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) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Implementation of cl_object for OSC layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
42 #define DEBUG_SUBSYSTEM S_OSC
44 #include "osc_cl_internal.h"
50 /*****************************************************************************
56 static struct lu_object *osc2lu(struct osc_object *osc)
58 return &osc->oo_cl.co_lu;
61 static struct osc_object *lu2osc(const struct lu_object *obj)
63 LINVRNT(osc_is_object(obj));
64 return container_of0(obj, struct osc_object, oo_cl.co_lu);
67 /*****************************************************************************
73 static int osc_object_init(const struct lu_env *env, struct lu_object *obj,
74 const struct lu_object_conf *conf)
76 struct osc_object *osc = lu2osc(obj);
77 const struct cl_object_conf *cconf = lu2cl_conf(conf);
79 osc->oo_oinfo = cconf->u.coc_oinfo;
80 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
81 mutex_init(&osc->oo_debug_mutex);
83 INIT_LIST_HEAD(&osc->oo_ready_item);
84 INIT_LIST_HEAD(&osc->oo_hp_ready_item);
85 INIT_LIST_HEAD(&osc->oo_write_item);
86 INIT_LIST_HEAD(&osc->oo_read_item);
88 osc->oo_root.rb_node = NULL;
89 INIT_LIST_HEAD(&osc->oo_hp_exts);
90 INIT_LIST_HEAD(&osc->oo_urgent_exts);
91 INIT_LIST_HEAD(&osc->oo_rpc_exts);
92 INIT_LIST_HEAD(&osc->oo_reading_exts);
93 atomic_set(&osc->oo_nr_reads, 0);
94 atomic_set(&osc->oo_nr_writes, 0);
95 spin_lock_init(&osc->oo_lock);
96 spin_lock_init(&osc->oo_tree_lock);
97 spin_lock_init(&osc->oo_ol_spin);
98 INIT_LIST_HEAD(&osc->oo_ol_list);
100 atomic_set(&osc->oo_nr_ios, 0);
101 init_waitqueue_head(&osc->oo_io_waitq);
103 cl_object_page_init(lu2cl(obj), sizeof(struct osc_page));
108 static void osc_object_free(const struct lu_env *env, struct lu_object *obj)
110 struct osc_object *osc = lu2osc(obj);
112 LASSERT(list_empty(&osc->oo_ready_item));
113 LASSERT(list_empty(&osc->oo_hp_ready_item));
114 LASSERT(list_empty(&osc->oo_write_item));
115 LASSERT(list_empty(&osc->oo_read_item));
117 LASSERT(osc->oo_root.rb_node == NULL);
118 LASSERT(list_empty(&osc->oo_hp_exts));
119 LASSERT(list_empty(&osc->oo_urgent_exts));
120 LASSERT(list_empty(&osc->oo_rpc_exts));
121 LASSERT(list_empty(&osc->oo_reading_exts));
122 LASSERT(atomic_read(&osc->oo_nr_reads) == 0);
123 LASSERT(atomic_read(&osc->oo_nr_writes) == 0);
124 LASSERT(list_empty(&osc->oo_ol_list));
125 LASSERT(atomic_read(&osc->oo_nr_ios) == 0);
128 OBD_SLAB_FREE_PTR(osc, osc_object_kmem);
131 int osc_lvb_print(const struct lu_env *env, void *cookie,
132 lu_printer_t p, const struct ost_lvb *lvb)
134 return (*p)(env, cookie, "size: "LPU64" mtime: "LPU64" atime: "LPU64" "
135 "ctime: "LPU64" blocks: "LPU64,
136 lvb->lvb_size, lvb->lvb_mtime, lvb->lvb_atime,
137 lvb->lvb_ctime, lvb->lvb_blocks);
140 static int osc_object_print(const struct lu_env *env, void *cookie,
141 lu_printer_t p, const struct lu_object *obj)
143 struct osc_object *osc = lu2osc(obj);
144 struct lov_oinfo *oinfo = osc->oo_oinfo;
145 struct osc_async_rc *ar = &oinfo->loi_ar;
147 (*p)(env, cookie, "id: "DOSTID" "
148 "idx: %d gen: %d kms_valid: %u kms "LPU64" "
149 "rc: %d force_sync: %d min_xid: "LPU64" ",
150 POSTID(&oinfo->loi_oi), oinfo->loi_ost_idx,
151 oinfo->loi_ost_gen, oinfo->loi_kms_valid, oinfo->loi_kms,
152 ar->ar_rc, ar->ar_force_sync, ar->ar_min_xid);
153 osc_lvb_print(env, cookie, p, &oinfo->loi_lvb);
158 static int osc_attr_get(const struct lu_env *env, struct cl_object *obj,
159 struct cl_attr *attr)
161 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
163 cl_lvb2attr(attr, &oinfo->loi_lvb);
164 attr->cat_kms = oinfo->loi_kms_valid ? oinfo->loi_kms : 0;
168 static int osc_attr_update(const struct lu_env *env, struct cl_object *obj,
169 const struct cl_attr *attr, unsigned valid)
171 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
172 struct ost_lvb *lvb = &oinfo->loi_lvb;
174 if (valid & CAT_SIZE)
175 lvb->lvb_size = attr->cat_size;
176 if (valid & CAT_MTIME)
177 lvb->lvb_mtime = attr->cat_mtime;
178 if (valid & CAT_ATIME)
179 lvb->lvb_atime = attr->cat_atime;
180 if (valid & CAT_CTIME)
181 lvb->lvb_ctime = attr->cat_ctime;
182 if (valid & CAT_BLOCKS)
183 lvb->lvb_blocks = attr->cat_blocks;
184 if (valid & CAT_KMS) {
185 CDEBUG(D_CACHE, "set kms from "LPU64"to "LPU64"\n",
186 oinfo->loi_kms, (__u64)attr->cat_kms);
187 loi_kms_set(oinfo, attr->cat_kms);
192 static int osc_object_glimpse(const struct lu_env *env,
193 const struct cl_object *obj, struct ost_lvb *lvb)
195 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
198 lvb->lvb_size = oinfo->loi_kms;
199 lvb->lvb_blocks = oinfo->loi_lvb.lvb_blocks;
203 static int osc_object_ast_clear(struct ldlm_lock *lock, void *data)
207 if (lock->l_ast_data == data)
208 lock->l_ast_data = NULL;
209 RETURN(LDLM_ITER_CONTINUE);
212 static int osc_object_prune(const struct lu_env *env, struct cl_object *obj)
214 struct osc_object *osc = cl2osc(obj);
215 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
217 /* DLM locks don't hold a reference of osc_object so we have to
218 * clear it before the object is being destroyed. */
219 ostid_build_res_name(&osc->oo_oinfo->loi_oi, resname);
220 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
221 osc_object_ast_clear, osc);
225 static int osc_object_fiemap(const struct lu_env *env, struct cl_object *obj,
226 struct ll_fiemap_info_key *fmkey,
227 struct fiemap *fiemap, size_t *buflen)
229 struct obd_export *exp = osc_export(cl2osc(obj));
230 struct ldlm_res_id resid;
231 union ldlm_policy_data policy;
232 struct lustre_handle lockh;
233 enum ldlm_mode mode = LCK_MINMODE;
234 struct ptlrpc_request *req;
235 struct fiemap *reply;
240 fmkey->lfik_oa.o_oi = cl2osc(obj)->oo_oinfo->loi_oi;
241 if (!(fmkey->lfik_fiemap.fm_flags & FIEMAP_FLAG_SYNC))
244 policy.l_extent.start = fmkey->lfik_fiemap.fm_start & PAGE_CACHE_MASK;
246 if (OBD_OBJECT_EOF - fmkey->lfik_fiemap.fm_length <=
247 fmkey->lfik_fiemap.fm_start + PAGE_CACHE_SIZE - 1)
248 policy.l_extent.end = OBD_OBJECT_EOF;
250 policy.l_extent.end = (fmkey->lfik_fiemap.fm_start +
251 fmkey->lfik_fiemap.fm_length +
252 PAGE_CACHE_SIZE - 1) & PAGE_CACHE_MASK;
254 ostid_build_res_name(&fmkey->lfik_oa.o_oi, &resid);
255 mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
256 LDLM_FL_BLOCK_GRANTED | LDLM_FL_LVB_READY,
257 &resid, LDLM_EXTENT, &policy,
258 LCK_PR | LCK_PW, &lockh, 0);
259 if (mode) { /* lock is cached on client */
260 if (mode != LCK_PR) {
261 ldlm_lock_addref(&lockh, LCK_PR);
262 ldlm_lock_decref(&lockh, LCK_PW);
264 } else { /* no cached lock, needs acquire lock on server side */
265 fmkey->lfik_oa.o_valid |= OBD_MD_FLFLAGS;
266 fmkey->lfik_oa.o_flags |= OBD_FL_SRVLOCK;
270 req = ptlrpc_request_alloc(class_exp2cliimp(exp),
271 &RQF_OST_GET_INFO_FIEMAP);
273 GOTO(drop_lock, rc = -ENOMEM);
275 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY, RCL_CLIENT,
277 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_CLIENT,
279 req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_VAL, RCL_SERVER,
282 rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
284 ptlrpc_request_free(req);
287 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
288 memcpy(tmp, fmkey, sizeof(*fmkey));
289 tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_VAL);
290 memcpy(tmp, fiemap, *buflen);
291 ptlrpc_request_set_replen(req);
293 rc = ptlrpc_queue_wait(req);
297 reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
299 GOTO(fini_req, rc = -EPROTO);
301 memcpy(fiemap, reply, *buflen);
303 ptlrpc_req_finished(req);
306 ldlm_lock_decref(&lockh, LCK_PR);
310 void osc_object_set_contended(struct osc_object *obj)
312 obj->oo_contention_time = cfs_time_current();
314 obj->oo_contended = 1;
317 void osc_object_clear_contended(struct osc_object *obj)
319 obj->oo_contended = 0;
322 int osc_object_is_contended(struct osc_object *obj)
324 struct osc_device *dev = lu2osc_dev(obj->oo_cl.co_lu.lo_dev);
325 int osc_contention_time = dev->od_contention_time;
326 cfs_time_t cur_time = cfs_time_current();
327 cfs_time_t retry_time;
329 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_OBJECT_CONTENTION))
332 if (!obj->oo_contended)
336 * I like copy-paste. the code is copied from
337 * ll_file_is_contended.
339 retry_time = cfs_time_add(obj->oo_contention_time,
340 cfs_time_seconds(osc_contention_time));
341 if (cfs_time_after(cur_time, retry_time)) {
342 osc_object_clear_contended(obj);
349 * Implementation of struct cl_object_operations::coo_req_attr_set() for osc
350 * layer. osc is responsible for struct obdo::o_id and struct obdo::o_seq
353 static void osc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
354 struct cl_req_attr *attr)
356 struct lov_oinfo *oinfo;
359 u64 flags = attr->cra_flags;
361 oinfo = cl2osc(obj)->oo_oinfo;
362 lvb = &oinfo->loi_lvb;
365 if ((flags & OBD_MD_FLMTIME) != 0) {
366 oa->o_mtime = lvb->lvb_mtime;
367 oa->o_valid |= OBD_MD_FLMTIME;
369 if ((flags & OBD_MD_FLATIME) != 0) {
370 oa->o_atime = lvb->lvb_atime;
371 oa->o_valid |= OBD_MD_FLATIME;
373 if ((flags & OBD_MD_FLCTIME) != 0) {
374 oa->o_ctime = lvb->lvb_ctime;
375 oa->o_valid |= OBD_MD_FLCTIME;
377 if (flags & OBD_MD_FLGROUP) {
378 ostid_set_seq(&oa->o_oi, ostid_seq(&oinfo->loi_oi));
379 oa->o_valid |= OBD_MD_FLGROUP;
381 if (flags & OBD_MD_FLID) {
382 ostid_set_id(&oa->o_oi, ostid_id(&oinfo->loi_oi));
383 oa->o_valid |= OBD_MD_FLID;
385 if (flags & OBD_MD_FLHANDLE) {
386 struct ldlm_lock *lock;
387 struct osc_page *opg;
389 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
390 lock = osc_dlmlock_at_pgoff(env, cl2osc(obj), osc_index(opg),
391 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_CANCELING);
392 if (lock == NULL && !opg->ops_srvlock) {
393 struct ldlm_resource *res;
394 struct ldlm_res_id *resname;
396 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
397 "uncovered page!\n");
399 resname = &osc_env_info(env)->oti_resname;
400 ostid_build_res_name(&oinfo->loi_oi, resname);
401 res = ldlm_resource_get(
402 osc_export(cl2osc(obj))->exp_obd->obd_namespace,
403 NULL, resname, LDLM_EXTENT, 0);
404 ldlm_resource_dump(D_ERROR, res);
406 libcfs_debug_dumpstack(NULL);
410 /* check for lockless io. */
412 oa->o_handle = lock->l_remote_handle;
413 oa->o_valid |= OBD_MD_FLHANDLE;
419 static const struct cl_object_operations osc_ops = {
420 .coo_page_init = osc_page_init,
421 .coo_lock_init = osc_lock_init,
422 .coo_io_init = osc_io_init,
423 .coo_attr_get = osc_attr_get,
424 .coo_attr_update = osc_attr_update,
425 .coo_glimpse = osc_object_glimpse,
426 .coo_prune = osc_object_prune,
427 .coo_fiemap = osc_object_fiemap,
428 .coo_req_attr_set = osc_req_attr_set
431 static const struct lu_object_operations osc_lu_obj_ops = {
432 .loo_object_init = osc_object_init,
433 .loo_object_release = NULL,
434 .loo_object_free = osc_object_free,
435 .loo_object_print = osc_object_print,
436 .loo_object_invariant = NULL
439 struct lu_object *osc_object_alloc(const struct lu_env *env,
440 const struct lu_object_header *unused,
441 struct lu_device *dev)
443 struct osc_object *osc;
444 struct lu_object *obj;
446 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
449 lu_object_init(obj, NULL, dev);
450 osc->oo_cl.co_ops = &osc_ops;
451 obj->lo_ops = &osc_lu_obj_ops;
457 int osc_object_invalidate(const struct lu_env *env, struct osc_object *osc)
459 struct l_wait_info lwi = { 0 };
462 CDEBUG(D_INODE, "Invalidate osc object: %p, # of active IOs: %d\n",
463 osc, atomic_read(&osc->oo_nr_ios));
465 l_wait_event(osc->oo_io_waitq, atomic_read(&osc->oo_nr_ios) == 0, &lwi);
467 /* Discard all dirty pages of this object. */
468 osc_cache_truncate_start(env, osc, 0, NULL);
470 /* Discard all caching pages */
471 osc_lock_discard_pages(env, osc, 0, CL_PAGE_EOF, CLM_WRITE);
473 /* Clear ast data of dlm lock. Do this after discarding all pages */
474 osc_object_prune(env, osc2cl(osc));