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 int orphan_key_to_fid(char *key, struct lu_fid *lf)
100 rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT,
101 (long long unsigned int *)&lf->f_seq, &lf->f_oid,
107 rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT_18,
108 (long long unsigned int *)&lf->f_seq, &lf->f_oid);
114 CERROR("can not parse orphan file name %s\n", key);
118 static inline void mdd_orphan_write_lock(const struct lu_env *env,
119 struct mdd_device *mdd)
122 struct dt_object *dor = mdd->mdd_orphans;
123 dor->do_ops->do_write_lock(env, dor, MOR_TGT_ORPHAN);
126 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
127 struct mdd_device *mdd)
130 struct dt_object *dor = mdd->mdd_orphans;
131 dor->do_ops->do_write_unlock(env, dor);
134 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
135 struct mdd_device *mdd,
136 struct mdd_object *obj,
140 struct dt_object *dor = mdd->mdd_orphans;
141 const struct lu_fid *lf = mdo2fid(obj);
142 struct dt_key *key = orph_key_fill(env, lf, op);
145 return dor->do_index_ops->dio_insert(env, dor,
151 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
152 struct mdd_device *mdd ,
156 struct dt_object *dor = mdd->mdd_orphans;
158 return dor->do_index_ops->dio_delete(env, dor,
163 static inline void mdd_orphan_ref_add(const struct lu_env *env,
164 struct mdd_device *mdd,
167 struct dt_object *dor = mdd->mdd_orphans;
168 dor->do_ops->do_ref_add(env, dor, th);
171 static inline void mdd_orphan_ref_del(const struct lu_env *env,
172 struct mdd_device *mdd,
175 struct dt_object *dor = mdd->mdd_orphans;
176 dor->do_ops->do_ref_del(env, dor, th);
180 int orph_declare_index_insert(const struct lu_env *env,
181 struct mdd_object *obj,
184 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
187 rc = dt_declare_insert(env, mdd->mdd_orphans, NULL, NULL, th);
191 rc = mdo_declare_ref_add(env, obj, th);
195 if (!S_ISDIR(mdd_object_type(obj)))
198 rc = mdo_declare_ref_add(env, obj, th);
202 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
206 rc = mdo_declare_index_delete(env, obj, dotdot, th);
210 rc = mdo_declare_index_insert(env, obj, NULL, dotdot, th);
215 static int orph_index_insert(const struct lu_env *env,
216 struct mdd_object *obj,
220 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
221 struct dt_object *dor = mdd->mdd_orphans;
222 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
223 struct dt_object *next = mdd_object_child(obj);
227 LASSERT(mdd_write_locked(env, obj) != 0);
228 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
229 LASSERT(obj->mod_count > 0);
231 mdd_orphan_write_lock(env, mdd);
233 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
237 mdo_ref_add(env, obj, th);
238 if (!S_ISDIR(mdd_object_type(obj)))
241 mdo_ref_add(env, obj, th);
242 mdd_orphan_ref_add(env, mdd, th);
244 /* try best to fixup directory, dont return errors
246 if (!dt_try_as_dir(env, next))
248 next->do_index_ops->dio_delete(env, next,
249 (const struct dt_key *)dotdot,
252 next->do_index_ops->dio_insert(env, next,
253 (struct dt_rec *)lf_dor,
254 (const struct dt_key *)dotdot,
259 obj->mod_flags |= ORPHAN_OBJ;
261 mdd_orphan_write_unlock(env, mdd);
267 * Destroy OSD object on MDD and associated OST objects.
269 * \param obj orphan object
270 * \param mdd used for sending llog msg to osts
275 static int orphan_object_kill(const struct lu_env *env,
276 struct mdd_object *obj,
277 struct mdd_device *mdd,
280 struct lu_attr *la = &mdd_env_info(env)->mti_la;
284 /* No need to lock this object as its recovery phase, and
285 * no other thread can access it. But we need to lock it
286 * as its precondition for osd api we using. */
288 mdo_ref_del(env, obj, th);
289 if (S_ISDIR(mdd_object_type(obj))) {
290 mdo_ref_del(env, obj, th);
291 mdd_orphan_ref_del(env, mdd, th);
293 /* regular file , cleanup linked ost objects */
294 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
296 rc = mdd_lov_destroy(env, mdd, obj, la);
298 mdo_destroy(env, obj, th);
302 int orph_declare_index_delete(const struct lu_env *env,
303 struct mdd_object *obj,
306 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
309 rc = dt_declare_delete(env, mdd->mdd_orphans, NULL, th);
313 rc = mdo_declare_ref_del(env, obj, th);
317 if (S_ISDIR(mdd_object_type(obj))) {
318 rc = mdo_declare_ref_del(env, obj, th);
322 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
328 static int orph_index_delete(const struct lu_env *env,
329 struct mdd_object *obj,
333 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
334 struct dt_object *dor = mdd->mdd_orphans;
340 LASSERT(mdd_write_locked(env, obj) != 0);
341 LASSERT(obj->mod_flags & ORPHAN_OBJ);
342 LASSERT(obj->mod_count == 0);
346 key = orph_key_fill(env, mdo2fid(obj), op);
347 mdd_orphan_write_lock(env, mdd);
349 rc = mdd_orphan_delete_obj(env, mdd, key, th);
352 key = orph_key_fill_18(env, mdo2fid(obj));
353 rc = mdd_orphan_delete_obj(env, mdd, key, th);
357 /* lov objects will be destroyed by caller */
358 mdo_ref_del(env, obj, th);
359 if (S_ISDIR(mdd_object_type(obj))) {
360 mdo_ref_del(env, obj, th);
361 mdd_orphan_ref_del(env, mdd, th);
363 obj->mod_flags &= ~ORPHAN_OBJ;
365 CERROR("could not delete object: rc = %d\n",rc);
368 mdd_orphan_write_unlock(env, mdd);
373 static int orphan_object_destroy(const struct lu_env *env,
374 struct mdd_object *obj,
377 struct thandle *th = NULL;
378 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
379 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
384 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
385 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
386 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
387 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
388 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
391 th = mdd_trans_create(env, mdd);
393 CERROR("Cannot get thandle\n");
396 rc = orph_declare_index_delete(env, obj, th);
400 rc = mdd_declare_object_kill(env, obj, ma, th);
404 rc = mdd_trans_start(env, mdd, th);
408 mdd_write_lock(env, obj, MOR_TGT_CHILD);
409 if (likely(obj->mod_count == 0)) {
410 mdd_orphan_write_lock(env, mdd);
411 rc = mdd_orphan_delete_obj(env, mdd, key, th);
413 orphan_object_kill(env, obj, mdd, th);
415 CERROR("could not delete object: rc = %d\n",rc);
416 mdd_orphan_write_unlock(env, mdd);
418 mdd_write_unlock(env, obj);
421 mdd_trans_stop(env, mdd, 0, th);
427 * Delete unused orphan with FID \a lf from PENDING directory
429 * \param mdd MDD device finishing recovery
430 * \param lf FID of file or directory to delete
431 * \param key cookie for this entry in index iterator
436 static int orph_key_test_and_del(const struct lu_env *env,
437 struct mdd_device *mdd,
441 struct mdd_object *mdo;
444 mdo = mdd_object_find(env, mdd, lf);
450 if (mdo->mod_count == 0) {
451 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
452 rc = orphan_object_destroy(env, mdo, key);
453 if (rc) /* so replay-single.sh test_37 works */
454 CERROR("%s: error unlinking orphan "DFID" from "
455 "PENDING: rc = %d\n",
456 mdd->mdd_obd_dev->obd_name, PFID(lf), rc);
458 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
459 if (likely(mdo->mod_count > 0)) {
460 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
461 PFID(lf), mdo->mod_count);
462 mdo->mod_flags |= ORPHAN_OBJ;
464 mdd_write_unlock(env, mdo);
467 mdd_object_put(env, mdo);
472 * delete unreferenced files and directories in the PENDING directory
474 * Files that remain in PENDING after client->MDS recovery has completed
475 * have to be referenced (opened) by some client during recovery, or they
476 * will be deleted here (for clients that did not complete recovery).
478 * \param mdd MDD device finishing recovery
483 static int orph_index_iterate(const struct lu_env *env,
484 struct mdd_device *mdd)
486 struct dt_object *dor = mdd->mdd_orphans;
487 char *mti_key = mdd_env_info(env)->mti_orph_key;
488 const struct dt_it_ops *iops;
498 /* In recovery phase, do not need for any lock here */
499 iops = &dor->do_index_ops->dio_it;
500 it = iops->init(env, dor, LUDA_64BITHASH, BYPASS_CAPA);
503 CERROR("%s: cannot clean PENDING: rc = %d\n",
504 mdd->mdd_obd_dev->obd_name, rc);
508 rc = iops->load(env, it, 0);
512 CERROR("%s: error loading iterator to clean PENDING\n",
513 mdd->mdd_obd_dev->obd_name);
514 /* Index contains no zero key? */
515 GOTO(out_put, rc = -EIO);
519 key = (void *)iops->key(env, it);
521 CERROR("%s: key failed when clean PENDING: rc = %ld\n",
522 mdd->mdd_obd_dev->obd_name, PTR_ERR(key));
525 key_sz = iops->key_size(env, it);
527 /* filter out "." and ".." entries from PENDING dir. */
531 memcpy(mti_key, key, key_sz);
534 if (orphan_key_to_fid(mti_key, &fid))
536 if (!fid_is_sane(&fid)) {
537 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
538 mdd->mdd_obd_dev->obd_name, PFID(&fid));
542 /* kill orphan object */
543 cookie = iops->store(env, it);
545 rc = orph_key_test_and_del(env, mdd, &fid,
546 (struct dt_key *)mti_key);
548 /* after index delete reset iterator */
550 result = iops->get(env, it, (const void *)"");
552 result = iops->load(env, it, cookie);
554 result = iops->next(env, it);
555 } while (result == 0);
557 GOTO(out_put, rc = 0);
567 * open the PENDING directory for device \a mdd
569 * The PENDING directory persistently tracks files and directories that were
570 * unlinked from the namespace (nlink == 0) but are still held open by clients.
571 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
572 * would not be able to recover and reopen those files. Instead, these inodes
573 * are linked into the PENDING directory on disk, and only deleted if all
574 * clients close them, or the MDS finishes client recovery without any client
575 * reopening them (i.e. former clients didn't join recovery).
576 * \param d mdd device being started.
579 * \retval -ve index operation error.
582 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
589 d = dt_store_open(env, mdd->mdd_child, "", orph_index_name, &fid);
591 mdd->mdd_orphans = d;
592 if (!dt_try_as_dir(env, d)) {
594 CERROR("\"%s\" is not an index! : rc = %d\n",
595 orph_index_name, rc);
598 CERROR("cannot find \"%s\" obj %d\n",
599 orph_index_name, (int)PTR_ERR(d));
606 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
609 if (mdd->mdd_orphans != NULL) {
610 lu_object_put(env, &mdd->mdd_orphans->do_lu);
611 mdd->mdd_orphans = NULL;
617 * Iterate orphan index to cleanup orphan objects after recovery is done.
618 * \param d mdd device in recovery.
620 int __mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
622 return orph_index_iterate(env, d);
626 * add an orphan \a obj to the orphan index.
627 * \param obj file or directory.
628 * \param th transaction for index insert.
630 * \pre obj nlink == 0 && obj->mod_count != 0
633 * \retval -ve index operation error.
635 int __mdd_orphan_add(const struct lu_env *env,
636 struct mdd_object *obj, struct thandle *th)
638 return orph_index_insert(env, obj, ORPH_OP_UNLINK, th);
642 * delete an orphan \a obj from orphan index.
643 * \param obj file or directory.
644 * \param th transaction for index deletion and object destruction.
646 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
649 * \retval -ve index operation error.
651 int __mdd_orphan_del(const struct lu_env *env,
652 struct mdd_object *obj, struct thandle *th)
654 return orph_index_delete(env, obj, ORPH_OP_UNLINK, th);