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[] = "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 *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 = mdo2fid(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, 1);
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, mdo2fid(obj));
110 rec->rec_fid = mdo2fid(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, MOR_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, 1);
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, mdo2fid(obj));
209 rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
213 rc = mdo_declare_ref_del(env, obj, th);
217 if (S_ISDIR(mdd_object_type(obj))) {
218 rc = mdo_declare_ref_del(env, obj, th);
222 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
229 * delete an orphan \a obj from orphan index.
230 * \param obj file or directory.
231 * \param th transaction for index deletion and object destruction.
233 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
236 * \retval -ve index operation error.
238 int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
241 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
242 struct dt_object *dor = mdd->mdd_orphans;
248 LASSERT(mdd_write_locked(env, obj) != 0);
249 LASSERT(obj->mod_flags & ORPHAN_OBJ);
250 LASSERT(obj->mod_count == 0);
254 key = mdd_orphan_key_fill(env, mdo2fid(obj));
255 dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
257 rc = dt_delete(env, mdd->mdd_orphans, key, th);
259 key = mdd_orphan_key_fill_20(env, mdo2fid(obj));
260 rc = dt_delete(env, mdd->mdd_orphans, key, th);
264 /* lov objects will be destroyed by caller */
265 mdo_ref_del(env, obj, th);
266 if (S_ISDIR(mdd_object_type(obj))) {
267 mdo_ref_del(env, obj, th);
268 dt_ref_del(env, mdd->mdd_orphans, th);
270 obj->mod_flags &= ~ORPHAN_OBJ;
272 CERROR("%s: could not delete orphan object "DFID": rc = %d\n",
273 mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)), rc);
276 dt_write_unlock(env, mdd->mdd_orphans);
281 static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
284 struct thandle *th = NULL;
285 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
289 th = mdd_trans_create(env, mdd);
292 if (rc != -EINPROGRESS)
293 CERROR("%s: cannot get orphan thandle: rc = %d\n",
294 mdd2obd_dev(mdd)->obd_name, rc);
298 rc = mdd_orphan_declare_delete(env, obj, th);
302 rc = mdo_declare_destroy(env, obj, th);
306 rc = mdd_trans_start(env, mdd, th);
310 mdd_write_lock(env, obj, MOR_TGT_CHILD);
311 if (likely(obj->mod_count == 0)) {
312 dt_write_lock(env, mdd->mdd_orphans, MOR_TGT_ORPHAN);
313 rc = dt_delete(env, mdd->mdd_orphans, key, th);
315 mdo_ref_del(env, obj, th);
316 if (S_ISDIR(mdd_object_type(obj))) {
317 mdo_ref_del(env, obj, th);
318 dt_ref_del(env, mdd->mdd_orphans, th);
320 rc = mdo_destroy(env, obj, th);
322 CERROR("%s: could not delete orphan "DFID": rc = %d\n",
323 mdd2obd_dev(mdd)->obd_name, PFID(mdo2fid(obj)),
326 dt_write_unlock(env, mdd->mdd_orphans);
328 mdd_write_unlock(env, obj);
331 rc = mdd_trans_stop(env, mdd, 0, th);
337 * Delete unused orphan with FID \a lf from PENDING directory
339 * \param mdd MDD device finishing recovery
340 * \param lf FID of file or directory to delete
341 * \param key cookie for this entry in index iterator
346 static int mdd_orphan_key_test_and_delete(const struct lu_env *env,
347 struct mdd_device *mdd,
348 struct lu_fid *lf, struct dt_key *key)
350 struct mdd_object *mdo;
353 mdo = mdd_object_find(env, mdd, lf);
359 if (mdo->mod_count == 0) {
360 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
361 rc = mdd_orphan_destroy(env, mdo, key);
362 if (rc) /* so replay-single.sh test_37 works */
363 CERROR("%s: error unlinking orphan "DFID": rc = %d\n",
364 mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
366 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
367 if (likely(mdo->mod_count > 0)) {
368 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
369 PFID(lf), mdo->mod_count);
370 mdo->mod_flags |= ORPHAN_OBJ;
372 mdd_write_unlock(env, mdo);
375 mdd_object_put(env, mdo);
380 * delete unreferenced files and directories in the PENDING directory
382 * Files that remain in PENDING after client->MDS recovery has completed
383 * have to be referenced (opened) by some client during recovery, or they
384 * will be deleted here (for clients that did not complete recovery).
386 * \param thread info about orphan cleanup thread
391 static int mdd_orphan_index_iterate(const struct lu_env *env,
392 struct mdd_generic_thread *thread)
394 struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
395 struct dt_object *dor = mdd->mdd_orphans;
396 struct lu_dirent *ent = &mdd_env_info(env)->mti_ent;
397 const struct dt_it_ops *iops;
405 iops = &dor->do_index_ops->dio_it;
406 it = iops->init(env, dor, LUDA_64BITHASH);
409 CERROR("%s: cannot clean '%s': rc = %d\n",
410 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name, rc);
414 rc = iops->load(env, it, 0);
418 CERROR("%s: error loading iterator to clean '%s'\n",
419 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name);
420 /* Index contains no zero key? */
421 GOTO(out_put, rc = -EIO);
425 if (thread->mgt_abort)
428 key_sz = iops->key_size(env, it);
429 /* filter out "." and ".." entries from PENDING dir. */
433 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
435 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
436 mdd2obd_dev(mdd)->obd_name, rc);
440 fid_le_to_cpu(&fid, &ent->lde_fid);
441 if (!fid_is_sane(&fid)) {
442 CERROR("%s: bad FID "DFID" cleaning '%s'\n",
443 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
444 mdd_orphan_index_name);
448 /* kill orphan object */
449 cookie = iops->store(env, it);
451 rc = mdd_orphan_key_test_and_delete(env, mdd, &fid,
452 (struct dt_key *)ent->lde_name);
454 /* after index delete reset iterator */
456 rc = iops->get(env, it, (const void *)"");
458 rc = iops->load(env, it, cookie);
460 rc = iops->next(env, it);
463 GOTO(out_put, rc = 0);
473 * open the PENDING directory for device \a mdd
475 * The PENDING directory persistently tracks files and directories that were
476 * unlinked from the namespace (nlink == 0) but are still held open by clients.
477 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
478 * would not be able to recover and reopen those files. Instead, these inodes
479 * are linked into the PENDING directory on disk, and only deleted if all
480 * clients close them, or the MDS finishes client recovery without any client
481 * reopening them (i.e. former clients didn't join recovery).
482 * \param d mdd device being started.
485 * \retval -ve index operation error.
488 int mdd_orphan_index_init(const struct lu_env *env, struct mdd_device *mdd)
496 /* create PENDING dir */
498 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
499 mdd_orphan_index_name, S_IFDIR | S_IRUGO |
500 S_IWUSR | S_IXUGO, &fid);
504 d = dt_locate(env, mdd->mdd_child, &fid);
507 LASSERT(lu_object_exists(&d->do_lu));
508 if (!dt_try_as_dir(env, d)) {
509 CERROR("%s: orphan dir '%s' is not an index: rc = %d\n",
510 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name, rc);
511 dt_object_put(env, d);
514 mdd->mdd_orphans = d;
518 void mdd_orphan_index_fini(const struct lu_env *env, struct mdd_device *mdd)
521 if (mdd->mdd_orphans != NULL) {
522 dt_object_put(env, mdd->mdd_orphans);
523 mdd->mdd_orphans = NULL;
528 static int mdd_orphan_cleanup_thread(void *args)
530 struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
531 struct lu_env *env = NULL;
535 complete(&thread->mgt_started);
539 GOTO(out, rc = -ENOMEM);
541 rc = lu_env_init(env, LCT_MD_THREAD);
545 rc = mdd_orphan_index_iterate(env, thread);
552 complete(&thread->mgt_finished);
557 * Iterate orphan index to cleanup orphan objects after recovery is done.
558 * \param d mdd device in recovery.
560 int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
565 OBD_ALLOC(name, MTI_NAME_MAXLEN);
569 snprintf(name, MTI_NAME_MAXLEN, "orph_%s", mdd2obd_dev(d)->obd_name);
571 rc = mdd_generic_thread_start(&d->mdd_orphan_cleanup_thread,
572 mdd_orphan_cleanup_thread, d, name);
575 CERROR("%s: start orphan cleanup thread failed: rc = %d\n",
576 mdd2obd_dev(d)->obd_name, rc);
578 OBD_FREE(name, MTI_NAME_MAXLEN);