1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/mdd/mdd_orphans.c
40 * Orphan handling code
42 * Author: Mike Pershin <tappro@clusterfs.com>
43 * Pravin B Shelar <pravin.shelar@sun.com>
47 # define EXPORT_SYMTAB
49 #define DEBUG_SUBSYSTEM S_MDS
52 #include <obd_class.h>
53 #include <lustre_ver.h>
54 #include <obd_support.h>
55 #include <lustre_fid.h>
56 #include "mdd_internal.h"
58 const char orph_index_name[] = "PENDING";
59 const char *dotdot = "..";
66 #define ORPHAN_FILE_NAME_FORMAT "%016llx:%08x:%08x:%2x"
67 #define ORPHAN_FILE_NAME_FORMAT_18 "%llx:%08x"
69 static struct dt_key* orph_key_fill(const struct lu_env *env,
70 const struct lu_fid *lf, __u32 op)
72 char *key = mdd_env_info(env)->mti_orph_key;
76 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT,
77 (long long unsigned int)fid_seq(lf),
78 fid_oid(lf), fid_ver(lf), op);
80 return (struct dt_key*) key;
85 static struct dt_key* orph_key_fill_18(const struct lu_env *env,
86 const struct lu_fid *lf)
88 char *key = mdd_env_info(env)->mti_orph_key;
92 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_18,
93 (unsigned long long)fid_seq(lf), fid_oid(lf));
95 return (struct dt_key*) key;
100 static int orphan_key_to_fid(char *key, struct lu_fid *lf)
105 rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT,
106 (long long unsigned int *)&lf->f_seq, &lf->f_oid,
112 rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT_18,
113 (long long unsigned int *)&lf->f_seq, &lf->f_oid);
119 CERROR("can not parse orphan file name %s\n", key);
123 static inline void mdd_orphan_write_lock(const struct lu_env *env,
124 struct mdd_device *mdd)
127 struct dt_object *dor = mdd->mdd_orphans;
128 dor->do_ops->do_write_lock(env, dor, MOR_TGT_ORPHAN);
131 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
132 struct mdd_device *mdd)
135 struct dt_object *dor = mdd->mdd_orphans;
136 dor->do_ops->do_write_unlock(env, dor);
139 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
140 struct mdd_device *mdd,
141 struct mdd_object *obj,
145 struct dt_object *dor = mdd->mdd_orphans;
146 const struct lu_fid *lf = mdo2fid(obj);
147 struct dt_key *key = orph_key_fill(env, lf, op);
150 return dor->do_index_ops->dio_insert(env, dor,
156 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
157 struct mdd_device *mdd ,
161 struct dt_object *dor = mdd->mdd_orphans;
163 return dor->do_index_ops->dio_delete(env, dor,
168 static inline void mdd_orphan_ref_add(const struct lu_env *env,
169 struct mdd_device *mdd,
172 struct dt_object *dor = mdd->mdd_orphans;
173 dor->do_ops->do_ref_add(env, dor, th);
176 static inline void mdd_orphan_ref_del(const struct lu_env *env,
177 struct mdd_device *mdd,
180 struct dt_object *dor = mdd->mdd_orphans;
181 dor->do_ops->do_ref_del(env, dor, th);
185 int orph_declare_index_insert(const struct lu_env *env,
186 struct mdd_object *obj,
189 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
192 rc = dt_declare_insert(env, mdd->mdd_orphans, NULL, NULL, th);
196 rc = mdo_declare_ref_add(env, obj, th);
200 if (!S_ISDIR(mdd_object_type(obj)))
203 rc = mdo_declare_ref_add(env, obj, th);
207 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
211 rc = mdo_declare_index_delete(env, obj, dotdot, th);
215 rc = mdo_declare_index_insert(env, obj, NULL, dotdot, th);
220 static int orph_index_insert(const struct lu_env *env,
221 struct mdd_object *obj,
225 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
226 struct dt_object *dor = mdd->mdd_orphans;
227 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
228 struct dt_object *next = mdd_object_child(obj);
232 LASSERT(mdd_write_locked(env, obj) != 0);
233 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
234 LASSERT(obj->mod_count > 0);
236 mdd_orphan_write_lock(env, mdd);
238 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
242 mdo_ref_add(env, obj, th);
243 if (!S_ISDIR(mdd_object_type(obj)))
246 mdo_ref_add(env, obj, th);
247 mdd_orphan_ref_add(env, mdd, th);
249 /* try best to fixup directory, dont return errors
251 if (!dt_try_as_dir(env, next))
253 next->do_index_ops->dio_delete(env, next,
254 (const struct dt_key *)dotdot,
257 next->do_index_ops->dio_insert(env, next,
258 (struct dt_rec *)lf_dor,
259 (const struct dt_key *)dotdot,
264 obj->mod_flags |= ORPHAN_OBJ;
266 mdd_orphan_write_unlock(env, mdd);
272 * Destroy OSD object on MDD and associated OST objects.
274 * \param obj orphan object
275 * \param mdd used for sending llog msg to osts
280 static int orphan_object_kill(const struct lu_env *env,
281 struct mdd_object *obj,
282 struct mdd_device *mdd,
285 struct lu_attr *la = &mdd_env_info(env)->mti_la;
289 /* No need to lock this object as its recovery phase, and
290 * no other thread can access it. But we need to lock it
291 * as its precondition for osd api we using. */
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 /* regular file , cleanup linked ost objects */
299 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
301 rc = mdd_lov_destroy(env, mdd, obj, la);
303 mdo_destroy(env, obj, th);
307 int orph_declare_index_delete(const struct lu_env *env,
308 struct mdd_object *obj,
311 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
314 rc = dt_declare_delete(env, mdd->mdd_orphans, NULL, th);
318 rc = mdo_declare_ref_del(env, obj, th);
322 if (S_ISDIR(mdd_object_type(obj))) {
323 rc = mdo_declare_ref_del(env, obj, th);
327 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
333 static int orph_index_delete(const struct lu_env *env,
334 struct mdd_object *obj,
338 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
339 struct dt_object *dor = mdd->mdd_orphans;
345 LASSERT(mdd_write_locked(env, obj) != 0);
346 LASSERT(obj->mod_flags & ORPHAN_OBJ);
347 LASSERT(obj->mod_count == 0);
351 key = orph_key_fill(env, mdo2fid(obj), op);
352 mdd_orphan_write_lock(env, mdd);
354 rc = mdd_orphan_delete_obj(env, mdd, key, th);
357 key = orph_key_fill_18(env, mdo2fid(obj));
358 rc = mdd_orphan_delete_obj(env, mdd, key, th);
362 /* lov objects will be destroyed by caller */
363 mdo_ref_del(env, obj, th);
364 if (S_ISDIR(mdd_object_type(obj))) {
365 mdo_ref_del(env, obj, th);
366 mdd_orphan_ref_del(env, mdd, th);
368 obj->mod_flags &= ~ORPHAN_OBJ;
370 CERROR("could not delete object: rc = %d\n",rc);
373 mdd_orphan_write_unlock(env, mdd);
378 static int orphan_object_destroy(const struct lu_env *env,
379 struct mdd_object *obj,
382 struct thandle *th = NULL;
383 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
384 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
389 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
390 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
391 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
392 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
393 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
396 th = mdd_trans_create(env, mdd);
398 CERROR("Cannot get thandle\n");
401 rc = orph_declare_index_delete(env, obj, th);
405 rc = mdd_declare_object_kill(env, obj, ma, th);
409 rc = mdd_trans_start(env, mdd, th);
413 mdd_write_lock(env, obj, MOR_TGT_CHILD);
414 if (likely(obj->mod_count == 0)) {
415 mdd_orphan_write_lock(env, mdd);
416 rc = mdd_orphan_delete_obj(env, mdd, key, th);
418 orphan_object_kill(env, obj, mdd, th);
420 CERROR("could not delete object: rc = %d\n",rc);
421 mdd_orphan_write_unlock(env, mdd);
423 mdd_write_unlock(env, obj);
426 mdd_trans_stop(env, mdd, 0, th);
432 * Delete unused orphan with FID \a lf from PENDING directory
434 * \param mdd MDD device finishing recovery
435 * \param lf FID of file or directory to delete
436 * \param key cookie for this entry in index iterator
441 static int orph_key_test_and_del(const struct lu_env *env,
442 struct mdd_device *mdd,
446 struct mdd_object *mdo;
449 mdo = mdd_object_find(env, mdd, lf);
455 if (mdo->mod_count == 0) {
456 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
457 rc = orphan_object_destroy(env, mdo, key);
458 if (rc) /* so replay-single.sh test_37 works */
459 CERROR("%s: error unlinking orphan "DFID" from "
460 "PENDING: rc = %d\n",
461 mdd->mdd_obd_dev->obd_name, PFID(lf), rc);
463 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
464 if (likely(mdo->mod_count > 0)) {
465 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
466 PFID(lf), mdo->mod_count);
467 mdo->mod_flags |= ORPHAN_OBJ;
469 mdd_write_unlock(env, mdo);
472 mdd_object_put(env, mdo);
477 * delete unreferenced files and directories in the PENDING directory
479 * Files that remain in PENDING after client->MDS recovery has completed
480 * have to be referenced (opened) by some client during recovery, or they
481 * will be deleted here (for clients that did not complete recovery).
483 * \param mdd MDD device finishing recovery
488 static int orph_index_iterate(const struct lu_env *env,
489 struct mdd_device *mdd)
491 struct dt_object *dor = mdd->mdd_orphans;
492 char *mti_key = mdd_env_info(env)->mti_orph_key;
493 const struct dt_it_ops *iops;
503 /* In recovery phase, do not need for any lock here */
504 iops = &dor->do_index_ops->dio_it;
505 it = iops->init(env, dor, LUDA_64BITHASH, BYPASS_CAPA);
508 CERROR("%s: cannot clean PENDING: rc = %d\n",
509 mdd->mdd_obd_dev->obd_name, rc);
513 rc = iops->load(env, it, 0);
517 CERROR("%s: error loading iterator to clean PENDING\n",
518 mdd->mdd_obd_dev->obd_name);
519 /* Index contains no zero key? */
520 GOTO(out_put, rc = -EIO);
524 key = (void *)iops->key(env, it);
526 CERROR("%s: key failed when clean PENDING: rc = %ld\n",
527 mdd->mdd_obd_dev->obd_name, PTR_ERR(key));
530 key_sz = iops->key_size(env, it);
532 /* filter out "." and ".." entries from PENDING dir. */
536 memcpy(mti_key, key, key_sz);
539 if (orphan_key_to_fid(mti_key, &fid))
541 if (!fid_is_sane(&fid)) {
542 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
543 mdd->mdd_obd_dev->obd_name, PFID(&fid));
547 /* kill orphan object */
548 cookie = iops->store(env, it);
550 rc = orph_key_test_and_del(env, mdd, &fid,
551 (struct dt_key *)mti_key);
553 /* after index delete reset iterator */
555 result = iops->get(env, it, (const void *)"");
557 result = iops->load(env, it, cookie);
559 result = iops->next(env, it);
560 } while (result == 0);
562 GOTO(out_put, rc = 0);
572 * open the PENDING directory for device \a mdd
574 * The PENDING directory persistently tracks files and directories that were
575 * unlinked from the namespace (nlink == 0) but are still held open by clients.
576 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
577 * would not be able to recover and reopen those files. Instead, these inodes
578 * are linked into the PENDING directory on disk, and only deleted if all
579 * clients close them, or the MDS finishes client recovery without any client
580 * reopening them (i.e. former clients didn't join recovery).
581 * \param d mdd device being started.
584 * \retval -ve index operation error.
587 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
594 d = dt_store_open(env, mdd->mdd_child, "", orph_index_name, &fid);
596 mdd->mdd_orphans = d;
597 if (!dt_try_as_dir(env, d)) {
599 CERROR("\"%s\" is not an index! : rc = %d\n",
600 orph_index_name, rc);
603 CERROR("cannot find \"%s\" obj %d\n",
604 orph_index_name, (int)PTR_ERR(d));
611 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
614 if (mdd->mdd_orphans != NULL) {
615 lu_object_put(env, &mdd->mdd_orphans->do_lu);
616 mdd->mdd_orphans = NULL;
622 * Iterate orphan index to cleanup orphan objects after recovery is done.
623 * \param d mdd device in recovery.
625 int __mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
627 return orph_index_iterate(env, d);
631 * add an orphan \a obj to the orphan index.
632 * \param obj file or directory.
633 * \param th transaction for index insert.
635 * \pre obj nlink == 0 && obj->mod_count != 0
638 * \retval -ve index operation error.
640 int __mdd_orphan_add(const struct lu_env *env,
641 struct mdd_object *obj, struct thandle *th)
643 return orph_index_insert(env, obj, ORPH_OP_UNLINK, th);
647 * delete an orphan \a obj from orphan index.
648 * \param obj file or directory.
649 * \param th transaction for index deletion and object destruction.
651 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
654 * \retval -ve index operation error.
656 int __mdd_orphan_del(const struct lu_env *env,
657 struct mdd_object *obj, struct thandle *th)
659 return orph_index_delete(env, obj, ORPH_OP_UNLINK, th);