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/
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 <obd_support.h>
45 #include <lustre_fid.h>
46 #include "mdd_internal.h"
48 static const char mdd_orphan_index_name[] = MDT_ORPHAN_DIR;
49 static const char dotdot[] = "..";
55 /* obsolete after 2.11, needed for upgrades from older 2.x versions */
56 #define ORPHAN_FILE_NAME_FORMAT_20 "%016llx:%08x:%08x:%2x"
58 static struct dt_key *mdd_orphan_key_fill(const struct lu_env *env,
59 const struct lu_fid *lf)
61 char *key = mdd_env_info(env)->mti_key;
64 snprintf(key, sizeof(mdd_env_info(env)->mti_key),
65 DFID_NOBRACE, PFID(lf));
67 return (struct dt_key *)key;
70 /* compatibility with orphan files created in versions before 2.11 */
71 static struct dt_key *mdd_orphan_key_fill_20(const struct lu_env *env,
72 const struct lu_fid *lf)
74 char *key = mdd_env_info(env)->mti_key;
77 snprintf(key, sizeof(mdd_env_info(env)->mti_key),
78 ORPHAN_FILE_NAME_FORMAT_20,
79 fid_seq(lf), fid_oid(lf), fid_ver(lf), ORPH_OP_UNLINK);
81 return (struct dt_key *)key;
84 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
85 struct mdd_device *mdd,
86 struct mdd_object *obj,
89 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
90 struct dt_object *dor = mdd->mdd_orphans;
91 const struct lu_fid *lf = mdd_object_fid(obj);
92 struct dt_key *key = mdd_orphan_key_fill(env, lf);
95 rec->rec_type = mdd_object_type(obj);
97 return dt_insert(env, dor, (const struct dt_rec *)rec, key, th);
100 int mdd_orphan_declare_insert(const struct lu_env *env, struct mdd_object *obj,
101 umode_t mode, struct thandle *th)
103 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
104 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
108 key = mdd_orphan_key_fill(env, mdd_object_fid(obj));
110 rec->rec_fid = mdd_object_fid(obj);
111 rec->rec_type = mode;
112 rc = dt_declare_insert(env, mdd->mdd_orphans,
113 (const struct dt_rec *)rec, key, th);
117 rc = mdo_declare_ref_add(env, obj, th);
124 rc = mdo_declare_ref_add(env, obj, th);
128 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
132 rc = mdo_declare_index_delete(env, obj, dotdot, th);
136 rc = mdo_declare_index_insert(env, obj,
137 lu_object_fid(&mdd->mdd_orphans->do_lu),
138 S_IFDIR, dotdot, th);
144 * add an orphan \a obj to the orphan index.
145 * \param obj file or directory.
146 * \param th transaction for index insert.
148 * \pre obj nlink == 0 && obj->mod_count != 0
151 * \retval -ve index operation error.
153 int mdd_orphan_insert(const struct lu_env *env, struct mdd_object *obj,
156 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
157 struct dt_object *dor = mdd->mdd_orphans;
158 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
159 struct dt_object *next = mdd_object_child(obj);
160 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
164 LASSERT(mdd_write_locked(env, obj) != 0);
165 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
167 dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
169 rc = mdd_orphan_insert_obj(env, mdd, obj, th);
173 mdo_ref_add(env, obj, th);
174 if (!S_ISDIR(mdd_object_type(obj)))
177 mdo_ref_add(env, obj, th);
178 dt_ref_add(env, mdd->mdd_orphans, th);
180 /* try best to fixup directory, do not return errors from here */
181 if (!dt_try_as_dir(env, next))
184 dt_delete(env, next, (const struct dt_key *)dotdot, th);
186 rec->rec_fid = lf_dor;
187 rec->rec_type = S_IFDIR;
188 dt_insert(env, next, (const struct dt_rec *)rec,
189 (const struct dt_key *)dotdot, th);
193 obj->mod_flags |= ORPHAN_OBJ;
195 dt_write_unlock(env, mdd->mdd_orphans);
200 int mdd_orphan_declare_delete(const struct lu_env *env, struct mdd_object *obj,
203 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
207 key = mdd_orphan_key_fill(env, mdd_object_fid(obj));
209 rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
213 if (!mdd_object_exists(obj))
216 rc = mdo_declare_ref_del(env, obj, th);
220 if (S_ISDIR(mdd_object_type(obj))) {
221 rc = mdo_declare_ref_del(env, obj, th);
225 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
232 * delete an orphan \a obj from orphan index.
233 * \param obj file or directory.
234 * \param th transaction for index deletion and object destruction.
236 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
239 * \retval -ve index operation error.
241 int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
244 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
245 struct dt_object *dor = mdd->mdd_orphans;
251 LASSERT(mdd_write_locked(env, obj) != 0);
252 LASSERT(obj->mod_flags & ORPHAN_OBJ);
253 LASSERT(obj->mod_count == 0);
257 key = mdd_orphan_key_fill(env, mdd_object_fid(obj));
258 dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
260 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ORPHAN_DELETE))
263 rc = dt_delete(env, mdd->mdd_orphans, key, th);
265 key = mdd_orphan_key_fill_20(env, mdd_object_fid(obj));
266 rc = dt_delete(env, mdd->mdd_orphans, key, th);
271 /* lov objects will be destroyed by caller */
272 mdo_ref_del(env, obj, th);
273 if (S_ISDIR(mdd_object_type(obj))) {
274 mdo_ref_del(env, obj, th);
275 dt_ref_del(env, mdd->mdd_orphans, th);
277 obj->mod_flags &= ~ORPHAN_OBJ;
279 CERROR("%s: could not delete orphan object "DFID": rc = %d\n",
280 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)), rc);
283 dt_write_unlock(env, mdd->mdd_orphans);
288 static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
291 struct thandle *th = NULL;
292 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
293 bool orphan_exists = true;
297 th = mdd_trans_create(env, mdd);
300 if (rc != -EINPROGRESS)
301 CERROR("%s: cannot get orphan thandle: rc = %d\n",
302 mdd2obd_dev(mdd)->obd_name, rc);
306 mdd_write_lock(env, obj, DT_TGT_CHILD);
307 rc = mdd_orphan_declare_delete(env, obj, th);
309 orphan_exists = false;
314 rc = mdo_declare_destroy(env, obj, th);
319 rc = mdd_trans_start(env, mdd, th);
323 if (likely(obj->mod_count == 0)) {
324 dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
325 rc = dt_delete(env, mdd->mdd_orphans, key, th);
327 key = mdd_orphan_key_fill_20(env, mdd_object_fid(obj));
328 rc = dt_delete(env, mdd->mdd_orphans, key, th);
331 CERROR("%s: could not delete orphan "DFID": rc = %d\n",
332 mdd_obj_dev_name(obj),
333 PFID(mdd_object_fid(obj)), rc);
334 } else if (orphan_exists) {
335 mdo_ref_del(env, obj, th);
336 if (S_ISDIR(mdd_object_type(obj))) {
337 mdo_ref_del(env, obj, th);
338 dt_ref_del(env, mdd->mdd_orphans, th);
340 rc = mdo_destroy(env, obj, th);
342 CWARN("%s: orphan %s "DFID" doesn't exist\n",
343 mdd_obj_dev_name(obj), (char *)key,
344 PFID(mdd_object_fid(obj)));
346 dt_write_unlock(env, mdd->mdd_orphans);
349 mdd_write_unlock(env, obj);
351 rc = mdd_trans_stop(env, mdd, 0, th);
357 * Delete unused orphan with FID \a lf from PENDING directory
359 * \param mdd MDD device finishing recovery
360 * \param lf FID of file or directory to delete
361 * \param key cookie for this entry in index iterator
366 static int mdd_orphan_key_test_and_delete(const struct lu_env *env,
367 struct mdd_device *mdd,
368 struct lu_fid *lf, struct dt_key *key)
370 struct mdd_object *mdo;
373 mdo = mdd_object_find(env, mdd, lf);
379 if (mdo->mod_count == 0) {
380 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
381 rc = mdd_orphan_destroy(env, mdo, key);
382 if (rc) /* below message checked in replay-single.sh test_37 */
383 CERROR("%s: error unlinking orphan "DFID": rc = %d\n",
384 mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
386 mdd_write_lock(env, mdo, DT_TGT_CHILD);
387 if (likely(mdo->mod_count > 0)) {
388 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
389 PFID(lf), mdo->mod_count);
390 mdo->mod_flags |= ORPHAN_OBJ;
392 mdd_write_unlock(env, mdo);
395 mdd_object_put(env, mdo);
400 * delete unreferenced files and directories in the PENDING directory
402 * Files that remain in PENDING after client->MDS recovery has completed
403 * have to be referenced (opened) by some client during recovery, or they
404 * will be deleted here (for clients that did not complete recovery).
406 * \param thread info about orphan cleanup thread
411 static int mdd_orphan_index_iterate(const struct lu_env *env,
412 struct mdd_generic_thread *thread)
414 struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
415 struct dt_object *dor = mdd->mdd_orphans;
416 struct lu_dirent *ent = &mdd_env_info(env)->mti_ent;
417 const struct dt_it_ops *iops;
425 iops = &dor->do_index_ops->dio_it;
426 it = iops->init(env, dor, LUDA_64BITHASH);
429 CERROR("%s: cannot clean '%s': rc = %d\n",
430 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name, rc);
434 rc = iops->load(env, it, 0);
438 CERROR("%s: error loading iterator to clean '%s'\n",
439 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name);
440 /* Index contains no zero key? */
441 GOTO(out_put, rc = -EIO);
445 if (thread->mgt_abort)
448 key_sz = iops->key_size(env, it);
449 /* filter out "." and ".." entries from PENDING dir. */
453 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
455 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
456 mdd2obd_dev(mdd)->obd_name, rc);
460 fid_le_to_cpu(&fid, &ent->lde_fid);
461 if (!fid_is_sane(&fid)) {
462 CERROR("%s: bad FID "DFID" cleaning '%s'\n",
463 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
464 mdd_orphan_index_name);
468 /* kill orphan object */
469 cookie = iops->store(env, it);
471 rc = mdd_orphan_key_test_and_delete(env, mdd, &fid,
472 (struct dt_key *)ent->lde_name);
474 /* after index delete reset iterator */
476 rc = iops->get(env, it, (const void *)"");
478 rc = iops->load(env, it, cookie);
480 rc = iops->next(env, it);
483 GOTO(out_put, rc = 0);
493 * open the PENDING directory for device \a mdd
495 * The PENDING directory persistently tracks files and directories that were
496 * unlinked from the namespace (nlink == 0) but are still held open by clients.
497 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
498 * would not be able to recover and reopen those files. Instead, these inodes
499 * are linked into the PENDING directory on disk, and only deleted if all
500 * clients close them, or the MDS finishes client recovery without any client
501 * reopening them (i.e. former clients didn't join recovery).
502 * \param d mdd device being started.
505 * \retval -ve index operation error.
508 int mdd_orphan_index_init(const struct lu_env *env, struct mdd_device *mdd)
516 /* create PENDING dir */
518 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
519 mdd_orphan_index_name, S_IFDIR | S_IRUGO |
520 S_IWUSR | S_IXUGO, &fid);
524 d = dt_locate(env, mdd->mdd_child, &fid);
527 LASSERT(lu_object_exists(&d->do_lu));
528 if (!dt_try_as_dir(env, d)) {
529 CERROR("%s: orphan dir '%s' is not an index: rc = %d\n",
530 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name, rc);
531 dt_object_put(env, d);
534 mdd->mdd_orphans = d;
538 void mdd_orphan_index_fini(const struct lu_env *env, struct mdd_device *mdd)
541 if (mdd->mdd_orphans != NULL) {
542 dt_object_put(env, mdd->mdd_orphans);
543 mdd->mdd_orphans = NULL;
548 static int mdd_orphan_cleanup_thread(void *args)
550 struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
551 struct lu_env *env = NULL;
555 complete(&thread->mgt_started);
559 GOTO(out, rc = -ENOMEM);
561 rc = lu_env_init(env, LCT_MD_THREAD);
565 rc = mdd_orphan_index_iterate(env, thread);
572 complete(&thread->mgt_finished);
577 * Iterate orphan index to cleanup orphan objects after recovery is done.
578 * \param d mdd device in recovery.
580 int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
585 OBD_ALLOC(name, MTI_NAME_MAXLEN);
589 snprintf(name, MTI_NAME_MAXLEN, "orph_%s", mdd2obd_dev(d)->obd_name);
591 rc = mdd_generic_thread_start(&d->mdd_orphan_cleanup_thread,
592 mdd_orphan_cleanup_thread, d, name);
595 CERROR("%s: start orphan cleanup thread failed: rc = %d\n",
596 mdd2obd_dev(d)->obd_name, rc);
598 OBD_FREE(name, MTI_NAME_MAXLEN);