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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdd/mdd_orphans.c
38 * Orphan handling code
40 * Author: Mike Pershin <tappro@clusterfs.com>
41 * Pravin B Shelar <pravin.shelar@sun.com>
44 #define DEBUG_SUBSYSTEM S_MDS
47 #include <obd_class.h>
48 #include <lustre_ver.h>
49 #include <obd_support.h>
50 #include <lustre_fid.h>
51 #include "mdd_internal.h"
53 const char orph_index_name[] = "PENDING";
54 const char *dotdot = "..";
61 #define ORPHAN_FILE_NAME_FORMAT "%016llx:%08x:%08x:%2x"
62 #define ORPHAN_FILE_NAME_FORMAT_18 "%llx:%08x"
64 static struct dt_key* orph_key_fill(const struct lu_env *env,
65 const struct lu_fid *lf, __u32 op)
67 char *key = mdd_env_info(env)->mti_orph_key;
71 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT,
72 (long long unsigned int)fid_seq(lf),
73 fid_oid(lf), fid_ver(lf), op);
75 return (struct dt_key*) key;
80 static struct dt_key* orph_key_fill_18(const struct lu_env *env,
81 const struct lu_fid *lf)
83 char *key = mdd_env_info(env)->mti_orph_key;
87 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_18,
88 (unsigned long long)fid_seq(lf), fid_oid(lf));
90 return (struct dt_key*) key;
95 static inline void mdd_orphan_write_lock(const struct lu_env *env,
96 struct mdd_device *mdd)
99 struct dt_object *dor = mdd->mdd_orphans;
100 dor->do_ops->do_write_lock(env, dor, MOR_TGT_ORPHAN);
103 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
104 struct mdd_device *mdd)
107 struct dt_object *dor = mdd->mdd_orphans;
108 dor->do_ops->do_write_unlock(env, dor);
111 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
112 struct mdd_device *mdd,
113 struct mdd_object *obj,
117 struct dt_object *dor = mdd->mdd_orphans;
118 const struct lu_fid *lf = mdo2fid(obj);
119 struct dt_key *key = orph_key_fill(env, lf, op);
122 return dor->do_index_ops->dio_insert(env, dor,
128 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
129 struct mdd_device *mdd ,
133 struct dt_object *dor = mdd->mdd_orphans;
135 return dor->do_index_ops->dio_delete(env, dor,
140 static inline void mdd_orphan_ref_add(const struct lu_env *env,
141 struct mdd_device *mdd,
144 struct dt_object *dor = mdd->mdd_orphans;
145 dor->do_ops->do_ref_add(env, dor, th);
148 static inline void mdd_orphan_ref_del(const struct lu_env *env,
149 struct mdd_device *mdd,
152 struct dt_object *dor = mdd->mdd_orphans;
153 dor->do_ops->do_ref_del(env, dor, th);
157 int orph_declare_index_insert(const struct lu_env *env,
158 struct mdd_object *obj,
161 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
164 rc = dt_declare_insert(env, mdd->mdd_orphans, NULL, NULL, th);
168 rc = mdo_declare_ref_add(env, obj, th);
172 if (!S_ISDIR(mdd_object_type(obj)))
175 rc = mdo_declare_ref_add(env, obj, th);
179 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
183 rc = mdo_declare_index_delete(env, obj, dotdot, th);
187 rc = mdo_declare_index_insert(env, obj, NULL, dotdot, th);
192 static int orph_index_insert(const struct lu_env *env,
193 struct mdd_object *obj,
197 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
198 struct dt_object *dor = mdd->mdd_orphans;
199 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
200 struct dt_object *next = mdd_object_child(obj);
204 LASSERT(mdd_write_locked(env, obj) != 0);
205 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
206 LASSERT(obj->mod_count > 0);
208 mdd_orphan_write_lock(env, mdd);
210 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
214 mdo_ref_add(env, obj, th);
215 if (!S_ISDIR(mdd_object_type(obj)))
218 mdo_ref_add(env, obj, th);
219 mdd_orphan_ref_add(env, mdd, th);
221 /* try best to fixup directory, dont return errors
223 if (!dt_try_as_dir(env, next))
225 next->do_index_ops->dio_delete(env, next,
226 (const struct dt_key *)dotdot,
229 next->do_index_ops->dio_insert(env, next,
230 (struct dt_rec *)lf_dor,
231 (const struct dt_key *)dotdot,
236 obj->mod_flags |= ORPHAN_OBJ;
238 mdd_orphan_write_unlock(env, mdd);
244 * Destroy OSD object on MDD and associated OST objects.
246 * \param obj orphan object
247 * \param mdd used for sending llog msg to osts
252 static int orphan_object_kill(const struct lu_env *env,
253 struct mdd_object *obj,
254 struct mdd_device *mdd,
257 struct lu_attr *la = &mdd_env_info(env)->mti_la;
261 /* No need to lock this object as its recovery phase, and
262 * no other thread can access it. But we need to lock it
263 * as its precondition for osd api we using. */
265 mdo_ref_del(env, obj, th);
266 if (S_ISDIR(mdd_object_type(obj))) {
267 mdo_ref_del(env, obj, th);
268 mdd_orphan_ref_del(env, mdd, th);
270 /* regular file , cleanup linked ost objects */
271 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
273 rc = mdd_lov_destroy(env, mdd, obj, la);
275 mdo_destroy(env, obj, th);
279 int orph_declare_index_delete(const struct lu_env *env,
280 struct mdd_object *obj,
283 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
286 rc = dt_declare_delete(env, mdd->mdd_orphans, NULL, th);
290 rc = mdo_declare_ref_del(env, obj, th);
294 if (S_ISDIR(mdd_object_type(obj))) {
295 rc = mdo_declare_ref_del(env, obj, th);
299 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
305 static int orph_index_delete(const struct lu_env *env,
306 struct mdd_object *obj,
310 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
311 struct dt_object *dor = mdd->mdd_orphans;
317 LASSERT(mdd_write_locked(env, obj) != 0);
318 LASSERT(obj->mod_flags & ORPHAN_OBJ);
319 LASSERT(obj->mod_count == 0);
323 key = orph_key_fill(env, mdo2fid(obj), op);
324 mdd_orphan_write_lock(env, mdd);
326 rc = mdd_orphan_delete_obj(env, mdd, key, th);
329 key = orph_key_fill_18(env, mdo2fid(obj));
330 rc = mdd_orphan_delete_obj(env, mdd, key, th);
334 /* lov objects will be destroyed by caller */
335 mdo_ref_del(env, obj, th);
336 if (S_ISDIR(mdd_object_type(obj))) {
337 mdo_ref_del(env, obj, th);
338 mdd_orphan_ref_del(env, mdd, th);
340 obj->mod_flags &= ~ORPHAN_OBJ;
342 CERROR("could not delete object: rc = %d\n",rc);
345 mdd_orphan_write_unlock(env, mdd);
350 static int orphan_object_destroy(const struct lu_env *env,
351 struct mdd_object *obj,
354 struct thandle *th = NULL;
355 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
356 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
361 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
362 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
363 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
364 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
365 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
368 th = mdd_trans_create(env, mdd);
370 CERROR("Cannot get thandle\n");
373 rc = orph_declare_index_delete(env, obj, th);
377 rc = mdd_declare_object_kill(env, obj, ma, th);
381 rc = mdd_trans_start(env, mdd, th);
385 mdd_write_lock(env, obj, MOR_TGT_CHILD);
386 if (likely(obj->mod_count == 0)) {
387 mdd_orphan_write_lock(env, mdd);
388 rc = mdd_orphan_delete_obj(env, mdd, key, th);
390 orphan_object_kill(env, obj, mdd, th);
392 CERROR("could not delete object: rc = %d\n",rc);
393 mdd_orphan_write_unlock(env, mdd);
395 mdd_write_unlock(env, obj);
398 mdd_trans_stop(env, mdd, 0, th);
404 * Delete unused orphan with FID \a lf from PENDING directory
406 * \param mdd MDD device finishing recovery
407 * \param lf FID of file or directory to delete
408 * \param key cookie for this entry in index iterator
413 static int orph_key_test_and_del(const struct lu_env *env,
414 struct mdd_device *mdd,
418 struct mdd_object *mdo;
421 mdo = mdd_object_find(env, mdd, lf);
427 if (mdo->mod_count == 0) {
428 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
429 rc = orphan_object_destroy(env, mdo, key);
430 if (rc) /* so replay-single.sh test_37 works */
431 CERROR("%s: error unlinking orphan "DFID" from "
432 "PENDING: rc = %d\n",
433 mdd->mdd_obd_dev->obd_name, PFID(lf), rc);
435 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
436 if (likely(mdo->mod_count > 0)) {
437 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
438 PFID(lf), mdo->mod_count);
439 mdo->mod_flags |= ORPHAN_OBJ;
441 mdd_write_unlock(env, mdo);
444 mdd_object_put(env, mdo);
449 * delete unreferenced files and directories in the PENDING directory
451 * Files that remain in PENDING after client->MDS recovery has completed
452 * have to be referenced (opened) by some client during recovery, or they
453 * will be deleted here (for clients that did not complete recovery).
455 * \param mdd MDD device finishing recovery
460 static int orph_index_iterate(const struct lu_env *env,
461 struct mdd_device *mdd)
463 struct dt_object *dor = mdd->mdd_orphans;
464 struct lu_dirent *ent = &mdd_env_info(env)->mti_orph_ent;
465 const struct dt_it_ops *iops;
473 /* In recovery phase, do not need for any lock here */
474 iops = &dor->do_index_ops->dio_it;
475 it = iops->init(env, dor, LUDA_64BITHASH, BYPASS_CAPA);
478 CERROR("%s: cannot clean PENDING: rc = %d\n",
479 mdd->mdd_obd_dev->obd_name, rc);
483 rc = iops->load(env, it, 0);
487 CERROR("%s: error loading iterator to clean PENDING\n",
488 mdd->mdd_obd_dev->obd_name);
489 /* Index contains no zero key? */
490 GOTO(out_put, rc = -EIO);
494 key_sz = iops->key_size(env, it);
495 /* filter out "." and ".." entries from PENDING dir. */
499 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
501 CERROR("%s: fail to get FID for orphan it: rc = %d\n",
502 mdd->mdd_obd_dev->obd_name, rc);
506 fid_le_to_cpu(&fid, &ent->lde_fid);
507 if (!fid_is_sane(&fid)) {
508 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
509 mdd->mdd_obd_dev->obd_name, PFID(&fid));
513 /* kill orphan object */
514 cookie = iops->store(env, it);
516 rc = orph_key_test_and_del(env, mdd, &fid,
517 (struct dt_key *)ent->lde_name);
519 /* after index delete reset iterator */
521 rc = iops->get(env, it, (const void *)"");
523 rc = iops->load(env, it, cookie);
525 rc = iops->next(env, it);
528 GOTO(out_put, rc = 0);
538 * open the PENDING directory for device \a mdd
540 * The PENDING directory persistently tracks files and directories that were
541 * unlinked from the namespace (nlink == 0) but are still held open by clients.
542 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
543 * would not be able to recover and reopen those files. Instead, these inodes
544 * are linked into the PENDING directory on disk, and only deleted if all
545 * clients close them, or the MDS finishes client recovery without any client
546 * reopening them (i.e. former clients didn't join recovery).
547 * \param d mdd device being started.
550 * \retval -ve index operation error.
553 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
560 d = dt_store_open(env, mdd->mdd_child, "", orph_index_name, &fid);
562 mdd->mdd_orphans = d;
563 if (!dt_try_as_dir(env, d)) {
565 CERROR("\"%s\" is not an index! : rc = %d\n",
566 orph_index_name, rc);
569 CERROR("cannot find \"%s\" obj %d\n",
570 orph_index_name, (int)PTR_ERR(d));
577 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
580 if (mdd->mdd_orphans != NULL) {
581 lu_object_put(env, &mdd->mdd_orphans->do_lu);
582 mdd->mdd_orphans = NULL;
588 * Iterate orphan index to cleanup orphan objects after recovery is done.
589 * \param d mdd device in recovery.
591 int __mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
593 return orph_index_iterate(env, d);
597 * add an orphan \a obj to the orphan index.
598 * \param obj file or directory.
599 * \param th transaction for index insert.
601 * \pre obj nlink == 0 && obj->mod_count != 0
604 * \retval -ve index operation error.
606 int __mdd_orphan_add(const struct lu_env *env,
607 struct mdd_object *obj, struct thandle *th)
609 return orph_index_insert(env, obj, ORPH_OP_UNLINK, th);
613 * delete an orphan \a obj from orphan index.
614 * \param obj file or directory.
615 * \param th transaction for index deletion and object destruction.
617 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
620 * \retval -ve index operation error.
622 int __mdd_orphan_del(const struct lu_env *env,
623 struct mdd_object *obj, struct thandle *th)
625 return orph_index_delete(env, obj, ORPH_OP_UNLINK, th);