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[] = "..";
56 #define ORPHAN_FILE_NAME_FORMAT "%016llx:%08x:%08x:%2x"
57 #define ORPHAN_FILE_NAME_FORMAT_18 "%llx:%08x"
59 static struct dt_key* orph_key_fill(const struct lu_env *env,
60 const struct lu_fid *lf, __u32 op)
62 char *key = mdd_env_info(env)->mti_key;
66 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT,
67 (long long unsigned int)fid_seq(lf),
68 fid_oid(lf), fid_ver(lf), op);
70 return (struct dt_key*) key;
75 static struct dt_key* orph_key_fill_18(const struct lu_env *env,
76 const struct lu_fid *lf)
78 char *key = mdd_env_info(env)->mti_key;
82 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_18,
83 (unsigned long long)fid_seq(lf), fid_oid(lf));
85 return (struct dt_key*) key;
90 static inline void mdd_orphan_write_lock(const struct lu_env *env,
91 struct mdd_device *mdd)
93 struct dt_object *dor = mdd->mdd_orphans;
94 dt_write_lock(env, dor, MOR_TGT_ORPHAN);
97 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
98 struct mdd_device *mdd)
100 struct dt_object *dor = mdd->mdd_orphans;
101 dt_write_unlock(env, dor);
104 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
105 struct mdd_device *mdd,
106 struct mdd_object *obj,
107 __u32 op, struct thandle *th)
109 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
110 struct dt_object *dor = mdd->mdd_orphans;
111 const struct lu_fid *lf = mdo2fid(obj);
112 struct dt_key *key = orph_key_fill(env, lf, op);
115 rec->rec_type = mdd_object_type(obj);
117 return dt_insert(env, dor, (const struct dt_rec *)rec, key, th, 1);
120 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
121 struct mdd_device *mdd ,
125 struct dt_object *dor = mdd->mdd_orphans;
127 return dt_delete(env, dor, key, th);
130 static inline int mdd_orphan_ref_add(const struct lu_env *env,
131 struct mdd_device *mdd,
134 struct dt_object *dor = mdd->mdd_orphans;
135 return dt_ref_add(env, dor, th);
138 static inline int mdd_orphan_ref_del(const struct lu_env *env,
139 struct mdd_device *mdd,
142 struct dt_object *dor = mdd->mdd_orphans;
143 return dt_ref_del(env, dor, th);
147 int orph_declare_index_insert(const struct lu_env *env,
148 struct mdd_object *obj,
149 umode_t mode, struct thandle *th)
151 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
152 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
156 key = orph_key_fill(env, mdo2fid(obj), ORPH_OP_UNLINK);
158 rec->rec_fid = mdo2fid(obj);
159 rec->rec_type = mode;
160 rc = dt_declare_insert(env, mdd->mdd_orphans,
161 (const struct dt_rec *)rec, key, th);
165 rc = mdo_declare_ref_add(env, obj, th);
172 rc = mdo_declare_ref_add(env, obj, th);
176 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
180 rc = mdo_declare_index_delete(env, obj, dotdot, th);
184 rc = mdo_declare_index_insert(env, obj,
185 lu_object_fid(&mdd->mdd_orphans->do_lu),
186 S_IFDIR, dotdot, th);
191 static int orph_index_insert(const struct lu_env *env,
192 struct mdd_object *obj,
193 __u32 op, struct thandle *th)
195 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
196 struct dt_object *dor = mdd->mdd_orphans;
197 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
198 struct dt_object *next = mdd_object_child(obj);
199 struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec;
203 LASSERT(mdd_write_locked(env, obj) != 0);
204 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
206 mdd_orphan_write_lock(env, mdd);
208 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
212 mdo_ref_add(env, obj, th);
213 if (!S_ISDIR(mdd_object_type(obj)))
216 mdo_ref_add(env, obj, th);
217 mdd_orphan_ref_add(env, mdd, th);
219 /* try best to fixup directory, dont return errors
221 if (!dt_try_as_dir(env, next))
224 dt_delete(env, next, (const struct dt_key *)dotdot, th);
226 rec->rec_fid = lf_dor;
227 rec->rec_type = S_IFDIR;
228 dt_insert(env, next, (const struct dt_rec *)rec,
229 (const struct dt_key *)dotdot, th, 1);
233 obj->mod_flags |= ORPHAN_OBJ;
235 mdd_orphan_write_unlock(env, mdd);
240 int orph_declare_index_delete(const struct lu_env *env,
241 struct mdd_object *obj,
245 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
249 key = orph_key_fill(env, mdo2fid(obj), ORPH_OP_UNLINK);
251 rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
255 rc = mdo_declare_ref_del(env, obj, th);
259 if (S_ISDIR(mdd_object_type(obj))) {
260 rc = mdo_declare_ref_del(env, obj, th);
264 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
270 static int orph_index_delete(const struct lu_env *env,
271 struct mdd_object *obj,
275 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
276 struct dt_object *dor = mdd->mdd_orphans;
282 LASSERT(mdd_write_locked(env, obj) != 0);
283 LASSERT(obj->mod_flags & ORPHAN_OBJ);
284 LASSERT(obj->mod_count == 0);
288 key = orph_key_fill(env, mdo2fid(obj), op);
289 mdd_orphan_write_lock(env, mdd);
291 rc = mdd_orphan_delete_obj(env, mdd, key, th);
294 key = orph_key_fill_18(env, mdo2fid(obj));
295 rc = mdd_orphan_delete_obj(env, mdd, key, th);
299 /* lov objects will be destroyed by caller */
300 mdo_ref_del(env, obj, th);
301 if (S_ISDIR(mdd_object_type(obj))) {
302 mdo_ref_del(env, obj, th);
303 mdd_orphan_ref_del(env, mdd, th);
305 obj->mod_flags &= ~ORPHAN_OBJ;
307 CERROR("could not delete object: rc = %d\n",rc);
310 mdd_orphan_write_unlock(env, mdd);
315 static int orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
318 struct thandle *th = NULL;
319 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
323 th = mdd_trans_create(env, mdd);
325 CERROR("Cannot get thandle\n");
329 rc = orph_declare_index_delete(env, obj, th);
333 rc = mdo_declare_destroy(env, obj, th);
337 rc = mdd_trans_start(env, mdd, th);
341 mdd_write_lock(env, obj, MOR_TGT_CHILD);
342 if (likely(obj->mod_count == 0)) {
343 mdd_orphan_write_lock(env, mdd);
344 rc = mdd_orphan_delete_obj(env, mdd, key, th);
346 mdo_ref_del(env, obj, th);
347 if (S_ISDIR(mdd_object_type(obj))) {
348 mdo_ref_del(env, obj, th);
349 mdd_orphan_ref_del(env, mdd, th);
351 rc = mdo_destroy(env, obj, th);
353 CERROR("could not delete object: rc = %d\n", rc);
354 mdd_orphan_write_unlock(env, mdd);
356 mdd_write_unlock(env, obj);
359 rc = mdd_trans_stop(env, mdd, 0, th);
365 * Delete unused orphan with FID \a lf from PENDING directory
367 * \param mdd MDD device finishing recovery
368 * \param lf FID of file or directory to delete
369 * \param key cookie for this entry in index iterator
374 static int orph_key_test_and_del(const struct lu_env *env,
375 struct mdd_device *mdd, struct lu_fid *lf,
378 struct mdd_object *mdo;
381 mdo = mdd_object_find(env, mdd, lf);
387 if (mdo->mod_count == 0) {
388 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
389 rc = orphan_destroy(env, mdo, key);
390 if (rc) /* so replay-single.sh test_37 works */
391 CERROR("%s: error unlinking orphan "DFID" from "
392 "PENDING: rc = %d\n",
393 mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
395 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
396 if (likely(mdo->mod_count > 0)) {
397 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
398 PFID(lf), mdo->mod_count);
399 mdo->mod_flags |= ORPHAN_OBJ;
401 mdd_write_unlock(env, mdo);
404 mdd_object_put(env, mdo);
409 * delete unreferenced files and directories in the PENDING directory
411 * Files that remain in PENDING after client->MDS recovery has completed
412 * have to be referenced (opened) by some client during recovery, or they
413 * will be deleted here (for clients that did not complete recovery).
415 * \param thread info about orphan cleanup thread
420 static int orph_index_iterate(const struct lu_env *env,
421 struct mdd_generic_thread *thread)
423 struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
424 struct dt_object *dor = mdd->mdd_orphans;
425 struct lu_dirent *ent = &mdd_env_info(env)->mti_ent;
426 const struct dt_it_ops *iops;
434 iops = &dor->do_index_ops->dio_it;
435 it = iops->init(env, dor, LUDA_64BITHASH);
438 CERROR("%s: cannot clean PENDING: rc = %d\n",
439 mdd2obd_dev(mdd)->obd_name, rc);
443 rc = iops->load(env, it, 0);
447 CERROR("%s: error loading iterator to clean PENDING\n",
448 mdd2obd_dev(mdd)->obd_name);
449 /* Index contains no zero key? */
450 GOTO(out_put, rc = -EIO);
454 if (thread->mgt_abort)
457 key_sz = iops->key_size(env, it);
458 /* filter out "." and ".." entries from PENDING dir. */
462 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
464 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
465 mdd2obd_dev(mdd)->obd_name, rc);
469 fid_le_to_cpu(&fid, &ent->lde_fid);
470 if (!fid_is_sane(&fid)) {
471 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
472 mdd2obd_dev(mdd)->obd_name, PFID(&fid));
476 /* kill orphan object */
477 cookie = iops->store(env, it);
479 rc = orph_key_test_and_del(env, mdd, &fid,
480 (struct dt_key *)ent->lde_name);
482 /* after index delete reset iterator */
484 rc = iops->get(env, it, (const void *)"");
486 rc = iops->load(env, it, cookie);
488 rc = iops->next(env, it);
491 GOTO(out_put, rc = 0);
501 * open the PENDING directory for device \a mdd
503 * The PENDING directory persistently tracks files and directories that were
504 * unlinked from the namespace (nlink == 0) but are still held open by clients.
505 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
506 * would not be able to recover and reopen those files. Instead, these inodes
507 * are linked into the PENDING directory on disk, and only deleted if all
508 * clients close them, or the MDS finishes client recovery without any client
509 * reopening them (i.e. former clients didn't join recovery).
510 * \param d mdd device being started.
513 * \retval -ve index operation error.
516 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
524 /* create PENDING dir */
526 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
527 orph_index_name, S_IFDIR | S_IRUGO |
528 S_IWUSR | S_IXUGO, &fid);
532 d = dt_locate(env, mdd->mdd_child, &fid);
535 LASSERT(lu_object_exists(&d->do_lu));
536 if (!dt_try_as_dir(env, d)) {
537 CERROR("%s: \"%s\" is not an index: rc = %d\n",
538 mdd2obd_dev(mdd)->obd_name, orph_index_name, rc);
539 dt_object_put(env, d);
542 mdd->mdd_orphans = d;
546 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
549 if (mdd->mdd_orphans != NULL) {
550 dt_object_put(env, mdd->mdd_orphans);
551 mdd->mdd_orphans = NULL;
556 static int __mdd_orphan_cleanup(void *args)
558 struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
559 struct lu_env *env = NULL;
563 complete(&thread->mgt_started);
567 GOTO(out, rc = -ENOMEM);
569 rc = lu_env_init(env, LCT_MD_THREAD);
573 rc = orph_index_iterate(env, thread);
580 complete(&thread->mgt_finished);
585 * Iterate orphan index to cleanup orphan objects after recovery is done.
586 * \param d mdd device in recovery.
588 int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
593 OBD_ALLOC(name, MTI_NAME_MAXLEN);
597 snprintf(name, MTI_NAME_MAXLEN, "orph_cleanup_%s",
598 mdd2obd_dev(d)->obd_name);
600 rc = mdd_generic_thread_start(&d->mdd_orph_cleanup_thread,
601 __mdd_orphan_cleanup, (void *)d, name);
604 CERROR("%s: start orphan cleanup thread failed:%d\n",
605 mdd2obd_dev(d)->obd_name, rc);
607 OBD_FREE(name, MTI_NAME_MAXLEN);
612 * add an orphan \a obj to the orphan index.
613 * \param obj file or directory.
614 * \param th transaction for index insert.
616 * \pre obj nlink == 0 && obj->mod_count != 0
619 * \retval -ve index operation error.
621 int __mdd_orphan_add(const struct lu_env *env,
622 struct mdd_object *obj, struct thandle *th)
624 return orph_index_insert(env, obj, ORPH_OP_UNLINK, th);
628 * delete an orphan \a obj from orphan index.
629 * \param obj file or directory.
630 * \param th transaction for index deletion and object destruction.
632 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
635 * \retval -ve index operation error.
637 int __mdd_orphan_del(const struct lu_env *env,
638 struct mdd_object *obj, struct thandle *th)
640 return orph_index_delete(env, obj, ORPH_OP_UNLINK, th);