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,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2012 Whamcloud, Inc.
26 * lustre/obdclass/local_storage.c
28 * Local storage for file/objects with fid generation. Works on top of OSD.
30 * Author: Mikhail Pershin <mike.pershin@intel.com>
33 #define DEBUG_SUBSYSTEM S_CLASS
35 #include "local_storage.h"
37 /* all initialized local storages on this node are linked on this */
38 static CFS_LIST_HEAD(ls_list_head);
39 static DEFINE_MUTEX(ls_list_mutex);
41 static int ls_object_init(const struct lu_env *env, struct lu_object *o,
42 const struct lu_object_conf *unused)
45 struct lu_object *below;
46 struct lu_device *under;
50 ls = container_of0(o->lo_dev, struct ls_device, ls_top_dev.dd_lu_dev);
51 under = &ls->ls_osd->dd_lu_dev;
52 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
56 lu_object_add(o, below);
61 static void ls_object_free(const struct lu_env *env, struct lu_object *o)
63 struct ls_object *obj = lu2ls_obj(o);
64 struct lu_object_header *h = o->lo_header;
66 dt_object_fini(&obj->ls_obj);
67 lu_object_header_fini(h);
71 struct lu_object_operations ls_lu_obj_ops = {
72 .loo_object_init = ls_object_init,
73 .loo_object_free = ls_object_free,
76 struct lu_object *ls_object_alloc(const struct lu_env *env,
77 const struct lu_object_header *_h,
80 struct lu_object_header *h;
91 lu_object_header_init(h);
92 dt_object_init(&o->ls_obj, h, d);
93 lu_object_add_top(h, l);
95 l->lo_ops = &ls_lu_obj_ops;
103 static struct lu_device_operations ls_lu_dev_ops = {
104 .ldo_object_alloc = ls_object_alloc
107 static struct ls_device *__ls_find_dev(struct dt_device *dev)
109 struct ls_device *ls, *ret = NULL;
111 cfs_list_for_each_entry(ls, &ls_list_head, ls_linkage) {
112 if (ls->ls_osd == dev) {
113 cfs_atomic_inc(&ls->ls_refcount);
121 struct ls_device *ls_find_dev(struct dt_device *dev)
123 struct ls_device *ls;
125 mutex_lock(&ls_list_mutex);
126 ls = __ls_find_dev(dev);
127 mutex_unlock(&ls_list_mutex);
132 static struct lu_device_type_operations ls_device_type_ops = {
137 static struct lu_device_type ls_lu_type = {
138 .ldt_name = "local_storage",
139 .ldt_ops = &ls_device_type_ops,
142 struct ls_device *ls_device_get(struct dt_device *dev)
144 struct ls_device *ls;
148 mutex_lock(&ls_list_mutex);
149 ls = __ls_find_dev(dev);
153 /* not found, then create */
156 GOTO(out_ls, ls = ERR_PTR(-ENOMEM));
158 cfs_atomic_set(&ls->ls_refcount, 1);
159 CFS_INIT_LIST_HEAD(&ls->ls_los_list);
160 mutex_init(&ls->ls_los_mutex);
164 LASSERT(dev->dd_lu_dev.ld_site);
165 lu_device_init(&ls->ls_top_dev.dd_lu_dev, &ls_lu_type);
166 ls->ls_top_dev.dd_lu_dev.ld_ops = &ls_lu_dev_ops;
167 ls->ls_top_dev.dd_lu_dev.ld_site = dev->dd_lu_dev.ld_site;
169 /* finally add ls to the list */
170 cfs_list_add(&ls->ls_linkage, &ls_list_head);
172 mutex_unlock(&ls_list_mutex);
176 void ls_device_put(const struct lu_env *env, struct ls_device *ls)
179 if (!cfs_atomic_dec_and_test(&ls->ls_refcount))
182 mutex_lock(&ls_list_mutex);
183 if (cfs_atomic_read(&ls->ls_refcount) == 0) {
184 LASSERT(cfs_list_empty(&ls->ls_los_list));
185 cfs_list_del(&ls->ls_linkage);
186 lu_site_purge(env, ls->ls_top_dev.dd_lu_dev.ld_site, ~0);
187 lu_device_fini(&ls->ls_top_dev.dd_lu_dev);
190 mutex_unlock(&ls_list_mutex);
194 * local file fid generation
196 int local_object_fid_generate(const struct lu_env *env,
197 struct local_oid_storage *los,
200 LASSERT(los->los_dev);
201 LASSERT(los->los_obj);
205 /* to make it unique after reboot we store
206 * the latest generated fid atomically with
207 * object creation see local_object_create() */
209 mutex_lock(&los->los_id_lock);
210 fid->f_seq = los->los_seq;
211 fid->f_oid = los->los_last_oid++;
213 mutex_unlock(&los->los_id_lock);
218 int local_object_declare_create(const struct lu_env *env,
219 struct local_oid_storage *los,
220 struct dt_object *o, struct lu_attr *attr,
221 struct dt_object_format *dof,
224 struct dt_thread_info *dti = dt_info(env);
229 /* update fid generation file */
231 LASSERT(dt_object_exists(los->los_obj));
232 rc = dt_declare_record_write(env, los->los_obj,
233 sizeof(struct los_ondisk), 0, th);
238 rc = dt_declare_create(env, o, attr, NULL, dof, th);
242 dti->dti_lb.lb_buf = NULL;
243 dti->dti_lb.lb_len = sizeof(dti->dti_lma);
244 rc = dt_declare_xattr_set(env, o, &dti->dti_lb, XATTR_NAME_LMA, 0, th);
249 int local_object_create(const struct lu_env *env,
250 struct local_oid_storage *los,
251 struct dt_object *o, struct lu_attr *attr,
252 struct dt_object_format *dof, struct thandle *th)
254 struct dt_thread_info *dti = dt_info(env);
255 struct los_ondisk losd;
260 rc = dt_create(env, o, attr, NULL, dof, th);
264 lustre_lma_init(&dti->dti_lma, lu_object_fid(&o->do_lu));
265 lustre_lma_swab(&dti->dti_lma);
266 dti->dti_lb.lb_buf = &dti->dti_lma;
267 dti->dti_lb.lb_len = sizeof(dti->dti_lma);
268 rc = dt_xattr_set(env, o, &dti->dti_lb, XATTR_NAME_LMA, 0, th,
274 LASSERT(los->los_obj);
275 LASSERT(dt_object_exists(los->los_obj));
277 /* many threads can be updated this, serialize
278 * them here to avoid the race where one thread
279 * takes the value first, but writes it last */
280 mutex_lock(&los->los_id_lock);
282 /* update local oid number on disk so that
283 * we know the last one used after reboot */
284 losd.lso_magic = cpu_to_le32(LOS_MAGIC);
285 losd.lso_next_oid = cpu_to_le32(los->los_last_oid);
288 dti->dti_lb.lb_buf = &losd;
289 dti->dti_lb.lb_len = sizeof(losd);
290 rc = dt_record_write(env, los->los_obj, &dti->dti_lb, &dti->dti_off,
292 mutex_unlock(&los->los_id_lock);
298 * Create local named object (file, directory or index) in parent directory.
300 struct dt_object *__local_file_create(const struct lu_env *env,
301 const struct lu_fid *fid,
302 struct local_oid_storage *los,
303 struct ls_device *ls,
304 struct dt_object *parent,
305 const char *name, struct lu_attr *attr,
306 struct dt_object_format *dof)
308 struct dt_thread_info *dti = dt_info(env);
309 struct dt_object *dto;
313 dto = ls_locate(env, ls, fid);
314 if (unlikely(IS_ERR(dto)))
317 LASSERT(dto != NULL);
318 if (dt_object_exists(dto))
319 GOTO(out, rc = -EEXIST);
321 th = dt_trans_create(env, ls->ls_osd);
323 GOTO(out, rc = PTR_ERR(th));
325 rc = local_object_declare_create(env, los, dto, attr, dof, th);
327 GOTO(trans_stop, rc);
329 if (dti->dti_dof.dof_type == DFT_DIR) {
330 dt_declare_ref_add(env, dto, th);
331 dt_declare_ref_add(env, parent, th);
334 rc = dt_declare_insert(env, parent, (void *)fid, (void *)name, th);
336 GOTO(trans_stop, rc);
338 rc = dt_trans_start_local(env, ls->ls_osd, th);
340 GOTO(trans_stop, rc);
342 dt_write_lock(env, dto, 0);
343 if (dt_object_exists(dto))
344 GOTO(unlock, rc = 0);
346 CDEBUG(D_OTHER, "create new object "DFID"\n",
347 PFID(lu_object_fid(&dto->do_lu)));
348 rc = local_object_create(env, los, dto, attr, dof, th);
351 LASSERT(dt_object_exists(dto));
353 if (dti->dti_dof.dof_type == DFT_DIR) {
354 if (!dt_try_as_dir(env, dto))
355 GOTO(destroy, rc = -ENOTDIR);
356 /* Add "." and ".." for newly created dir */
357 rc = dt_insert(env, dto, (void *)fid, (void *)".", th,
361 dt_ref_add(env, dto, th);
362 rc = dt_insert(env, dto, (void *)lu_object_fid(&parent->do_lu),
363 (void *)"..", th, BYPASS_CAPA, 1);
368 dt_write_lock(env, parent, 0);
369 rc = dt_insert(env, parent, (const struct dt_rec *)fid,
370 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
371 if (dti->dti_dof.dof_type == DFT_DIR)
372 dt_ref_add(env, parent, th);
373 dt_write_unlock(env, parent);
378 dt_destroy(env, dto, th);
380 dt_write_unlock(env, dto);
382 dt_trans_stop(env, ls->ls_osd, th);
385 lu_object_put_nocache(env, &dto->do_lu);
388 struct lu_fid dti_fid;
389 /* since local files FIDs are not in OI the directory entry
390 * is used to get inode number/generation, we need to do lookup
391 * again to cache this data after create */
392 rc = dt_lookup_dir(env, parent, name, &dti_fid);
399 * Look up and create (if it does not exist) a local named file or directory in
402 struct dt_object *local_file_find_or_create(const struct lu_env *env,
403 struct local_oid_storage *los,
404 struct dt_object *parent,
405 const char *name, __u32 mode)
407 struct dt_thread_info *dti = dt_info(env);
408 struct dt_object *dto;
413 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
415 /* name is found, get the object */
416 dto = ls_locate(env, dt2ls_dev(los->los_dev), &dti->dti_fid);
417 else if (rc != -ENOENT)
420 rc = local_object_fid_generate(env, los, &dti->dti_fid);
424 /* create the object */
425 dti->dti_attr.la_valid = LA_MODE;
426 dti->dti_attr.la_mode = mode;
427 dti->dti_dof.dof_type = dt_mode_to_dft(mode & S_IFMT);
428 dto = __local_file_create(env, &dti->dti_fid, los,
429 dt2ls_dev(los->los_dev),
430 parent, name, &dti->dti_attr,
436 EXPORT_SYMBOL(local_file_find_or_create);
438 struct dt_object *local_file_find_or_create_with_fid(const struct lu_env *env,
439 struct dt_device *dt,
440 const struct lu_fid *fid,
441 struct dt_object *parent,
445 struct dt_thread_info *dti = dt_info(env);
446 struct dt_object *dto;
451 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
453 /* name is found, get the object */
454 if (!lu_fid_eq(fid, &dti->dti_fid))
455 dto = ERR_PTR(-EINVAL);
457 dto = dt_locate(env, dt, fid);
458 } else if (rc != -ENOENT) {
461 struct ls_device *ls;
463 ls = ls_device_get(dt);
465 dto = ERR_PTR(PTR_ERR(ls));
467 /* create the object */
468 dti->dti_attr.la_valid = LA_MODE;
469 dti->dti_attr.la_mode = mode;
470 dti->dti_dof.dof_type = dt_mode_to_dft(mode & S_IFMT);
471 dto = __local_file_create(env, fid, NULL, ls, parent,
472 name, &dti->dti_attr,
474 /* ls_device_put() will finalize the ls device, we
475 * have to open the object in other device stack */
477 dti->dti_fid = dto->do_lu.lo_header->loh_fid;
478 lu_object_put_nocache(env, &dto->do_lu);
479 dto = dt_locate(env, dt, &dti->dti_fid);
481 ls_device_put(env, ls);
486 EXPORT_SYMBOL(local_file_find_or_create_with_fid);
489 * Look up and create (if it does not exist) a local named index file in parent
492 struct dt_object *local_index_find_or_create(const struct lu_env *env,
493 struct local_oid_storage *los,
494 struct dt_object *parent,
495 const char *name, __u32 mode,
496 const struct dt_index_features *ft)
498 struct dt_thread_info *dti = dt_info(env);
499 struct dt_object *dto;
504 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
506 /* name is found, get the object */
507 dto = ls_locate(env, dt2ls_dev(los->los_dev), &dti->dti_fid);
508 } else if (rc != -ENOENT) {
511 rc = local_object_fid_generate(env, los, &dti->dti_fid);
515 /* create the object */
516 dti->dti_attr.la_valid = LA_MODE;
517 dti->dti_attr.la_mode = mode;
518 dti->dti_dof.dof_type = DFT_INDEX;
519 dti->dti_dof.u.dof_idx.di_feat = ft;
520 dto = __local_file_create(env, &dti->dti_fid, los,
521 dt2ls_dev(los->los_dev),
522 parent, name, &dti->dti_attr,
529 EXPORT_SYMBOL(local_index_find_or_create);
532 local_index_find_or_create_with_fid(const struct lu_env *env,
533 struct dt_device *dt,
534 const struct lu_fid *fid,
535 struct dt_object *parent,
536 const char *name, __u32 mode,
537 const struct dt_index_features *ft)
539 struct dt_thread_info *dti = dt_info(env);
540 struct dt_object *dto;
545 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
547 /* name is found, get the object */
548 if (!lu_fid_eq(fid, &dti->dti_fid))
549 dto = ERR_PTR(-EINVAL);
551 dto = dt_locate(env, dt, fid);
552 } else if (rc != -ENOENT) {
555 struct ls_device *ls;
557 ls = ls_device_get(dt);
559 dto = ERR_PTR(PTR_ERR(ls));
561 /* create the object */
562 dti->dti_attr.la_valid = LA_MODE;
563 dti->dti_attr.la_mode = mode;
564 dti->dti_dof.dof_type = DFT_INDEX;
565 dti->dti_dof.u.dof_idx.di_feat = ft;
566 dto = __local_file_create(env, fid, NULL, ls, parent,
567 name, &dti->dti_attr,
569 /* ls_device_put() will finalize the ls device, we
570 * have to open the object in other device stack */
572 dti->dti_fid = dto->do_lu.lo_header->loh_fid;
573 lu_object_put_nocache(env, &dto->do_lu);
574 dto = dt_locate(env, dt, &dti->dti_fid);
576 ls_device_put(env, ls);
581 EXPORT_SYMBOL(local_index_find_or_create_with_fid);
583 struct local_oid_storage *dt_los_find(struct ls_device *ls, __u64 seq)
585 struct local_oid_storage *los, *ret = NULL;
587 cfs_list_for_each_entry(los, &ls->ls_los_list, los_list) {
588 if (los->los_seq == seq) {
589 cfs_atomic_inc(&los->los_refcount);
597 void dt_los_put(struct local_oid_storage *los)
599 if (cfs_atomic_dec_and_test(&los->los_refcount))
600 /* should never happen, only local_oid_storage_fini should
601 * drop refcount to zero */
607 * Initialize local OID storage for required sequence.
608 * That may be needed for services that uses local files and requires
609 * dynamic OID allocation for them.
611 * Per each sequence we have an object with 'first_fid' identificator
612 * containing the counter for OIDs of locally created files with that
615 * It is used now by llog subsystem and MGS for NID tables
617 * Function gets first_fid to create counter object.
618 * All dynamic fids will be generated with the same sequence and incremented
621 * Returned local_oid_storage is in-memory representaion of OID storage
623 int local_oid_storage_init(const struct lu_env *env, struct dt_device *dev,
624 const struct lu_fid *first_fid,
625 struct local_oid_storage **los)
627 struct dt_thread_info *dti = dt_info(env);
628 struct ls_device *ls;
629 struct los_ondisk losd;
630 struct dt_object *root = NULL;
631 struct dt_object *o = NULL;
637 ls = ls_device_get(dev);
641 mutex_lock(&ls->ls_los_mutex);
642 *los = dt_los_find(ls, fid_seq(first_fid));
646 /* not found, then create */
649 GOTO(out, rc = -ENOMEM);
651 cfs_atomic_set(&(*los)->los_refcount, 1);
652 mutex_init(&(*los)->los_id_lock);
653 (*los)->los_dev = &ls->ls_top_dev;
654 cfs_atomic_inc(&ls->ls_refcount);
655 cfs_list_add(&(*los)->los_list, &ls->ls_los_list);
657 rc = dt_root_get(env, dev, &dti->dti_fid);
661 root = ls_locate(env, ls, &dti->dti_fid);
663 GOTO(out_los, rc = PTR_ERR(root));
665 snprintf(dti->dti_buf, sizeof(dti->dti_buf), "seq-%Lx-lastid",
667 rc = dt_lookup_dir(env, root, dti->dti_buf, &dti->dti_fid);
668 if (rc != 0 && rc != -ENOENT)
671 /* initialize data allowing to generate new fids,
672 * literally we need a sequence */
674 o = ls_locate(env, ls, &dti->dti_fid);
676 o = ls_locate(env, ls, first_fid);
678 GOTO(out_los, rc = PTR_ERR(o));
680 dt_write_lock(env, o, 0);
681 if (!dt_object_exists(o)) {
682 LASSERT(rc == -ENOENT);
684 th = dt_trans_create(env, dev);
686 GOTO(out_lock, rc = PTR_ERR(th));
688 dti->dti_attr.la_valid = LA_MODE | LA_TYPE;
689 dti->dti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
690 dti->dti_dof.dof_type = dt_mode_to_dft(S_IFREG);
692 rc = dt_declare_create(env, o, &dti->dti_attr, NULL,
697 rc = dt_declare_insert(env, root,
698 (const struct dt_rec *)lu_object_fid(&o->do_lu),
699 (const struct dt_key *)dti->dti_buf,
704 dti->dti_lb.lb_buf = NULL;
705 dti->dti_lb.lb_len = sizeof(dti->dti_lma);
706 rc = dt_declare_xattr_set(env, o, &dti->dti_lb, XATTR_NAME_LMA,
711 rc = dt_declare_record_write(env, o, sizeof(losd), 0, th);
715 rc = dt_trans_start_local(env, dev, th);
719 LASSERT(!dt_object_exists(o));
720 rc = dt_create(env, o, &dti->dti_attr, NULL, &dti->dti_dof, th);
723 LASSERT(dt_object_exists(o));
725 lustre_lma_init(&dti->dti_lma, lu_object_fid(&o->do_lu));
726 lustre_lma_swab(&dti->dti_lma);
727 dti->dti_lb.lb_buf = &dti->dti_lma;
728 dti->dti_lb.lb_len = sizeof(dti->dti_lma);
729 rc = dt_xattr_set(env, o, &dti->dti_lb, XATTR_NAME_LMA, 0,
734 losd.lso_magic = cpu_to_le32(LOS_MAGIC);
735 losd.lso_next_oid = cpu_to_le32(fid_oid(first_fid) + 1);
738 dti->dti_lb.lb_buf = &losd;
739 dti->dti_lb.lb_len = sizeof(losd);
740 rc = dt_record_write(env, o, &dti->dti_lb, &dti->dti_off, th);
743 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 3, 90, 0)
744 #error "fix this before release"
747 * there is one technical debt left in Orion:
748 * proper hanlding of named vs no-name objects.
749 * Llog objects have name always as they are placed in O/d/...
751 if (fid_seq(lu_object_fid(&o->do_lu)) != FID_SEQ_LLOG) {
752 rc = dt_insert(env, root,
753 (const struct dt_rec *)first_fid,
754 (const struct dt_key *)dti->dti_buf,
760 dt_trans_stop(env, dev, th);
763 dti->dti_lb.lb_buf = &losd;
764 dti->dti_lb.lb_len = sizeof(losd);
765 rc = dt_record_read(env, o, &dti->dti_lb, &dti->dti_off);
766 if (rc == 0 && le32_to_cpu(losd.lso_magic) != LOS_MAGIC) {
767 CERROR("local storage file "DFID" is corrupted\n",
773 dt_write_unlock(env, o);
775 if (root != NULL && !IS_ERR(root))
776 lu_object_put_nocache(env, &root->do_lu);
779 cfs_list_del(&(*los)->los_list);
780 cfs_atomic_dec(&ls->ls_refcount);
783 if (o != NULL && !IS_ERR(o))
784 lu_object_put_nocache(env, &o->do_lu);
786 (*los)->los_seq = fid_seq(first_fid);
787 (*los)->los_last_oid = le32_to_cpu(losd.lso_next_oid);
791 mutex_unlock(&ls->ls_los_mutex);
792 ls_device_put(env, ls);
795 EXPORT_SYMBOL(local_oid_storage_init);
797 void local_oid_storage_fini(const struct lu_env *env,
798 struct local_oid_storage *los)
800 struct ls_device *ls;
802 if (!cfs_atomic_dec_and_test(&los->los_refcount))
806 LASSERT(los->los_dev);
807 ls = dt2ls_dev(los->los_dev);
809 mutex_lock(&ls->ls_los_mutex);
810 if (cfs_atomic_read(&los->los_refcount) == 0) {
812 lu_object_put_nocache(env, &los->los_obj->do_lu);
813 cfs_list_del(&los->los_list);
816 mutex_unlock(&ls->ls_los_mutex);
817 ls_device_put(env, ls);
819 EXPORT_SYMBOL(local_oid_storage_fini);