ERR_OR_NULL(spobj))
4631 mdd_object_put(env, spobj);
4632 if (!IS_ERR_OR_NULL(tpobj))
4633 mdd_object_put(env, tpobj);
4639 static int mdd_declare_1sd_collapse(const struct lu_env *env,
4640 struct mdd_object *pobj,
4641 struct mdd_object *obj,
4642 struct mdd_object *stripe,
4643 struct lu_attr *attr,
4644 struct mdd_xattrs *xattrs,
4645 struct md_layout_change *mlc,
4646 struct lu_name *lname,
4647 struct thandle *handle)
4651 mlc->mlc_opc = MD_LAYOUT_DETACH;
4652 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4656 rc = mdo_declare_index_insert(env, stripe, mdd_object_fid(pobj),
4657 S_IFDIR, dotdot, handle);
4661 rc = mdd_foreach_xattr(env, stripe, xattrs, handle,
4662 mdo_declare_xattr_set);
4666 rc = mdo_declare_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4670 rc = mdo_declare_attr_set(env, stripe, attr, handle);
4674 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
4678 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(stripe),
4679 attr->la_mode, lname->ln_name, handle);
4683 rc = mdo_declare_ref_del(env, obj, handle);
4687 rc = mdo_declare_ref_del(env, obj, handle);
4691 rc = mdo_declare_destroy(env, obj, handle);
4698 /* transform one-stripe directory to a plain directory */
4699 static int mdd_1sd_collapse(const struct lu_env *env,
4700 struct mdd_object *pobj,
4701 struct mdd_object *obj,
4702 struct mdd_object *stripe,
4703 struct lu_attr *attr,
4704 struct mdd_xattrs *xattrs,
4705 struct md_layout_change *mlc,
4706 struct lu_name *lname,
4707 struct thandle *handle)
4713 /* replace 1-stripe directory with its stripe */
4714 mlc->mlc_opc = MD_LAYOUT_DETACH;
4716 mdd_write_lock(env, obj, DT_SRC_PARENT);
4717 rc = mdo_layout_change(env, obj, mlc, handle);
4718 mdd_write_unlock(env, obj);
4722 mdd_write_lock(env, pobj, DT_SRC_PARENT);
4723 mdd_write_lock(env, obj, DT_SRC_CHILD);
4725 /* insert dotdot to stripe which points to parent */
4726 rc = __mdd_index_insert_only(env, stripe, mdd_object_fid(pobj),
4727 S_IFDIR, dotdot, handle);
4731 rc = mdd_foreach_xattr(env, stripe, xattrs, handle, mdo_xattr_set);
4736 rc = mdo_xattr_del(env, stripe, XATTR_NAME_LMV, handle);
4740 /* don't set nlink from parent */
4741 attr->la_valid &= ~LA_NLINK;
4743 rc = mdo_attr_set(env, stripe, attr, handle);
4747 /* delete dir name from parent */
4748 rc = __mdd_index_delete_only(env, pobj, lname->ln_name, handle);
4752 /* insert stripe to parent with dir name */
4753 rc = __mdd_index_insert_only(env, pobj, mdd_object_fid(stripe),
4754 attr->la_mode, lname->ln_name, handle);
4758 /* destroy dir obj */
4759 rc = mdo_ref_del(env, obj, handle);
4763 rc = mdo_ref_del(env, obj, handle);
4767 rc = mdo_destroy(env, obj, handle);
4773 mdd_write_unlock(env, obj);
4774 mdd_write_unlock(env, pobj);
4780 * shrink directory stripes after migration/merge
4782 int mdd_dir_layout_shrink(const struct lu_env *env,
4783 struct md_object *md_obj,
4784 struct md_layout_change *mlc)
4786 struct mdd_device *mdd = mdo2mdd(md_obj);
4787 struct mdd_thread_info *info = mdd_env_info(env);
4788 struct mdd_object *obj = md2mdd_obj(md_obj);
4789 struct mdd_object *pobj = NULL;
4790 struct mdd_object *stripe = NULL;
4791 struct lu_attr *attr = &info->mdi_pattr;
4792 struct lu_fid *fid = &info->mdi_fid2;
4793 struct lu_name lname = { NULL };
4794 struct lu_buf lmv_buf = { NULL };
4795 struct mdd_xattrs xattrs;
4796 struct lmv_mds_md_v1 *lmv;
4797 struct lmv_user_md *lmu;
4798 struct thandle *handle;
4803 rc = mdd_la_get(env, obj, attr);
4807 if (!S_ISDIR(attr->la_mode))
4810 rc = mdd_stripe_get(env, obj, &lmv_buf, XATTR_NAME_LMV);
4814 lmv = lmv_buf.lb_buf;
4815 if (!lmv_is_sane(lmv))
4818 lmu = mlc->mlc_buf.lb_buf;
4820 /* adjust the default value '0' to '1' */
4821 if (lmu->lum_stripe_count == 0)
4822 lmu->lum_stripe_count = cpu_to_le32(1);
4824 /* these were checked in MDT */
4825 LASSERT(le32_to_cpu(lmu->lum_stripe_count) <
4826 le32_to_cpu(lmv->lmv_stripe_count));
4827 LASSERT(!lmv_is_splitting(lmv));
4828 LASSERT(lmv_is_migrating(lmv) || lmv_is_merging(lmv));
4830 mdd_xattrs_init(&xattrs);
4832 /* if dir stripe count will be shrunk to 1, it needs to be transformed
4833 * to a plain dir, which will cause FID change and namespace update.
4835 if (le32_to_cpu(lmu->lum_stripe_count) == 1) {
4836 struct linkea_data *ldata = &info->mdi_link_data;
4837 char *filename = info->mdi_name;
4839 rc = mdd_links_read(env, obj, ldata);
4843 if (ldata->ld_leh->leh_reccount > 1)
4844 GOTO(out, rc = -EINVAL);
4846 linkea_first_entry(ldata);
4848 GOTO(out, rc = -ENODATA);
4850 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname,
4853 /* Note: lname might miss \0 at the end */
4854 snprintf(filename, sizeof(info->mdi_name), "%.*s",
4855 lname.ln_namelen, lname.ln_name);
4856 lname.ln_name = filename;
4858 pobj = mdd_object_find(env, mdd, fid);
4865 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
4867 stripe = mdd_object_find(env, mdd, fid);
4868 if (IS_ERR(stripe)) {
4869 mdd_object_put(env, pobj);
4871 GOTO(out, rc = PTR_ERR(stripe));
4874 if (!lmv_is_fixed(lmv))
4875 rc = mdd_xattrs_migrate_prep(env, &xattrs, obj, false,
4879 handle = mdd_trans_create(env, mdd);
4881 GOTO(out, rc = PTR_ERR(handle));
4883 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4884 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4886 GOTO(stop_trans, rc);
4888 if (le32_to_cpu(lmu->lum_stripe_count) == 1 && !lmv_is_fixed(lmv)) {
4889 rc = mdd_declare_1sd_collapse(env, pobj, obj, stripe, attr,
4890 &xattrs, mlc, &lname, handle);
4892 GOTO(stop_trans, rc);
4895 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL, NULL,
4898 GOTO(stop_trans, rc);
4900 rc = mdd_trans_start(env, mdd, handle);
4902 GOTO(stop_trans, rc);
4904 mdd_write_lock(env, obj, DT_SRC_PARENT);
4905 mlc->mlc_opc = MD_LAYOUT_SHRINK;
4906 rc = mdo_layout_change(env, obj, mlc, handle);
4907 mdd_write_unlock(env, obj);
4909 GOTO(stop_trans, rc);
4911 if (le32_to_cpu(lmu->lum_stripe_count) == 1 && !lmv_is_fixed(lmv)) {
4912 rc = mdd_1sd_collapse(env, pobj, obj, stripe, attr, &xattrs,
4913 mlc, &lname, handle);
4915 GOTO(stop_trans, rc);
4918 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
4919 XATTR_NAME_LMV, handle);
4920 GOTO(stop_trans, rc);
4923 rc = mdd_trans_stop(env, mdd, rc, handle);
4925 mdd_xattrs_fini(&xattrs);
4927 mdd_object_put(env, stripe);
4928 mdd_object_put(env, pobj);
4930 lu_buf_free(&lmv_buf);
4934 static int mdd_dir_declare_split_plain(const struct lu_env *env,
4935 struct mdd_device *mdd,
4936 struct mdd_object *pobj,
4937 struct mdd_object *obj,
4938 struct mdd_object *tobj,
4939 struct mdd_xattrs *xattrs,
4940 struct md_layout_change *mlc,
4941 struct dt_allocation_hint *hint,
4942 struct thandle *handle)
4944 struct mdd_thread_info *info = mdd_env_info(env);
4945 const struct lu_name *lname = mlc->mlc_name;
4946 struct lu_attr *la = &info->mdi_la_for_fix;
4947 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
4948 struct linkea_data *ldata = &info->mdi_link_data;
4949 struct lmv_mds_md_v1 *lmv;
4953 mlc->mlc_opc = MD_LAYOUT_DETACH;
4954 rc = mdo_declare_layout_change(env, obj, mlc, handle);
4958 memset(ldata, 0, sizeof(*ldata));
4959 rc = mdd_linkea_prepare(env, obj, NULL, NULL, mdd_object_fid(pobj),
4960 lname, 1, 0, ldata);
4964 count = lum->lum_stripe_count;
4965 lum->lum_stripe_count = 0;
4966 /* don't set default LMV since it will become a striped dir */
4967 lum->lum_max_inherit = LMV_INHERIT_NONE;
4968 mdd_object_make_hint(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
4970 rc = mdd_declare_create(env, mdo2mdd(&pobj->mod_obj), pobj, tobj,
4971 lname, mlc->mlc_attr, handle, mlc->mlc_spec,
4972 ldata, NULL, NULL, NULL, hint);
4976 /* tobj mode will be used in lod_declare_xattr_set(), but it's not
4979 tobj->mod_obj.mo_lu.lo_header->loh_attr |= S_IFDIR;
4981 lmv = (typeof(lmv))info->mdi_key;
4982 memset(lmv, 0, sizeof(*lmv));
4983 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
4984 lmv->lmv_stripe_count = cpu_to_le32(1);
4985 lmv->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_DEFAULT);
4986 fid_le_to_cpu(&lmv->lmv_stripe_fids[0], mdd_object_fid(obj));
4988 mlc->mlc_opc = MD_LAYOUT_ATTACH;
4989 mlc->mlc_buf.lb_buf = lmv;
4990 mlc->mlc_buf.lb_len = lmv_mds_md_size(1, LMV_MAGIC_V1);
4991 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
4995 rc = mdd_foreach_xattr(env, tobj, xattrs, handle,
4996 mdo_declare_xattr_set);
5000 lum->lum_stripe_count = count;
5001 mlc->mlc_opc = MD_LAYOUT_SPLIT;
5002 rc = mdo_declare_layout_change(env, tobj, mlc, handle);
5006 rc = mdo_declare_index_delete(env, pobj, lname->ln_name, handle);
5010 rc = mdo_declare_index_insert(env, pobj, mdd_object_fid(tobj),
5011 S_IFDIR, lname->ln_name, handle);
5015 la->la_valid = LA_CTIME | LA_MTIME;
5016 rc = mdo_declare_attr_set(env, obj, la, handle);
5020 rc = mdo_declare_attr_set(env, pobj, la, handle);
5024 rc = mdd_declare_changelog_store(env, mdd, CL_MIGRATE, lname, NULL,
5030 * plain directory split:
5031 * 1. create \a tobj as plain directory.
5032 * 2. append \a obj as first stripe of \a tobj.
5033 * 3. migrate xattrs from \a obj to \a tobj.
5034 * 4. split \a tobj to specific stripe count.
5036 static int mdd_dir_split_plain(const struct lu_env *env,
5037 struct mdd_device *mdd,
5038 struct mdd_object *pobj,
5039 struct mdd_object *obj,
5040 struct mdd_object *tobj,
5041 struct mdd_xattrs *xattrs,
5042 struct md_layout_change *mlc,
5043 struct dt_allocation_hint *hint,
5044 struct thandle *handle)
5046 struct mdd_thread_info *info = mdd_env_info(env);
5047 struct lu_attr *pattr = &info->mdi_pattr;
5048 struct lu_attr *la = &info->mdi_la_for_fix;
5049 const struct lu_name *lname = mlc->mlc_name;
5050 struct linkea_data *ldata = &info->mdi_link_data;
5055 /* copy linkea out and set on target later */
5056 rc = mdd_links_read(env, obj, ldata);
5060 mlc->mlc_opc = MD_LAYOUT_DETACH;
5061 rc = mdo_layout_change(env, obj, mlc, handle);
5065 /* don't set nlink from obj */
5066 mlc->mlc_attr->la_valid &= ~LA_NLINK;
5068 rc = mdd_create_object(env, pobj, tobj, mlc->mlc_attr, mlc->mlc_spec,
5069 NULL, NULL, NULL, hint, handle, false);
5073 rc = mdd_foreach_xattr(env, tobj, xattrs, handle, mdo_xattr_set);
5077 rc = mdd_links_write(env, tobj, ldata, handle);
5081 rc = __mdd_index_delete(env, pobj, lname->ln_name, true, handle);
5085 rc = __mdd_index_insert(env, pobj, mdd_object_fid(tobj), S_IFDIR,
5086 lname->ln_name, handle);
5090 la->la_ctime = la->la_mtime = mlc->mlc_attr->la_mtime;
5091 la->la_valid = LA_CTIME | LA_MTIME;
5093 mdd_write_lock(env, obj, DT_SRC_CHILD);
5094 rc = mdd_update_time(env, tobj, mlc->mlc_attr, la, handle);
5095 mdd_write_unlock(env, obj);
5099 rc = mdd_la_get(env, pobj, pattr);
5103 la->la_valid = LA_CTIME | LA_MTIME;
5105 mdd_write_lock(env, pobj, DT_SRC_PARENT);
5106 rc = mdd_update_time(env, pobj, pattr, la, handle);
5107 mdd_write_unlock(env, pobj);
5111 /* FID changes, record it as CL_MIGRATE */
5112 rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, tobj,
5113 mdd_object_fid(pobj), mdd_object_fid(obj),
5114 mdd_object_fid(pobj), lname, lname, handle);
5118 int mdd_dir_layout_split(const struct lu_env *env, struct md_object *o,
5119 struct md_layout_change *mlc)
5121 struct mdd_thread_info *info = mdd_env_info(env);
5122 struct mdd_device *mdd = mdo2mdd(o);
5123 struct mdd_object *obj = md2mdd_obj(o);
5124 struct mdd_object *pobj = md2mdd_obj(mlc->mlc_parent);
5125 struct mdd_object *tobj = md2mdd_obj(mlc->mlc_target);
5126 struct dt_allocation_hint *hint = &info->mdi_hint;
5127 bool is_plain = false;
5128 struct mdd_xattrs xattrs;
5129 struct thandle *handle;
5134 LASSERT(S_ISDIR(mdd_object_type(obj)));
5136 rc = mdo_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LMV);
5142 mdd_xattrs_init(&xattrs);
5144 rc = mdd_xattrs_migrate_prep(env, &xattrs, obj, true, true);
5146 handle = mdd_trans_create(env, mdd);
5148 GOTO(out, rc = PTR_ERR(handle));
5151 rc = mdd_dir_declare_split_plain(env, mdd, pobj, obj, tobj,
5152 &xattrs, mlc, hint, handle);
5154 mlc->mlc_opc = MD_LAYOUT_SPLIT;
5155 rc = mdo_declare_layout_change(env, obj, mlc, handle);
5157 GOTO(stop_trans, rc);
5159 rc = mdd_declare_changelog_store(env, mdd, CL_LAYOUT, NULL,
5163 GOTO(stop_trans, rc);
5165 rc = mdd_trans_start(env, mdd, handle);
5167 GOTO(stop_trans, rc);
5170 rc = mdd_dir_split_plain(env, mdd, pobj, obj, tobj, &xattrs,
5173 mdd_write_lock(env, obj, DT_TGT_CHILD);
5174 rc = mdo_xattr_set(env, obj, NULL, XATTR_NAME_LMV,
5175 LU_XATTR_CREATE, handle);
5176 mdd_write_unlock(env, obj);
5178 GOTO(stop_trans, rc);
5180 rc = mdd_changelog_data_store_xattr(env, mdd, CL_LAYOUT, 0, obj,
5181 XATTR_NAME_LMV, handle);
5184 GOTO(stop_trans, rc);
5189 rc = mdd_trans_stop(env, mdd, rc, handle);
5191 mdd_xattrs_fini(&xattrs);
5196 const struct md_dir_operations mdd_dir_ops = {
5197 .mdo_is_subdir = mdd_is_subdir,
5198 .mdo_lookup = mdd_lookup,
5199 .mdo_create = mdd_create,
5200 .mdo_rename = mdd_rename,
5201 .mdo_link = mdd_link,
5202 .mdo_unlink = mdd_unlink,
5203 .mdo_create_data = mdd_create_data,
5204 .mdo_migrate = mdd_migrate,