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 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/mdd/mdd_orphans.c
41 * Orphan handling code
43 * Author: Mike Pershin <tappro@clusterfs.com>
44 * Pravin B Shelar <pravin.shelar@sun.com>
48 # define EXPORT_SYMTAB
50 #define DEBUG_SUBSYSTEM S_MDS
53 #include <obd_class.h>
54 #include <lustre_ver.h>
55 #include <obd_support.h>
56 #include <lustre_fid.h>
57 #include "mdd_internal.h"
59 const char orph_index_name[] = "PENDING";
60 const char *dotdot = "..";
67 #define ORPHAN_FILE_NAME_FORMAT "%016llx:%08x:%08x:%2x"
68 #define ORPHAN_FILE_NAME_FORMAT_18 "%llx:%08x"
70 static struct dt_key* orph_key_fill(const struct lu_env *env,
71 const struct lu_fid *lf, __u32 op)
73 char *key = mdd_env_info(env)->mti_orph_key;
77 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT,
78 (long long unsigned int)fid_seq(lf),
79 fid_oid(lf), fid_ver(lf), op);
81 return (struct dt_key*) key;
86 static struct dt_key* orph_key_fill_18(const struct lu_env *env,
87 const struct lu_fid *lf)
89 char *key = mdd_env_info(env)->mti_orph_key;
93 rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_18,
94 (unsigned long long)fid_seq(lf), fid_oid(lf));
96 return (struct dt_key*) key;
101 static int orphan_key_to_fid(char *key, struct lu_fid *lf)
106 rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT,
107 (long long unsigned int *)&lf->f_seq, &lf->f_oid,
113 rc = sscanf(key, ORPHAN_FILE_NAME_FORMAT_18,
114 (long long unsigned int *)&lf->f_seq, &lf->f_oid);
120 CERROR("can not parse orphan file name %s\n", key);
124 static inline void mdd_orphan_write_lock(const struct lu_env *env,
125 struct mdd_device *mdd)
128 struct dt_object *dor = mdd->mdd_orphans;
129 dor->do_ops->do_write_lock(env, dor, MOR_TGT_ORPHAN);
132 static inline void mdd_orphan_write_unlock(const struct lu_env *env,
133 struct mdd_device *mdd)
136 struct dt_object *dor = mdd->mdd_orphans;
137 dor->do_ops->do_write_unlock(env, dor);
140 static inline int mdd_orphan_insert_obj(const struct lu_env *env,
141 struct mdd_device *mdd,
142 struct mdd_object *obj,
146 struct dt_object *dor = mdd->mdd_orphans;
147 const struct lu_fid *lf = mdo2fid(obj);
148 struct dt_key *key = orph_key_fill(env, lf, op);
151 return dor->do_index_ops->dio_insert(env, dor,
157 static inline int mdd_orphan_delete_obj(const struct lu_env *env,
158 struct mdd_device *mdd ,
162 struct dt_object *dor = mdd->mdd_orphans;
164 return dor->do_index_ops->dio_delete(env, dor,
169 static inline void mdd_orphan_ref_add(const struct lu_env *env,
170 struct mdd_device *mdd,
173 struct dt_object *dor = mdd->mdd_orphans;
174 dor->do_ops->do_ref_add(env, dor, th);
177 static inline void mdd_orphan_ref_del(const struct lu_env *env,
178 struct mdd_device *mdd,
181 struct dt_object *dor = mdd->mdd_orphans;
182 dor->do_ops->do_ref_del(env, dor, th);
186 int orph_declare_index_insert(const struct lu_env *env,
187 struct mdd_object *obj,
190 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
193 rc = dt_declare_insert(env, mdd->mdd_orphans, NULL, NULL, th);
197 rc = mdo_declare_ref_add(env, obj, th);
201 if (!S_ISDIR(mdd_object_type(obj)))
204 rc = mdo_declare_ref_add(env, obj, th);
208 rc = dt_declare_ref_add(env, mdd->mdd_orphans, th);
212 rc = mdo_declare_index_delete(env, obj, dotdot, th);
216 rc = mdo_declare_index_insert(env, obj, NULL, dotdot, th);
221 static int orph_index_insert(const struct lu_env *env,
222 struct mdd_object *obj,
226 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
227 struct dt_object *dor = mdd->mdd_orphans;
228 const struct lu_fid *lf_dor = lu_object_fid(&dor->do_lu);
229 struct dt_object *next = mdd_object_child(obj);
233 LASSERT(mdd_write_locked(env, obj) != 0);
234 LASSERT(!(obj->mod_flags & ORPHAN_OBJ));
235 LASSERT(obj->mod_count > 0);
237 mdd_orphan_write_lock(env, mdd);
239 rc = mdd_orphan_insert_obj(env, mdd, obj, op, th);
243 mdo_ref_add(env, obj, th);
244 if (!S_ISDIR(mdd_object_type(obj)))
247 mdo_ref_add(env, obj, th);
248 mdd_orphan_ref_add(env, mdd, th);
250 /* try best to fixup directory, dont return errors
252 if (!dt_try_as_dir(env, next))
254 next->do_index_ops->dio_delete(env, next,
255 (const struct dt_key *)dotdot,
258 next->do_index_ops->dio_insert(env, next,
259 (struct dt_rec *)lf_dor,
260 (const struct dt_key *)dotdot,
265 obj->mod_flags |= ORPHAN_OBJ;
267 mdd_orphan_write_unlock(env, mdd);
273 * Destroy OSD object on MDD and associated OST objects.
275 * \param obj orphan object
276 * \param mdd used for sending llog msg to osts
281 static int orphan_object_kill(const struct lu_env *env,
282 struct mdd_object *obj,
283 struct mdd_device *mdd,
286 struct lu_attr *la = &mdd_env_info(env)->mti_la;
290 /* No need to lock this object as its recovery phase, and
291 * no other thread can access it. But we need to lock it
292 * as its precondition for osd api we using. */
294 mdo_ref_del(env, obj, th);
295 if (S_ISDIR(mdd_object_type(obj))) {
296 mdo_ref_del(env, obj, th);
297 mdd_orphan_ref_del(env, mdd, th);
299 /* regular file , cleanup linked ost objects */
300 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
302 rc = mdd_lov_destroy(env, mdd, obj, la);
304 mdo_destroy(env, obj, th);
308 int orph_declare_index_delete(const struct lu_env *env,
309 struct mdd_object *obj,
312 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
315 rc = dt_declare_delete(env, mdd->mdd_orphans, NULL, th);
319 rc = mdo_declare_ref_del(env, obj, th);
323 if (S_ISDIR(mdd_object_type(obj))) {
324 rc = mdo_declare_ref_del(env, obj, th);
328 rc = dt_declare_ref_del(env, mdd->mdd_orphans, th);
334 static int orph_index_delete(const struct lu_env *env,
335 struct mdd_object *obj,
339 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
340 struct dt_object *dor = mdd->mdd_orphans;
346 LASSERT(mdd_write_locked(env, obj) != 0);
347 LASSERT(obj->mod_flags & ORPHAN_OBJ);
348 LASSERT(obj->mod_count == 0);
352 key = orph_key_fill(env, mdo2fid(obj), op);
353 mdd_orphan_write_lock(env, mdd);
355 rc = mdd_orphan_delete_obj(env, mdd, key, th);
358 key = orph_key_fill_18(env, mdo2fid(obj));
359 rc = mdd_orphan_delete_obj(env, mdd, key, th);
363 /* lov objects will be destroyed by caller */
364 mdo_ref_del(env, obj, th);
365 if (S_ISDIR(mdd_object_type(obj))) {
366 mdo_ref_del(env, obj, th);
367 mdd_orphan_ref_del(env, mdd, th);
369 obj->mod_flags &= ~ORPHAN_OBJ;
371 CERROR("could not delete object: rc = %d\n",rc);
374 mdd_orphan_write_unlock(env, mdd);
379 static int orphan_object_destroy(const struct lu_env *env,
380 struct mdd_object *obj,
383 struct thandle *th = NULL;
384 struct mdd_device *mdd = mdo2mdd(&obj->mod_obj);
385 struct md_attr *ma = &mdd_env_info(env)->mti_ma;
390 ma->ma_lmm_size = mdd_lov_mdsize(env, mdd);
391 ma->ma_lmm = mdd_max_lmm_get(env, mdd);
392 ma->ma_cookie_size = mdd_lov_cookiesize(env, mdd);
393 ma->ma_cookie = mdd_max_cookie_get(env, mdd);
394 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
397 th = mdd_trans_create(env, mdd);
399 CERROR("Cannot get thandle\n");
402 rc = orph_declare_index_delete(env, obj, th);
406 rc = mdd_declare_object_kill(env, obj, ma, th);
410 rc = mdd_trans_start(env, mdd, th);
414 mdd_write_lock(env, obj, MOR_TGT_CHILD);
415 if (likely(obj->mod_count == 0)) {
416 mdd_orphan_write_lock(env, mdd);
417 rc = mdd_orphan_delete_obj(env, mdd, key, th);
419 orphan_object_kill(env, obj, mdd, th);
421 CERROR("could not delete object: rc = %d\n",rc);
422 mdd_orphan_write_unlock(env, mdd);
424 mdd_write_unlock(env, obj);
427 mdd_trans_stop(env, mdd, 0, th);
433 * Delete unused orphan with FID \a lf from PENDING directory
435 * \param mdd MDD device finishing recovery
436 * \param lf FID of file or directory to delete
437 * \param key cookie for this entry in index iterator
442 static int orph_key_test_and_del(const struct lu_env *env,
443 struct mdd_device *mdd,
447 struct mdd_object *mdo;
450 mdo = mdd_object_find(env, mdd, lf);
456 if (mdo->mod_count == 0) {
457 CDEBUG(D_HA, "Found orphan "DFID", delete it\n", PFID(lf));
458 rc = orphan_object_destroy(env, mdo, key);
459 if (rc) /* so replay-single.sh test_37 works */
460 CERROR("%s: error unlinking orphan "DFID" from "
461 "PENDING: rc = %d\n",
462 mdd->mdd_obd_dev->obd_name, PFID(lf), rc);
464 mdd_write_lock(env, mdo, MOR_TGT_CHILD);
465 if (likely(mdo->mod_count > 0)) {
466 CDEBUG(D_HA, "Found orphan "DFID" count %d, skip it\n",
467 PFID(lf), mdo->mod_count);
468 mdo->mod_flags |= ORPHAN_OBJ;
470 mdd_write_unlock(env, mdo);
473 mdd_object_put(env, mdo);
478 * delete unreferenced files and directories in the PENDING directory
480 * Files that remain in PENDING after client->MDS recovery has completed
481 * have to be referenced (opened) by some client during recovery, or they
482 * will be deleted here (for clients that did not complete recovery).
484 * \param mdd MDD device finishing recovery
489 static int orph_index_iterate(const struct lu_env *env,
490 struct mdd_device *mdd)
492 struct dt_object *dor = mdd->mdd_orphans;
493 char *mti_key = mdd_env_info(env)->mti_orph_key;
494 const struct dt_it_ops *iops;
504 /* In recovery phase, do not need for any lock here */
505 iops = &dor->do_index_ops->dio_it;
506 it = iops->init(env, dor, LUDA_64BITHASH, BYPASS_CAPA);
509 CERROR("%s: cannot clean PENDING: rc = %d\n",
510 mdd->mdd_obd_dev->obd_name, rc);
514 rc = iops->load(env, it, 0);
518 CERROR("%s: error loading iterator to clean PENDING\n",
519 mdd->mdd_obd_dev->obd_name);
520 /* Index contains no zero key? */
521 GOTO(out_put, rc = -EIO);
525 key = (void *)iops->key(env, it);
527 CERROR("%s: key failed when clean PENDING: rc = %ld\n",
528 mdd->mdd_obd_dev->obd_name, PTR_ERR(key));
531 key_sz = iops->key_size(env, it);
533 /* filter out "." and ".." entries from PENDING dir. */
537 memcpy(mti_key, key, key_sz);
540 if (orphan_key_to_fid(mti_key, &fid))
542 if (!fid_is_sane(&fid)) {
543 CERROR("%s: bad FID "DFID" cleaning PENDING\n",
544 mdd->mdd_obd_dev->obd_name, PFID(&fid));
548 /* kill orphan object */
549 cookie = iops->store(env, it);
551 rc = orph_key_test_and_del(env, mdd, &fid,
552 (struct dt_key *)mti_key);
554 /* after index delete reset iterator */
556 result = iops->get(env, it, (const void *)"");
558 result = iops->load(env, it, cookie);
560 result = iops->next(env, it);
561 } while (result == 0);
563 GOTO(out_put, rc = 0);
573 * open the PENDING directory for device \a mdd
575 * The PENDING directory persistently tracks files and directories that were
576 * unlinked from the namespace (nlink == 0) but are still held open by clients.
577 * Those inodes shouldn't be deleted if the MDS crashes, because the clients
578 * would not be able to recover and reopen those files. Instead, these inodes
579 * are linked into the PENDING directory on disk, and only deleted if all
580 * clients close them, or the MDS finishes client recovery without any client
581 * reopening them (i.e. former clients didn't join recovery).
582 * \param d mdd device being started.
585 * \retval -ve index operation error.
588 int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
595 d = dt_store_open(env, mdd->mdd_child, "", orph_index_name, &fid);
597 mdd->mdd_orphans = d;
598 if (!dt_try_as_dir(env, d)) {
600 CERROR("\"%s\" is not an index! : rc = %d\n",
601 orph_index_name, rc);
604 CERROR("cannot find \"%s\" obj %d\n",
605 orph_index_name, (int)PTR_ERR(d));
612 void orph_index_fini(const struct lu_env *env, struct mdd_device *mdd)
615 if (mdd->mdd_orphans != NULL) {
616 lu_object_put(env, &mdd->mdd_orphans->do_lu);
617 mdd->mdd_orphans = NULL;
623 * Iterate orphan index to cleanup orphan objects after recovery is done.
624 * \param d mdd device in recovery.
626 int __mdd_orphan_cleanup(const struct lu_env *env, struct mdd_device *d)
628 return orph_index_iterate(env, d);
632 * add an orphan \a obj to the orphan index.
633 * \param obj file or directory.
634 * \param th transaction for index insert.
636 * \pre obj nlink == 0 && obj->mod_count != 0
639 * \retval -ve index operation error.
641 int __mdd_orphan_add(const struct lu_env *env,
642 struct mdd_object *obj, struct thandle *th)
644 return orph_index_insert(env, obj, ORPH_OP_UNLINK, th);
648 * delete an orphan \a obj from orphan index.
649 * \param obj file or directory.
650 * \param th transaction for index deletion and object destruction.
652 * \pre obj->mod_count == 0 && ORPHAN_OBJ is set for obj.
655 * \retval -ve index operation error.
657 int __mdd_orphan_del(const struct lu_env *env,
658 struct mdd_object *obj, struct thandle *th)
660 return orph_index_delete(env, obj, ORPH_OP_UNLINK, th);