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) 2017, DDN Storage Corporation.
26 * Persistent Client Cache
28 * PCC is a new framework which provides a group of local cache on Lustre
29 * client side. It works in two modes: RW-PCC enables a read-write cache on the
30 * local SSDs of a single client; RO-PCC provides a read-only cache on the
31 * local SSDs of multiple clients. Less overhead is visible to the applications
32 * and network latencies and lock conflicts can be significantly reduced.
34 * For RW-PCC, no global namespace will be provided. Each client uses its own
35 * local storage as a cache for itself. Local file system is used to manage
36 * the data on local caches. Cached I/O is directed to local file system while
37 * normal I/O is directed to OSTs. RW-PCC uses HSM for data synchronization.
38 * It uses HSM copytool to restore file from local caches to Lustre OSTs. Each
39 * PCC has a copytool instance running with unique archive number. Any remote
40 * access from another Lustre client would trigger the data synchronization. If
41 * a client with RW-PCC goes offline, the cached data becomes inaccessible for
42 * other client temporarily. And after the RW-PCC client reboots and the
43 * copytool restarts, the data will be accessible again.
45 * Following is what will happen in different conditions for RW-PCC:
47 * > When file is being created on RW-PCC
49 * A normal HSM released file is created on MDT;
50 * An empty mirror file is created on local cache;
51 * The HSM status of the Lustre file will be set to archived and released;
52 * The archive number will be set to the proper value.
54 * > When file is being prefetched to RW-PCC
56 * An file is copied to the local cache;
57 * The HSM status of the Lustre file will be set to archived and released;
58 * The archive number will be set to the proper value.
60 * > When file is being accessed from PCC
62 * Data will be read directly from local cache;
63 * Metadata will be read from MDT, except file size;
64 * File size will be got from local cache.
66 * > When PCC cached file is being accessed on another client
68 * RW-PCC cached files are automatically restored when a process on another
69 * client tries to read or modify them. The corresponding I/O will block
70 * waiting for the released file to be restored. This is transparent to the
73 * For RW-PCC, when a file is being created, a rule-based policy is used to
74 * determine whether it will be cached. Rule-based caching of newly created
75 * files can determine which file can use a cache on PCC directly without any
78 * RW-PCC design can accelerate I/O intensive applications with one-to-one
79 * mappings between files and accessing clients. However, in several use cases,
80 * files will never be updated, but need to be read simultaneously from many
81 * clients. RO-PCC implements a read-only caching on Lustre clients using
82 * SSDs. RO-PCC is based on the same framework as RW-PCC, expect
83 * that no HSM mechanism is used.
85 * The main advantages to use this SSD cache on the Lustre clients via PCC
87 * - The I/O stack becomes much simpler for the cached data, as there is no
88 * interference with I/Os from other clients, which enables easier
89 * performance optimizations;
90 * - The requirements on the HW inside the client nodes are small, any kind of
91 * SSDs or even HDDs can be used as cache devices;
92 * - Caching reduces the pressure on the object storage targets (OSTs), as
93 * small or random I/Os can be regularized to big sequential I/Os and
94 * temporary files do not even need to be flushed to OSTs.
96 * PCC can accelerate applications with certain I/O patterns:
97 * - small-sized random writes (< 1MB) from a single client
98 * - repeated read of data that is larger than RAM
99 * - clients with high network latency
101 * Author: Li Xi <lixi@ddn.com>
102 * Author: Qian Yingjin <qian@ddn.com>
105 #define DEBUG_SUBSYSTEM S_LLITE
108 #include <linux/namei.h>
109 #include <linux/file.h>
110 #include <lustre_compat.h>
111 #include "llite_internal.h"
113 struct kmem_cache *pcc_inode_slab;
115 int pcc_super_init(struct pcc_super *super)
119 super->pccs_cred = cred = prepare_creds();
123 /* Never override disk quota limits or use reserved space */
124 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
125 init_rwsem(&super->pccs_rw_sem);
126 INIT_LIST_HEAD(&super->pccs_datasets);
127 super->pccs_generation = 1;
132 /* Rule based auto caching */
133 static void pcc_id_list_free(struct list_head *id_list)
135 struct pcc_match_id *id, *n;
137 list_for_each_entry_safe(id, n, id_list, pmi_linkage) {
138 list_del_init(&id->pmi_linkage);
143 static void pcc_fname_list_free(struct list_head *fname_list)
145 struct pcc_match_fname *fname, *n;
147 list_for_each_entry_safe(fname, n, fname_list, pmf_linkage) {
148 OBD_FREE(fname->pmf_name, strlen(fname->pmf_name) + 1);
149 list_del_init(&fname->pmf_linkage);
154 static void pcc_expression_free(struct pcc_expression *expr)
156 LASSERT(expr->pe_field >= PCC_FIELD_UID &&
157 expr->pe_field < PCC_FIELD_MAX);
158 switch (expr->pe_field) {
161 case PCC_FIELD_PROJID:
162 pcc_id_list_free(&expr->pe_cond);
164 case PCC_FIELD_FNAME:
165 pcc_fname_list_free(&expr->pe_cond);
173 static void pcc_conjunction_free(struct pcc_conjunction *conjunction)
175 struct pcc_expression *expression, *n;
177 LASSERT(list_empty(&conjunction->pc_linkage));
178 list_for_each_entry_safe(expression, n,
179 &conjunction->pc_expressions,
181 list_del_init(&expression->pe_linkage);
182 pcc_expression_free(expression);
184 OBD_FREE_PTR(conjunction);
187 static void pcc_rule_conds_free(struct list_head *cond_list)
189 struct pcc_conjunction *conjunction, *n;
191 list_for_each_entry_safe(conjunction, n, cond_list, pc_linkage) {
192 list_del_init(&conjunction->pc_linkage);
193 pcc_conjunction_free(conjunction);
197 static void pcc_cmd_fini(struct pcc_cmd *cmd)
199 if (cmd->pccc_cmd == PCC_ADD_DATASET) {
200 if (!list_empty(&cmd->u.pccc_add.pccc_conds))
201 pcc_rule_conds_free(&cmd->u.pccc_add.pccc_conds);
202 if (cmd->u.pccc_add.pccc_conds_str)
203 OBD_FREE(cmd->u.pccc_add.pccc_conds_str,
204 strlen(cmd->u.pccc_add.pccc_conds_str) + 1);
208 #define PCC_DISJUNCTION_DELIM (',')
209 #define PCC_CONJUNCTION_DELIM ('&')
210 #define PCC_EXPRESSION_DELIM ('=')
213 pcc_fname_list_add(struct cfs_lstr *id, struct list_head *fname_list)
215 struct pcc_match_fname *fname;
217 OBD_ALLOC_PTR(fname);
221 OBD_ALLOC(fname->pmf_name, id->ls_len + 1);
222 if (fname->pmf_name == NULL) {
227 memcpy(fname->pmf_name, id->ls_str, id->ls_len);
228 list_add_tail(&fname->pmf_linkage, fname_list);
233 pcc_fname_list_parse(char *str, int len, struct list_head *fname_list)
243 INIT_LIST_HEAD(fname_list);
245 rc = cfs_gettok(&src, ' ', &res);
250 rc = pcc_fname_list_add(&res, fname_list);
255 pcc_fname_list_free(fname_list);
260 pcc_id_list_parse(char *str, int len, struct list_head *id_list,
269 if (type != PCC_FIELD_UID && type != PCC_FIELD_GID &&
270 type != PCC_FIELD_PROJID)
275 INIT_LIST_HEAD(id_list);
277 struct pcc_match_id *id;
280 if (cfs_gettok(&src, ' ', &res) == 0)
281 GOTO(out, rc = -EINVAL);
283 if (!cfs_str2num_check(res.ls_str, res.ls_len,
284 &id_val, 0, (u32)~0U))
285 GOTO(out, rc = -EINVAL);
289 GOTO(out, rc = -ENOMEM);
292 list_add_tail(&id->pmi_linkage, id_list);
296 pcc_id_list_free(id_list);
301 pcc_check_field(struct cfs_lstr *field, char *str)
303 int len = strlen(str);
305 return (field->ls_len == len &&
306 strncmp(field->ls_str, str, len) == 0);
310 pcc_expression_parse(struct cfs_lstr *src, struct list_head *cond_list)
312 struct pcc_expression *expr;
313 struct cfs_lstr field;
320 rc = cfs_gettok(src, PCC_EXPRESSION_DELIM, &field);
321 if (rc == 0 || src->ls_len <= 2 || src->ls_str[0] != '{' ||
322 src->ls_str[src->ls_len - 1] != '}')
323 GOTO(out, rc = -EINVAL);
325 /* Skip '{' and '}' */
329 if (pcc_check_field(&field, "uid")) {
330 if (pcc_id_list_parse(src->ls_str,
334 GOTO(out, rc = -EINVAL);
335 expr->pe_field = PCC_FIELD_UID;
336 } else if (pcc_check_field(&field, "gid")) {
337 if (pcc_id_list_parse(src->ls_str,
341 GOTO(out, rc = -EINVAL);
342 expr->pe_field = PCC_FIELD_GID;
343 } else if (pcc_check_field(&field, "projid")) {
344 if (pcc_id_list_parse(src->ls_str,
347 PCC_FIELD_PROJID) < 0)
348 GOTO(out, rc = -EINVAL);
349 expr->pe_field = PCC_FIELD_PROJID;
350 } else if (pcc_check_field(&field, "fname")) {
351 if (pcc_fname_list_parse(src->ls_str,
354 GOTO(out, rc = -EINVAL);
355 expr->pe_field = PCC_FIELD_FNAME;
357 GOTO(out, rc = -EINVAL);
360 list_add_tail(&expr->pe_linkage, cond_list);
368 pcc_conjunction_parse(struct cfs_lstr *src, struct list_head *cond_list)
370 struct pcc_conjunction *conjunction;
371 struct cfs_lstr expr;
374 OBD_ALLOC_PTR(conjunction);
375 if (conjunction == NULL)
378 INIT_LIST_HEAD(&conjunction->pc_expressions);
379 list_add_tail(&conjunction->pc_linkage, cond_list);
381 while (src->ls_str) {
382 rc = cfs_gettok(src, PCC_CONJUNCTION_DELIM, &expr);
387 rc = pcc_expression_parse(&expr,
388 &conjunction->pc_expressions);
395 static int pcc_conds_parse(char *str, int len, struct list_head *cond_list)
403 INIT_LIST_HEAD(cond_list);
405 rc = cfs_gettok(&src, PCC_DISJUNCTION_DELIM, &res);
410 rc = pcc_conjunction_parse(&res, cond_list);
417 static int pcc_id_parse(struct pcc_cmd *cmd, const char *id)
421 OBD_ALLOC(cmd->u.pccc_add.pccc_conds_str, strlen(id) + 1);
422 if (cmd->u.pccc_add.pccc_conds_str == NULL)
425 memcpy(cmd->u.pccc_add.pccc_conds_str, id, strlen(id));
427 rc = pcc_conds_parse(cmd->u.pccc_add.pccc_conds_str,
428 strlen(cmd->u.pccc_add.pccc_conds_str),
429 &cmd->u.pccc_add.pccc_conds);
437 pcc_parse_value_pair(struct pcc_cmd *cmd, char *buffer)
444 key = strsep(&val, "=");
445 if (val == NULL || strlen(val) == 0)
448 /* Key of the value pair */
449 if (strcmp(key, "rwid") == 0) {
450 rc = kstrtoul(val, 10, &id);
455 cmd->u.pccc_add.pccc_rwid = id;
456 } else if (strcmp(key, "roid") == 0) {
457 rc = kstrtoul(val, 10, &id);
462 cmd->u.pccc_add.pccc_roid = id;
463 } else if (strcmp(key, "auto_attach") == 0) {
464 rc = kstrtoul(val, 10, &id);
468 cmd->u.pccc_add.pccc_flags &= ~PCC_DATASET_AUTO_ATTACH;
469 } else if (strcmp(key, "open_attach") == 0) {
470 rc = kstrtoul(val, 10, &id);
474 cmd->u.pccc_add.pccc_flags &= ~PCC_DATASET_OPEN_ATTACH;
475 } else if (strcmp(key, "io_attach") == 0) {
476 rc = kstrtoul(val, 10, &id);
480 cmd->u.pccc_add.pccc_flags &= ~PCC_DATASET_IO_ATTACH;
481 } else if (strcmp(key, "stat_attach") == 0) {
482 rc = kstrtoul(val, 10, &id);
486 cmd->u.pccc_add.pccc_flags &= ~PCC_DATASET_STAT_ATTACH;
487 } else if (strcmp(key, "rwpcc") == 0) {
488 rc = kstrtoul(val, 10, &id);
492 cmd->u.pccc_add.pccc_flags |= PCC_DATASET_RWPCC;
493 } else if (strcmp(key, "ropcc") == 0) {
494 rc = kstrtoul(val, 10, &id);
498 cmd->u.pccc_add.pccc_flags |= PCC_DATASET_ROPCC;
507 pcc_parse_value_pairs(struct pcc_cmd *cmd, char *buffer)
513 switch (cmd->pccc_cmd) {
514 case PCC_ADD_DATASET:
515 /* Enable auto attach by default */
516 cmd->u.pccc_add.pccc_flags |= PCC_DATASET_AUTO_ATTACH;
518 case PCC_DEL_DATASET:
526 while (val != NULL && strlen(val) != 0) {
527 token = strsep(&val, " ");
528 rc = pcc_parse_value_pair(cmd, token);
533 switch (cmd->pccc_cmd) {
534 case PCC_ADD_DATASET:
535 if (cmd->u.pccc_add.pccc_flags & PCC_DATASET_RWPCC &&
536 cmd->u.pccc_add.pccc_flags & PCC_DATASET_ROPCC)
539 * By default, a PCC backend can provide caching service for
540 * both RW-PCC and RO-PCC.
542 if ((cmd->u.pccc_add.pccc_flags & PCC_DATASET_PCC_ALL) == 0)
543 cmd->u.pccc_add.pccc_flags |= PCC_DATASET_PCC_ALL;
545 /* For RW-PCC, the value of @rwid must be non zero. */
546 if (cmd->u.pccc_add.pccc_flags & PCC_DATASET_RWPCC &&
547 cmd->u.pccc_add.pccc_rwid == 0)
551 case PCC_DEL_DATASET:
561 pcc_dataset_rule_fini(struct pcc_match_rule *rule)
563 if (!list_empty(&rule->pmr_conds))
564 pcc_rule_conds_free(&rule->pmr_conds);
565 LASSERT(rule->pmr_conds_str != NULL);
566 OBD_FREE(rule->pmr_conds_str, strlen(rule->pmr_conds_str) + 1);
570 pcc_dataset_rule_init(struct pcc_match_rule *rule, struct pcc_cmd *cmd)
574 LASSERT(cmd->u.pccc_add.pccc_conds_str);
575 OBD_ALLOC(rule->pmr_conds_str,
576 strlen(cmd->u.pccc_add.pccc_conds_str) + 1);
577 if (rule->pmr_conds_str == NULL)
580 memcpy(rule->pmr_conds_str,
581 cmd->u.pccc_add.pccc_conds_str,
582 strlen(cmd->u.pccc_add.pccc_conds_str));
584 INIT_LIST_HEAD(&rule->pmr_conds);
585 if (!list_empty(&cmd->u.pccc_add.pccc_conds))
586 rc = pcc_conds_parse(rule->pmr_conds_str,
587 strlen(rule->pmr_conds_str),
591 pcc_dataset_rule_fini(rule);
598 pcc_id_list_match(struct list_head *id_list, __u32 id_val)
600 struct pcc_match_id *id;
602 list_for_each_entry(id, id_list, pmi_linkage) {
603 if (id->pmi_id == id_val)
610 cfs_match_wildcard(const char *pattern, const char *content)
612 if (*pattern == '\0' && *content == '\0')
615 if (*pattern == '*' && *(pattern + 1) != '\0' && *content == '\0')
618 while (*pattern == *content) {
621 if (*pattern == '\0' && *content == '\0')
624 if (*pattern == '*' && *(pattern + 1) != '\0' &&
630 return (cfs_match_wildcard(pattern + 1, content) ||
631 cfs_match_wildcard(pattern, content + 1));
637 pcc_fname_list_match(struct list_head *fname_list, const char *name)
639 struct pcc_match_fname *fname;
641 list_for_each_entry(fname, fname_list, pmf_linkage) {
642 if (cfs_match_wildcard(fname->pmf_name, name))
649 pcc_expression_match(struct pcc_expression *expr, struct pcc_matcher *matcher)
651 switch (expr->pe_field) {
653 return pcc_id_list_match(&expr->pe_cond, matcher->pm_uid);
655 return pcc_id_list_match(&expr->pe_cond, matcher->pm_gid);
656 case PCC_FIELD_PROJID:
657 return pcc_id_list_match(&expr->pe_cond, matcher->pm_projid);
658 case PCC_FIELD_FNAME:
659 return pcc_fname_list_match(&expr->pe_cond,
660 matcher->pm_name->name);
667 pcc_conjunction_match(struct pcc_conjunction *conjunction,
668 struct pcc_matcher *matcher)
670 struct pcc_expression *expr;
673 list_for_each_entry(expr, &conjunction->pc_expressions, pe_linkage) {
674 matched = pcc_expression_match(expr, matcher);
683 pcc_cond_match(struct pcc_match_rule *rule, struct pcc_matcher *matcher)
685 struct pcc_conjunction *conjunction;
688 list_for_each_entry(conjunction, &rule->pmr_conds, pc_linkage) {
689 matched = pcc_conjunction_match(conjunction, matcher);
698 pcc_dataset_match_get(struct pcc_super *super, struct pcc_matcher *matcher)
700 struct pcc_dataset *dataset;
701 struct pcc_dataset *selected = NULL;
703 down_read(&super->pccs_rw_sem);
704 list_for_each_entry(dataset, &super->pccs_datasets, pccd_linkage) {
705 if (!(dataset->pccd_flags & PCC_DATASET_RWPCC))
708 if (pcc_cond_match(&dataset->pccd_rule, matcher)) {
709 atomic_inc(&dataset->pccd_refcount);
714 up_read(&super->pccs_rw_sem);
716 CDEBUG(D_CACHE, "PCC create, matched %s - %d:%d:%d:%s\n",
717 dataset->pccd_rule.pmr_conds_str,
718 matcher->pm_uid, matcher->pm_gid,
719 matcher->pm_projid, matcher->pm_name->name);
725 * pcc_dataset_add - Add a Cache policy to control which files need be
726 * cached and where it will be cached.
728 * @super: superblock of pcc
732 pcc_dataset_add(struct pcc_super *super, struct pcc_cmd *cmd)
734 char *pathname = cmd->pccc_pathname;
735 struct pcc_dataset *dataset;
736 struct pcc_dataset *tmp;
740 OBD_ALLOC_PTR(dataset);
744 rc = kern_path(pathname, LOOKUP_DIRECTORY, &dataset->pccd_path);
746 OBD_FREE_PTR(dataset);
749 strncpy(dataset->pccd_pathname, pathname, PATH_MAX);
750 dataset->pccd_rwid = cmd->u.pccc_add.pccc_rwid;
751 dataset->pccd_roid = cmd->u.pccc_add.pccc_roid;
752 dataset->pccd_flags = cmd->u.pccc_add.pccc_flags;
753 atomic_set(&dataset->pccd_refcount, 1);
755 rc = pcc_dataset_rule_init(&dataset->pccd_rule, cmd);
757 pcc_dataset_put(dataset);
761 down_write(&super->pccs_rw_sem);
762 list_for_each_entry(tmp, &super->pccs_datasets, pccd_linkage) {
763 if (strcmp(tmp->pccd_pathname, pathname) == 0 ||
764 (dataset->pccd_rwid != 0 &&
765 dataset->pccd_rwid == tmp->pccd_rwid) ||
766 (dataset->pccd_roid != 0 &&
767 dataset->pccd_roid == tmp->pccd_roid)) {
773 list_add(&dataset->pccd_linkage, &super->pccs_datasets);
774 up_write(&super->pccs_rw_sem);
777 pcc_dataset_put(dataset);
785 pcc_dataset_get(struct pcc_super *super, enum lu_pcc_type type, __u32 id)
787 struct pcc_dataset *dataset;
788 struct pcc_dataset *selected = NULL;
794 * archive ID (read-write ID) or read-only ID is unique in the list,
795 * we just return last added one as first priority.
797 down_read(&super->pccs_rw_sem);
798 list_for_each_entry(dataset, &super->pccs_datasets, pccd_linkage) {
799 if (type == LU_PCC_READWRITE && (dataset->pccd_rwid != id ||
800 !(dataset->pccd_flags & PCC_DATASET_RWPCC)))
802 atomic_inc(&dataset->pccd_refcount);
806 up_read(&super->pccs_rw_sem);
808 CDEBUG(D_CACHE, "matched id %u, PCC mode %d\n", id, type);
814 pcc_dataset_put(struct pcc_dataset *dataset)
816 if (atomic_dec_and_test(&dataset->pccd_refcount)) {
817 pcc_dataset_rule_fini(&dataset->pccd_rule);
818 path_put(&dataset->pccd_path);
819 OBD_FREE_PTR(dataset);
824 pcc_dataset_del(struct pcc_super *super, char *pathname)
826 struct list_head *l, *tmp;
827 struct pcc_dataset *dataset;
830 down_write(&super->pccs_rw_sem);
831 list_for_each_safe(l, tmp, &super->pccs_datasets) {
832 dataset = list_entry(l, struct pcc_dataset, pccd_linkage);
833 if (strcmp(dataset->pccd_pathname, pathname) == 0) {
834 list_del_init(&dataset->pccd_linkage);
835 pcc_dataset_put(dataset);
836 super->pccs_generation++;
841 up_write(&super->pccs_rw_sem);
846 pcc_dataset_dump(struct pcc_dataset *dataset, struct seq_file *m)
848 seq_printf(m, "%s:\n", dataset->pccd_pathname);
849 seq_printf(m, " rwid: %u\n", dataset->pccd_rwid);
850 seq_printf(m, " flags: %x\n", dataset->pccd_flags);
851 seq_printf(m, " autocache: %s\n", dataset->pccd_rule.pmr_conds_str);
855 pcc_super_dump(struct pcc_super *super, struct seq_file *m)
857 struct pcc_dataset *dataset;
859 down_read(&super->pccs_rw_sem);
860 list_for_each_entry(dataset, &super->pccs_datasets, pccd_linkage) {
861 pcc_dataset_dump(dataset, m);
863 up_read(&super->pccs_rw_sem);
867 static void pcc_remove_datasets(struct pcc_super *super)
869 struct pcc_dataset *dataset, *tmp;
871 down_write(&super->pccs_rw_sem);
872 list_for_each_entry_safe(dataset, tmp,
873 &super->pccs_datasets, pccd_linkage) {
874 list_del(&dataset->pccd_linkage);
875 pcc_dataset_put(dataset);
877 super->pccs_generation++;
878 up_write(&super->pccs_rw_sem);
881 void pcc_super_fini(struct pcc_super *super)
883 pcc_remove_datasets(super);
884 put_cred(super->pccs_cred);
887 static bool pathname_is_valid(const char *pathname)
889 /* Needs to be absolute path */
890 if (pathname == NULL || strlen(pathname) == 0 ||
891 strlen(pathname) >= PATH_MAX || pathname[0] != '/')
896 static struct pcc_cmd *
897 pcc_cmd_parse(char *buffer, unsigned long count)
899 static struct pcc_cmd *cmd;
906 GOTO(out, rc = -ENOMEM);
908 /* clear all setting */
909 if (strncmp(buffer, "clear", 5) == 0) {
910 cmd->pccc_cmd = PCC_CLEAR_ALL;
915 token = strsep(&val, " ");
916 if (val == NULL || strlen(val) == 0)
917 GOTO(out_free_cmd, rc = -EINVAL);
919 /* Type of the command */
920 if (strcmp(token, "add") == 0)
921 cmd->pccc_cmd = PCC_ADD_DATASET;
922 else if (strcmp(token, "del") == 0)
923 cmd->pccc_cmd = PCC_DEL_DATASET;
925 GOTO(out_free_cmd, rc = -EINVAL);
927 /* Pathname of the dataset */
928 token = strsep(&val, " ");
929 if ((val == NULL && cmd->pccc_cmd != PCC_DEL_DATASET) ||
930 !pathname_is_valid(token))
931 GOTO(out_free_cmd, rc = -EINVAL);
932 cmd->pccc_pathname = token;
934 if (cmd->pccc_cmd == PCC_ADD_DATASET) {
938 val = strrchr(token, '}');
940 GOTO(out_free_cmd, rc = -EINVAL);
946 } else if (*val == ' ') {
950 GOTO(out_free_cmd, rc = -EINVAL);
953 rc = pcc_id_parse(cmd, token);
955 GOTO(out_free_cmd, rc);
957 rc = pcc_parse_value_pairs(cmd, val);
959 GOTO(out_cmd_fini, rc = -EINVAL);
972 int pcc_cmd_handle(char *buffer, unsigned long count,
973 struct pcc_super *super)
978 cmd = pcc_cmd_parse(buffer, count);
982 switch (cmd->pccc_cmd) {
983 case PCC_ADD_DATASET:
984 rc = pcc_dataset_add(super, cmd);
986 case PCC_DEL_DATASET:
987 rc = pcc_dataset_del(super, cmd->pccc_pathname);
990 pcc_remove_datasets(super);
1002 static inline void pcc_inode_lock(struct inode *inode)
1004 mutex_lock(&ll_i2info(inode)->lli_pcc_lock);
1007 static inline void pcc_inode_unlock(struct inode *inode)
1009 mutex_unlock(&ll_i2info(inode)->lli_pcc_lock);
1012 static void pcc_inode_init(struct pcc_inode *pcci, struct ll_inode_info *lli)
1014 pcci->pcci_lli = lli;
1015 lli->lli_pcc_inode = pcci;
1016 atomic_set(&pcci->pcci_refcount, 0);
1017 pcci->pcci_type = LU_PCC_NONE;
1018 pcci->pcci_layout_gen = CL_LAYOUT_GEN_NONE;
1019 atomic_set(&pcci->pcci_active_ios, 0);
1020 init_waitqueue_head(&pcci->pcci_waitq);
1023 static void pcc_inode_fini(struct pcc_inode *pcci)
1025 struct ll_inode_info *lli = pcci->pcci_lli;
1027 path_put(&pcci->pcci_path);
1028 pcci->pcci_type = LU_PCC_NONE;
1029 OBD_SLAB_FREE_PTR(pcci, pcc_inode_slab);
1030 lli->lli_pcc_inode = NULL;
1033 static void pcc_inode_get(struct pcc_inode *pcci)
1035 atomic_inc(&pcci->pcci_refcount);
1038 static void pcc_inode_put(struct pcc_inode *pcci)
1040 if (atomic_dec_and_test(&pcci->pcci_refcount))
1041 pcc_inode_fini(pcci);
1044 void pcc_inode_free(struct inode *inode)
1046 struct pcc_inode *pcci = ll_i2pcci(inode);
1049 WARN_ON(atomic_read(&pcci->pcci_refcount) > 1);
1050 pcc_inode_put(pcci);
1056 * As Andreas suggested, we'd better use new layout to
1058 * (fid->f_oid >> 16 & oxFFFF)/FID
1060 #define MAX_PCC_DATABASE_PATH (6 * 5 + FID_NOBRACE_LEN + 1)
1061 static int pcc_fid2dataset_path(char *buf, int sz, struct lu_fid *fid)
1063 return snprintf(buf, sz, "%04x/%04x/%04x/%04x/%04x/%04x/"
1065 (fid)->f_oid & 0xFFFF,
1066 (fid)->f_oid >> 16 & 0xFFFF,
1067 (unsigned int)((fid)->f_seq & 0xFFFF),
1068 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
1069 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
1070 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
1074 static inline const struct cred *pcc_super_cred(struct super_block *sb)
1076 return ll_s2sbi(sb)->ll_pcc_super.pccs_cred;
1079 void pcc_file_init(struct pcc_file *pccf)
1081 pccf->pccf_file = NULL;
1082 pccf->pccf_type = LU_PCC_NONE;
1085 static inline bool pcc_auto_attach_enabled(enum pcc_dataset_flags flags,
1086 enum pcc_io_type iot)
1088 if (iot == PIT_OPEN)
1089 return flags & PCC_DATASET_OPEN_ATTACH;
1090 if (iot == PIT_GETATTR)
1091 return flags & PCC_DATASET_STAT_ATTACH;
1093 return flags & PCC_DATASET_AUTO_ATTACH;
1096 static const char pcc_xattr_layout[] = XATTR_USER_PREFIX "PCC.layout";
1098 static int pcc_layout_xattr_set(struct pcc_inode *pcci, __u32 gen)
1100 struct dentry *pcc_dentry = pcci->pcci_path.dentry;
1101 struct ll_inode_info *lli = pcci->pcci_lli;
1106 if (!(lli->lli_pcc_dsflags & PCC_DATASET_AUTO_ATTACH))
1109 #ifndef HAVE_VFS_SETXATTR
1110 if (!pcc_dentry->d_inode->i_op->setxattr)
1113 rc = pcc_dentry->d_inode->i_op->setxattr(pcc_dentry, pcc_xattr_layout,
1114 &gen, sizeof(gen), 0);
1116 rc = __vfs_setxattr(pcc_dentry, pcc_dentry->d_inode, pcc_xattr_layout,
1117 &gen, sizeof(gen), 0);
1122 static int pcc_get_layout_info(struct inode *inode, struct cl_layout *clt)
1125 struct ll_inode_info *lli = ll_i2info(inode);
1134 env = cl_env_get(&refcheck);
1136 RETURN(PTR_ERR(env));
1138 rc = cl_object_layout_get(env, lli->lli_clob, clt);
1140 CDEBUG(D_INODE, "Cannot get layout for "DFID"\n",
1141 PFID(ll_inode2fid(inode)));
1143 cl_env_put(env, &refcheck);
1147 static int pcc_fid2dataset_fullpath(char *buf, int sz, struct lu_fid *fid,
1148 struct pcc_dataset *dataset)
1150 return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
1152 dataset->pccd_pathname,
1153 (fid)->f_oid & 0xFFFF,
1154 (fid)->f_oid >> 16 & 0xFFFF,
1155 (unsigned int)((fid)->f_seq & 0xFFFF),
1156 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
1157 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
1158 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
1162 /* Must be called with pcci->pcci_lock held */
1163 static void pcc_inode_attach_init(struct pcc_dataset *dataset,
1164 struct pcc_inode *pcci,
1165 struct dentry *dentry,
1166 enum lu_pcc_type type)
1168 pcci->pcci_path.mnt = mntget(dataset->pccd_path.mnt);
1169 pcci->pcci_path.dentry = dentry;
1170 LASSERT(atomic_read(&pcci->pcci_refcount) == 0);
1171 atomic_set(&pcci->pcci_refcount, 1);
1172 pcci->pcci_type = type;
1173 pcci->pcci_attr_valid = false;
1176 static inline void pcc_inode_dsflags_set(struct ll_inode_info *lli,
1177 struct pcc_dataset *dataset)
1179 lli->lli_pcc_generation = ll_info2pccs(lli)->pccs_generation;
1180 lli->lli_pcc_dsflags = dataset->pccd_flags;
1183 static void pcc_inode_attach_set(struct pcc_super *super,
1184 struct pcc_dataset *dataset,
1185 struct ll_inode_info *lli,
1186 struct pcc_inode *pcci,
1187 struct dentry *dentry,
1188 enum lu_pcc_type type)
1190 pcc_inode_init(pcci, lli);
1191 pcc_inode_attach_init(dataset, pcci, dentry, type);
1192 down_read(&super->pccs_rw_sem);
1193 pcc_inode_dsflags_set(lli, dataset);
1194 up_read(&super->pccs_rw_sem);
1197 static inline void pcc_layout_gen_set(struct pcc_inode *pcci,
1200 pcci->pcci_layout_gen = gen;
1203 static inline bool pcc_inode_has_layout(struct pcc_inode *pcci)
1205 return pcci->pcci_layout_gen != CL_LAYOUT_GEN_NONE;
1208 static int pcc_try_dataset_attach(struct inode *inode, __u32 gen,
1209 enum lu_pcc_type type,
1210 struct pcc_dataset *dataset,
1213 struct ll_inode_info *lli = ll_i2info(inode);
1214 struct pcc_inode *pcci = lli->lli_pcc_inode;
1215 const struct cred *old_cred;
1216 struct dentry *pcc_dentry;
1224 if (type == LU_PCC_READWRITE &&
1225 !(dataset->pccd_flags & PCC_DATASET_RWPCC))
1228 OBD_ALLOC(pathname, PATH_MAX);
1229 if (pathname == NULL)
1232 pcc_fid2dataset_fullpath(pathname, PATH_MAX, &lli->lli_fid, dataset);
1234 old_cred = override_creds(pcc_super_cred(inode->i_sb));
1235 rc = kern_path(pathname, LOOKUP_FOLLOW, &path);
1237 /* ignore this error */
1240 pcc_dentry = path.dentry;
1241 #ifndef HAVE_VFS_SETXATTR
1242 if (!pcc_dentry->d_inode->i_op->getxattr)
1243 /* ignore this error */
1244 GOTO(out_put_path, rc = 0);
1246 rc = pcc_dentry->d_inode->i_op->getxattr(pcc_dentry, pcc_xattr_layout,
1247 &pcc_gen, sizeof(pcc_gen));
1249 rc = __vfs_getxattr(pcc_dentry, pcc_dentry->d_inode, pcc_xattr_layout,
1250 &pcc_gen, sizeof(pcc_gen));
1254 /* ignore this error */
1255 GOTO(out_put_path, rc = 0);
1258 /* The file is still valid cached in PCC, attach it immediately. */
1259 if (pcc_gen == gen) {
1260 CDEBUG(D_CACHE, DFID" L.Gen (%d) consistent, auto attached.\n",
1261 PFID(&lli->lli_fid), gen);
1263 OBD_SLAB_ALLOC_PTR_GFP(pcci, pcc_inode_slab, GFP_NOFS);
1265 GOTO(out_put_path, rc = -ENOMEM);
1267 pcc_inode_init(pcci, lli);
1269 pcc_inode_attach_init(dataset, pcci, pcc_dentry, type);
1272 * This happened when a file was once attached into
1273 * PCC, and some processes keep this file opened
1274 * (pcci->refcount > 1) and corresponding PCC file
1275 * without any I/O activity, and then this file was
1276 * detached by the manual detach command or the
1277 * revocation of the layout lock (i.e. cached LRU lock
1280 pcc_inode_get(pcci);
1281 pcci->pcci_type = type;
1283 pcc_inode_dsflags_set(lli, dataset);
1284 pcc_layout_gen_set(pcci, gen);
1290 revert_creds(old_cred);
1291 OBD_FREE(pathname, PATH_MAX);
1295 static int pcc_try_datasets_attach(struct inode *inode, enum pcc_io_type iot,
1296 __u32 gen, enum lu_pcc_type type,
1299 struct pcc_super *super = &ll_i2sbi(inode)->ll_pcc_super;
1300 struct ll_inode_info *lli = ll_i2info(inode);
1301 struct pcc_dataset *dataset = NULL, *tmp;
1306 down_read(&super->pccs_rw_sem);
1307 list_for_each_entry_safe(dataset, tmp,
1308 &super->pccs_datasets, pccd_linkage) {
1309 if (!pcc_auto_attach_enabled(dataset->pccd_flags, iot))
1312 rc = pcc_try_dataset_attach(inode, gen, type, dataset, cached);
1313 if (rc < 0 || (!rc && *cached))
1318 * Update the saved dataset flags for the inode accordingly if failed.
1320 if (!rc && !*cached) {
1322 * Currently auto attach strategy for a PCC backend is
1323 * unchangeable once once it was added into the PCC datasets on
1324 * a client as the support to change auto attach strategy is
1325 * not implemented yet.
1328 * If tried to attach from one PCC backend:
1329 * @lli_pcc_generation > 0:
1330 * 1) The file was once attached into PCC, but now the
1331 * corresponding PCC backend should be removed from the client;
1332 * 2) The layout generation was changed, the data has been
1334 * 3) The corresponding PCC copy is not existed on PCC
1335 * @lli_pcc_generation == 0:
1336 * The file is never attached into PCC but in a HSM released
1337 * state, or once attached into PCC but the inode was evicted
1338 * from icache later.
1339 * Set the saved dataset flags with PCC_DATASET_NONE. Then this
1340 * file will skip from the candidates to try auto attach until
1341 * the file is attached into PCC again.
1343 * If the file was never attached into PCC, or once attached but
1344 * its inode was evicted from icache (lli_pcc_generation == 0),
1345 * or the corresponding dataset was removed from the client,
1346 * set the saved dataset flags with PCC_DATASET_NONE.
1348 * TODO: If the file was once attached into PCC but not try to
1349 * auto attach due to the change of the configuration parameters
1350 * for this dataset (i.e. change from auto attach enabled to
1351 * auto attach disabled for this dataset), update the saved
1352 * dataset flags with the found one.
1354 lli->lli_pcc_dsflags = PCC_DATASET_NONE;
1356 up_read(&super->pccs_rw_sem);
1362 * TODO: For RW-PCC, it is desirable to store HSM info as a layout (LU-10606).
1363 * Thus the client can get archive ID from the layout directly. When try to
1364 * attach the file automatically which is in HSM released state (according to
1365 * LOV_PATTERN_F_RELEASED in the layout), it can determine whether the file is
1366 * valid cached on PCC more precisely according to the @rwid (archive ID) in
1367 * the PCC dataset and the archive ID in HSM attrs.
1369 static int pcc_try_auto_attach(struct inode *inode, bool *cached,
1370 enum pcc_io_type iot)
1372 struct pcc_super *super = &ll_i2sbi(inode)->ll_pcc_super;
1373 struct cl_layout clt = {
1375 .cl_is_released = false,
1377 struct ll_inode_info *lli = ll_i2info(inode);
1384 * Quick check whether there is PCC device.
1386 if (list_empty(&super->pccs_datasets))
1390 * The file layout lock was cancelled. And this open does not
1391 * obtain valid layout lock from MDT (i.e. the file is being
1394 if (iot == PIT_OPEN) {
1395 if (ll_layout_version_get(lli) == CL_LAYOUT_GEN_NONE)
1398 rc = ll_layout_refresh(inode, &gen);
1403 rc = pcc_get_layout_info(inode, &clt);
1407 if (iot != PIT_OPEN && gen != clt.cl_layout_gen) {
1408 CDEBUG(D_CACHE, DFID" layout changed from %d to %d.\n",
1409 PFID(ll_inode2fid(inode)), gen, clt.cl_layout_gen);
1413 if (clt.cl_is_released)
1414 rc = pcc_try_datasets_attach(inode, iot, clt.cl_layout_gen,
1415 LU_PCC_READWRITE, cached);
1420 static inline bool pcc_may_auto_attach(struct inode *inode,
1421 enum pcc_io_type iot)
1423 struct ll_inode_info *lli = ll_i2info(inode);
1424 struct pcc_super *super = ll_i2pccs(inode);
1426 /* Known the file was not in any PCC backend. */
1427 if (lli->lli_pcc_dsflags & PCC_DATASET_NONE)
1431 * lli_pcc_generation == 0 means that the file was never attached into
1432 * PCC, or may be once attached into PCC but detached as the inode is
1433 * evicted from icache (i.e. "echo 3 > /proc/sys/vm/drop_caches" or
1434 * icache shrinking due to the memory pressure), which will cause the
1435 * file detach from PCC when releasing the inode from icache.
1436 * In either case, we still try to attach.
1438 /* lli_pcc_generation == 0, or the PCC setting was changed,
1439 * or there is no PCC setup on the client and the try will return
1440 * immediately in pcc_try_auto_attach().
1442 if (super->pccs_generation != lli->lli_pcc_generation)
1445 /* The cached setting @lli_pcc_dsflags is valid */
1446 if (iot == PIT_OPEN)
1447 return lli->lli_pcc_dsflags & PCC_DATASET_OPEN_ATTACH;
1449 if (iot == PIT_GETATTR)
1450 return lli->lli_pcc_dsflags & PCC_DATASET_STAT_ATTACH;
1452 return lli->lli_pcc_dsflags & PCC_DATASET_IO_ATTACH;
1455 int pcc_file_open(struct inode *inode, struct file *file)
1457 struct pcc_inode *pcci;
1458 struct ll_inode_info *lli = ll_i2info(inode);
1459 struct ll_file_data *fd = file->private_data;
1460 struct pcc_file *pccf = &fd->fd_pcc_file;
1461 struct file *pcc_file;
1464 bool cached = false;
1469 if (!S_ISREG(inode->i_mode))
1472 pcc_inode_lock(inode);
1473 pcci = ll_i2pcci(inode);
1475 if (lli->lli_pcc_state & PCC_STATE_FL_ATTACHING)
1476 GOTO(out_unlock, rc = 0);
1478 if (!pcci || !pcc_inode_has_layout(pcci)) {
1479 if (pcc_may_auto_attach(inode, PIT_OPEN))
1480 rc = pcc_try_auto_attach(inode, &cached, PIT_OPEN);
1482 if (rc < 0 || !cached)
1483 GOTO(out_unlock, rc);
1486 pcci = ll_i2pcci(inode);
1489 pcc_inode_get(pcci);
1490 WARN_ON(pccf->pccf_file);
1492 path = &pcci->pcci_path;
1493 dname = &path->dentry->d_name;
1494 CDEBUG(D_CACHE, "opening pcc file '%.*s'\n", dname->len,
1497 pcc_file = dentry_open(path, file->f_flags,
1498 pcc_super_cred(inode->i_sb));
1499 if (IS_ERR_OR_NULL(pcc_file)) {
1500 rc = pcc_file == NULL ? -EINVAL : PTR_ERR(pcc_file);
1501 pcc_inode_put(pcci);
1503 pccf->pccf_file = pcc_file;
1504 pccf->pccf_type = pcci->pcci_type;
1508 pcc_inode_unlock(inode);
1512 void pcc_file_release(struct inode *inode, struct file *file)
1514 struct pcc_inode *pcci;
1515 struct ll_file_data *fd = file->private_data;
1516 struct pcc_file *pccf;
1522 if (!S_ISREG(inode->i_mode) || fd == NULL)
1525 pccf = &fd->fd_pcc_file;
1526 pcc_inode_lock(inode);
1527 if (pccf->pccf_file == NULL)
1530 pcci = ll_i2pcci(inode);
1532 path = &pcci->pcci_path;
1533 dname = &path->dentry->d_name;
1534 CDEBUG(D_CACHE, "releasing pcc file \"%.*s\"\n", dname->len,
1536 pcc_inode_put(pcci);
1537 fput(pccf->pccf_file);
1538 pccf->pccf_file = NULL;
1540 pcc_inode_unlock(inode);
1544 static void pcc_io_init(struct inode *inode, enum pcc_io_type iot, bool *cached)
1546 struct pcc_inode *pcci;
1548 pcc_inode_lock(inode);
1549 pcci = ll_i2pcci(inode);
1550 if (pcci && pcc_inode_has_layout(pcci)) {
1551 LASSERT(atomic_read(&pcci->pcci_refcount) > 0);
1552 atomic_inc(&pcci->pcci_active_ios);
1556 if (pcc_may_auto_attach(inode, iot)) {
1557 (void) pcc_try_auto_attach(inode, cached, iot);
1559 pcci = ll_i2pcci(inode);
1560 LASSERT(atomic_read(&pcci->pcci_refcount) > 0);
1561 atomic_inc(&pcci->pcci_active_ios);
1565 pcc_inode_unlock(inode);
1568 static void pcc_io_fini(struct inode *inode)
1570 struct pcc_inode *pcci = ll_i2pcci(inode);
1572 LASSERT(pcci && atomic_read(&pcci->pcci_active_ios) > 0);
1573 if (atomic_dec_and_test(&pcci->pcci_active_ios))
1574 wake_up_all(&pcci->pcci_waitq);
1579 __pcc_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1581 struct file *file = iocb->ki_filp;
1583 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1584 return file->f_op->read_iter(iocb, iter);
1590 iov_for_each(iov, i, *iter) {
1593 res = file->f_op->aio_read(iocb, &iov, 1, iocb->ki_pos);
1594 if (-EIOCBQUEUED == res)
1595 res = wait_on_sync_kiocb(iocb);
1603 if (res < iov.iov_len)
1608 iov_iter_advance(iter, bytes);
1613 ssize_t pcc_file_read_iter(struct kiocb *iocb,
1614 struct iov_iter *iter, bool *cached)
1616 struct file *file = iocb->ki_filp;
1617 struct ll_file_data *fd = file->private_data;
1618 struct pcc_file *pccf = &fd->fd_pcc_file;
1619 struct inode *inode = file_inode(file);
1624 if (pccf->pccf_file == NULL) {
1629 pcc_io_init(inode, PIT_READ, cached);
1633 iocb->ki_filp = pccf->pccf_file;
1634 /* generic_file_aio_read does not support ext4-dax,
1635 * __pcc_file_read_iter uses ->aio_read hook directly
1636 * to add support for ext4-dax.
1638 result = __pcc_file_read_iter(iocb, iter);
1639 iocb->ki_filp = file;
1646 __pcc_file_write_iter(struct kiocb *iocb, struct iov_iter *iter)
1648 struct file *file = iocb->ki_filp;
1650 #ifdef HAVE_FILE_OPERATIONS_READ_WRITE_ITER
1651 return file->f_op->write_iter(iocb, iter);
1657 iov_for_each(iov, i, *iter) {
1660 res = file->f_op->aio_write(iocb, &iov, 1, iocb->ki_pos);
1661 if (-EIOCBQUEUED == res)
1662 res = wait_on_sync_kiocb(iocb);
1670 if (res < iov.iov_len)
1675 iov_iter_advance(iter, bytes);
1680 ssize_t pcc_file_write_iter(struct kiocb *iocb,
1681 struct iov_iter *iter, bool *cached)
1683 struct file *file = iocb->ki_filp;
1684 struct ll_file_data *fd = file->private_data;
1685 struct pcc_file *pccf = &fd->fd_pcc_file;
1686 struct inode *inode = file_inode(file);
1691 if (pccf->pccf_file == NULL) {
1696 if (pccf->pccf_type != LU_PCC_READWRITE) {
1701 pcc_io_init(inode, PIT_WRITE, cached);
1705 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_PCC_FAKE_ERROR))
1706 GOTO(out, result = -ENOSPC);
1708 iocb->ki_filp = pccf->pccf_file;
1710 /* Since __pcc_file_write_iter makes write calls via
1711 * the normal vfs interface to the local PCC file system,
1712 * the inode lock is not needed.
1714 result = __pcc_file_write_iter(iocb, iter);
1715 iocb->ki_filp = file;
1721 int pcc_inode_setattr(struct inode *inode, struct iattr *attr,
1725 const struct cred *old_cred;
1726 struct iattr attr2 = *attr;
1727 struct dentry *pcc_dentry;
1728 struct pcc_inode *pcci;
1732 if (!S_ISREG(inode->i_mode)) {
1737 pcc_io_init(inode, PIT_SETATTR, cached);
1741 attr2.ia_valid = attr->ia_valid & (ATTR_SIZE | ATTR_ATIME |
1742 ATTR_ATIME_SET | ATTR_MTIME | ATTR_MTIME_SET |
1743 ATTR_CTIME | ATTR_UID | ATTR_GID);
1744 pcci = ll_i2pcci(inode);
1745 pcc_dentry = pcci->pcci_path.dentry;
1746 inode_lock(pcc_dentry->d_inode);
1747 old_cred = override_creds(pcc_super_cred(inode->i_sb));
1748 rc = pcc_dentry->d_inode->i_op->setattr(pcc_dentry, &attr2);
1749 revert_creds(old_cred);
1750 inode_unlock(pcc_dentry->d_inode);
1756 int pcc_inode_getattr(struct inode *inode, bool *cached)
1758 struct ll_inode_info *lli = ll_i2info(inode);
1759 const struct cred *old_cred;
1768 if (!S_ISREG(inode->i_mode)) {
1773 pcc_io_init(inode, PIT_GETATTR, cached);
1777 old_cred = override_creds(pcc_super_cred(inode->i_sb));
1778 rc = ll_vfs_getattr(&ll_i2pcci(inode)->pcci_path, &stat);
1779 revert_creds(old_cred);
1783 ll_inode_size_lock(inode);
1784 if (ll_file_test_and_clear_flag(lli, LLIF_UPDATE_ATIME) ||
1785 inode->i_atime.tv_sec < lli->lli_atime)
1786 inode->i_atime.tv_sec = lli->lli_atime;
1788 inode->i_mtime.tv_sec = lli->lli_mtime;
1789 inode->i_ctime.tv_sec = lli->lli_ctime;
1791 atime = inode->i_atime.tv_sec;
1792 mtime = inode->i_mtime.tv_sec;
1793 ctime = inode->i_ctime.tv_sec;
1795 if (atime < stat.atime.tv_sec)
1796 atime = stat.atime.tv_sec;
1798 if (ctime < stat.ctime.tv_sec)
1799 ctime = stat.ctime.tv_sec;
1801 if (mtime < stat.mtime.tv_sec)
1802 mtime = stat.mtime.tv_sec;
1804 i_size_write(inode, stat.size);
1805 inode->i_blocks = stat.blocks;
1807 inode->i_atime.tv_sec = atime;
1808 inode->i_mtime.tv_sec = mtime;
1809 inode->i_ctime.tv_sec = ctime;
1811 ll_inode_size_unlock(inode);
1817 ssize_t pcc_file_splice_read(struct file *in_file, loff_t *ppos,
1818 struct pipe_inode_info *pipe,
1819 size_t count, unsigned int flags,
1822 struct inode *inode = file_inode(in_file);
1823 struct ll_file_data *fd = in_file->private_data;
1824 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
1833 if (!file_inode(pcc_file)->i_fop->splice_read)
1836 pcc_io_init(inode, PIT_SPLICE_READ, cached);
1840 result = file_inode(pcc_file)->i_fop->splice_read(pcc_file,
1848 int pcc_fsync(struct file *file, loff_t start, loff_t end,
1849 int datasync, bool *cached)
1851 struct inode *inode = file_inode(file);
1852 struct ll_file_data *fd = file->private_data;
1853 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
1863 pcc_io_init(inode, PIT_FSYNC, cached);
1867 rc = file_inode(pcc_file)->i_fop->fsync(pcc_file,
1868 start, end, datasync);
1874 int pcc_file_mmap(struct file *file, struct vm_area_struct *vma,
1877 struct inode *inode = file_inode(file);
1878 struct ll_file_data *fd = file->private_data;
1879 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
1880 struct pcc_inode *pcci;
1885 if (!pcc_file || !file_inode(pcc_file)->i_fop->mmap) {
1890 pcc_inode_lock(inode);
1891 pcci = ll_i2pcci(inode);
1892 if (pcci && pcc_inode_has_layout(pcci)) {
1893 LASSERT(atomic_read(&pcci->pcci_refcount) > 1);
1895 vma->vm_file = pcc_file;
1896 rc = file_inode(pcc_file)->i_fop->mmap(pcc_file, vma);
1897 vma->vm_file = file;
1898 /* Save the vm ops of backend PCC */
1899 vma->vm_private_data = (void *)vma->vm_ops;
1903 pcc_inode_unlock(inode);
1908 void pcc_vm_open(struct vm_area_struct *vma)
1910 struct pcc_inode *pcci;
1911 struct file *file = vma->vm_file;
1912 struct inode *inode = file_inode(file);
1913 struct ll_file_data *fd = file->private_data;
1914 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
1915 struct vm_operations_struct *pcc_vm_ops = vma->vm_private_data;
1919 if (!pcc_file || !pcc_vm_ops || !pcc_vm_ops->open)
1922 pcc_inode_lock(inode);
1923 pcci = ll_i2pcci(inode);
1924 if (pcci && pcc_inode_has_layout(pcci)) {
1925 vma->vm_file = pcc_file;
1926 pcc_vm_ops->open(vma);
1927 vma->vm_file = file;
1929 pcc_inode_unlock(inode);
1933 void pcc_vm_close(struct vm_area_struct *vma)
1935 struct file *file = vma->vm_file;
1936 struct inode *inode = file_inode(file);
1937 struct ll_file_data *fd = file->private_data;
1938 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
1939 struct vm_operations_struct *pcc_vm_ops = vma->vm_private_data;
1943 if (!pcc_file || !pcc_vm_ops || !pcc_vm_ops->close)
1946 pcc_inode_lock(inode);
1947 /* Layout lock maybe revoked here */
1948 vma->vm_file = pcc_file;
1949 pcc_vm_ops->close(vma);
1950 vma->vm_file = file;
1951 pcc_inode_unlock(inode);
1955 int pcc_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
1958 struct page *page = vmf->page;
1959 struct mm_struct *mm = vma->vm_mm;
1960 struct file *file = vma->vm_file;
1961 struct inode *inode = file_inode(file);
1962 struct ll_file_data *fd = file->private_data;
1963 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
1964 struct vm_operations_struct *pcc_vm_ops = vma->vm_private_data;
1969 if (!pcc_file || !pcc_vm_ops) {
1974 if (!pcc_vm_ops->page_mkwrite &&
1975 page->mapping == pcc_file->f_mapping) {
1977 "%s: PCC backend fs not support ->page_mkwrite()\n",
1978 ll_i2sbi(inode)->ll_fsname);
1979 pcc_ioctl_detach(inode, PCC_DETACH_OPT_UNCACHE);
1980 up_read(&mm->mmap_sem);
1982 RETURN(VM_FAULT_RETRY | VM_FAULT_NOPAGE);
1984 /* Pause to allow for a race with concurrent detach */
1985 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_PCC_MKWRITE_PAUSE, cfs_fail_val);
1987 pcc_io_init(inode, PIT_PAGE_MKWRITE, cached);
1989 /* This happens when the file is detached from PCC after got
1990 * the fault page via ->fault() on the inode of the PCC copy.
1991 * Here it can not simply fall back to normal Lustre I/O path.
1992 * The reason is that the address space of fault page used by
1993 * ->page_mkwrite() is still the one of PCC inode. In the
1994 * normal Lustre ->page_mkwrite() I/O path, it will be wrongly
1995 * handled as the address space of the fault page is not
1996 * consistent with the one of the Lustre inode (though the
1997 * fault page was truncated).
1998 * As the file is detached from PCC, the fault page must
1999 * be released frist, and retry the mmap write (->fault() and
2001 * We use an ugly and tricky method by returning
2002 * VM_FAULT_NOPAGE | VM_FAULT_RETRY to the caller
2003 * __do_page_fault and retry the memory fault handling.
2005 if (page->mapping == pcc_file->f_mapping) {
2007 up_read(&mm->mmap_sem);
2008 RETURN(VM_FAULT_RETRY | VM_FAULT_NOPAGE);
2015 * This fault injection can also be used to simulate -ENOSPC and
2016 * -EDQUOT failure of underlying PCC backend fs.
2018 if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_PCC_DETACH_MKWRITE)) {
2020 pcc_ioctl_detach(inode, PCC_DETACH_OPT_UNCACHE);
2021 up_read(&mm->mmap_sem);
2022 RETURN(VM_FAULT_RETRY | VM_FAULT_NOPAGE);
2025 vma->vm_file = pcc_file;
2026 #ifdef HAVE_VM_OPS_USE_VM_FAULT_ONLY
2027 rc = pcc_vm_ops->page_mkwrite(vmf);
2029 rc = pcc_vm_ops->page_mkwrite(vma, vmf);
2031 vma->vm_file = file;
2037 int pcc_fault(struct vm_area_struct *vma, struct vm_fault *vmf,
2040 struct file *file = vma->vm_file;
2041 struct inode *inode = file_inode(file);
2042 struct ll_file_data *fd = file->private_data;
2043 struct file *pcc_file = fd->fd_pcc_file.pccf_file;
2044 struct vm_operations_struct *pcc_vm_ops = vma->vm_private_data;
2049 if (!pcc_file || !pcc_vm_ops || !pcc_vm_ops->fault) {
2054 pcc_io_init(inode, PIT_FAULT, cached);
2058 vma->vm_file = pcc_file;
2059 #ifdef HAVE_VM_OPS_USE_VM_FAULT_ONLY
2060 rc = pcc_vm_ops->fault(vmf);
2062 rc = pcc_vm_ops->fault(vma, vmf);
2064 vma->vm_file = file;
2070 static void pcc_layout_wait(struct pcc_inode *pcci)
2073 while (atomic_read(&pcci->pcci_active_ios) > 0) {
2074 CDEBUG(D_CACHE, "Waiting for IO completion: %d\n",
2075 atomic_read(&pcci->pcci_active_ios));
2076 wait_event_idle(pcci->pcci_waitq,
2077 atomic_read(&pcci->pcci_active_ios) == 0);
2081 static void __pcc_layout_invalidate(struct pcc_inode *pcci)
2083 pcci->pcci_type = LU_PCC_NONE;
2084 pcc_layout_gen_set(pcci, CL_LAYOUT_GEN_NONE);
2085 pcc_layout_wait(pcci);
2088 void pcc_layout_invalidate(struct inode *inode)
2090 struct pcc_inode *pcci;
2094 pcc_inode_lock(inode);
2095 pcci = ll_i2pcci(inode);
2096 if (pcci && pcc_inode_has_layout(pcci)) {
2097 LASSERT(atomic_read(&pcci->pcci_refcount) > 0);
2098 __pcc_layout_invalidate(pcci);
2100 CDEBUG(D_CACHE, "Invalidate "DFID" layout gen %d\n",
2101 PFID(&ll_i2info(inode)->lli_fid), pcci->pcci_layout_gen);
2103 pcc_inode_put(pcci);
2105 pcc_inode_unlock(inode);
2110 static int pcc_inode_remove(struct inode *inode, struct dentry *pcc_dentry)
2114 rc = ll_vfs_unlink(pcc_dentry->d_parent->d_inode, pcc_dentry);
2116 CWARN("%s: failed to unlink PCC file %.*s, rc = %d\n",
2117 ll_i2sbi(inode)->ll_fsname, pcc_dentry->d_name.len,
2118 pcc_dentry->d_name.name, rc);
2123 /* Create directory under base if directory does not exist */
2124 static struct dentry *
2125 pcc_mkdir(struct dentry *base, const char *name, umode_t mode)
2128 struct dentry *dentry;
2129 struct inode *dir = base->d_inode;
2132 dentry = lookup_one_len(name, base, strlen(name));
2136 if (d_is_positive(dentry))
2139 rc = vfs_mkdir(dir, dentry, mode);
2142 dentry = ERR_PTR(rc);
2150 static struct dentry *
2151 pcc_mkdir_p(struct dentry *root, char *path, umode_t mode)
2153 char *ptr, *entry_name;
2154 struct dentry *parent;
2155 struct dentry *child = ERR_PTR(-EINVAL);
2162 parent = dget(root);
2163 while ((ptr = strchr(ptr, '/')) != NULL) {
2165 child = pcc_mkdir(parent, entry_name, mode);
2179 /* Create file under base. If file already exist, return failure */
2180 static struct dentry *
2181 pcc_create(struct dentry *base, const char *name, umode_t mode)
2184 struct dentry *dentry;
2185 struct inode *dir = base->d_inode;
2188 dentry = lookup_one_len(name, base, strlen(name));
2192 if (d_is_positive(dentry))
2195 rc = vfs_create(dir, dentry, mode, false);
2198 dentry = ERR_PTR(rc);
2206 static int __pcc_inode_create(struct pcc_dataset *dataset,
2208 struct dentry **dentry)
2211 struct dentry *base;
2212 struct dentry *child;
2215 OBD_ALLOC(path, MAX_PCC_DATABASE_PATH);
2219 pcc_fid2dataset_path(path, MAX_PCC_DATABASE_PATH, fid);
2221 base = pcc_mkdir_p(dataset->pccd_path.dentry, path, 0);
2227 snprintf(path, MAX_PCC_DATABASE_PATH, DFID_NOBRACE, PFID(fid));
2228 child = pcc_create(base, path, 0);
2229 if (IS_ERR(child)) {
2230 rc = PTR_ERR(child);
2238 OBD_FREE(path, MAX_PCC_DATABASE_PATH);
2243 * Reset uid, gid or size for the PCC copy masked by @valid.
2244 * TODO: Set the project ID for PCC copy.
2246 int pcc_inode_reset_iattr(struct dentry *dentry, unsigned int valid,
2247 kuid_t uid, kgid_t gid, loff_t size)
2249 struct inode *inode = dentry->d_inode;
2255 attr.ia_valid = valid;
2258 attr.ia_size = size;
2261 rc = notify_change(dentry, &attr, NULL);
2262 inode_unlock(inode);
2267 int pcc_inode_create(struct super_block *sb, struct pcc_dataset *dataset,
2268 struct lu_fid *fid, struct dentry **pcc_dentry)
2270 const struct cred *old_cred;
2273 old_cred = override_creds(pcc_super_cred(sb));
2274 rc = __pcc_inode_create(dataset, fid, pcc_dentry);
2275 revert_creds(old_cred);
2279 int pcc_inode_create_fini(struct inode *inode, struct pcc_create_attach *pca)
2281 struct dentry *pcc_dentry = pca->pca_dentry;
2282 struct pcc_super *super = ll_i2pccs(inode);
2283 const struct cred *old_cred;
2284 struct pcc_inode *pcci;
2289 if (!pca->pca_dataset)
2293 GOTO(out_dataset_put, rc = 0);
2295 LASSERT(pcc_dentry);
2297 old_cred = override_creds(super->pccs_cred);
2298 pcc_inode_lock(inode);
2299 LASSERT(ll_i2pcci(inode) == NULL);
2300 OBD_SLAB_ALLOC_PTR_GFP(pcci, pcc_inode_slab, GFP_NOFS);
2302 GOTO(out_put, rc = -ENOMEM);
2304 rc = pcc_inode_reset_iattr(pcc_dentry, ATTR_UID | ATTR_GID,
2305 old_cred->suid, old_cred->sgid, 0);
2309 pcc_inode_attach_set(super, pca->pca_dataset, ll_i2info(inode),
2310 pcci, pcc_dentry, LU_PCC_READWRITE);
2312 rc = pcc_layout_xattr_set(pcci, 0);
2314 (void) pcc_inode_remove(inode, pcci->pcci_path.dentry);
2315 pcc_inode_put(pcci);
2316 GOTO(out_unlock, rc);
2319 /* Set the layout generation of newly created file with 0 */
2320 pcc_layout_gen_set(pcci, 0);
2324 (void) pcc_inode_remove(inode, pcc_dentry);
2328 OBD_SLAB_FREE_PTR(pcci, pcc_inode_slab);
2331 pcc_inode_unlock(inode);
2332 revert_creds(old_cred);
2334 pcc_dataset_put(pca->pca_dataset);
2338 void pcc_create_attach_cleanup(struct super_block *sb,
2339 struct pcc_create_attach *pca)
2341 if (!pca->pca_dataset)
2344 if (pca->pca_dentry) {
2345 const struct cred *old_cred;
2348 old_cred = override_creds(pcc_super_cred(sb));
2349 rc = ll_vfs_unlink(pca->pca_dentry->d_parent->d_inode,
2352 CWARN("failed to unlink PCC file %.*s, rc = %d\n",
2353 pca->pca_dentry->d_name.len,
2354 pca->pca_dentry->d_name.name, rc);
2355 /* ignore the unlink failure */
2356 revert_creds(old_cred);
2357 dput(pca->pca_dentry);
2360 pcc_dataset_put(pca->pca_dataset);
2363 static int pcc_filp_write(struct file *filp, const void *buf, ssize_t count,
2369 size = cfs_kernel_write(filp, buf, count, offset);
2378 static ssize_t pcc_copy_data(struct file *src, struct file *dst)
2382 loff_t pos, offset = 0;
2383 size_t buf_len = 1048576;
2388 OBD_ALLOC_LARGE(buf, buf_len);
2393 if (signal_pending(current))
2394 GOTO(out_free, rc = -EINTR);
2397 rc2 = cfs_kernel_read(src, buf, buf_len, &pos);
2399 GOTO(out_free, rc = rc2);
2404 rc = pcc_filp_write(dst, buf, rc2, &pos);
2412 OBD_FREE_LARGE(buf, buf_len);
2416 static int pcc_attach_allowed_check(struct inode *inode)
2418 struct ll_inode_info *lli = ll_i2info(inode);
2419 struct pcc_inode *pcci;
2424 pcc_inode_lock(inode);
2425 if (lli->lli_pcc_state & PCC_STATE_FL_ATTACHING)
2426 GOTO(out_unlock, rc = -EBUSY);
2428 pcci = ll_i2pcci(inode);
2429 if (pcci && pcc_inode_has_layout(pcci))
2430 GOTO(out_unlock, rc = -EEXIST);
2432 lli->lli_pcc_state |= PCC_STATE_FL_ATTACHING;
2434 pcc_inode_unlock(inode);
2438 int pcc_readwrite_attach(struct file *file, struct inode *inode,
2441 struct pcc_dataset *dataset;
2442 struct ll_inode_info *lli = ll_i2info(inode);
2443 struct pcc_super *super = ll_i2pccs(inode);
2444 struct pcc_inode *pcci;
2445 const struct cred *old_cred;
2446 struct dentry *dentry;
2447 struct file *pcc_filp;
2454 rc = pcc_attach_allowed_check(inode);
2458 dataset = pcc_dataset_get(&ll_i2sbi(inode)->ll_pcc_super,
2459 LU_PCC_READWRITE, archive_id);
2460 if (dataset == NULL)
2463 old_cred = override_creds(super->pccs_cred);
2464 rc = __pcc_inode_create(dataset, &lli->lli_fid, &dentry);
2466 GOTO(out_dataset_put, rc);
2468 path.mnt = dataset->pccd_path.mnt;
2469 path.dentry = dentry;
2470 pcc_filp = dentry_open(&path, O_WRONLY | O_LARGEFILE, current_cred());
2471 if (IS_ERR_OR_NULL(pcc_filp)) {
2472 rc = pcc_filp == NULL ? -EINVAL : PTR_ERR(pcc_filp);
2473 GOTO(out_dentry, rc);
2476 rc = pcc_inode_reset_iattr(dentry, ATTR_UID | ATTR_GID,
2477 old_cred->uid, old_cred->gid, 0);
2481 ret = pcc_copy_data(file, pcc_filp);
2483 GOTO(out_fput, rc = ret);
2486 * It must to truncate the PCC copy to the same size of the Lustre
2487 * copy after copy data. Otherwise, it may get wrong file size after
2488 * re-attach a file. See LU-13023 for details.
2490 rc = pcc_inode_reset_iattr(dentry, ATTR_SIZE, KUIDT_INIT(0),
2491 KGIDT_INIT(0), ret);
2495 /* Pause to allow for a race with concurrent HSM remove */
2496 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_PCC_ATTACH_PAUSE, cfs_fail_val);
2498 pcc_inode_lock(inode);
2499 pcci = ll_i2pcci(inode);
2501 OBD_SLAB_ALLOC_PTR_GFP(pcci, pcc_inode_slab, GFP_NOFS);
2503 GOTO(out_unlock, rc = -ENOMEM);
2505 pcc_inode_attach_set(super, dataset, lli, pcci,
2506 dentry, LU_PCC_READWRITE);
2508 pcc_inode_unlock(inode);
2513 (void) pcc_inode_remove(inode, dentry);
2517 pcc_dataset_put(dataset);
2518 revert_creds(old_cred);
2523 int pcc_readwrite_attach_fini(struct file *file, struct inode *inode,
2524 __u32 gen, bool lease_broken, int rc,
2527 struct ll_inode_info *lli = ll_i2info(inode);
2528 const struct cred *old_cred;
2529 struct pcc_inode *pcci;
2534 old_cred = override_creds(pcc_super_cred(inode->i_sb));
2535 pcc_inode_lock(inode);
2536 pcci = ll_i2pcci(inode);
2537 if (rc || lease_broken) {
2538 if (attached && pcci)
2539 pcc_inode_put(pcci);
2541 GOTO(out_unlock, rc);
2544 /* PCC inode may be released due to layout lock revocatioin */
2546 GOTO(out_unlock, rc = -ESTALE);
2549 rc = pcc_layout_xattr_set(pcci, gen);
2553 LASSERT(lli->lli_pcc_state & PCC_STATE_FL_ATTACHING);
2554 rc = ll_layout_refresh(inode, &gen2);
2557 pcc_layout_gen_set(pcci, gen);
2560 DFID" layout changed from %d to %d.\n",
2561 PFID(ll_inode2fid(inode)), gen, gen2);
2562 GOTO(out_put, rc = -ESTALE);
2568 (void) pcc_inode_remove(inode, pcci->pcci_path.dentry);
2569 pcc_inode_put(pcci);
2572 lli->lli_pcc_state &= ~PCC_STATE_FL_ATTACHING;
2573 pcc_inode_unlock(inode);
2574 revert_creds(old_cred);
2578 static int pcc_hsm_remove(struct inode *inode)
2580 struct hsm_user_request *hur;
2587 rc = ll_layout_restore(inode, 0, OBD_OBJECT_EOF);
2589 CDEBUG(D_CACHE, DFID" RESTORE failure: %d\n",
2590 PFID(&ll_i2info(inode)->lli_fid), rc);
2594 ll_layout_refresh(inode, &gen);
2596 len = sizeof(struct hsm_user_request) +
2597 sizeof(struct hsm_user_item);
2598 OBD_ALLOC(hur, len);
2602 hur->hur_request.hr_action = HUA_REMOVE;
2603 hur->hur_request.hr_archive_id = 0;
2604 hur->hur_request.hr_flags = 0;
2605 memcpy(&hur->hur_user_item[0].hui_fid, &ll_i2info(inode)->lli_fid,
2606 sizeof(hur->hur_user_item[0].hui_fid));
2607 hur->hur_user_item[0].hui_extent.offset = 0;
2608 hur->hur_user_item[0].hui_extent.length = OBD_OBJECT_EOF;
2609 hur->hur_request.hr_itemcount = 1;
2610 rc = obd_iocontrol(LL_IOC_HSM_REQUEST, ll_i2sbi(inode)->ll_md_exp,
2613 CDEBUG(D_CACHE, DFID" HSM REMOVE failure: %d\n",
2614 PFID(&ll_i2info(inode)->lli_fid), rc);
2620 int pcc_ioctl_detach(struct inode *inode, __u32 opt)
2622 struct ll_inode_info *lli = ll_i2info(inode);
2623 struct pcc_inode *pcci;
2624 bool hsm_remove = false;
2629 pcc_inode_lock(inode);
2630 pcci = lli->lli_pcc_inode;
2631 if (!pcci || lli->lli_pcc_state & PCC_STATE_FL_ATTACHING ||
2632 !pcc_inode_has_layout(pcci))
2633 GOTO(out_unlock, rc = 0);
2635 LASSERT(atomic_read(&pcci->pcci_refcount) > 0);
2637 if (pcci->pcci_type == LU_PCC_READWRITE) {
2638 if (opt == PCC_DETACH_OPT_UNCACHE) {
2641 * The file will be removed from PCC, set the flags
2642 * with PCC_DATASET_NONE even the later removal of the
2645 lli->lli_pcc_dsflags = PCC_DATASET_NONE;
2648 __pcc_layout_invalidate(pcci);
2649 pcc_inode_put(pcci);
2653 pcc_inode_unlock(inode);
2655 const struct cred *old_cred;
2657 old_cred = override_creds(pcc_super_cred(inode->i_sb));
2658 rc = pcc_hsm_remove(inode);
2659 revert_creds(old_cred);
2665 int pcc_ioctl_state(struct file *file, struct inode *inode,
2666 struct lu_pcc_state *state)
2672 int buf_len = sizeof(state->pccs_path);
2673 struct ll_file_data *fd = file->private_data;
2674 struct pcc_file *pccf = &fd->fd_pcc_file;
2675 struct pcc_inode *pcci;
2682 OBD_ALLOC(buf, buf_len);
2686 pcc_inode_lock(inode);
2687 pcci = ll_i2pcci(inode);
2689 state->pccs_type = LU_PCC_NONE;
2690 GOTO(out_unlock, rc = 0);
2693 count = atomic_read(&pcci->pcci_refcount);
2695 state->pccs_type = LU_PCC_NONE;
2696 state->pccs_open_count = 0;
2697 GOTO(out_unlock, rc = 0);
2700 if (pcc_inode_has_layout(pcci))
2702 if (pccf->pccf_file != NULL)
2704 state->pccs_type = pcci->pcci_type;
2705 state->pccs_open_count = count;
2706 state->pccs_flags = ll_i2info(inode)->lli_pcc_state;
2707 path = dentry_path_raw(pcci->pcci_path.dentry, buf, buf_len);
2709 GOTO(out_unlock, rc = PTR_ERR(path));
2711 if (strlcpy(state->pccs_path, path, buf_len) >= buf_len)
2712 GOTO(out_unlock, rc = -ENAMETOOLONG);
2715 pcc_inode_unlock(inode);
2716 OBD_FREE(buf, buf_len);