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, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdd/mdd_orphans.c
34 * Orphan handling code
36 * Author: Mike Pershin <tappro@clusterfs.com>
37 * Pravin B Shelar <pravin.shelar@sun.com>
40 #define DEBUG_SUBSYSTEM S_MDS
43 #include <obd_class.h>
44 #include <lustre_ver.h>
45 #include <obd_support.h>
46 #include <lustre_fid.h>
47 #include "mdd_internal.h"
49 const char orph_index_name[] = "PENDING";
50 static const char dotdot[] = "..";
57 #define ORPHAN_FILE_NAME_FORMAT "%016llx:%08x:%08x:%2x"
58 #define ORPHAN_FILE_NAME_FORMAT_18 "%llx:%08x"
60 static struct dt_key* orph_key_fill(const struct lu_env *env,
61 const struct lu_fid *lf, __u32 op)
63 char *key = mdd_env_info(env)->mti_key;
67 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT,
68 (long long unsigned int)fid_seq(lf),
69 fid_oid(lf), fid_ver(lf), op);
71 return (struct dt_key*) key;
76 static struct dt_key* orph_key_fill_18(const struct lu_env *env,
77 const struct lu_fid *lf)
79 char *key = mdd_env_info(env)->mti_key;
83 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_18,
84 (unsigned long long)fid_seq(lf), fid_oid(lf));
86 return (struct dt_key*) key;
91 static inline void mdd_orphan_write_lock(const struct lu_env *env,
92 struct mdd_device *mdd)
94 struct dt_object *dor = mdd->mdd_orphans;
95 dt_write_lock(env, dor, MOR_TGT_ORPHAN);
98 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
99 struct mdd_device *mdd)
101 struct dt_object *dor = mdd->mdd_orphans;
102 dt_write_unlock(env, dor);
105 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
106 struct mdd_device *mdd,
107 struct mdd_object *obj,
108 __u32 op, struct thandle *th)
110 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
111 struct dt_object *dor = mdd->mdd_orphans;
112 const struct lu_fid *lf = mdo2fid(obj);
113 struct dt_key *key = orph_key_fill(env, lf, op);
116 rec->rec_type = mdd_object_type(obj);
118 return dt_insert(env, dor, (const struct dt_rec *)rec, key, th, 1);
121 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
122 struct mdd_device *mdd ,
126 struct dt_object *dor = mdd->mdd_orphans;
128 return dt_delete(env, dor, key, th);
131 static inline int mdd_orphan_ref_add(const struct lu_env *env,
132 struct mdd_device *mdd,
135 struct dt_object *dor = mdd->mdd_orphans;
136 return dt_ref_add(env, dor, th);
139 static inline int mdd_orphan_ref_del(const struct lu_env *env,
140 struct mdd_device *mdd,
143 struct dt_object *dor = mdd->mdd_orphans;
144 return dt_ref_del(env, dor, th);
148 int orph_declare_index_insert(const struct lu_env *env,
149 struct mdd_object *obj,
150 umode_t mode, struct thandle *th)
152 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
153 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
157 key = orph_key_fill(env, mdo2fid(obj), ORPH_OP_UNLINK);
159 rec->rec_fid = mdo2fid(obj);
160 rec->rec_type = mode;
161 rc = dt_declare_insert(env, mdd->mdd_orphans,
162 (const struct dt_rec *)rec, key, th);
166 rc = mdo_declare_ref_add(env, obj, th);
173 rc = mdo_declare_ref_add(env, obj, th);
177 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
181 rc = mdo_declare_index_delete(env, obj, dotdot, th);
185 rc = mdo_declare_index_insert(env, obj,
186 lu_object_fid(&mdd->mdd_orphans->do_lu),
187 S_IFDIR, dotdot, th);
192 static int orph_index_insert(const struct lu_env *env,
193 struct mdd_object *obj,
194 __u32 op, struct thandle *th)
196 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
197 struct dt_object *dor = mdd->mdd_orphans;
198 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
199 struct dt_object *next = mdd_object_child(obj);
200 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
204 LASSERT(mdd_write_locked(env, obj) != 0);
205 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
207 mdd_orphan_write_lock(env, mdd);
209 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
213 mdo_ref_add(env, obj, th);
214 if (!S_ISDIR(mdd_object_type(obj)))
217 mdo_ref_add(env, obj, th);
218 mdd_orphan_ref_add(env, mdd, th);
220 /* try best to fixup directory, dont return errors
222 if (!dt_try_as_dir(env, next))
225 dt_delete(env, next, (const struct dt_key *)dotdot, th);
227 rec->rec_fid = lf_dor;
228 rec->rec_type = S_IFDIR;
229 dt_insert(env, next, (const struct dt_rec *)rec,
230 (const struct dt_key *)dotdot, th, 1);
234 obj->mod_flags |= ORPHAN_OBJ;
236 mdd_orphan_write_unlock(env, mdd);
241 int orph_declare_index_delete(const struct lu_env *env,
242 struct mdd_object *obj,
246 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
250 key = orph_key_fill(env, mdo2fid(obj), ORPH_OP_UNLINK);
252 rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
256 rc = mdo_declare_ref_del(env, obj, th);
260 if (S_ISDIR(mdd_object_type(obj))) {
261 rc = mdo_declare_ref_del(env, obj, th);
265 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
271 static int orph_index_delete(const struct lu_env *env,
272 struct mdd_object *obj,
276 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
277 struct dt_object *dor = mdd->mdd_orphans;
283 LASSERT(mdd_write_locked(env, obj) != 0);
284 LASSERT(obj->mod_flags & ORPHAN_OBJ);
285 LASSERT(obj->mod_count == 0);
289 key = orph_key_fill(env, mdo2fid(obj), op);
290 mdd_orphan_write_lock(env, mdd);
292 rc = mdd_orphan_delete_obj(env, mdd, key, th);
295 key = orph_key_fill_18(env, mdo2fid(obj));
296 rc = mdd_orphan_delete_obj(env, mdd, key, th);
300 /* lov objects will be destroyed by caller */
301 mdo_ref_del(env, obj, th);
302 if (S_ISDIR(mdd_object_type(obj))) {
303 mdo_ref_del(env, obj, th);
304 mdd_orphan_ref_del(env, mdd, th);
306 obj->mod_flags &= ~ORPHAN_OBJ;
308 CERROR("could not delete object: rc = %d\n",rc);
311 mdd_orphan_write_unlock(env, mdd);
316 static int orphan_object_destroy(const struct lu_env *env,
317 struct mdd_object *obj,
320 struct thandle *th = NULL;
321 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
325 th = mdd_trans_create(env, mdd);
327 CERROR("Cannot get thandle\n");
331 rc = orph_declare_index_delete(env, obj, th);
335 rc = mdo_declare_destroy(env, obj, th);
339 rc = mdd_trans_start(env, mdd, th);
343 mdd_write_lock(env, obj, MOR_TGT_CHILD);
344 if (likely(obj->mod_count == 0)) {
345 mdd_orphan_write_lock(env, mdd);
346 rc = mdd_orphan_delete_obj(env, mdd, key, th);
348 mdo_ref_del(env, obj, th);
349 if (S_ISDIR(mdd_object_type(obj))) {
350 mdo_ref_del(env, obj, th);
351 mdd_orphan_ref_del(env, mdd, th);
353 rc = mdo_destroy(env, obj, th);
355 CERROR("could not delete object: rc = %d\n", rc);
356 mdd_orphan_write_unlock(env, mdd);
358 mdd_write_unlock(env, obj);
361 rc = mdd_trans_stop(env, mdd, 0, th);
367 * Delete unused orphan with FID \a lf from PENDING directory
369 * \param mdd MDD device finishing recovery
370 * \param lf FID of file or directory to delete
371 * \param key cookie for this entry in index iterator
376 static int orph_key_test_and_del(const struct lu_env *env,
377 struct mdd_device *mdd,
381 struct mdd_object *mdo;
384 mdo = mdd_object_find(env, mdd, lf);
390 if (mdo->mod_count == 0) {
391 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
392 rc = orphan_object_destroy(env, mdo, key);
393 if (rc) /* so replay-single.sh test_37 works */
394 CERROR("%s: error unlinking orphan "DFID" from "
395 "PENDING: rc = %d\n",
396 mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
398 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
399 if (likely(mdo->mod_count > 0)) {
400 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
401 PFID(lf), mdo->mod_count);
402 mdo->mod_flags |= ORPHAN_OBJ;
404 mdd_write_unlock(env, mdo);
407 mdd_object_put(env, mdo);
412 * delete unreferenced files and directories in the PENDING directory
414 * Files that remain in PENDING after client->MDS recovery has completed
415 * have to be referenced (opened) by some client during recovery, or they
416 * will be deleted here (for clients that did not complete recovery).
418 * \param thread info about orphan cleanup thread
423 static int orph_index_iterate(const struct lu_env *env,
424 struct mdd_generic_thread *thread)
426 struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
427 struct dt_object *dor = mdd->mdd_orphans;
428 struct lu_dirent *ent = &mdd_env_info(env)->mti_ent;
429 const struct dt_it_ops *iops;
437 iops = &dor->do_index_ops->dio_it;
438 it = iops->init(env, dor, LUDA_64BITHASH);
441 CERROR("%s: cannot clean PENDING: rc = %d\n",
442 mdd2obd_dev(mdd)->obd_name, rc);
446 rc = iops->load(env, it, 0);
450 CERROR("%s: error loading iterator to clean PENDING\n",
451 mdd2obd_dev(mdd)->obd_name);
452 /* Index contains no zero key? */
453 GOTO(out_put, rc = -EIO);
457 if (thread->mgt_abort)
460 key_sz = iops->key_size(env, it);
461 /* filter out "." and ".." entries from PENDING dir. */
465 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
467 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
468 mdd2obd_dev(mdd)->obd_name, rc);
472 fid_le_to_cpu(&fid, &ent->lde_fid);
473 if (!fid_is_sane(&fid)) {
474 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
475 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
479 /* kill orphan object */
480 cookie = iops->store(env, it);
482 rc = orph_key_test_and_del(env, mdd, &fid,
483 (struct dt_key *)ent->lde_name);
485 /* after index delete reset iterator */
487 rc = iops->get(env, it, (const void *)"");
489 rc = iops->load(env, it, cookie);
491 rc = iops->next(env, it);
494 GOTO(out_put, rc = 0);
504 * open the PENDING directory for device \a mdd
506 * The PENDING directory persistently tracks files and directories that were
507 * unlinked from the namespace (nlink == 0) but are still held open by clients.
508 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
509 * would not be able to recover and reopen those files. Instead, these inodes
510 * are linked into the PENDING directory on disk, and only deleted if all
511 * clients close them, or the MDS finishes client recovery without any client
512 * reopening them (i.e. former clients didn't join recovery).
513 * \param d mdd device being started.
516 * \retval -ve index operation error.
519 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
527 /* create PENDING dir */
529 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
530 orph_index_name, S_IFDIR | S_IRUGO |
531 S_IWUSR | S_IXUGO, &fid);
535 d = dt_locate(env, mdd->mdd_child, &fid);
538 LASSERT(lu_object_exists(&d->do_lu));
539 if (!dt_try_as_dir(env, d)) {
540 CERROR("%s: \"%s\" is not an index: rc = %d\n",
541 mdd2obd_dev(mdd)->obd_name, orph_index_name, rc);
542 dt_object_put(env, d);
545 mdd->mdd_orphans = d;
549 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
552 if (mdd->mdd_orphans != NULL) {
553 dt_object_put(env, mdd->mdd_orphans);
554 mdd->mdd_orphans = NULL;
559 static int __mdd_orphan_cleanup(void *args)
561 struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
562 struct lu_env *env = NULL;
566 complete(&thread->mgt_started);
570 GOTO(out, rc = -ENOMEM);
572 rc = lu_env_init(env, LCT_MD_THREAD);
576 rc = orph_index_iterate(env, thread);
583 complete(&thread->mgt_finished);
588 * Iterate orphan index to cleanup orphan objects after recovery is done.
589 * \param d mdd device in recovery.
591 int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
596 OBD_ALLOC(name, MTI_NAME_MAXLEN);
600 snprintf(name, MTI_NAME_MAXLEN, "orph_cleanup_%s",
601 mdd2obd_dev(d)->obd_name);
603 rc = mdd_generic_thread_start(&d->mdd_orph_cleanup_thread,
604 __mdd_orphan_cleanup, (void *)d, name);
607 CERROR("%s: start orphan cleanup thread failed:%d\n",
608 mdd2obd_dev(d)->obd_name, rc);
610 OBD_FREE(name, MTI_NAME_MAXLEN);
615 * add an orphan \a obj to the orphan index.
616 * \param obj file or directory.
617 * \param th transaction for index insert.
619 * \pre obj nlink == 0 && obj->mod_count != 0
622 * \retval -ve index operation error.
624 int __mdd_orphan_add(const struct lu_env *env,
625 struct mdd_object *obj, struct thandle *th)
627 return orph_index_insert(env, obj, ORPH_OP_UNLINK, th);
631 * delete an orphan \a obj from orphan index.
632 * \param obj file or directory.
633 * \param th transaction for index deletion and object destruction.
635 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
638 * \retval -ve index operation error.
640 int __mdd_orphan_del(const struct lu_env *env,
641 struct mdd_object *obj, struct thandle *th)
643 return orph_index_delete(env, obj, ORPH_OP_UNLINK, th);