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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * Implementation of cl_object for OSC layer.
34 * Author: Nikita Danilov <nikita.danilov@sun.com>
35 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
38 #define DEBUG_SUBSYSTEM S_OSC
39 #include <lustre_osc.h>
41 #include "osc_internal.h"
47 /*****************************************************************************
52 static void osc_obj_build_res_name(struct osc_object *osc,
53 struct ldlm_res_id *resname)
55 ostid_build_res_name(&osc->oo_oinfo->loi_oi, resname);
58 static const struct osc_object_operations osc_object_ops = {
59 .oto_build_res_name = osc_obj_build_res_name,
60 .oto_dlmlock_at_pgoff = osc_obj_dlmlock_at_pgoff,
63 int osc_object_init(const struct lu_env *env, struct lu_object *obj,
64 const struct lu_object_conf *conf)
66 struct osc_object *osc = lu2osc(obj);
67 const struct cl_object_conf *cconf = lu2cl_conf(conf);
69 osc->oo_oinfo = cconf->u.coc_oinfo;
70 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
71 mutex_init(&osc->oo_debug_mutex);
73 INIT_LIST_HEAD(&osc->oo_ready_item);
74 INIT_LIST_HEAD(&osc->oo_hp_ready_item);
75 INIT_LIST_HEAD(&osc->oo_write_item);
76 INIT_LIST_HEAD(&osc->oo_read_item);
78 osc->oo_root.rb_node = NULL;
79 INIT_LIST_HEAD(&osc->oo_hp_exts);
80 INIT_LIST_HEAD(&osc->oo_urgent_exts);
81 INIT_LIST_HEAD(&osc->oo_full_exts);
82 INIT_LIST_HEAD(&osc->oo_reading_exts);
83 atomic_set(&osc->oo_nr_reads, 0);
84 atomic_set(&osc->oo_nr_writes, 0);
85 spin_lock_init(&osc->oo_lock);
86 spin_lock_init(&osc->oo_tree_lock);
87 spin_lock_init(&osc->oo_ol_spin);
88 INIT_LIST_HEAD(&osc->oo_ol_list);
90 atomic_set(&osc->oo_nr_ios, 0);
91 init_waitqueue_head(&osc->oo_io_waitq);
93 LASSERT(osc->oo_obj_ops != NULL);
95 cl_object_page_init(lu2cl(obj), sizeof(struct osc_page));
99 EXPORT_SYMBOL(osc_object_init);
101 void osc_object_free(const struct lu_env *env, struct lu_object *obj)
103 struct osc_object *osc = lu2osc(obj);
105 LASSERT(list_empty(&osc->oo_ready_item));
106 LASSERT(list_empty(&osc->oo_hp_ready_item));
107 LASSERT(list_empty(&osc->oo_write_item));
108 LASSERT(list_empty(&osc->oo_read_item));
110 LASSERT(osc->oo_root.rb_node == NULL);
111 LASSERT(list_empty(&osc->oo_hp_exts));
112 LASSERT(list_empty(&osc->oo_urgent_exts));
113 LASSERT(list_empty(&osc->oo_full_exts));
114 LASSERT(list_empty(&osc->oo_reading_exts));
115 LASSERT(atomic_read(&osc->oo_nr_reads) == 0);
116 LASSERT(atomic_read(&osc->oo_nr_writes) == 0);
117 LASSERT(list_empty(&osc->oo_ol_list));
118 LASSERT(atomic_read(&osc->oo_nr_ios) == 0);
121 /* osc doen't contain an lu_object_header, so we don't need call_rcu */
122 OBD_SLAB_FREE_PTR(osc, osc_object_kmem);
124 EXPORT_SYMBOL(osc_object_free);
126 int osc_lvb_print(const struct lu_env *env, void *cookie,
127 lu_printer_t p, const struct ost_lvb *lvb)
129 return (*p)(env, cookie, "size: %llu mtime: %llu atime: %llu "
130 "ctime: %llu blocks: %llu",
131 lvb->lvb_size, lvb->lvb_mtime, lvb->lvb_atime,
132 lvb->lvb_ctime, lvb->lvb_blocks);
134 EXPORT_SYMBOL(osc_lvb_print);
136 int osc_object_print(const struct lu_env *env, void *cookie,
137 lu_printer_t p, const struct lu_object *obj)
139 struct osc_object *osc = lu2osc(obj);
140 struct lov_oinfo *oinfo = osc->oo_oinfo;
141 struct osc_async_rc *ar = &oinfo->loi_ar;
143 (*p)(env, cookie, "id: "DOSTID" "
144 "idx: %d gen: %d kms_valid: %u kms %llu "
145 "rc: %d force_sync: %d min_xid: %llu ",
146 POSTID(&oinfo->loi_oi), oinfo->loi_ost_idx,
147 oinfo->loi_ost_gen, oinfo->loi_kms_valid, oinfo->loi_kms,
148 ar->ar_rc, ar->ar_force_sync, ar->ar_min_xid);
149 osc_lvb_print(env, cookie, p, &oinfo->loi_lvb);
152 EXPORT_SYMBOL(osc_object_print);
155 int osc_attr_get(const struct lu_env *env, struct cl_object *obj,
156 struct cl_attr *attr)
158 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
160 cl_lvb2attr(attr, &oinfo->loi_lvb);
161 attr->cat_kms = oinfo->loi_kms_valid ? oinfo->loi_kms : 0;
164 EXPORT_SYMBOL(osc_attr_get);
166 int osc_attr_update(const struct lu_env *env, struct cl_object *obj,
167 const struct cl_attr *attr, unsigned valid)
169 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
170 struct ost_lvb *lvb = &oinfo->loi_lvb;
172 if (valid & CAT_SIZE)
173 lvb->lvb_size = attr->cat_size;
174 if (valid & CAT_MTIME)
175 lvb->lvb_mtime = attr->cat_mtime;
176 if (valid & CAT_ATIME)
177 lvb->lvb_atime = attr->cat_atime;
178 if (valid & CAT_CTIME)
179 lvb->lvb_ctime = attr->cat_ctime;
180 if (valid & CAT_BLOCKS)
181 lvb->lvb_blocks = attr->cat_blocks;
182 if (valid & CAT_KMS) {
183 CDEBUG(D_CACHE, "set kms from %lluto %llu\n",
184 oinfo->loi_kms, (__u64)attr->cat_kms);
185 loi_kms_set(oinfo, attr->cat_kms);
189 EXPORT_SYMBOL(osc_attr_update);
191 int osc_object_glimpse(const struct lu_env *env, const struct cl_object *obj,
194 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
196 lvb->lvb_size = oinfo->loi_kms;
197 lvb->lvb_blocks = oinfo->loi_lvb.lvb_blocks;
200 EXPORT_SYMBOL(osc_object_glimpse);
202 static int osc_object_ast_clear(struct ldlm_lock *lock, void *data)
204 struct osc_object *osc = (struct osc_object *)data;
205 struct ost_lvb *lvb = lock->l_lvb_data;
206 struct lov_oinfo *oinfo;
209 if (lock->l_ast_data == data) {
210 lock->l_ast_data = NULL;
212 LASSERT(osc != NULL);
213 LASSERT(osc->oo_oinfo != NULL);
214 LASSERT(lvb != NULL);
216 /* Updates lvb in lock by the cached oinfo */
217 oinfo = osc->oo_oinfo;
219 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
220 "%llu %llu %llu by oinfo size %llu blocks %llu "
221 "[cma]time %llu %llu %llu", lvb->lvb_size,
222 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
223 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
224 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
225 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
226 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
228 cl_object_attr_lock(&osc->oo_cl);
229 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
230 cl_object_attr_unlock(&osc->oo_cl);
231 ldlm_clear_lvb_cached(lock);
233 RETURN(LDLM_ITER_CONTINUE);
236 int osc_object_prune(const struct lu_env *env, struct cl_object *obj)
238 struct osc_object *osc = cl2osc(obj);
239 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
241 /* DLM locks don't hold a reference of osc_object so we have to
242 * clear it before the object is being destroyed. */
243 osc_build_res_name(osc, resname);
244 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
245 osc_object_ast_clear, osc);
248 EXPORT_SYMBOL(osc_object_prune);
250 static int osc_object_fiemap(const struct lu_env *env, struct cl_object *obj,
251 struct ll_fiemap_info_key *fmkey,
252 struct fiemap *fiemap, size_t *buflen)
254 struct obd_export *exp = osc_export(cl2osc(obj));
255 struct ldlm_res_id resid;
256 union ldlm_policy_data policy;
257 struct lustre_handle lockh;
258 enum ldlm_mode mode = LCK_MINMODE;
259 struct ptlrpc_request *req;
260 struct fiemap *reply;
265 fmkey->lfik_oa.o_oi = cl2osc(obj)->oo_oinfo->loi_oi;
266 if (!(fmkey->lfik_fiemap.fm_flags & FIEMAP_FLAG_SYNC))
269 policy.l_extent.start = fmkey->lfik_fiemap.fm_start & PAGE_MASK;
271 if (OBD_OBJECT_EOF - fmkey->lfik_fiemap.fm_length <=
272 fmkey->lfik_fiemap.fm_start + PAGE_SIZE - 1)
273 policy.l_extent.end = OBD_OBJECT_EOF;
275 policy.l_extent.end = (fmkey->lfik_fiemap.fm_start +
276 fmkey->lfik_fiemap.fm_length +
277 PAGE_SIZE - 1) & PAGE_MASK;
279 ostid_build_res_name(&fmkey->lfik_oa.o_oi, &resid);
280 mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
281 LDLM_FL_BLOCK_GRANTED | LDLM_FL_LVB_READY,
282 &resid, LDLM_EXTENT, &policy,
283 LCK_PR | LCK_PW, &lockh);
284 if (mode) { /* lock is cached on client */
285 if (mode != LCK_PR) {
286 ldlm_lock_addref(&lockh, LCK_PR);
287 ldlm_lock_decref(&lockh, LCK_PW);
289 } else { /* no cached lock, needs acquire lock on server side */
290 fmkey->lfik_oa.o_valid |= OBD_MD_FLFLAGS;
291 fmkey->lfik_oa.o_flags |= OBD_FL_SRVLOCK;
295 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
296 &RQF_OST_GET_INFO_FIEMAP);
298 GOTO(drop_lock, rc = -ENOMEM);
300 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY, RCL_CLIENT,
302 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_CLIENT,
304 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_SERVER,
307 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
309 ptlrpc_request_free(req);
312 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
313 memcpy(tmp, fmkey, sizeof(*fmkey));
314 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
315 memcpy(tmp, fiemap, *buflen);
316 ptlrpc_request_set_replen(req);
318 rc = ptlrpc_queue_wait(req);
322 reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
324 GOTO(fini_req, rc = -EPROTO);
326 memcpy(fiemap, reply, *buflen);
328 ptlrpc_req_finished(req);
331 ldlm_lock_decref(&lockh, LCK_PR);
335 int osc_object_is_contended(struct osc_object *obj)
337 struct osc_device *dev = lu2osc_dev(obj->oo_cl.co_lu.lo_dev);
338 time64_t osc_contention_time = dev->od_contention_time;
341 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_OBJECT_CONTENTION))
344 if (!obj->oo_contended)
348 * I like copy-paste. the code is copied from
349 * ll_file_is_contended.
351 retry_time = ktime_add_ns(obj->oo_contention_time,
352 osc_contention_time * NSEC_PER_SEC);
353 if (ktime_after(ktime_get(), retry_time)) {
354 osc_object_clear_contended(obj);
359 EXPORT_SYMBOL(osc_object_is_contended);
362 * Implementation of struct cl_object_operations::coo_req_attr_set() for osc
363 * layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq
366 static void osc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
367 struct cl_req_attr *attr)
369 struct lov_oinfo *oinfo;
372 u64 flags = attr->cra_flags;
374 oinfo = cl2osc(obj)->oo_oinfo;
375 lvb = &oinfo->loi_lvb;
378 if ((flags & OBD_MD_FLMTIME) != 0) {
379 oa->o_mtime = lvb->lvb_mtime;
380 oa->o_valid |= OBD_MD_FLMTIME;
383 * I don't understand this part, what for OSC resets atime just
384 * set by VVP layer to 0 so that OST gets 0 instead of actual
385 * atime, bzzz. please inspect this place with extra care.
387 if ((flags & OBD_MD_FLATIME) && lvb->lvb_atime > oa->o_atime) {
388 oa->o_atime = lvb->lvb_atime;
389 oa->o_valid |= OBD_MD_FLATIME;
391 if ((flags & OBD_MD_FLCTIME) != 0) {
392 oa->o_ctime = lvb->lvb_ctime;
393 oa->o_valid |= OBD_MD_FLCTIME;
395 if (flags & OBD_MD_FLGROUP) {
396 ostid_set_seq(&oa->o_oi, ostid_seq(&oinfo->loi_oi));
397 oa->o_valid |= OBD_MD_FLGROUP;
399 if (flags & OBD_MD_FLID) {
402 rc = ostid_set_id(&oa->o_oi, ostid_id(&oinfo->loi_oi));
404 CERROR("Bad %llu to set " DOSTID " : rc %d\n",
405 (unsigned long long)ostid_id(&oinfo->loi_oi),
406 POSTID(&oa->o_oi), rc);
408 oa->o_valid |= OBD_MD_FLID;
410 if (flags & OBD_MD_FLHANDLE) {
411 struct ldlm_lock *lock;
412 struct osc_page *opg;
414 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
415 lock = osc_dlmlock_at_pgoff(env, cl2osc(obj), osc_index(opg),
416 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_CANCELING);
417 if (lock == NULL && !opg->ops_srvlock) {
418 struct ldlm_resource *res;
419 struct ldlm_res_id *resname;
421 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
422 "uncovered page!\n");
424 resname = &osc_env_info(env)->oti_resname;
425 ostid_build_res_name(&oinfo->loi_oi, resname);
426 res = ldlm_resource_get(
427 osc_export(cl2osc(obj))->exp_obd->obd_namespace,
428 NULL, resname, LDLM_EXTENT, 0);
429 ldlm_resource_dump(D_ERROR, res);
431 libcfs_debug_dumpstack(NULL);
435 /* check for lockless io. */
437 oa->o_handle = lock->l_remote_handle;
438 oa->o_valid |= OBD_MD_FLHANDLE;
444 static const struct cl_object_operations osc_ops = {
445 .coo_page_init = osc_page_init,
446 .coo_lock_init = osc_lock_init,
447 .coo_io_init = osc_io_init,
448 .coo_attr_get = osc_attr_get,
449 .coo_attr_update = osc_attr_update,
450 .coo_glimpse = osc_object_glimpse,
451 .coo_prune = osc_object_prune,
452 .coo_fiemap = osc_object_fiemap,
453 .coo_req_attr_set = osc_req_attr_set
456 static const struct lu_object_operations osc_lu_obj_ops = {
457 .loo_object_init = osc_object_init,
458 .loo_object_release = NULL,
459 .loo_object_free = osc_object_free,
460 .loo_object_print = osc_object_print,
461 .loo_object_invariant = NULL
464 struct lu_object *osc_object_alloc(const struct lu_env *env,
465 const struct lu_object_header *unused,
466 struct lu_device *dev)
468 struct osc_object *osc;
469 struct lu_object *obj;
471 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
474 lu_object_init(obj, NULL, dev);
475 osc->oo_cl.co_ops = &osc_ops;
476 obj->lo_ops = &osc_lu_obj_ops;
477 osc->oo_obj_ops = &osc_object_ops;
483 int osc_object_invalidate(const struct lu_env *env, struct osc_object *osc)
487 CDEBUG(D_INODE, "Invalidate osc object: %p, # of active IOs: %d\n",
488 osc, atomic_read(&osc->oo_nr_ios));
490 wait_event_idle(osc->oo_io_waitq, atomic_read(&osc->oo_nr_ios) == 0);
492 /* Discard all dirty pages of this object. */
493 osc_cache_truncate_start(env, osc, 0, NULL);
495 /* Discard all caching pages */
496 osc_lock_discard_pages(env, osc, 0, CL_PAGE_EOF, true);
498 /* Clear ast data of dlm lock. Do this after discarding all pages */
499 osc_object_prune(env, osc2cl(osc));
503 EXPORT_SYMBOL(osc_object_invalidate);