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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/obd.h>
42 #elif defined(__APPLE__)
43 #include <darwin/obd.h>
44 #elif defined(__WINNT__)
45 #include <winnt/obd.h>
47 #error Unsupported operating system.
50 #define IOC_OSC_TYPE 'h'
51 #define IOC_OSC_MIN_NR 20
52 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
53 #define IOC_OSC_MAX_NR 50
55 #define IOC_MDC_TYPE 'i'
56 #define IOC_MDC_MIN_NR 20
57 #define IOC_MDC_MAX_NR 50
59 #include <lustre/lustre_idl.h>
60 #ifdef HAVE_SERVER_SUPPORT
61 # include <lu_target.h>
64 #include <lustre_lib.h>
65 #include <lustre_export.h>
66 #include <lustre_fld.h>
67 #include <lustre_capa.h>
69 #include <libcfs/bitmap.h>
72 #define MAX_OBD_DEVICES 8192
80 struct lov_oinfo { /* per-stripe data structure */
81 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
82 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
83 int loi_ost_gen; /* generation of this loi_ost_idx */
85 unsigned long loi_kms_valid:1;
86 __u64 loi_kms; /* known minimum size */
87 struct ost_lvb loi_lvb;
88 struct osc_async_rc loi_ar;
90 #define loi_id loi_oi.oi_id
91 #define loi_seq loi_oi.oi_seq
93 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
96 oinfo->loi_kms_valid = 1;
99 static inline void loi_init(struct lov_oinfo *loi)
103 struct lov_stripe_md {
104 cfs_atomic_t lsm_refc;
106 pid_t lsm_lock_owner; /* debugging */
108 /* maximum possible file size, might change as OSTs status changes,
109 * e.g. disconnected, deactivated */
112 /* Public members. */
113 struct ost_id lw_object_oid; /* lov object id/seq */
115 /* LOV-private members start here -- only for use in lov/. */
117 __u32 lw_stripe_size; /* size of the stripe */
118 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
119 __u16 lw_stripe_count; /* number of objects being striped over */
120 __u16 lw_layout_gen; /* generation of the layout */
121 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
124 struct lov_oinfo *lsm_oinfo[0];
127 #define lsm_object_oid lsm_wire.lw_object_oid
128 #define lsm_object_id lsm_wire.lw_object_oid.oi_id
129 #define lsm_object_seq lsm_wire.lw_object_oid.oi_seq
130 #define lsm_magic lsm_wire.lw_magic
131 #define lsm_layout_gen lsm_wire.lw_layout_gen
132 #define lsm_stripe_size lsm_wire.lw_stripe_size
133 #define lsm_pattern lsm_wire.lw_pattern
134 #define lsm_stripe_count lsm_wire.lw_stripe_count
135 #define lsm_pool_name lsm_wire.lw_pool_name
139 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
141 /* obd info for a particular level (lov, osc). */
143 /* Lock policy. It keeps an extent which is specific for a particular
144 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
145 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
146 ldlm_policy_data_t oi_policy;
147 /* Flags used for set request specific flags:
148 - while lock handling, the flags obtained on the enqueue
149 request are set here.
150 - while stats, the flags used for control delay/resend.
151 - while setattr, the flags used for distinguish punch operation
154 /* Lock handle specific for every OSC lock. */
155 struct lustre_handle *oi_lockh;
156 /* lsm data specific for every OSC. */
157 struct lov_stripe_md *oi_md;
158 /* obdo data specific for every OSC, if needed at all. */
160 /* statfs data specific for every OSC, if needed at all. */
161 struct obd_statfs *oi_osfs;
162 /* An update callback which is called to update some data on upper
163 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
164 * request in osc level for enqueue requests. It is also possible to
165 * update some caller data from LOV layer if needed. */
166 obd_enqueue_update_f oi_cb_up;
167 /* oss capability, its type is obd_capa in client to avoid copy.
168 * in contrary its type is lustre_capa in OSS. */
170 /* transfer jobid from ost_sync() to filter_sync()... */
174 /* compare all relevant fields. */
175 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
176 struct lov_stripe_md *m2)
179 * ->lsm_wire contains padding, but it should be zeroed out during
182 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
185 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
186 struct lov_stripe_md *lsm)
188 if (lsm->lsm_magic != lum->lmm_magic)
190 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
191 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
193 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
194 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
196 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
197 (lsm->lsm_pattern != lum->lmm_pattern))
199 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
200 (strncmp(lsm->lsm_pool_name,
201 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
202 LOV_MAXPOOLNAME) != 0))
207 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
209 struct lov_user_md *lum)
211 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
214 *lmm_magic = lumv3->lmm_magic;
216 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
217 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
218 *lmm_magic = LOV_USER_MAGIC_V1;
219 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
220 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
222 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
223 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
225 lustre_swab_lov_user_md_v3(lumv3);
226 *lmm_magic = LOV_USER_MAGIC_V3;
227 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
229 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
230 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
236 void lov_stripe_lock(struct lov_stripe_md *md);
237 void lov_stripe_unlock(struct lov_stripe_md *md);
240 cfs_list_t typ_chain;
241 struct obd_ops *typ_dt_ops;
242 struct md_ops *typ_md_ops;
243 cfs_proc_dir_entry_t *typ_procroot;
246 struct lu_device_type *typ_lu;
247 spinlock_t obd_type_lock;
257 /* Individual type definitions */
259 struct ost_server_data;
261 struct osd_properties {
262 size_t osd_max_ea_size;
265 #define OBT_MAGIC 0xBDDECEAE
266 /* hold common fields for "target" device */
267 struct obd_device_target {
270 struct super_block *obt_sb;
271 /** last_rcvd file */
272 struct file *obt_rcvd_filp;
273 #ifdef HAVE_SERVER_SUPPORT
274 struct lu_target *obt_lut;
276 __u64 obt_mount_count;
277 struct rw_semaphore obt_rwsem;
278 struct vfsmount *obt_vfsmnt;
279 struct file *obt_health_check_filp;
280 struct osd_properties obt_osd_properties;
281 struct obd_job_stats obt_jobstats;
286 LLOG_CONFIG_ORIG_CTXT = 0,
287 LLOG_CONFIG_REPL_CTXT,
288 LLOG_MDS_OST_ORIG_CTXT,
289 LLOG_MDS_OST_REPL_CTXT,
296 LLOG_LOVEA_ORIG_CTXT,
297 LLOG_LOVEA_REPL_CTXT,
298 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
299 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
300 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
304 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
306 struct filter_subdirs {
307 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
317 /* NB this field MUST be first */
318 struct obd_device_target fo_obt;
319 const char *fo_fstype;
322 cfs_dentry_t *fo_dentry_O;
323 cfs_dentry_t **fo_dentry_O_groups;
324 struct filter_subdirs *fo_dentry_O_sub;
325 struct mutex fo_init_lock; /* group initialization lock*/
326 int fo_committed_group;
328 spinlock_t fo_objidlock; /* protect fo_lastobjid */
330 unsigned long fo_destroys_in_progress;
331 struct mutex fo_create_locks[FILTER_SUBDIR_COUNT];
333 cfs_list_t fo_export_list;
336 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
337 obd_size fo_tot_granted; /* all values in bytes */
338 obd_size fo_tot_pending;
339 int fo_tot_granted_clients;
341 obd_size fo_readcache_max_filesize;
342 spinlock_t fo_flags_lock;
343 unsigned int fo_read_cache:1, /**< enable read-only cache */
344 fo_writethrough_cache:1,/**< read cache writes */
345 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
346 fo_raid_degraded:1;/**< RAID device degraded */
348 struct obd_import *fo_mdc_imp;
349 struct obd_uuid fo_mdc_uuid;
350 struct lustre_handle fo_mdc_conn;
351 struct file **fo_last_objid_files;
352 __u64 *fo_last_objids; /* last created objid for groups,
353 * protected by fo_objidlock */
355 struct mutex fo_alloc_lock;
357 cfs_atomic_t fo_r_in_flight;
358 cfs_atomic_t fo_w_in_flight;
361 * per-filter pool of kiobuf's allocated by filter_common_setup() and
362 * torn down by filter_cleanup().
364 * This pool contains kiobuf used by
365 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
368 * Locking: protected by internal lock of cfs_hash, pool can be
369 * found from this hash table by t_id of ptlrpc_thread.
371 struct cfs_hash *fo_iobuf_hash;
373 cfs_list_t fo_llog_list;
374 spinlock_t fo_llog_list_lock;
376 struct brw_stats fo_filter_stats;
378 int fo_fmd_max_num; /* per exp filter_mod_data */
379 int fo_fmd_max_age; /* jiffies to fmd expiry */
380 unsigned long fo_syncjournal:1, /* sync journal on writes */
381 fo_sync_lock_cancel:2;/* sync on lock cancel */
385 rwlock_t fo_sptlrpc_lock;
386 struct sptlrpc_rule_set fo_sptlrpc_rset;
388 /* capability related */
389 unsigned int fo_fl_oss_capa;
390 cfs_list_t fo_capa_keys;
391 cfs_hlist_head_t *fo_capa_hash;
392 struct llog_commit_master *fo_lcm;
396 struct timeout_item {
397 enum timeout_event ti_event;
398 cfs_time_t ti_timeout;
401 cfs_list_t ti_obd_list;
405 #define OSC_MAX_RIF_DEFAULT 8
406 #define MDS_OSC_MAX_RIF_DEFAULT 50
407 #define OSC_MAX_RIF_MAX 256
408 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
409 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
410 #define OSC_DEFAULT_RESENDS 10
412 /* possible values for fo_sync_lock_cancel */
414 NEVER_SYNC_ON_CANCEL = 0,
415 BLOCKING_SYNC_ON_CANCEL = 1,
416 ALWAYS_SYNC_ON_CANCEL = 2,
417 NUM_SYNC_ON_CANCEL_STATES
420 #define MDC_MAX_RIF_DEFAULT 8
421 #define MDC_MAX_RIF_MAX 512
426 struct rw_semaphore cl_sem;
427 struct obd_uuid cl_target_uuid;
428 struct obd_import *cl_import; /* ptlrpc connection state */
430 /* max_mds_easize is purely a performance thing so we don't have to
431 * call obd_size_diskmd() all the time. */
432 int cl_default_mds_easize;
433 int cl_max_mds_easize;
434 int cl_max_mds_cookiesize;
436 enum lustre_sec_part cl_sp_me;
437 enum lustre_sec_part cl_sp_to;
438 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
440 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
441 void *cl_llcd_offset;
443 /* the grant values are protected by loi_list_lock below */
444 long cl_dirty; /* all _dirty_ in bytes */
445 long cl_dirty_max; /* allowed w/o rpc */
446 long cl_dirty_transit; /* dirty synchronous */
447 long cl_avail_grant; /* bytes of credit for ost */
448 long cl_lost_grant; /* lost credits (trunc) */
450 /* since we allocate grant by blocks, we don't know how many grant will
451 * be used to add a page into cache. As a solution, we reserve maximum
452 * grant before trying to dirty a page and unreserve the rest.
453 * See osc_{reserve|unreserve}_grant for details. */
454 long cl_reserved_grant;
455 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
456 cfs_time_t cl_next_shrink_grant; /* jiffies */
457 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
458 int cl_grant_shrink_interval; /* seconds */
460 /* A chunk is an optimal size used by osc_extent to determine
461 * the extent size. A chunk is max(CFS_PAGE_SIZE, OST block size) */
464 int cl_extent_tax; /* extent overhead, by bytes */
466 /* keep track of objects that have lois that contain pages which
467 * have been queued for async brw. this lock also protects the
468 * lists of osc_client_pages that hang off of the loi */
470 * ->cl_loi_list_lock protects consistency of
471 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
472 * ->ap_completion() call-backs are executed under this lock. As we
473 * cannot guarantee that these call-backs never block on all platforms
474 * (as a matter of fact they do block on Mac OS X), type of
475 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
476 * and blocking mutex on Mac OS X. (Alternative is to make this lock
477 * blocking everywhere, but we don't want to slow down fast-path of
478 * our main platform.)
480 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
481 * with client_obd_list_{un,}lock() and
482 * client_obd_list_lock_{init,done}() functions.
484 * NB by Jinshan: though field names are still _loi_, but actually
485 * osc_object{}s are in the list.
487 client_obd_lock_t cl_loi_list_lock;
488 cfs_list_t cl_loi_ready_list;
489 cfs_list_t cl_loi_hp_ready_list;
490 cfs_list_t cl_loi_write_list;
491 cfs_list_t cl_loi_read_list;
494 /* just a sum of the loi/lop pending numbers to be exported by /proc */
495 cfs_atomic_t cl_pending_w_pages;
496 cfs_atomic_t cl_pending_r_pages;
497 __u32 cl_max_pages_per_rpc;
498 int cl_max_rpcs_in_flight;
499 struct obd_histogram cl_read_rpc_hist;
500 struct obd_histogram cl_write_rpc_hist;
501 struct obd_histogram cl_read_page_hist;
502 struct obd_histogram cl_write_page_hist;
503 struct obd_histogram cl_read_offset_hist;
504 struct obd_histogram cl_write_offset_hist;
506 /* lru for osc caching pages */
507 struct cl_client_cache *cl_cache;
508 cfs_list_t cl_lru_osc; /* member of cl_cache->ccc_lru */
509 cfs_atomic_t *cl_lru_left;
510 cfs_atomic_t cl_lru_busy;
511 cfs_atomic_t cl_lru_shrinkers;
512 cfs_atomic_t cl_lru_in_list;
513 cfs_list_t cl_lru_list; /* lru page list */
514 client_obd_lock_t cl_lru_list_lock; /* page list protector */
516 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
517 cfs_atomic_t cl_destroy_in_flight;
518 cfs_waitq_t cl_destroy_waitq;
520 struct mdc_rpc_lock *cl_rpc_lock;
521 struct mdc_rpc_lock *cl_close_lock;
524 struct semaphore cl_mgc_sem;
525 struct vfsmount *cl_mgc_vfsmnt;
526 struct dentry *cl_mgc_configs_dir;
527 cfs_atomic_t cl_mgc_refcount;
528 struct obd_export *cl_mgc_mgsexp;
530 /* checksumming for data sent over the network */
531 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
532 /* supported checksum types that are worked out at connect time */
533 __u32 cl_supp_cksum_types;
534 /* checksum algorithm to be used */
535 cksum_type_t cl_cksum_type;
537 /* also protected by the poorly named _loi_list_lock lock above */
538 struct osc_async_rc cl_ar;
540 /* used by quotacheck when the servers are older than 2.4 */
541 int cl_qchk_stat; /* quotacheck stat of the peer */
542 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
543 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
544 #warning "please consider removing quotacheck compatibility code"
547 /* sequence manager */
548 struct lu_client_seq *cl_seq;
550 cfs_atomic_t cl_resends; /* resend count */
552 /* ptlrpc work for writeback in ptlrpcd context */
553 void *cl_writeback_work;
554 /* hash tables for osc_quota_info */
555 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
557 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
567 struct obd_device_target eo_obt;
571 struct lustre_handle eo_nl_lock;
572 cfs_atomic_t eo_prep;
576 struct ptlrpc_service *ost_service;
577 struct ptlrpc_service *ost_create_service;
578 struct ptlrpc_service *ost_io_service;
579 struct ptlrpc_service *ost_seq_service;
580 struct mutex ost_health_mutex;
583 struct echo_client_obd {
584 struct obd_export *ec_exp; /* the local connection to osc/lov */
586 cfs_list_t ec_objects;
593 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
594 cfs_list_t lqo_oss_list; /* link to lov_qos */
595 __u64 lqo_bavail; /* total bytes avail on OSS */
596 __u64 lqo_penalty; /* current penalty */
597 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
598 time_t lqo_used; /* last used time, seconds */
599 __u32 lqo_ost_count; /* number of osts on this oss */
603 struct lov_qos_oss *ltq_oss; /* oss info */
604 __u64 ltq_penalty; /* current penalty */
605 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
606 __u64 ltq_weight; /* net weighting */
607 time_t ltq_used; /* last used time, seconds */
608 unsigned int ltq_usable:1; /* usable for striping */
611 /* Generic subset of OSTs */
613 __u32 *op_array; /* array of index of
615 unsigned int op_count; /* number of OSTs in the array */
616 unsigned int op_size; /* allocated size of lp_array */
617 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
620 /* Round-robin allocator data */
622 __u32 lqr_start_idx; /* start index of new inode */
623 __u32 lqr_offset_idx; /* aliasing for start_idx */
624 int lqr_start_count; /* reseed counter */
625 struct ost_pool lqr_pool; /* round-robin optimized list */
626 unsigned long lqr_dirty:1; /* recalc round-robin list */
629 /* allow statfs data caching for 1 second */
630 #define OBD_STATFS_CACHE_SECONDS 1
632 struct lov_statfs_data {
633 struct obd_info lsd_oi;
634 struct obd_statfs lsd_statfs;
636 /* Stripe placement optimization */
638 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
639 struct rw_semaphore lq_rw_sem;
640 __u32 lq_active_oss_count;
641 unsigned int lq_prio_free; /* priority for free space */
642 unsigned int lq_threshold_rr;/* priority for rr */
643 struct lov_qos_rr lq_rr; /* round robin qos data */
644 unsigned long lq_dirty:1, /* recalc qos data */
645 lq_same_space:1,/* the ost's all have approx.
646 the same space avail */
647 lq_reset:1, /* zero current penalties */
648 lq_statfs_in_progress:1; /* statfs op in
650 /* qos statfs data */
651 struct lov_statfs_data *lq_statfs_data;
652 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
653 * requests completion */
656 struct lov_tgt_desc {
658 struct obd_uuid ltd_uuid;
659 struct obd_device *ltd_obd;
660 struct obd_export *ltd_exp;
661 struct ltd_qos ltd_qos; /* qos info per target */
663 __u32 ltd_index; /* index in lov_obd->tgts */
664 unsigned long ltd_active:1,/* is this target up for requests */
665 ltd_activate:1,/* should target be activated */
666 ltd_reap:1; /* should this target be deleted */
670 #define pool_tgt_size(_p) _p->pool_obds.op_size
671 #define pool_tgt_count(_p) _p->pool_obds.op_count
672 #define pool_tgt_array(_p) _p->pool_obds.op_array
673 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
676 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
677 struct ost_pool pool_obds; /* pool members */
678 cfs_atomic_t pool_refcount; /* pool ref. counter */
679 struct lov_qos_rr pool_rr; /* round robin qos */
680 cfs_hlist_node_t pool_hash; /* access by poolname */
681 cfs_list_t pool_list; /* serial access */
682 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
683 struct obd_device *pool_lobd; /* obd of the lov/lod to which
684 * this pool belongs */
688 struct lov_desc desc;
689 struct lov_tgt_desc **lov_tgts; /* sparse array */
690 struct ost_pool lov_packed; /* all OSTs in a packed
692 struct mutex lov_lock;
693 struct obd_connect_data lov_ocd;
694 cfs_atomic_t lov_refcount;
695 __u32 lov_tgt_count; /* how many OBD's */
696 __u32 lov_active_tgt_count; /* how many active */
697 __u32 lov_death_row;/* tgts scheduled to be deleted */
698 __u32 lov_tgt_size; /* size of tgts array */
701 cfs_hash_t *lov_pools_hash_body; /* used for key access */
702 cfs_list_t lov_pool_list; /* used for sequential access */
703 cfs_proc_dir_entry_t *lov_pool_proc_entry;
704 enum lustre_sec_part lov_sp_me;
706 /* Cached LRU pages from upper layer */
710 struct lmv_tgt_desc {
711 struct obd_uuid ltd_uuid;
712 struct obd_export *ltd_exp;
714 struct mutex ltd_fid_mutex;
715 unsigned long ltd_active:1; /* target up for requests */
718 enum placement_policy {
719 PLACEMENT_CHAR_POLICY = 0,
720 PLACEMENT_NID_POLICY = 1,
721 PLACEMENT_INVAL_POLICY = 2,
725 typedef enum placement_policy placement_policy_t;
729 struct lu_client_fld lmv_fld;
731 placement_policy_t lmv_placement;
732 struct lmv_desc desc;
733 struct obd_uuid cluuid;
734 struct obd_export *exp;
736 struct mutex init_mutex;
743 int tgts_size; /* size of tgts array */
744 struct lmv_tgt_desc **tgts;
746 struct obd_connect_data conn_data;
749 struct niobuf_local {
750 __u64 lnb_file_offset;
751 __u32 lnb_page_offset;
755 cfs_dentry_t *dentry;
760 #define LUSTRE_FLD_NAME "fld"
761 #define LUSTRE_SEQ_NAME "seq"
763 #define LUSTRE_MDD_NAME "mdd"
764 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
765 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
766 #define LUSTRE_VVP_NAME "vvp"
767 #define LUSTRE_LMV_NAME "lmv"
768 #define LUSTRE_SLP_NAME "slp"
769 #define LUSTRE_LOD_NAME "lod"
770 #define LUSTRE_OSP_NAME "osp"
771 #define LUSTRE_LWP_NAME "lwp"
773 /* obd device type names */
774 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
775 #define LUSTRE_MDS_NAME "mds"
776 #define LUSTRE_MDT_NAME "mdt"
777 #define LUSTRE_MDC_NAME "mdc"
778 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
779 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
780 #define LUSTRE_OSC_NAME "osc"
781 #define LUSTRE_LOV_NAME "lov"
782 #define LUSTRE_MGS_NAME "mgs"
783 #define LUSTRE_MGC_NAME "mgc"
785 #define LUSTRE_ECHO_NAME "obdecho"
786 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
787 #define LUSTRE_QMT_NAME "qmt"
789 /* Constant obd names (post-rename) */
790 #define LUSTRE_MDS_OBDNAME "MDS"
791 #define LUSTRE_OSS_OBDNAME "OSS"
792 #define LUSTRE_MGS_OBDNAME "MGS"
793 #define LUSTRE_MGC_OBDNAME "MGC"
795 static inline int is_osp_on_mdt(char *name)
799 ptr = strrchr(name, '-');
801 CERROR("%s is not a obdname\n", name);
805 /* 1.8 OSC/OSP name on MDT is fsname-OSTxxxx-osc */
806 if (strncmp(ptr + 1, "osc", 3) == 0)
809 if (strncmp(ptr + 1, "MDT", 3) != 0)
812 while (*(--ptr) != '-' && ptr != name);
817 if (strncmp(ptr + 1, LUSTRE_OSP_NAME, strlen(LUSTRE_OSP_NAME)) != 0 &&
818 strncmp(ptr + 1, LUSTRE_OSC_NAME, strlen(LUSTRE_OSC_NAME)) != 0)
824 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
825 #define N_LOCAL_TEMP_PAGE 0x10000000
827 struct obd_trans_info {
830 /* Only used on the server side for tracking acks. */
831 struct oti_req_ack_lock {
832 struct lustre_handle lock;
836 struct llog_cookie oti_onecookie;
837 struct llog_cookie *oti_logcookies;
839 /** synchronous write is needed */
840 unsigned long oti_sync_write:1;
842 /* initial thread handling transaction */
843 struct ptlrpc_thread * oti_thread;
846 __u64 oti_pre_version;
850 struct obd_uuid *oti_ost_uuid;
853 static inline void oti_init(struct obd_trans_info *oti,
854 struct ptlrpc_request *req)
858 memset(oti, 0, sizeof(*oti));
863 oti->oti_xid = req->rq_xid;
864 /** VBR: take versions from request */
865 if (req->rq_reqmsg != NULL &&
866 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
867 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
868 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
869 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
872 /** called from mds_create_objects */
873 if (req->rq_repmsg != NULL)
874 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
875 oti->oti_thread = req->rq_svc_thread;
876 if (req->rq_reqmsg != NULL)
877 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
880 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
885 if (num_cookies == 1)
886 oti->oti_logcookies = &oti->oti_onecookie;
888 OBD_ALLOC_LARGE(oti->oti_logcookies,
889 num_cookies * sizeof(oti->oti_onecookie));
891 oti->oti_numcookies = num_cookies;
894 static inline void oti_free_cookies(struct obd_trans_info *oti)
896 if (!oti || !oti->oti_logcookies)
899 if (oti->oti_logcookies == &oti->oti_onecookie)
900 LASSERT(oti->oti_numcookies == 1);
902 OBD_FREE_LARGE(oti->oti_logcookies,
903 oti->oti_numcookies*sizeof(oti->oti_onecookie));
904 oti->oti_logcookies = NULL;
905 oti->oti_numcookies = 0;
909 * Events signalled through obd_notify() upcall-chain.
911 enum obd_notify_event {
914 /* Device connect start */
916 /* Device activated */
918 /* Device deactivated */
920 /* Device disconnected */
922 /* Connect data for import were changed */
925 OBD_NOTIFY_SYNC_NONBLOCK,
927 /* Configuration event */
929 /* Administratively deactivate/activate event */
930 OBD_NOTIFY_DEACTIVATE,
934 /* bit-mask flags for config events */
936 CONFIG_LOG = 0x1, /* finished processing config log */
937 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
938 CONFIG_TARGET = 0x4 /* one target is added */
942 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
943 * and liblustre being main examples).
945 struct obd_notify_upcall {
946 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
947 enum obd_notify_event ev, void *owner, void *data);
948 /* Opaque datum supplied by upper layer listener */
952 struct target_recovery_data {
953 svc_handler_t trd_recovery_handler;
954 pid_t trd_processing_task;
955 struct completion trd_starting;
956 struct completion trd_finishing;
960 * In HEAD for CMD, the object is created in group number which is 3>=
961 * or indexing starts from 3. To test this assertions are added to disallow
962 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
963 * object in group 0 needs to be allowed.
964 * So for interop mode following changes needs to be done:
965 * 1. No need to assert on group 0 or allow group 0
966 * 2. The group number indexing starts from 0 instead of 3
969 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
971 static inline __u64 objseq_to_mdsno(obd_seq seq)
973 LASSERT_SEQ_IS_MDT(seq);
974 if (seq == FID_SEQ_OST_MDT0)
976 return seq - FID_SEQ_OST_MDT1 + 1;
979 static inline int mdt_to_obd_objseq(int mdtid)
982 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
986 return FID_SEQ_OST_MDT1 + mdtid - 1;
990 struct obd_llog_group {
993 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
994 cfs_waitq_t olg_waitq;
996 struct obd_export *olg_exp;
997 int olg_initializing;
998 struct mutex olg_cat_processing;
1001 /* corresponds to one of the obd's */
1002 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1003 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1006 struct obd_type *obd_type;
1009 /* common and UUID name of this device */
1010 char obd_name[MAX_OBD_NAME];
1011 struct obd_uuid obd_uuid;
1013 struct lu_device *obd_lu_dev;
1016 /* bitfield modification is protected by obd_dev_lock */
1017 unsigned long obd_attached:1, /* finished attach */
1018 obd_set_up:1, /* finished setup */
1019 obd_recovering:1, /* there are recoverable clients */
1020 obd_abort_recovery:1,/* recovery expired */
1021 obd_version_recov:1, /* obd uses version checking */
1022 obd_replayable:1, /* recovery is enabled; inform clients */
1023 obd_no_transno:1, /* no committed-transno notification */
1024 obd_no_recov:1, /* fail instead of retry messages */
1025 obd_stopping:1, /* started cleanup */
1026 obd_starting:1, /* started setup */
1027 obd_force:1, /* cleanup with > 0 obd refcount */
1028 obd_fail:1, /* cleanup with failover */
1029 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1030 obd_no_conn:1, /* deny new connections */
1031 obd_inactive:1, /* device active/inactive
1032 * (for /proc/status only!!) */
1033 obd_no_ir:1, /* no imperative recovery. */
1034 obd_process_conf:1; /* device is processing mgs config */
1035 /* use separate field as it is set in interrupt to don't mess with
1036 * protection of other bits using _bh lock */
1037 unsigned long obd_recovery_expired:1;
1038 /* uuid-export hash body */
1039 cfs_hash_t *obd_uuid_hash;
1040 /* nid-export hash body */
1041 cfs_hash_t *obd_nid_hash;
1042 /* nid stats body */
1043 cfs_hash_t *obd_nid_stats_hash;
1044 cfs_list_t obd_nid_stats;
1045 cfs_atomic_t obd_refcount;
1046 cfs_waitq_t obd_refcount_waitq;
1047 cfs_list_t obd_exports;
1048 cfs_list_t obd_unlinked_exports;
1049 cfs_list_t obd_delayed_exports;
1050 int obd_num_exports;
1051 spinlock_t obd_nid_lock;
1052 struct ldlm_namespace *obd_namespace;
1053 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1054 /* a spinlock is OK for what we do now, may need a semaphore later */
1055 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
1056 struct mutex obd_dev_mutex;
1057 __u64 obd_last_committed;
1058 struct fsfilt_operations *obd_fsops;
1059 spinlock_t obd_osfs_lock;
1060 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1062 struct lvfs_run_ctxt obd_lvfs_ctxt;
1063 struct obd_llog_group obd_olg; /* default llog group */
1064 struct obd_device *obd_observer;
1065 struct rw_semaphore obd_observer_link_sem;
1066 struct obd_notify_upcall obd_upcall;
1067 struct obd_export *obd_self_export;
1068 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1069 cfs_list_t obd_exports_timed;
1070 time_t obd_eviction_timer; /* for ping evictor */
1072 int obd_max_recoverable_clients;
1073 cfs_atomic_t obd_connected_clients;
1074 int obd_stale_clients;
1075 int obd_delayed_clients;
1076 /* this lock protects all recovery list_heads, timer and
1077 * obd_next_recovery_transno value */
1078 spinlock_t obd_recovery_task_lock;
1079 __u64 obd_next_recovery_transno;
1080 int obd_replayed_requests;
1081 int obd_requests_queued_for_recovery;
1082 cfs_waitq_t obd_next_transno_waitq;
1083 /* protected by obd_recovery_task_lock */
1084 cfs_timer_t obd_recovery_timer;
1085 time_t obd_recovery_start; /* seconds */
1086 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1087 int obd_recovery_time_hard;
1088 int obd_recovery_timeout;
1089 int obd_recovery_ir_factor;
1091 /* new recovery stuff from CMD2 */
1092 struct target_recovery_data obd_recovery_data;
1093 int obd_replayed_locks;
1094 cfs_atomic_t obd_req_replay_clients;
1095 cfs_atomic_t obd_lock_replay_clients;
1096 /* all lists are protected by obd_recovery_task_lock */
1097 cfs_list_t obd_req_replay_queue;
1098 cfs_list_t obd_lock_replay_queue;
1099 cfs_list_t obd_final_req_queue;
1100 int obd_recovery_stage;
1103 struct obd_device_target obt;
1104 struct filter_obd filter;
1105 struct client_obd cli;
1107 struct echo_client_obd echo_client;
1108 struct echo_obd echo;
1112 /* Fields used by LProcFS */
1113 unsigned int obd_cntr_base;
1114 struct lprocfs_stats *obd_stats;
1116 unsigned int md_cntr_base;
1117 struct lprocfs_stats *md_stats;
1119 cfs_proc_dir_entry_t *obd_proc_entry;
1120 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1121 cfs_proc_dir_entry_t *obd_svc_procroot;
1122 struct lprocfs_stats *obd_svc_stats;
1123 cfs_atomic_t obd_evict_inprogress;
1124 cfs_waitq_t obd_evict_inprogress_waitq;
1125 cfs_list_t obd_evict_list; /* protected with pet_lock */
1128 * Ldlm pool part. Save last calculated SLV and Limit.
1130 rwlock_t obd_pool_lock;
1135 * A list of outstanding class_incref()'s against this obd. For
1138 struct lu_ref obd_reference;
1140 int obd_conn_inprogress;
1143 #define OBD_LLOG_FL_SENDNOW 0x0001
1144 #define OBD_LLOG_FL_EXIT 0x0002
1146 enum obd_cleanup_stage {
1147 /* Special case hack for MDS LOVs */
1149 /* can be directly mapped to .ldto_device_fini() */
1150 OBD_CLEANUP_EXPORTS,
1153 /* get/set_info keys */
1154 #define KEY_ASYNC "async"
1155 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1156 #define KEY_BLOCKSIZE "blocksize"
1157 #define KEY_CAPA_KEY "capa_key"
1158 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1159 #define KEY_FID2PATH "fid2path"
1160 #define KEY_CHECKSUM "checksum"
1161 #define KEY_CLEAR_FS "clear_fs"
1162 #define KEY_CONN_DATA "conn_data"
1163 #define KEY_EVICT_BY_NID "evict_by_nid"
1164 #define KEY_FIEMAP "fiemap"
1165 #define KEY_FLUSH_CTX "flush_ctx"
1166 #define KEY_GRANT_SHRINK "grant_shrink"
1167 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1168 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1169 #define KEY_INIT_RECOV "initial_recov"
1170 #define KEY_INTERMDS "inter_mds"
1171 #define KEY_LAST_ID "last_id"
1172 #define KEY_LAST_FID "last_fid"
1173 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1174 #define KEY_LOVDESC "lovdesc"
1175 #define KEY_LOV_IDX "lov_idx"
1176 #define KEY_MAX_EASIZE "max_easize"
1177 #define KEY_MDS_CONN "mds_conn"
1178 #define KEY_MGSSEC "mgssec"
1179 #define KEY_NEXT_ID "next_id"
1180 #define KEY_READ_ONLY "read-only"
1181 #define KEY_REGISTER_TARGET "register_target"
1182 #define KEY_SET_FS "set_fs"
1183 #define KEY_TGT_COUNT "tgt_count"
1184 /* KEY_SET_INFO in lustre_idl.h */
1185 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1186 #define KEY_CONNECT_FLAG "connect_flags"
1187 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1189 #define KEY_CACHE_SET "cache_set"
1190 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
1194 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1195 #define IT_OPEN (1 << 0)
1196 #define IT_CREAT (1 << 1)
1197 #define IT_READDIR (1 << 2)
1198 #define IT_GETATTR (1 << 3)
1199 #define IT_LOOKUP (1 << 4)
1200 #define IT_UNLINK (1 << 5)
1201 #define IT_TRUNC (1 << 6)
1202 #define IT_GETXATTR (1 << 7)
1203 #define IT_EXEC (1 << 8)
1204 #define IT_PIN (1 << 9)
1205 #define IT_LAYOUT (1 << 10)
1206 #define IT_QUOTA_DQACQ (1 << 11)
1207 #define IT_QUOTA_CONN (1 << 12)
1209 static inline int it_to_lock_mode(struct lookup_intent *it)
1211 /* CREAT needs to be tested before open (both could be set) */
1212 if (it->it_op & IT_CREAT)
1214 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1218 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1223 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1224 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1225 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1226 struct lu_fid op_fid4; /* to the operation locks. */
1227 mdsno_t op_mds; /* what mds server open will go to */
1228 struct lustre_handle op_handle;
1229 obd_time op_mod_time;
1230 const char *op_name;
1233 struct lmv_stripe_md *op_mea1;
1234 struct lmv_stripe_md *op_mea2;
1235 __u32 op_suppgids[2];
1241 /* iattr fields and blocks. */
1242 struct iattr op_attr;
1244 unsigned int op_attr_flags;
1247 loff_t op_attr_blocks;
1249 /* Size-on-MDS epoch and flags. */
1254 struct obd_capa *op_capa1;
1255 struct obd_capa *op_capa2;
1257 /* Various operation flags. */
1260 /* Operation type */
1263 /* Used by readdir */
1266 /* Used by readdir */
1269 /* used to transfer info between the stacks of MD client
1270 * see enum op_cli_flags */
1275 CLI_SET_MEA = 1 << 0,
1276 CLI_RM_ENTRY = 1 << 1,
1279 struct md_enqueue_info;
1280 /* metadata stat-ahead */
1281 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1282 struct md_enqueue_info *minfo,
1285 /* seq client type */
1287 LUSTRE_SEQ_METADATA = 1,
1291 struct md_enqueue_info {
1292 struct md_op_data mi_data;
1293 struct lookup_intent mi_it;
1294 struct lustre_handle mi_lockh;
1295 struct inode *mi_dir;
1296 md_enqueue_cb_t mi_cb;
1298 unsigned int mi_generation;
1302 cfs_module_t *o_owner;
1303 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1304 void *karg, void *uarg);
1305 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1306 __u32 keylen, void *key, __u32 *vallen, void *val,
1307 struct lov_stripe_md *lsm);
1308 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1309 __u32 keylen, void *key,
1310 __u32 vallen, void *val,
1311 struct ptlrpc_request_set *set);
1312 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1313 int (*o_detach)(struct obd_device *dev);
1314 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1315 int (*o_precleanup)(struct obd_device *dev,
1316 enum obd_cleanup_stage cleanup_stage);
1317 int (*o_cleanup)(struct obd_device *dev);
1318 int (*o_process_config)(struct obd_device *dev, obd_count len,
1320 int (*o_postrecov)(struct obd_device *dev);
1321 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1323 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1324 /* connect to the target device with given connection
1325 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1326 * granted by the target, which are guaranteed to be a subset of flags
1327 * asked for. If @ocd == NULL, use default parameters. */
1328 int (*o_connect)(const struct lu_env *env,
1329 struct obd_export **exp, struct obd_device *src,
1330 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1332 int (*o_reconnect)(const struct lu_env *env,
1333 struct obd_export *exp, struct obd_device *src,
1334 struct obd_uuid *cluuid,
1335 struct obd_connect_data *ocd,
1337 int (*o_disconnect)(struct obd_export *exp);
1339 /* Initialize/finalize fids infrastructure. */
1340 int (*o_fid_init)(struct obd_device *obd,
1341 struct obd_export *exp, enum lu_cli_type type);
1342 int (*o_fid_fini)(struct obd_device *obd);
1344 /* Allocate new fid according to passed @hint. */
1345 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1346 struct md_op_data *op_data);
1349 * Object with @fid is getting deleted, we may want to do something
1352 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1353 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1354 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1355 __u64 max_age, struct ptlrpc_request_set *set);
1356 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1357 struct lov_stripe_md *mem_src);
1358 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1359 struct lov_mds_md *disk_src, int disk_len);
1360 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1362 /* FIXME: add fid capability support for create & destroy! */
1363 int (*o_precreate)(struct obd_export *exp);
1364 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1365 struct obdo *oa, struct lov_stripe_md **ea,
1366 struct obd_trans_info *oti);
1367 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1368 struct lov_stripe_md **ea,
1369 struct obd_trans_info *oti);
1370 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1371 struct obdo *oa, struct lov_stripe_md *ea,
1372 struct obd_trans_info *oti, struct obd_export *md_exp,
1374 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1375 struct obd_info *oinfo, struct obd_trans_info *oti);
1376 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1377 struct obd_trans_info *oti,
1378 struct ptlrpc_request_set *rqset);
1379 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1380 struct obd_info *oinfo);
1381 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1382 struct ptlrpc_request_set *set);
1383 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1384 obd_count oa_bufs, struct brw_page *pgarr,
1385 struct obd_trans_info *oti);
1386 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1387 struct ost_lvb *lvb, int kms_only);
1388 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1389 obd_off size, int shrink);
1390 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1391 struct obd_info *oinfo, struct obd_trans_info *oti,
1392 struct ptlrpc_request_set *rqset);
1393 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1394 struct obd_info *oinfo, obd_size start, obd_size end,
1395 struct ptlrpc_request_set *set);
1396 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1397 struct lov_stripe_md *src, obd_size start,
1398 obd_size end, struct obd_trans_info *oti);
1399 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1400 struct lustre_handle *srconn, struct lov_stripe_md *src,
1401 obd_size start, obd_size end, struct obd_trans_info *);
1402 int (*o_iterate)(struct lustre_handle *conn,
1403 int (*)(obd_id, obd_seq, void *),
1404 obd_id *startid, obd_seq seq, void *data);
1405 int (*o_preprw)(const struct lu_env *env, int cmd,
1406 struct obd_export *exp, struct obdo *oa, int objcount,
1407 struct obd_ioobj *obj, struct niobuf_remote *remote,
1408 int *nr_pages, struct niobuf_local *local,
1409 struct obd_trans_info *oti, struct lustre_capa *capa);
1410 int (*o_commitrw)(const struct lu_env *env, int cmd,
1411 struct obd_export *exp, struct obdo *oa,
1412 int objcount, struct obd_ioobj *obj,
1413 struct niobuf_remote *remote, int pages,
1414 struct niobuf_local *local,
1415 struct obd_trans_info *oti, int rc);
1416 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1417 struct ldlm_enqueue_info *einfo,
1418 struct ptlrpc_request_set *rqset);
1419 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1420 ldlm_iterator_t it, void *data);
1421 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1422 ldlm_iterator_t it, void *data);
1423 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1424 __u32 mode, struct lustre_handle *);
1425 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1426 ldlm_cancel_flags_t flags, void *opaque);
1427 int (*o_init_export)(struct obd_export *exp);
1428 int (*o_destroy_export)(struct obd_export *exp);
1429 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1430 int cmd, obd_off *);
1432 /* llog related obd_methods */
1433 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1434 struct obd_device *disk_obd, int *idx);
1435 int (*o_llog_finish)(struct obd_device *obd, int count);
1436 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1438 /* metadata-only methods */
1439 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1440 struct obd_capa *, struct obd_client_handle *, int flag);
1441 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1443 int (*o_import_event)(struct obd_device *, struct obd_import *,
1444 enum obd_import_event);
1446 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1447 enum obd_notify_event ev, void *data);
1449 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1450 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1453 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1454 struct obd_quotactl *);
1455 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1456 struct obd_quotactl *);
1458 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1461 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1462 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1463 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1465 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1467 void (*o_getref)(struct obd_device *obd);
1468 void (*o_putref)(struct obd_device *obd);
1470 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1471 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1472 * Also, add a wrapper function in include/linux/obd_class.h. */
1476 LUSTRE_OPC_MKDIR = (1 << 0),
1477 LUSTRE_OPC_SYMLINK = (1 << 1),
1478 LUSTRE_OPC_MKNOD = (1 << 2),
1479 LUSTRE_OPC_CREATE = (1 << 3),
1480 LUSTRE_OPC_ANY = (1 << 4)
1483 /* lmv structures */
1484 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1485 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1486 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1488 #define MAX_HASH_SIZE_32 0x7fffffffUL
1489 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1490 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1493 struct mdt_body *body;
1494 struct lov_stripe_md *lsm;
1495 struct lmv_stripe_md *mea;
1496 #ifdef CONFIG_FS_POSIX_ACL
1497 struct posix_acl *posix_acl;
1499 struct mdt_remote_perm *remote_perm;
1500 struct obd_capa *mds_capa;
1501 struct obd_capa *oss_capa;
1504 struct md_open_data {
1505 struct obd_client_handle *mod_och;
1506 struct ptlrpc_request *mod_open_req;
1507 struct ptlrpc_request *mod_close_req;
1508 cfs_atomic_t mod_refcount;
1511 struct lookup_intent;
1514 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1515 struct obd_capa **);
1516 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1517 ldlm_iterator_t, void *);
1518 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1519 ldlm_iterator_t, void *);
1520 int (*m_close)(struct obd_export *, struct md_op_data *,
1521 struct md_open_data *, struct ptlrpc_request **);
1522 int (*m_create)(struct obd_export *, struct md_op_data *,
1523 const void *, int, int, __u32, __u32, cfs_cap_t,
1524 __u64, struct ptlrpc_request **);
1525 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1526 struct md_open_data *);
1527 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1528 struct lookup_intent *, struct md_op_data *,
1529 struct lustre_handle *, void *, int,
1530 struct ptlrpc_request **, __u64);
1531 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1532 struct ptlrpc_request **);
1533 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1534 struct ptlrpc_request **);
1535 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1536 void *, int, struct lookup_intent *, int,
1537 struct ptlrpc_request **,
1538 ldlm_blocking_callback, __u64);
1539 int (*m_link)(struct obd_export *, struct md_op_data *,
1540 struct ptlrpc_request **);
1541 int (*m_rename)(struct obd_export *, struct md_op_data *,
1542 const char *, int, const char *, int,
1543 struct ptlrpc_request **);
1544 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1545 const struct lu_fid *,
1546 struct ptlrpc_request **);
1547 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1548 int , void *, int, struct ptlrpc_request **,
1549 struct md_open_data **mod);
1550 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1551 struct obd_capa *, struct ptlrpc_request **);
1552 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1553 struct page **, struct ptlrpc_request **);
1555 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1556 struct ptlrpc_request **);
1558 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1559 struct obd_capa *, obd_valid, const char *,
1560 const char *, int, int, int, __u32,
1561 struct ptlrpc_request **);
1563 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1564 struct obd_capa *, obd_valid, const char *,
1565 const char *, int, int, int,
1566 struct ptlrpc_request **);
1568 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1570 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1571 struct obd_export *, struct obd_export *,
1572 struct lustre_md *);
1574 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1576 int (*m_set_open_replay_data)(struct obd_export *,
1577 struct obd_client_handle *,
1578 struct ptlrpc_request *);
1579 int (*m_clear_open_replay_data)(struct obd_export *,
1580 struct obd_client_handle *);
1581 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1583 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1584 const struct lu_fid *, ldlm_type_t,
1585 ldlm_policy_data_t *, ldlm_mode_t,
1586 struct lustre_handle *);
1588 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1589 ldlm_policy_data_t *, ldlm_mode_t,
1590 ldlm_cancel_flags_t flags, void *opaque);
1591 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1592 renew_capa_cb_t cb);
1593 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1594 const struct req_msg_field *, struct obd_capa **);
1596 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1597 struct obd_capa *, __u32,
1598 struct ptlrpc_request **);
1600 int (*m_intent_getattr_async)(struct obd_export *,
1601 struct md_enqueue_info *,
1602 struct ldlm_enqueue_info *);
1604 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1605 struct lu_fid *, __u64 *bits);
1608 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1609 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1610 * wrapper function in include/linux/obd_class.h.
1614 struct lsm_operations {
1615 void (*lsm_free)(struct lov_stripe_md *);
1616 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1617 struct obd_export *md_exp);
1618 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1620 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1622 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1623 __u16 *stripe_count);
1624 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1625 struct lov_mds_md *lmm);
1628 extern const struct lsm_operations lsm_v1_ops;
1629 extern const struct lsm_operations lsm_v3_ops;
1630 static inline const struct lsm_operations *lsm_op_find(int magic)
1638 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1643 /* Requests for obd_extent_calc() */
1644 #define OBD_CALC_STRIPE_START 1
1645 #define OBD_CALC_STRIPE_END 2
1647 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1649 return oinfo->oi_capa;
1652 static inline struct md_open_data *obd_mod_alloc(void)
1654 struct md_open_data *mod;
1658 cfs_atomic_set(&mod->mod_refcount, 1);
1662 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1663 #define obd_mod_put(mod) \
1665 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1666 if ((mod)->mod_open_req) \
1667 ptlrpc_req_finished((mod)->mod_open_req); \
1668 OBD_FREE_PTR(mod); \
1672 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1673 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1675 /* return 1 if client should be resend request */
1676 static inline int client_should_resend(int resend, struct client_obd *cli)
1678 return cfs_atomic_read(&cli->cl_resends) ?
1679 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1683 * Return device name for this device
1685 * XXX: lu_device is declared before obd_device, while a pointer pointing
1686 * back to obd_device in lu_device, so this helper function defines here
1687 * instead of in lu_object.h
1689 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1691 return lu_dev->ld_obd->obd_name;
1694 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1699 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1702 /* caller does not care of idx */
1706 /* volatile file, the MDT can be set from name */
1707 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1708 /* if no MDT is specified, use std way */
1709 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1711 /* test for no MDT idx case */
1712 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1713 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1717 /* we have an idx, read it */
1718 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1719 *idx = strtoul(start, &end, 0);
1721 * no digit, no trailing :, negative value
1723 if (((*idx == 0) && (end == start)) ||
1724 (*end != ':') || (*idx < 0))
1729 /* bad format of mdt idx, we cannot return an error
1730 * to caller so we use hash algo */
1731 CERROR("Bad volatile file name format: %s\n",
1732 name + LUSTRE_VOLATILE_HDR_LEN);
1736 static inline int cli_brw_size(struct obd_device *obd)
1738 LASSERT(obd != NULL);
1739 return obd->u.cli.cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
1742 #endif /* __OBD_H */