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/
31 * lustre/mdd/mdd_orphans.c
33 * Orphan handling code
35 * Author: Mike Pershin <tappro@clusterfs.com>
36 * Pravin B Shelar <pravin.shelar@sun.com>
39 #define DEBUG_SUBSYSTEM S_MDS
42 #include <obd_class.h>
43 #include <obd_support.h>
44 #include <lustre_fid.h>
45 #include "mdd_internal.h"
47 static const char mdd_orphan_index_name[] = MDT_ORPHAN_DIR;
48 static const char dotdot[] = "..";
54 /* obsolete after 2.11, needed for upgrades from older 2.x versions */
55 #define ORPHAN_FILE_NAME_FORMAT_20 "%016llx:%08x:%08x:%2x"
57 static struct dt_key *mdd_orphan_key_fill(const struct lu_env *env,
58 const struct lu_fid *lf)
60 char *key = mdd_env_info(env)->mdi_key;
63 if (!(MDI_KEEP_KEY & mdd_env_info(env)->mdi_flags))
64 snprintf(key, sizeof(mdd_env_info(env)->mdi_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)->mdi_key;
77 if (!(MDI_KEEP_KEY & mdd_env_info(env)->mdi_flags))
78 snprintf(key, sizeof(mdd_env_info(env)->mdi_key),
79 ORPHAN_FILE_NAME_FORMAT_20,
80 fid_seq(lf), fid_oid(lf), fid_ver(lf),
83 return (struct dt_key *)key;
86 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
87 struct mdd_device *mdd,
88 struct mdd_object *obj,
91 struct dt_insert_rec *rec = &mdd_env_info(env)->mdi_dt_rec;
92 struct dt_object *dor = mdd->mdd_orphans;
93 const struct lu_fid *lf = mdd_object_fid(obj);
94 struct dt_key *key = mdd_orphan_key_fill(env, lf);
97 rec->rec_type = mdd_object_type(obj);
99 return dt_insert(env, dor, (const struct dt_rec *)rec, key, th);
102 int mdd_orphan_declare_insert(const struct lu_env *env, struct mdd_object *obj,
103 umode_t mode, struct thandle *th)
105 struct dt_insert_rec *rec = &mdd_env_info(env)->mdi_dt_rec;
106 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
110 key = mdd_orphan_key_fill(env, mdd_object_fid(obj));
112 rec->rec_fid = mdd_object_fid(obj);
113 rec->rec_type = mode;
114 rc = dt_declare_insert(env, mdd->mdd_orphans,
115 (const struct dt_rec *)rec, key, th);
119 rc = mdo_declare_ref_add(env, obj, th);
126 rc = mdo_declare_ref_add(env, obj, th);
130 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
134 rc = mdo_declare_index_delete(env, obj, dotdot, th);
138 rc = mdo_declare_index_insert(env, obj,
139 lu_object_fid(&mdd->mdd_orphans->do_lu),
140 S_IFDIR, dotdot, th);
146 * add an orphan \a obj to the orphan index.
147 * \param obj file or directory.
148 * \param th transaction for index insert.
150 * \pre obj nlink == 0 && obj->mod_count != 0
153 * \retval -ve index operation error.
155 int mdd_orphan_insert(const struct lu_env *env, struct mdd_object *obj,
158 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
159 struct dt_object *dor = mdd->mdd_orphans;
160 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
161 struct dt_object *next = mdd_object_child(obj);
162 struct dt_insert_rec *rec = &mdd_env_info(env)->mdi_dt_rec;
166 LASSERT(mdd_write_locked(env, obj) != 0);
167 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
169 dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
171 rc = mdd_orphan_insert_obj(env, mdd, obj, th);
175 mdo_ref_add(env, obj, th);
176 if (!S_ISDIR(mdd_object_type(obj)))
179 mdo_ref_add(env, obj, th);
180 dt_ref_add(env, mdd->mdd_orphans, th);
182 /* try best to fixup directory, do not return errors from here */
183 if (!dt_try_as_dir(env, next))
186 dt_delete(env, next, (const struct dt_key *)dotdot, th);
188 rec->rec_fid = lf_dor;
189 rec->rec_type = S_IFDIR;
190 dt_insert(env, next, (const struct dt_rec *)rec,
191 (const struct dt_key *)dotdot, th);
195 obj->mod_flags |= ORPHAN_OBJ;
197 dt_write_unlock(env, mdd->mdd_orphans);
202 int mdd_orphan_declare_delete(const struct lu_env *env, struct mdd_object *obj,
205 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
209 key = mdd_orphan_key_fill(env, mdd_object_fid(obj));
211 rc = dt_declare_delete(env, mdd->mdd_orphans, key, th);
215 if (!mdd_object_exists(obj))
218 rc = mdo_declare_ref_del(env, obj, th);
222 if (S_ISDIR(mdd_object_type(obj))) {
223 rc = mdo_declare_ref_del(env, obj, th);
227 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
234 * delete an orphan \a obj from orphan index.
235 * \param obj file or directory.
236 * \param th transaction for index deletion and object destruction.
238 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
241 * \retval -ve index operation error.
243 int mdd_orphan_delete(const struct lu_env *env, struct mdd_object *obj,
246 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
247 struct dt_object *dor = mdd->mdd_orphans;
253 LASSERT(mdd_write_locked(env, obj) != 0);
254 LASSERT(obj->mod_flags & ORPHAN_OBJ);
255 LASSERT(obj->mod_count == 0);
259 key = mdd_orphan_key_fill(env, mdd_object_fid(obj));
260 dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
262 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ORPHAN_DELETE))
265 rc = dt_delete(env, mdd->mdd_orphans, key, th);
267 key = mdd_orphan_key_fill_20(env, mdd_object_fid(obj));
268 rc = dt_delete(env, mdd->mdd_orphans, key, th);
273 /* lov objects will be destroyed by caller */
274 mdo_ref_del(env, obj, th);
275 if (S_ISDIR(mdd_object_type(obj))) {
276 mdo_ref_del(env, obj, th);
277 dt_ref_del(env, mdd->mdd_orphans, th);
279 obj->mod_flags &= ~ORPHAN_OBJ;
281 CERROR("%s: could not delete orphan object "DFID": rc = %d\n",
282 mdd_obj_dev_name(obj), PFID(mdd_object_fid(obj)), rc);
285 dt_write_unlock(env, mdd->mdd_orphans);
290 static int mdd_orphan_destroy(const struct lu_env *env, struct mdd_object *obj,
293 struct thandle *th = NULL;
294 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
298 th = mdd_trans_create(env, mdd);
301 if (rc != -EINPROGRESS)
302 CERROR("%s: cannot get orphan thandle: rc = %d\n",
303 mdd2obd_dev(mdd)->obd_name, rc);
307 rc = mdd_orphan_declare_delete(env, obj, th);
308 if (rc && rc != -ENOENT)
312 rc = mdo_declare_destroy(env, obj, th);
313 if (rc && rc != -ENOENT)
317 rc = mdd_trans_start(env, mdd, th);
321 mdd_write_lock(env, obj, DT_TGT_CHILD);
322 if (likely(obj->mod_count == 0)) {
323 dt_write_lock(env, mdd->mdd_orphans, DT_TGT_ORPHAN);
324 rc = dt_delete(env, mdd->mdd_orphans, key, th);
325 /* We should remove object even dt_delete failed */
326 if (mdd_object_exists(obj) &&
327 !lu_object_is_dying(obj->mod_obj.mo_lu.lo_header)) {
328 mdo_ref_del(env, obj, th);
329 if (S_ISDIR(mdd_object_type(obj))) {
330 mdo_ref_del(env, obj, th);
331 dt_ref_del(env, mdd->mdd_orphans, th);
333 rc1 = mdo_destroy(env, obj, th);
335 dt_write_unlock(env, mdd->mdd_orphans);
337 mdd_write_unlock(env, obj);
339 mdd_trans_stop(env, mdd, 0, th);
341 RETURN(rc ? rc : rc1);
345 * Delete unused orphan with FID \a lf from PENDING directory
347 * \param mdd MDD device finishing recovery
348 * \param lf FID of file or directory to delete
349 * \param key cookie for this entry in index iterator
354 static int mdd_orphan_key_test_and_delete(const struct lu_env *env,
355 struct mdd_device *mdd,
356 struct lu_fid *lf, struct dt_key *key)
358 struct mdd_object *mdo;
361 mdo = mdd_object_find(env, mdd, lf);
367 if (mdo->mod_count == 0) {
368 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
369 rc = mdd_orphan_destroy(env, mdo, key);
370 if (rc) /* below message checked in replay-single.sh test_37 */
371 CERROR("%s: error unlinking orphan "DFID": rc = %d\n",
372 mdd2obd_dev(mdd)->obd_name, PFID(lf), rc);
374 mdd_write_lock(env, mdo, DT_TGT_CHILD);
375 if (likely(mdo->mod_count > 0)) {
376 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
377 PFID(lf), mdo->mod_count);
378 mdo->mod_flags |= ORPHAN_OBJ;
380 mdd_write_unlock(env, mdo);
383 mdd_object_put(env, mdo);
388 * delete unreferenced files and directories in the PENDING directory
390 * Files that remain in PENDING after client->MDS recovery has completed
391 * have to be referenced (opened) by some client during recovery, or they
392 * will be deleted here (for clients that did not complete recovery).
394 * \param thread info about orphan cleanup thread
399 static int mdd_orphan_index_iterate(const struct lu_env *env,
400 struct mdd_generic_thread *thread)
402 struct mdd_device *mdd = (struct mdd_device *)thread->mgt_data;
403 struct dt_object *dor = mdd->mdd_orphans;
404 struct lu_dirent *ent = &mdd_env_info(env)->mdi_ent;
405 const struct dt_it_ops *iops;
412 iops = &dor->do_index_ops->dio_it;
413 it = iops->init(env, dor, LUDA_64BITHASH);
416 CERROR("%s: cannot clean '%s': rc = %d\n",
417 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name, rc);
421 rc = iops->load(env, it, 0);
425 CERROR("%s: error loading iterator to clean '%s'\n",
426 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name);
427 /* Index contains no zero key? */
428 GOTO(out_put, rc = -EIO);
431 mdd_env_info(env)->mdi_flags |= MDI_KEEP_KEY;
433 if (thread->mgt_abort)
436 key_sz = iops->key_size(env, it);
437 /* filter out "." and ".." entries from PENDING dir. */
441 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
443 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
444 mdd2obd_dev(mdd)->obd_name, rc);
448 fid_le_to_cpu(&fid, &ent->lde_fid);
449 if (!fid_is_sane(&fid)) {
450 CERROR("%s: bad FID "DFID" cleaning '%s'\n",
451 mdd2obd_dev(mdd)->obd_name, PFID(&fid),
452 mdd_orphan_index_name);
456 /* kill orphan object */
458 rc = mdd_orphan_key_test_and_delete(env, mdd, &fid,
459 (struct dt_key *)ent->lde_name);
460 /* after index delete reset iterator */
462 rc = iops->get(env, it, (const void *)"");
464 rc = iops->next(env, it);
467 GOTO(out_put, rc = 0);
477 * open the PENDING directory for device \a mdd
479 * The PENDING directory persistently tracks files and directories that were
480 * unlinked from the namespace (nlink == 0) but are still held open by clients.
481 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
482 * would not be able to recover and reopen those files. Instead, these inodes
483 * are linked into the PENDING directory on disk, and only deleted if all
484 * clients close them, or the MDS finishes client recovery without any client
485 * reopening them (i.e. former clients didn't join recovery).
486 * \param d mdd device being started.
489 * \retval -ve index operation error.
492 int mdd_orphan_index_init(const struct lu_env *env, struct mdd_device *mdd)
500 /* create PENDING dir */
502 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
503 mdd_orphan_index_name, S_IFDIR | S_IRUGO |
504 S_IWUSR | S_IXUGO, &fid);
508 d = dt_locate(env, mdd->mdd_child, &fid);
511 LASSERT(lu_object_exists(&d->do_lu));
512 if (!dt_try_as_dir(env, d)) {
513 CERROR("%s: orphan dir '%s' is not an index: rc = %d\n",
514 mdd2obd_dev(mdd)->obd_name, mdd_orphan_index_name, rc);
515 dt_object_put(env, d);
518 mdd->mdd_orphans = d;
522 void mdd_orphan_index_fini(const struct lu_env *env, struct mdd_device *mdd)
525 if (mdd->mdd_orphans != NULL) {
526 dt_object_put(env, mdd->mdd_orphans);
527 mdd->mdd_orphans = NULL;
532 static int mdd_orphan_cleanup_thread(void *args)
534 struct mdd_generic_thread *thread = (struct mdd_generic_thread *)args;
535 struct lu_env *env = NULL;
539 complete(&thread->mgt_started);
543 GOTO(out, rc = -ENOMEM);
545 rc = lu_env_init(env, LCT_MD_THREAD);
549 rc = mdd_orphan_index_iterate(env, thread);
556 complete(&thread->mgt_finished);
561 * Iterate orphan index to cleanup orphan objects after recovery is done.
562 * \param d mdd device in recovery.
564 int mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
569 OBD_ALLOC(name, MTI_NAME_MAXLEN);
573 snprintf(name, MTI_NAME_MAXLEN, "orph_%s", mdd2obd_dev(d)->obd_name);
575 rc = mdd_generic_thread_start(&d->mdd_orphan_cleanup_thread,
576 mdd_orphan_cleanup_thread, d, name);
579 CERROR("%s: start orphan cleanup thread failed: rc = %d\n",
580 mdd2obd_dev(d)->obd_name, rc);
582 OBD_FREE(name, MTI_NAME_MAXLEN);