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, Intel Corporation.
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);
267 LASSERT(los->los_obj);
268 LASSERT(dt_object_exists(los->los_obj));
270 /* many threads can be updated this, serialize
271 * them here to avoid the race where one thread
272 * takes the value first, but writes it last */
273 mutex_lock(&los->los_id_lock);
275 /* update local oid number on disk so that
276 * we know the last one used after reboot */
277 losd.lso_magic = cpu_to_le32(LOS_MAGIC);
278 losd.lso_next_oid = cpu_to_le32(los->los_last_oid);
281 dti->dti_lb.lb_buf = &losd;
282 dti->dti_lb.lb_len = sizeof(losd);
283 rc = dt_record_write(env, los->los_obj, &dti->dti_lb, &dti->dti_off,
285 mutex_unlock(&los->los_id_lock);
291 * Create local named object (file, directory or index) in parent directory.
293 struct dt_object *__local_file_create(const struct lu_env *env,
294 const struct lu_fid *fid,
295 struct local_oid_storage *los,
296 struct ls_device *ls,
297 struct dt_object *parent,
298 const char *name, struct lu_attr *attr,
299 struct dt_object_format *dof)
301 struct dt_thread_info *dti = dt_info(env);
302 struct dt_object *dto;
306 dto = ls_locate(env, ls, fid);
307 if (unlikely(IS_ERR(dto)))
310 LASSERT(dto != NULL);
311 if (dt_object_exists(dto))
312 GOTO(out, rc = -EEXIST);
314 th = dt_trans_create(env, ls->ls_osd);
316 GOTO(out, rc = PTR_ERR(th));
318 rc = local_object_declare_create(env, los, dto, attr, dof, th);
320 GOTO(trans_stop, rc);
322 if (dti->dti_dof.dof_type == DFT_DIR) {
323 dt_declare_ref_add(env, dto, th);
324 dt_declare_ref_add(env, parent, th);
327 rc = dt_declare_insert(env, parent, (void *)fid, (void *)name, th);
329 GOTO(trans_stop, rc);
331 rc = dt_trans_start_local(env, ls->ls_osd, th);
333 GOTO(trans_stop, rc);
335 dt_write_lock(env, dto, 0);
336 if (dt_object_exists(dto))
337 GOTO(unlock, rc = 0);
339 CDEBUG(D_OTHER, "create new object "DFID"\n",
340 PFID(lu_object_fid(&dto->do_lu)));
341 rc = local_object_create(env, los, dto, attr, dof, th);
344 LASSERT(dt_object_exists(dto));
346 if (dti->dti_dof.dof_type == DFT_DIR) {
347 if (!dt_try_as_dir(env, dto))
348 GOTO(destroy, rc = -ENOTDIR);
349 /* Add "." and ".." for newly created dir */
350 rc = dt_insert(env, dto, (void *)fid, (void *)".", th,
354 dt_ref_add(env, dto, th);
355 rc = dt_insert(env, dto, (void *)lu_object_fid(&parent->do_lu),
356 (void *)"..", th, BYPASS_CAPA, 1);
361 dt_write_lock(env, parent, 0);
362 rc = dt_insert(env, parent, (const struct dt_rec *)fid,
363 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
364 if (dti->dti_dof.dof_type == DFT_DIR)
365 dt_ref_add(env, parent, th);
366 dt_write_unlock(env, parent);
371 dt_destroy(env, dto, th);
373 dt_write_unlock(env, dto);
375 dt_trans_stop(env, ls->ls_osd, th);
378 lu_object_put_nocache(env, &dto->do_lu);
385 * Look up and create (if it does not exist) a local named file or directory in
388 struct dt_object *local_file_find_or_create(const struct lu_env *env,
389 struct local_oid_storage *los,
390 struct dt_object *parent,
391 const char *name, __u32 mode)
393 struct dt_thread_info *dti = dt_info(env);
394 struct dt_object *dto;
399 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
401 /* name is found, get the object */
402 dto = ls_locate(env, dt2ls_dev(los->los_dev), &dti->dti_fid);
403 else if (rc != -ENOENT)
406 rc = local_object_fid_generate(env, los, &dti->dti_fid);
410 /* create the object */
411 dti->dti_attr.la_valid = LA_MODE;
412 dti->dti_attr.la_mode = mode;
413 dti->dti_dof.dof_type = dt_mode_to_dft(mode & S_IFMT);
414 dto = __local_file_create(env, &dti->dti_fid, los,
415 dt2ls_dev(los->los_dev),
416 parent, name, &dti->dti_attr,
422 EXPORT_SYMBOL(local_file_find_or_create);
424 struct dt_object *local_file_find_or_create_with_fid(const struct lu_env *env,
425 struct dt_device *dt,
426 const struct lu_fid *fid,
427 struct dt_object *parent,
431 struct dt_thread_info *dti = dt_info(env);
432 struct dt_object *dto;
437 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
439 /* name is found, get the object */
440 if (!lu_fid_eq(fid, &dti->dti_fid))
441 dto = ERR_PTR(-EINVAL);
443 dto = dt_locate(env, dt, fid);
444 } else if (rc != -ENOENT) {
447 struct ls_device *ls;
449 ls = ls_device_get(dt);
451 dto = ERR_PTR(PTR_ERR(ls));
453 /* create the object */
454 dti->dti_attr.la_valid = LA_MODE;
455 dti->dti_attr.la_mode = mode;
456 dti->dti_dof.dof_type = dt_mode_to_dft(mode & S_IFMT);
457 dto = __local_file_create(env, fid, NULL, ls, parent,
458 name, &dti->dti_attr,
460 /* ls_device_put() will finalize the ls device, we
461 * have to open the object in other device stack */
463 dti->dti_fid = dto->do_lu.lo_header->loh_fid;
464 lu_object_put_nocache(env, &dto->do_lu);
465 dto = dt_locate(env, dt, &dti->dti_fid);
467 ls_device_put(env, ls);
472 EXPORT_SYMBOL(local_file_find_or_create_with_fid);
475 * Look up and create (if it does not exist) a local named index file in parent
478 struct dt_object *local_index_find_or_create(const struct lu_env *env,
479 struct local_oid_storage *los,
480 struct dt_object *parent,
481 const char *name, __u32 mode,
482 const struct dt_index_features *ft)
484 struct dt_thread_info *dti = dt_info(env);
485 struct dt_object *dto;
490 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
492 /* name is found, get the object */
493 dto = ls_locate(env, dt2ls_dev(los->los_dev), &dti->dti_fid);
494 } else if (rc != -ENOENT) {
497 rc = local_object_fid_generate(env, los, &dti->dti_fid);
501 /* create the object */
502 dti->dti_attr.la_valid = LA_MODE;
503 dti->dti_attr.la_mode = mode;
504 dti->dti_dof.dof_type = DFT_INDEX;
505 dti->dti_dof.u.dof_idx.di_feat = ft;
506 dto = __local_file_create(env, &dti->dti_fid, los,
507 dt2ls_dev(los->los_dev),
508 parent, name, &dti->dti_attr,
515 EXPORT_SYMBOL(local_index_find_or_create);
518 local_index_find_or_create_with_fid(const struct lu_env *env,
519 struct dt_device *dt,
520 const struct lu_fid *fid,
521 struct dt_object *parent,
522 const char *name, __u32 mode,
523 const struct dt_index_features *ft)
525 struct dt_thread_info *dti = dt_info(env);
526 struct dt_object *dto;
531 rc = dt_lookup_dir(env, parent, name, &dti->dti_fid);
533 /* name is found, get the object */
534 if (!lu_fid_eq(fid, &dti->dti_fid))
535 dto = ERR_PTR(-EINVAL);
537 dto = dt_locate(env, dt, fid);
538 } else if (rc != -ENOENT) {
541 struct ls_device *ls;
543 ls = ls_device_get(dt);
545 dto = ERR_PTR(PTR_ERR(ls));
547 /* create the object */
548 dti->dti_attr.la_valid = LA_MODE;
549 dti->dti_attr.la_mode = mode;
550 dti->dti_dof.dof_type = DFT_INDEX;
551 dti->dti_dof.u.dof_idx.di_feat = ft;
552 dto = __local_file_create(env, fid, NULL, ls, parent,
553 name, &dti->dti_attr,
555 /* ls_device_put() will finalize the ls device, we
556 * have to open the object in other device stack */
558 dti->dti_fid = dto->do_lu.lo_header->loh_fid;
559 lu_object_put_nocache(env, &dto->do_lu);
560 dto = dt_locate(env, dt, &dti->dti_fid);
562 ls_device_put(env, ls);
567 EXPORT_SYMBOL(local_index_find_or_create_with_fid);
569 struct local_oid_storage *dt_los_find(struct ls_device *ls, __u64 seq)
571 struct local_oid_storage *los, *ret = NULL;
573 cfs_list_for_each_entry(los, &ls->ls_los_list, los_list) {
574 if (los->los_seq == seq) {
575 cfs_atomic_inc(&los->los_refcount);
583 void dt_los_put(struct local_oid_storage *los)
585 if (cfs_atomic_dec_and_test(&los->los_refcount))
586 /* should never happen, only local_oid_storage_fini should
587 * drop refcount to zero */
593 * Initialize local OID storage for required sequence.
594 * That may be needed for services that uses local files and requires
595 * dynamic OID allocation for them.
597 * Per each sequence we have an object with 'first_fid' identificator
598 * containing the counter for OIDs of locally created files with that
601 * It is used now by llog subsystem and MGS for NID tables
603 * Function gets first_fid to create counter object.
604 * All dynamic fids will be generated with the same sequence and incremented
607 * Returned local_oid_storage is in-memory representaion of OID storage
609 int local_oid_storage_init(const struct lu_env *env, struct dt_device *dev,
610 const struct lu_fid *first_fid,
611 struct local_oid_storage **los)
613 struct dt_thread_info *dti = dt_info(env);
614 struct ls_device *ls;
615 struct los_ondisk losd;
616 struct dt_object *root = NULL;
617 struct dt_object *o = NULL;
623 ls = ls_device_get(dev);
627 mutex_lock(&ls->ls_los_mutex);
628 *los = dt_los_find(ls, fid_seq(first_fid));
632 /* not found, then create */
635 GOTO(out, rc = -ENOMEM);
637 cfs_atomic_set(&(*los)->los_refcount, 1);
638 mutex_init(&(*los)->los_id_lock);
639 (*los)->los_dev = &ls->ls_top_dev;
640 cfs_atomic_inc(&ls->ls_refcount);
641 cfs_list_add(&(*los)->los_list, &ls->ls_los_list);
643 rc = dt_root_get(env, dev, &dti->dti_fid);
647 root = ls_locate(env, ls, &dti->dti_fid);
649 GOTO(out_los, rc = PTR_ERR(root));
651 snprintf(dti->dti_buf, sizeof(dti->dti_buf), "seq-%Lx-lastid",
653 rc = dt_lookup_dir(env, root, dti->dti_buf, &dti->dti_fid);
654 if (rc != 0 && rc != -ENOENT)
657 /* initialize data allowing to generate new fids,
658 * literally we need a sequence */
660 o = ls_locate(env, ls, &dti->dti_fid);
662 o = ls_locate(env, ls, first_fid);
664 GOTO(out_los, rc = PTR_ERR(o));
666 dt_write_lock(env, o, 0);
667 if (!dt_object_exists(o)) {
668 LASSERT(rc == -ENOENT);
670 th = dt_trans_create(env, dev);
672 GOTO(out_lock, rc = PTR_ERR(th));
674 dti->dti_attr.la_valid = LA_MODE | LA_TYPE;
675 dti->dti_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
676 dti->dti_dof.dof_type = dt_mode_to_dft(S_IFREG);
678 rc = dt_declare_create(env, o, &dti->dti_attr, NULL,
683 rc = dt_declare_insert(env, root,
684 (const struct dt_rec *)lu_object_fid(&o->do_lu),
685 (const struct dt_key *)dti->dti_buf,
690 dti->dti_lb.lb_buf = NULL;
691 dti->dti_lb.lb_len = sizeof(dti->dti_lma);
692 rc = dt_declare_xattr_set(env, o, &dti->dti_lb, XATTR_NAME_LMA,
697 rc = dt_declare_record_write(env, o, sizeof(losd), 0, th);
701 rc = dt_trans_start_local(env, dev, th);
705 LASSERT(!dt_object_exists(o));
706 rc = dt_create(env, o, &dti->dti_attr, NULL, &dti->dti_dof, th);
709 LASSERT(dt_object_exists(o));
711 losd.lso_magic = cpu_to_le32(LOS_MAGIC);
712 losd.lso_next_oid = cpu_to_le32(fid_oid(first_fid) + 1);
715 dti->dti_lb.lb_buf = &losd;
716 dti->dti_lb.lb_len = sizeof(losd);
717 rc = dt_record_write(env, o, &dti->dti_lb, &dti->dti_off, th);
720 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 3, 90, 0)
721 #error "fix this before release"
724 * there is one technical debt left in Orion:
725 * proper hanlding of named vs no-name objects.
726 * Llog objects have name always as they are placed in O/d/...
728 if (fid_seq(lu_object_fid(&o->do_lu)) != FID_SEQ_LLOG) {
729 rc = dt_insert(env, root,
730 (const struct dt_rec *)first_fid,
731 (const struct dt_key *)dti->dti_buf,
737 dt_trans_stop(env, dev, th);
740 dti->dti_lb.lb_buf = &losd;
741 dti->dti_lb.lb_len = sizeof(losd);
742 rc = dt_record_read(env, o, &dti->dti_lb, &dti->dti_off);
743 if (rc == 0 && le32_to_cpu(losd.lso_magic) != LOS_MAGIC) {
744 CERROR("local storage file "DFID" is corrupted\n",
750 dt_write_unlock(env, o);
752 if (root != NULL && !IS_ERR(root))
753 lu_object_put_nocache(env, &root->do_lu);
756 cfs_list_del(&(*los)->los_list);
757 cfs_atomic_dec(&ls->ls_refcount);
760 if (o != NULL && !IS_ERR(o))
761 lu_object_put_nocache(env, &o->do_lu);
763 (*los)->los_seq = fid_seq(first_fid);
764 (*los)->los_last_oid = le32_to_cpu(losd.lso_next_oid);
768 mutex_unlock(&ls->ls_los_mutex);
769 ls_device_put(env, ls);
772 EXPORT_SYMBOL(local_oid_storage_init);
774 void local_oid_storage_fini(const struct lu_env *env,
775 struct local_oid_storage *los)
777 struct ls_device *ls;
779 if (!cfs_atomic_dec_and_test(&los->los_refcount))
783 LASSERT(los->los_dev);
784 ls = dt2ls_dev(los->los_dev);
786 mutex_lock(&ls->ls_los_mutex);
787 if (cfs_atomic_read(&los->los_refcount) == 0) {
789 lu_object_put_nocache(env, &los->los_obj->do_lu);
790 cfs_list_del(&los->los_list);
793 mutex_unlock(&ls->ls_los_mutex);
794 ls_device_put(env, ls);
796 EXPORT_SYMBOL(local_oid_storage_fini);