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 <obd_support.h>
45 #include <lustre_fid.h>
46 #include "mdd_internal.h"
48 const char orph_index_name[] = "PENDING";
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 *orph_key_fill(const struct lu_env *env,
59 const struct lu_fid *lf)
61 char *key = mdd_env_info(env)->mti_key;
65 rc = snprintf(key, NAME_MAX + 1, DFID_NOBRACE, PFID(lf));
67 return (struct dt_key *)key;
72 /* compatibility with orphan files created in versions before 2.11 */
73 static struct dt_key *orph_key_fill_20(const struct lu_env *env,
74 const struct lu_fid *lf)
76 char *key = mdd_env_info(env)->mti_key;
80 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_20,
81 fid_seq(lf), fid_oid(lf), fid_ver(lf), ORPH_OP_UNLINK);
83 return (struct dt_key *)key;
88 static inline void mdd_orphan_write_lock(const struct lu_env *env,
89 struct mdd_device *mdd)
91 struct dt_object *dor = mdd->mdd_orphans;
92 dt_write_lock(env, dor, MOR_TGT_ORPHAN);
95 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
96 struct mdd_device *mdd)
98 struct dt_object *dor = mdd->mdd_orphans;
99 dt_write_unlock(env, dor);
102 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
103 struct mdd_device *mdd,
104 struct mdd_object *obj,
105 __u32 op, struct thandle *th)
107 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
108 struct dt_object *dor = mdd->mdd_orphans;
109 const struct lu_fid *lf = mdo2fid(obj);
110 struct dt_key *key = orph_key_fill(env, lf, op);
113 rec->rec_type = mdd_object_type(obj);
115 return dt_insert(env, dor, (const struct dt_rec *)rec, key, th, 1);
118 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
119 struct mdd_device *mdd ,
123 struct dt_object *dor = mdd->mdd_orphans;
125 return dt_delete(env, dor, key, th);
128 static inline int mdd_orphan_ref_add(const struct lu_env *env,
129 struct mdd_device *mdd,
132 struct dt_object *dor = mdd->mdd_orphans;
133 return dt_ref_add(env, dor, th);
136 static inline int mdd_orphan_ref_del(const struct lu_env *env,
137 struct mdd_device *mdd,
140 struct dt_object *dor = mdd->mdd_orphans;
141 return dt_ref_del(env, dor, th);
145 int orph_declare_index_insert(const struct lu_env *env, struct mdd_object *obj,
146 umode_t mode, struct thandle *th)
148 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
149 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
153 key = orph_key_fill(env, mdo2fid(obj));
155 rec->rec_fid = mdo2fid(obj);
156 rec->rec_type = mode;
157 rc = dt_declare_insert(env, mdd->mdd_orphans,
158 (const struct dt_rec *)rec, key, th);
162 rc = mdo_declare_ref_add(env, obj, th);
169 rc = mdo_declare_ref_add(env, obj, th);
173 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
177 rc = mdo_declare_index_delete(env, obj, dotdot, th);
181 rc = mdo_declare_index_insert(env, obj,
182 lu_object_fid(&mdd->mdd_orphans->do_lu),
183 S_IFDIR, dotdot, th);
188 static int orph_index_insert(const struct lu_env *env,
189 struct mdd_object *obj,
190 __u32 op, struct thandle *th)
192 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
193 struct dt_object *dor = mdd->mdd_orphans;
194 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
195 struct dt_object *next = mdd_object_child(obj);
196 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
200 LASSERT(mdd_write_locked(env, obj) != 0);
201 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
203 mdd_orphan_write_lock(env, mdd);
205 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
209 mdo_ref_add(env, obj, th);
210 if (!S_ISDIR(mdd_object_type(obj)))
213 mdo_ref_add(env, obj, th);
214 mdd_orphan_ref_add(env, mdd, th);
216 /* try best to fixup directory, dont return errors
218 if (!dt_try_as_dir(env, next))
221 dt_delete(env, next, (const struct dt_key *)dotdot, th);
223 rec->rec_fid = lf_dor;
224 rec->rec_type = S_IFDIR;
225 dt_insert(env, next, (const struct dt_rec *)rec,
226 (const struct dt_key *)dotdot, th, 1);
230 obj->mod_flags |= ORPHAN_OBJ;
232 mdd_orphan_write_unlock(env, mdd);
237 int orph_declare_index_delete(const struct lu_env *env, struct mdd_object *obj,
240 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
244 key = orph_key_fill(env, mdo2fid(obj));
246 rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
250 rc = mdo_declare_ref_del(env, obj, th);
254 if (S_ISDIR(mdd_object_type(obj))) {
255 rc = mdo_declare_ref_del(env, obj, th);
259 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
265 static int orph_index_delete(const struct lu_env *env, struct mdd_object *obj,
266 __u32 op, struct thandle *th)
268 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
269 struct dt_object *dor = mdd->mdd_orphans;
275 LASSERT(mdd_write_locked(env, obj) != 0);
276 LASSERT(obj->mod_flags & ORPHAN_OBJ);
277 LASSERT(obj->mod_count == 0);
281 key = orph_key_fill(env, mdo2fid(obj), op);
282 mdd_orphan_write_lock(env, mdd);
284 rc = mdd_orphan_delete_obj(env, mdd, key, th);
287 key = orph_key_fill_20(env, mdo2fid(obj));
288 rc = mdd_orphan_delete_obj(env, mdd, key, th);
292 /* lov objects will be destroyed by caller */
293 mdo_ref_del(env, obj, th);
294 if (S_ISDIR(mdd_object_type(obj))) {
295 mdo_ref_del(env, obj, th);
296 mdd_orphan_ref_del(env, mdd, th);
298 obj->mod_flags &= ~ORPHAN_OBJ;
300 CERROR("%s: could not delete orphan object "DFID": rc = %d\n",
301 mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)), rc);
304 mdd_orphan_write_unlock(env, mdd);
309 static int orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
312 struct thandle *th = NULL;
313 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
317 th = mdd_trans_create(env, mdd);
319 CERROR("Cannot get thandle\n");
323 rc = orph_declare_index_delete(env, obj, th);
327 rc = mdo_declare_destroy(env, obj, th);
331 rc = mdd_trans_start(env, mdd, th);
335 mdd_write_lock(env, obj, MOR_TGT_CHILD);
336 if (likely(obj->mod_count == 0)) {
337 mdd_orphan_write_lock(env, mdd);
338 rc = mdd_orphan_delete_obj(env, mdd, key, th);
340 mdo_ref_del(env, obj, th);
341 if (S_ISDIR(mdd_object_type(obj))) {
342 mdo_ref_del(env, obj, th);
343 mdd_orphan_ref_del(env, mdd, th);
345 rc = mdo_destroy(env, obj, th);
347 CERROR("could not delete object: rc = %d\n", rc);
348 mdd_orphan_write_unlock(env, mdd);
350 mdd_write_unlock(env, obj);
353 rc = mdd_trans_stop(env, mdd, 0, th);
359 * Delete unused orphan with FID \a lf from PENDING directory
361 * \param mdd MDD device finishing recovery
362 * \param lf FID of file or directory to delete
363 * \param key cookie for this entry in index iterator
368 static int orph_key_test_and_del(const struct lu_env *env,
369 struct mdd_device *mdd, struct lu_fid *lf,
372 struct mdd_object *mdo;
375 mdo = mdd_object_find(env, mdd, lf);
381 if (mdo->mod_count == 0) {
382 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
383 rc = orphan_destroy(env, mdo, key);
384 if (rc) /* so replay-single.sh test_37 works */
385 CERROR("%s: error unlinking orphan "DFID" from "
386 "PENDING: rc = %d\n",
387 mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
389 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
390 if (likely(mdo->mod_count > 0)) {
391 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
392 PFID(lf), mdo->mod_count);
393 mdo->mod_flags |= ORPHAN_OBJ;
395 mdd_write_unlock(env, mdo);
398 mdd_object_put(env, mdo);
403 * delete unreferenced files and directories in the PENDING directory
405 * Files that remain in PENDING after client->MDS recovery has completed
406 * have to be referenced (opened) by some client during recovery, or they
407 * will be deleted here (for clients that did not complete recovery).
409 * \param thread info about orphan cleanup thread
414 static int orph_index_iterate(const struct lu_env *env,
415 struct mdd_generic_thread *thread)
417 struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
418 struct dt_object *dor = mdd->mdd_orphans;
419 struct lu_dirent *ent = &mdd_env_info(env)->mti_ent;
420 const struct dt_it_ops *iops;
428 iops = &dor->do_index_ops->dio_it;
429 it = iops->init(env, dor, LUDA_64BITHASH);
432 CERROR("%s: cannot clean PENDING: rc = %d\n",
433 mdd2obd_dev(mdd)->obd_name, rc);
437 rc = iops->load(env, it, 0);
441 CERROR("%s: error loading iterator to clean PENDING\n",
442 mdd2obd_dev(mdd)->obd_name);
443 /* Index contains no zero key? */
444 GOTO(out_put, rc = -EIO);
448 if (thread->mgt_abort)
451 key_sz = iops->key_size(env, it);
452 /* filter out "." and ".." entries from PENDING dir. */
456 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
458 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
459 mdd2obd_dev(mdd)->obd_name, rc);
463 fid_le_to_cpu(&fid, &ent->lde_fid);
464 if (!fid_is_sane(&fid)) {
465 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
466 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
470 /* kill orphan object */
471 cookie = iops->store(env, it);
473 rc = orph_key_test_and_del(env, mdd, &fid,
474 (struct dt_key *)ent->lde_name);
476 /* after index delete reset iterator */
478 rc = iops->get(env, it, (const void *)"");
480 rc = iops->load(env, it, cookie);
482 rc = iops->next(env, it);
485 GOTO(out_put, rc = 0);
495 * open the PENDING directory for device \a mdd
497 * The PENDING directory persistently tracks files and directories that were
498 * unlinked from the namespace (nlink == 0) but are still held open by clients.
499 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
500 * would not be able to recover and reopen those files. Instead, these inodes
501 * are linked into the PENDING directory on disk, and only deleted if all
502 * clients close them, or the MDS finishes client recovery without any client
503 * reopening them (i.e. former clients didn't join recovery).
504 * \param d mdd device being started.
507 * \retval -ve index operation error.
510 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
518 /* create PENDING dir */
520 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
521 orph_index_name, S_IFDIR | S_IRUGO |
522 S_IWUSR | S_IXUGO, &fid);
526 d = dt_locate(env, mdd->mdd_child, &fid);
529 LASSERT(lu_object_exists(&d->do_lu));
530 if (!dt_try_as_dir(env, d)) {
531 CERROR("%s: \"%s\" is not an index: rc = %d\n",
532 mdd2obd_dev(mdd)->obd_name, orph_index_name, rc);
533 dt_object_put(env, d);
536 mdd->mdd_orphans = d;
540 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
543 if (mdd->mdd_orphans != NULL) {
544 dt_object_put(env, mdd->mdd_orphans);
545 mdd->mdd_orphans = NULL;
550 static int __mdd_orphan_cleanup(void *args)
552 struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
553 struct lu_env *env = NULL;
557 complete(&thread->mgt_started);
561 GOTO(out, rc = -ENOMEM);
563 rc = lu_env_init(env, LCT_MD_THREAD);
567 rc = orph_index_iterate(env, thread);
574 complete(&thread->mgt_finished);
579 * Iterate orphan index to cleanup orphan objects after recovery is done.
580 * \param d mdd device in recovery.
582 int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
587 OBD_ALLOC(name, MTI_NAME_MAXLEN);
591 snprintf(name, MTI_NAME_MAXLEN, "orph_cleanup_%s",
592 mdd2obd_dev(d)->obd_name);
594 rc = mdd_generic_thread_start(&d->mdd_orph_cleanup_thread,
595 __mdd_orphan_cleanup, (void *)d, name);
598 CERROR("%s: start orphan cleanup thread failed:%d\n",
599 mdd2obd_dev(d)->obd_name, rc);
601 OBD_FREE(name, MTI_NAME_MAXLEN);
606 * add an orphan \a obj to the orphan index.
607 * \param obj file or directory.
608 * \param th transaction for index insert.
610 * \pre obj nlink == 0 && obj->mod_count != 0
613 * \retval -ve index operation error.
615 int __mdd_orphan_add(const struct lu_env *env, struct mdd_object *obj,
618 return orph_index_insert(env, obj, th);
622 * delete an orphan \a obj from orphan index.
623 * \param obj file or directory.
624 * \param th transaction for index deletion and object destruction.
626 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
629 * \retval -ve index operation error.
631 int __mdd_orphan_del(const struct lu_env *env, struct mdd_object *obj,
634 return orph_index_delete(env, obj, th);