1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
43 #if defined(__linux__)
44 #include <linux/obd.h>
45 #elif defined(__APPLE__)
46 #include <darwin/obd.h>
47 #elif defined(__WINNT__)
48 #include <winnt/obd.h>
50 #error Unsupported operating system.
53 #define IOC_OSC_TYPE 'h'
54 #define IOC_OSC_MIN_NR 20
55 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
56 #define IOC_OSC_MAX_NR 50
58 #define IOC_MDC_TYPE 'i'
59 #define IOC_MDC_MIN_NR 20
60 #define IOC_MDC_MAX_NR 50
62 #include <lustre/lustre_idl.h>
63 #include <lu_target.h>
65 #include <lustre_lib.h>
66 #include <lustre_export.h>
67 #include <lustre_quota.h>
68 #include <lustre_fld.h>
69 #include <lustre_capa.h>
71 #include <libcfs/bitmap.h>
74 #define MAX_OBD_DEVICES 8192
76 /* this is really local to the OSC */
77 struct loi_oap_pages {
78 cfs_list_t lop_pending;
79 cfs_list_t lop_urgent;
80 cfs_list_t lop_pending_group;
90 struct lov_oinfo { /* per-stripe data structure */
91 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
92 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
93 int loi_ost_gen; /* generation of this loi_ost_idx */
95 /* used by the osc to keep track of what objects to build into rpcs */
96 struct loi_oap_pages loi_read_lop;
97 struct loi_oap_pages loi_write_lop;
98 cfs_list_t loi_ready_item;
99 cfs_list_t loi_hp_ready_item;
100 cfs_list_t loi_write_item;
101 cfs_list_t loi_read_item;
103 unsigned long loi_kms_valid:1;
104 __u64 loi_kms; /* known minimum size */
105 struct ost_lvb loi_lvb;
106 struct osc_async_rc loi_ar;
108 #define loi_id loi_oi.oi_id
109 #define loi_seq loi_oi.oi_seq
111 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
113 oinfo->loi_kms = kms;
114 oinfo->loi_kms_valid = 1;
117 static inline void loi_init(struct lov_oinfo *loi)
119 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
120 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
121 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
122 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
123 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
124 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
125 CFS_INIT_LIST_HEAD(&loi->loi_ready_item);
126 CFS_INIT_LIST_HEAD(&loi->loi_hp_ready_item);
127 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
128 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
131 struct lov_stripe_md {
132 cfs_spinlock_t lsm_lock;
133 pid_t lsm_lock_owner; /* debugging */
135 /* maximum possible file size, might change as OSTs status changes,
136 * e.g. disconnected, deactivated */
139 /* Public members. */
140 __u64 lw_object_id; /* lov object id */
141 __u64 lw_object_seq; /* lov object seq */
143 /* LOV-private members start here -- only for use in lov/. */
145 __u32 lw_stripe_size; /* size of the stripe */
146 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
147 __u16 lw_stripe_count; /* number of objects being striped over */
148 __u16 lw_layout_gen; /* generation of the layout */
149 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
152 struct lov_oinfo *lsm_oinfo[0];
155 #define lsm_object_id lsm_wire.lw_object_id
156 #define lsm_object_seq lsm_wire.lw_object_seq
157 #define lsm_magic lsm_wire.lw_magic
158 #define lsm_layout_gen lsm_wire.lw_layout_gen
159 #define lsm_stripe_size lsm_wire.lw_stripe_size
160 #define lsm_pattern lsm_wire.lw_pattern
161 #define lsm_stripe_count lsm_wire.lw_stripe_count
162 #define lsm_pool_name lsm_wire.lw_pool_name
166 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
168 /* obd info for a particular level (lov, osc). */
170 /* Lock policy. It keeps an extent which is specific for a particular
171 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
172 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
173 ldlm_policy_data_t oi_policy;
174 /* Flags used for set request specific flags:
175 - while lock handling, the flags obtained on the enqueue
176 request are set here.
177 - while stats, the flags used for control delay/resend.
178 - while setattr, the flags used for distinguish punch operation
181 /* Lock handle specific for every OSC lock. */
182 struct lustre_handle *oi_lockh;
183 /* lsm data specific for every OSC. */
184 struct lov_stripe_md *oi_md;
185 /* obdo data specific for every OSC, if needed at all. */
187 /* statfs data specific for every OSC, if needed at all. */
188 struct obd_statfs *oi_osfs;
189 /* An update callback which is called to update some data on upper
190 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
191 * request in osc level for enqueue requests. It is also possible to
192 * update some caller data from LOV layer if needed. */
193 obd_enqueue_update_f oi_cb_up;
194 /* oss capability, its type is obd_capa in client to avoid copy.
195 * in contrary its type is lustre_capa in OSS. */
199 /* compare all relevant fields. */
200 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
201 struct lov_stripe_md *m2)
204 * ->lsm_wire contains padding, but it should be zeroed out during
207 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
210 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
211 struct lov_stripe_md *lsm)
213 if (lsm->lsm_magic != lum->lmm_magic)
215 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
216 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
218 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
219 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
221 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
222 (lsm->lsm_pattern != lum->lmm_pattern))
224 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
225 (strncmp(lsm->lsm_pool_name,
226 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
227 LOV_MAXPOOLNAME) != 0))
232 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
234 struct lov_user_md *lum)
236 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
239 *lmm_magic = lumv3->lmm_magic;
241 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
242 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
243 *lmm_magic = LOV_USER_MAGIC_V1;
244 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
245 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
247 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
248 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
250 lustre_swab_lov_user_md_v3(lumv3);
251 *lmm_magic = LOV_USER_MAGIC_V3;
252 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
254 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
255 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
261 void lov_stripe_lock(struct lov_stripe_md *md);
262 void lov_stripe_unlock(struct lov_stripe_md *md);
265 cfs_list_t typ_chain;
266 struct obd_ops *typ_dt_ops;
267 struct md_ops *typ_md_ops;
268 cfs_proc_dir_entry_t *typ_procroot;
271 struct lu_device_type *typ_lu;
272 cfs_spinlock_t obd_type_lock;
282 /* Individual type definitions */
284 struct ost_server_data;
286 struct osd_properties {
287 size_t osd_max_ea_size;
290 #define OBT_MAGIC 0xBDDECEAE
291 /* hold common fields for "target" device */
292 struct obd_device_target {
295 struct super_block *obt_sb;
296 /** last_rcvd file */
297 struct file *obt_rcvd_filp;
298 struct lu_target *obt_lut;
299 __u64 obt_mount_count;
300 cfs_semaphore_t obt_quotachecking;
301 struct lustre_quota_ctxt obt_qctxt;
302 lustre_quota_version_t obt_qfmt;
303 cfs_rw_semaphore_t obt_rwsem;
304 struct vfsmount *obt_vfsmnt;
305 struct file *obt_health_check_filp;
306 struct osd_properties obt_osd_properties;
311 LLOG_CONFIG_ORIG_CTXT = 0,
312 LLOG_CONFIG_REPL_CTXT,
313 LLOG_MDS_OST_ORIG_CTXT,
314 LLOG_MDS_OST_REPL_CTXT,
321 LLOG_LOVEA_ORIG_CTXT,
322 LLOG_LOVEA_REPL_CTXT,
323 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
324 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
325 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
329 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
331 struct filter_subdirs {
332 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
342 /* NB this field MUST be first */
343 struct obd_device_target fo_obt;
344 const char *fo_fstype;
347 cfs_dentry_t *fo_dentry_O;
348 cfs_dentry_t **fo_dentry_O_groups;
349 struct filter_subdirs *fo_dentry_O_sub;
350 cfs_semaphore_t fo_init_lock; /* group initialization lock */
351 int fo_committed_group;
353 cfs_spinlock_t fo_objidlock; /* protect fo_lastobjid */
355 unsigned long fo_destroys_in_progress;
356 cfs_semaphore_t fo_create_locks[FILTER_SUBDIR_COUNT];
358 cfs_list_t fo_export_list;
361 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
362 obd_size fo_tot_granted; /* all values in bytes */
363 obd_size fo_tot_pending;
364 int fo_tot_granted_clients;
366 obd_size fo_readcache_max_filesize;
367 cfs_spinlock_t fo_flags_lock;
368 unsigned int fo_read_cache:1, /**< enable read-only cache */
369 fo_writethrough_cache:1,/**< read cache writes */
370 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
371 fo_raid_degraded:1;/**< RAID device degraded */
373 struct obd_import *fo_mdc_imp;
374 struct obd_uuid fo_mdc_uuid;
375 struct lustre_handle fo_mdc_conn;
376 struct file **fo_last_objid_files;
377 __u64 *fo_last_objids; /* last created objid for groups,
378 * protected by fo_objidlock */
380 cfs_semaphore_t fo_alloc_lock;
382 cfs_atomic_t fo_r_in_flight;
383 cfs_atomic_t fo_w_in_flight;
386 * per-filter pool of kiobuf's allocated by filter_common_setup() and
387 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
388 * which ->fo_iobuf_count were allocated.
390 * This pool contains kiobuf used by
391 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
394 * Locking: none, each OST thread uses only one element, determined by
395 * its "ordinal number", ->t_id.
397 struct filter_iobuf **fo_iobuf_pool;
400 cfs_list_t fo_llog_list;
401 cfs_spinlock_t fo_llog_list_lock;
403 struct brw_stats fo_filter_stats;
404 struct lustre_quota_ctxt fo_quota_ctxt;
405 cfs_spinlock_t fo_quotacheck_lock;
406 cfs_atomic_t fo_quotachecking;
408 int fo_fmd_max_num; /* per exp filter_mod_data */
409 int fo_fmd_max_age; /* jiffies to fmd expiry */
410 unsigned long fo_syncjournal:1, /* sync journal on writes */
411 fo_sync_lock_cancel:2;/* sync on lock cancel */
415 cfs_rwlock_t fo_sptlrpc_lock;
416 struct sptlrpc_rule_set fo_sptlrpc_rset;
418 /* capability related */
419 unsigned int fo_fl_oss_capa;
420 cfs_list_t fo_capa_keys;
421 cfs_hlist_head_t *fo_capa_hash;
422 struct llog_commit_master *fo_lcm;
426 struct timeout_item {
427 enum timeout_event ti_event;
428 cfs_time_t ti_timeout;
431 cfs_list_t ti_obd_list;
435 #define OSC_MAX_RIF_DEFAULT 8
436 #define OSC_MAX_RIF_MAX 256
437 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
438 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
439 #define OSC_DEFAULT_RESENDS 10
441 /* possible values for fo_sync_lock_cancel */
443 NEVER_SYNC_ON_CANCEL = 0,
444 BLOCKING_SYNC_ON_CANCEL = 1,
445 ALWAYS_SYNC_ON_CANCEL = 2,
446 NUM_SYNC_ON_CANCEL_STATES
449 #define MDC_MAX_RIF_DEFAULT 8
450 #define MDC_MAX_RIF_MAX 512
455 cfs_rw_semaphore_t cl_sem;
456 struct obd_uuid cl_target_uuid;
457 struct obd_import *cl_import; /* ptlrpc connection state */
459 /* max_mds_easize is purely a performance thing so we don't have to
460 * call obd_size_diskmd() all the time. */
461 int cl_default_mds_easize;
462 int cl_max_mds_easize;
463 int cl_max_mds_cookiesize;
465 enum lustre_sec_part cl_sp_me;
466 enum lustre_sec_part cl_sp_to;
467 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
469 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
470 void *cl_llcd_offset;
472 /* the grant values are protected by loi_list_lock below */
473 long cl_dirty; /* all _dirty_ in bytes */
474 long cl_dirty_max; /* allowed w/o rpc */
475 long cl_dirty_transit; /* dirty synchronous */
476 long cl_avail_grant; /* bytes of credit for ost */
477 long cl_lost_grant; /* lost credits (trunc) */
478 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
479 cfs_time_t cl_next_shrink_grant; /* jiffies */
480 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
481 cfs_semaphore_t cl_grant_sem; /*grant shrink list cfs_semaphore*/
482 int cl_grant_shrink_interval; /* seconds */
484 /* keep track of objects that have lois that contain pages which
485 * have been queued for async brw. this lock also protects the
486 * lists of osc_client_pages that hang off of the loi */
488 * ->cl_loi_list_lock protects consistency of
489 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
490 * ->ap_completion() call-backs are executed under this lock. As we
491 * cannot guarantee that these call-backs never block on all platforms
492 * (as a matter of fact they do block on Mac OS X), type of
493 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
494 * and blocking mutex on Mac OS X. (Alternative is to make this lock
495 * blocking everywhere, but we don't want to slow down fast-path of
496 * our main platform.)
498 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
499 * with client_obd_list_{un,}lock() and
500 * client_obd_list_lock_{init,done}() functions.
502 client_obd_lock_t cl_loi_list_lock;
503 cfs_list_t cl_loi_ready_list;
504 cfs_list_t cl_loi_hp_ready_list;
505 cfs_list_t cl_loi_write_list;
506 cfs_list_t cl_loi_read_list;
509 /* just a sum of the loi/lop pending numbers to be exported by /proc */
510 int cl_pending_w_pages;
511 int cl_pending_r_pages;
512 int cl_max_pages_per_rpc;
513 int cl_max_rpcs_in_flight;
514 struct obd_histogram cl_read_rpc_hist;
515 struct obd_histogram cl_write_rpc_hist;
516 struct obd_histogram cl_read_page_hist;
517 struct obd_histogram cl_write_page_hist;
518 struct obd_histogram cl_read_offset_hist;
519 struct obd_histogram cl_write_offset_hist;
521 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
522 cfs_atomic_t cl_destroy_in_flight;
523 cfs_waitq_t cl_destroy_waitq;
525 struct mdc_rpc_lock *cl_rpc_lock;
526 struct mdc_rpc_lock *cl_close_lock;
527 struct osc_creator cl_oscc;
530 cfs_semaphore_t cl_mgc_sem;
531 struct vfsmount *cl_mgc_vfsmnt;
532 struct dentry *cl_mgc_configs_dir;
533 cfs_atomic_t cl_mgc_refcount;
534 struct obd_export *cl_mgc_mgsexp;
536 /* checksumming for data sent over the network */
537 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
538 /* supported checksum types that are worked out at connect time */
539 __u32 cl_supp_cksum_types;
540 /* checksum algorithm to be used */
541 cksum_type_t cl_cksum_type;
543 /* also protected by the poorly named _loi_list_lock lock above */
544 struct osc_async_rc cl_ar;
546 /* used by quotacheck */
547 int cl_qchk_stat; /* quotacheck stat of the peer */
549 /* sequence manager */
550 struct lu_client_seq *cl_seq;
552 cfs_atomic_t cl_resends; /* resend count */
554 /* ptlrpc work for writeback in ptlrpcd context */
555 void *cl_writeback_work;
557 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
559 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
562 struct obd_device_target mgs_obt;
563 struct ptlrpc_service *mgs_service;
564 struct vfsmount *mgs_vfsmnt;
565 struct super_block *mgs_sb;
566 struct dentry *mgs_configs_dir;
567 cfs_list_t mgs_fs_db_list;
568 cfs_semaphore_t mgs_sem;
569 cfs_proc_dir_entry_t *mgs_proc_live;
570 cfs_time_t mgs_start_time;
574 /* NB this field MUST be first */
575 struct obd_device_target mds_obt;
576 struct ptlrpc_service *mds_service;
577 struct ptlrpc_service *mds_setattr_service;
578 struct ptlrpc_service *mds_readpage_service;
580 int mds_max_cookiesize;
582 unsigned long mds_atime_diff;
583 cfs_semaphore_t mds_epoch_sem;
584 struct ll_fid mds_rootfid;
585 cfs_dentry_t *mds_pending_dir;
586 cfs_dentry_t *mds_logs_dir;
587 cfs_dentry_t *mds_objects_dir;
588 struct llog_handle *mds_cfg_llh;
589 struct obd_device *mds_lov_obd;
590 struct obd_uuid mds_lov_uuid;
592 struct obd_export *mds_lov_exp;
593 struct lov_desc mds_lov_desc;
596 /* mark pages dirty for write. */
597 cfs_bitmap_t *mds_lov_page_dirty;
598 /* array for store pages with obd_id */
599 void **mds_lov_page_array;
600 /* file for store objid */
601 struct file *mds_lov_objid_filp;
602 __u32 mds_lov_objid_count;
603 __u32 mds_lov_objid_max_index;
604 __u32 mds_lov_objid_lastpage;
605 __u32 mds_lov_objid_lastidx;
608 struct lustre_quota_info mds_quota_info;
609 cfs_rw_semaphore_t mds_qonoff_sem;
610 cfs_semaphore_t mds_health_sem;
611 unsigned long mds_fl_user_xattr:1,
613 mds_evict_ost_nids:1,
617 mds_fl_target:1; /* mds have one or
620 struct upcall_cache *mds_identity_cache;
622 /* for capability keys update */
623 struct lustre_capa_key *mds_capa_keys;
624 cfs_rw_semaphore_t mds_notify_lock;
628 extern __u32 mds_max_ost_index;
630 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
632 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
634 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
636 extern int mds_lov_init_objids(struct obd_device *obd);
637 extern void mds_lov_destroy_objids(struct obd_device *obd);
647 struct obd_device_target eo_obt;
649 cfs_spinlock_t eo_lock;
651 struct lustre_handle eo_nl_lock;
652 cfs_atomic_t eo_prep;
656 struct ptlrpc_service *ost_service;
657 struct ptlrpc_service *ost_create_service;
658 struct ptlrpc_service *ost_io_service;
659 cfs_semaphore_t ost_health_sem;
662 struct echo_client_obd {
663 struct obd_export *ec_exp; /* the local connection to osc/lov */
664 cfs_spinlock_t ec_lock;
665 cfs_list_t ec_objects;
672 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
673 cfs_list_t lqo_oss_list; /* link to lov_qos */
674 __u64 lqo_bavail; /* total bytes avail on OSS */
675 __u64 lqo_penalty; /* current penalty */
676 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
677 time_t lqo_used; /* last used time, seconds */
678 __u32 lqo_ost_count; /* number of osts on this oss */
682 struct lov_qos_oss *ltq_oss; /* oss info */
683 __u64 ltq_penalty; /* current penalty */
684 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
685 __u64 ltq_weight; /* net weighting */
686 time_t ltq_used; /* last used time, seconds */
687 unsigned int ltq_usable:1; /* usable for striping */
690 /* Generic subset of OSTs */
692 __u32 *op_array; /* array of index of
694 unsigned int op_count; /* number of OSTs in the array */
695 unsigned int op_size; /* allocated size of lp_array */
696 cfs_rw_semaphore_t op_rw_sem; /* to protect ost_pool use */
699 /* Round-robin allocator data */
701 __u32 lqr_start_idx; /* start index of new inode */
702 __u32 lqr_offset_idx; /* aliasing for start_idx */
703 int lqr_start_count; /* reseed counter */
704 struct ost_pool lqr_pool; /* round-robin optimized list */
705 unsigned long lqr_dirty:1; /* recalc round-robin list */
708 /* allow statfs data caching for 1 second */
709 #define OBD_STATFS_CACHE_SECONDS 1
711 struct lov_statfs_data {
712 struct obd_info lsd_oi;
713 struct obd_statfs lsd_statfs;
715 /* Stripe placement optimization */
717 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
718 cfs_rw_semaphore_t lq_rw_sem;
719 __u32 lq_active_oss_count;
720 unsigned int lq_prio_free; /* priority for free space */
721 unsigned int lq_threshold_rr;/* priority for rr */
722 struct lov_qos_rr lq_rr; /* round robin qos data */
723 unsigned long lq_dirty:1, /* recalc qos data */
724 lq_same_space:1,/* the ost's all have approx.
725 the same space avail */
726 lq_reset:1, /* zero current penalties */
727 lq_statfs_in_progress:1; /* statfs op in
729 /* qos statfs data */
730 struct lov_statfs_data *lq_statfs_data;
731 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
732 * requests completion */
735 struct lov_tgt_desc {
737 struct obd_uuid ltd_uuid;
738 struct obd_device *ltd_obd;
739 struct obd_export *ltd_exp;
740 struct ltd_qos ltd_qos; /* qos info per target */
742 __u32 ltd_index; /* index in lov_obd->tgts */
743 unsigned long ltd_active:1,/* is this target up for requests */
744 ltd_activate:1,/* should target be activated */
745 ltd_reap:1; /* should this target be deleted */
749 #define pool_tgt_size(_p) _p->pool_obds.op_size
750 #define pool_tgt_count(_p) _p->pool_obds.op_count
751 #define pool_tgt_array(_p) _p->pool_obds.op_array
752 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
753 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
756 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
757 struct ost_pool pool_obds; /* pool members */
758 cfs_atomic_t pool_refcount; /* pool ref. counter */
759 struct lov_qos_rr pool_rr; /* round robin qos */
760 cfs_hlist_node_t pool_hash; /* access by poolname */
761 cfs_list_t pool_list; /* serial access */
762 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
763 struct lov_obd *pool_lov; /* lov obd to which this
768 struct lov_desc desc;
769 struct lov_tgt_desc **lov_tgts; /* sparse array */
770 struct ost_pool lov_packed; /* all OSTs in a packed
772 cfs_semaphore_t lov_lock;
773 struct obd_connect_data lov_ocd;
774 struct lov_qos lov_qos; /* qos info per lov */
775 cfs_atomic_t lov_refcount;
776 __u32 lov_tgt_count; /* how many OBD's */
777 __u32 lov_active_tgt_count; /* how many active */
778 __u32 lov_death_row;/* tgts scheduled to be deleted */
779 __u32 lov_tgt_size; /* size of tgts array */
782 cfs_hash_t *lov_pools_hash_body; /* used for key access */
783 cfs_list_t lov_pool_list; /* used for sequential access */
784 cfs_proc_dir_entry_t *lov_pool_proc_entry;
785 enum lustre_sec_part lov_sp_me;
788 struct lmv_tgt_desc {
789 struct obd_uuid ltd_uuid;
790 struct obd_export *ltd_exp;
791 int ltd_active; /* is this target up for requests */
793 cfs_semaphore_t ltd_fid_sem;
796 enum placement_policy {
797 PLACEMENT_CHAR_POLICY = 0,
798 PLACEMENT_NID_POLICY = 1,
799 PLACEMENT_INVAL_POLICY = 2,
803 typedef enum placement_policy placement_policy_t;
807 struct lu_client_fld lmv_fld;
808 cfs_spinlock_t lmv_lock;
809 placement_policy_t lmv_placement;
810 struct lmv_desc desc;
811 struct obd_uuid cluuid;
812 struct obd_export *exp;
819 cfs_semaphore_t init_sem;
821 struct lmv_tgt_desc *tgts;
824 struct obd_connect_data *datas;
827 struct obd_connect_data conn_data;
830 struct niobuf_local {
835 cfs_dentry_t *dentry;
840 #define LUSTRE_FLD_NAME "fld"
841 #define LUSTRE_SEQ_NAME "seq"
843 #define LUSTRE_CMM_NAME "cmm"
844 #define LUSTRE_MDD_NAME "mdd"
845 #define LUSTRE_OSD_NAME "osd-ldiskfs"
846 #define LUSTRE_VVP_NAME "vvp"
847 #define LUSTRE_LMV_NAME "lmv"
848 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
849 #define LUSTRE_SLP_NAME "slp"
851 /* obd device type names */
852 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
853 #define LUSTRE_MDS_NAME "mds"
854 #define LUSTRE_MDT_NAME "mdt"
855 #define LUSTRE_MDC_NAME "mdc"
856 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
857 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
858 #define LUSTRE_OSC_NAME "osc"
859 #define LUSTRE_LOV_NAME "lov"
860 #define LUSTRE_MGS_NAME "mgs"
861 #define LUSTRE_MGC_NAME "mgc"
863 #define LUSTRE_CACHEOBD_NAME "cobd"
864 #define LUSTRE_ECHO_NAME "obdecho"
865 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
867 /* Constant obd names (post-rename) */
868 #define LUSTRE_MDS_OBDNAME "MDS"
869 #define LUSTRE_OSS_OBDNAME "OSS"
870 #define LUSTRE_MGS_OBDNAME "MGS"
871 #define LUSTRE_MGC_OBDNAME "MGC"
873 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
874 #define N_LOCAL_TEMP_PAGE 0x10000000
876 struct obd_trans_info {
879 /* Only used on the server side for tracking acks. */
880 struct oti_req_ack_lock {
881 struct lustre_handle lock;
885 struct llog_cookie oti_onecookie;
886 struct llog_cookie *oti_logcookies;
889 /* initial thread handling transaction */
890 struct ptlrpc_thread * oti_thread;
893 __u64 oti_pre_version;
895 struct obd_uuid *oti_ost_uuid;
898 static inline void oti_init(struct obd_trans_info *oti,
899 struct ptlrpc_request *req)
903 memset(oti, 0, sizeof(*oti));
908 oti->oti_xid = req->rq_xid;
909 /** VBR: take versions from request */
910 if (req->rq_reqmsg != NULL &&
911 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
912 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
913 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
914 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
917 /** called from mds_create_objects */
918 if (req->rq_repmsg != NULL)
919 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
920 oti->oti_thread = req->rq_svc_thread;
921 if (req->rq_reqmsg != NULL)
922 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
925 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
930 if (num_cookies == 1)
931 oti->oti_logcookies = &oti->oti_onecookie;
933 OBD_ALLOC_LARGE(oti->oti_logcookies,
934 num_cookies * sizeof(oti->oti_onecookie));
936 oti->oti_numcookies = num_cookies;
939 static inline void oti_free_cookies(struct obd_trans_info *oti)
941 if (!oti || !oti->oti_logcookies)
944 if (oti->oti_logcookies == &oti->oti_onecookie)
945 LASSERT(oti->oti_numcookies == 1);
947 OBD_FREE_LARGE(oti->oti_logcookies,
948 oti->oti_numcookies*sizeof(oti->oti_onecookie));
949 oti->oti_logcookies = NULL;
950 oti->oti_numcookies = 0;
954 * Events signalled through obd_notify() upcall-chain.
956 enum obd_notify_event {
959 /* Device connect start */
961 /* Device activated */
963 /* Device deactivated */
965 /* Device disconnected */
967 /* Connect data for import were changed */
970 OBD_NOTIFY_SYNC_NONBLOCK,
972 /* Configuration event */
974 /* Trigger quota recovery */
976 /* Administratively deactivate/activate event */
977 OBD_NOTIFY_DEACTIVATE,
981 /* bit-mask flags for config events */
983 CONFIG_LOG = 0x1, /* finished processing config log */
984 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
985 CONFIG_TARGET = 0x4 /* one target is added */
989 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
990 * and liblustre being main examples).
992 struct obd_notify_upcall {
993 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
994 enum obd_notify_event ev, void *owner, void *data);
995 /* Opaque datum supplied by upper layer listener */
999 struct target_recovery_data {
1000 svc_handler_t trd_recovery_handler;
1001 pid_t trd_processing_task;
1002 cfs_completion_t trd_starting;
1003 cfs_completion_t trd_finishing;
1007 * In HEAD for CMD, the object is created in group number which is 3>=
1008 * or indexing starts from 3. To test this assertions are added to disallow
1009 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
1010 * object in group 0 needs to be allowed.
1011 * So for interop mode following changes needs to be done:
1012 * 1. No need to assert on group 0 or allow group 0
1013 * 2. The group number indexing starts from 0 instead of 3
1016 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
1018 static inline __u64 objseq_to_mdsno(obd_seq seq)
1020 LASSERT_SEQ_IS_MDT(seq);
1021 if (seq == FID_SEQ_OST_MDT0)
1023 return seq - FID_SEQ_OST_MDT1 + 1;
1026 static inline int mdt_to_obd_objseq(int mdtid)
1029 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
1033 return FID_SEQ_OST_MDT1 + mdtid - 1;
1037 struct obd_llog_group {
1038 cfs_list_t olg_list;
1040 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
1041 cfs_waitq_t olg_waitq;
1042 cfs_spinlock_t olg_lock;
1043 struct obd_export *olg_exp;
1044 int olg_initializing;
1045 cfs_semaphore_t olg_cat_processing;
1048 /* corresponds to one of the obd's */
1049 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1050 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1053 struct obd_type *obd_type;
1056 /* common and UUID name of this device */
1057 char obd_name[MAX_OBD_NAME];
1058 struct obd_uuid obd_uuid;
1060 struct lu_device *obd_lu_dev;
1063 /* bitfield modification is protected by obd_dev_lock */
1064 unsigned long obd_attached:1, /* finished attach */
1065 obd_set_up:1, /* finished setup */
1066 obd_recovering:1, /* there are recoverable clients */
1067 obd_abort_recovery:1,/* recovery expired */
1068 obd_version_recov:1, /* obd uses version checking */
1069 obd_replayable:1, /* recovery is enabled; inform clients */
1070 obd_no_transno:1, /* no committed-transno notification */
1071 obd_no_recov:1, /* fail instead of retry messages */
1072 obd_stopping:1, /* started cleanup */
1073 obd_starting:1, /* started setup */
1074 obd_force:1, /* cleanup with > 0 obd refcount */
1075 obd_fail:1, /* cleanup with failover */
1076 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1077 obd_no_conn:1, /* deny new connections */
1078 obd_inactive:1, /* device active/inactive
1079 * (for /proc/status only!!) */
1080 obd_no_ir:1, /* no imperative recovery. */
1081 obd_process_conf:1; /* device is processing mgs config */
1082 /* use separate field as it is set in interrupt to don't mess with
1083 * protection of other bits using _bh lock */
1084 unsigned long obd_recovery_expired:1;
1085 /* uuid-export hash body */
1086 cfs_hash_t *obd_uuid_hash;
1087 /* nid-export hash body */
1088 cfs_hash_t *obd_nid_hash;
1089 /* nid stats body */
1090 cfs_hash_t *obd_nid_stats_hash;
1091 cfs_list_t obd_nid_stats;
1092 cfs_atomic_t obd_refcount;
1093 cfs_waitq_t obd_refcount_waitq;
1094 cfs_list_t obd_exports;
1095 cfs_list_t obd_unlinked_exports;
1096 cfs_list_t obd_delayed_exports;
1097 int obd_num_exports;
1098 cfs_spinlock_t obd_nid_lock;
1099 struct ldlm_namespace *obd_namespace;
1100 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1101 /* a spinlock is OK for what we do now, may need a semaphore later */
1102 cfs_spinlock_t obd_dev_lock; /* protects obd bitfield above */
1103 cfs_semaphore_t obd_dev_sem;
1104 __u64 obd_last_committed;
1105 struct fsfilt_operations *obd_fsops;
1106 cfs_spinlock_t obd_osfs_lock;
1107 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1109 struct lvfs_run_ctxt obd_lvfs_ctxt;
1110 struct obd_llog_group obd_olg; /* default llog group */
1111 struct obd_device *obd_observer;
1112 cfs_rw_semaphore_t obd_observer_link_sem;
1113 struct obd_notify_upcall obd_upcall;
1114 struct obd_export *obd_self_export;
1115 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1116 cfs_list_t obd_exports_timed;
1117 time_t obd_eviction_timer; /* for ping evictor */
1119 int obd_max_recoverable_clients;
1120 cfs_atomic_t obd_connected_clients;
1121 int obd_stale_clients;
1122 int obd_delayed_clients;
1123 /* this lock protects all recovery list_heads, timer and
1124 * obd_next_recovery_transno value */
1125 cfs_spinlock_t obd_recovery_task_lock;
1126 __u64 obd_next_recovery_transno;
1127 int obd_replayed_requests;
1128 int obd_requests_queued_for_recovery;
1129 cfs_waitq_t obd_next_transno_waitq;
1130 /* protected by obd_recovery_task_lock */
1131 cfs_timer_t obd_recovery_timer;
1132 time_t obd_recovery_start; /* seconds */
1133 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1134 int obd_recovery_time_hard;
1135 int obd_recovery_timeout;
1136 int obd_recovery_ir_factor;
1138 /* new recovery stuff from CMD2 */
1139 struct target_recovery_data obd_recovery_data;
1140 int obd_replayed_locks;
1141 cfs_atomic_t obd_req_replay_clients;
1142 cfs_atomic_t obd_lock_replay_clients;
1143 /* all lists are protected by obd_recovery_task_lock */
1144 cfs_list_t obd_req_replay_queue;
1145 cfs_list_t obd_lock_replay_queue;
1146 cfs_list_t obd_final_req_queue;
1147 int obd_recovery_stage;
1150 struct obd_device_target obt;
1151 struct filter_obd filter;
1153 struct client_obd cli;
1155 struct echo_client_obd echo_client;
1156 struct echo_obd echo;
1161 /* Fields used by LProcFS */
1162 unsigned int obd_cntr_base;
1163 struct lprocfs_stats *obd_stats;
1165 unsigned int md_cntr_base;
1166 struct lprocfs_stats *md_stats;
1168 cfs_proc_dir_entry_t *obd_proc_entry;
1169 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1170 cfs_proc_dir_entry_t *obd_svc_procroot;
1171 struct lprocfs_stats *obd_svc_stats;
1172 cfs_atomic_t obd_evict_inprogress;
1173 cfs_waitq_t obd_evict_inprogress_waitq;
1174 cfs_list_t obd_evict_list; /* protected with pet_lock */
1177 * Ldlm pool part. Save last calculated SLV and Limit.
1179 cfs_rwlock_t obd_pool_lock;
1184 * A list of outstanding class_incref()'s against this obd. For
1187 struct lu_ref obd_reference;
1190 #define OBD_LLOG_FL_SENDNOW 0x0001
1192 enum obd_cleanup_stage {
1193 /* Special case hack for MDS LOVs */
1195 /* can be directly mapped to .ldto_device_fini() */
1196 OBD_CLEANUP_EXPORTS,
1199 /* get/set_info keys */
1200 #define KEY_ASYNC "async"
1201 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1202 #define KEY_BLOCKSIZE "blocksize"
1203 #define KEY_CAPA_KEY "capa_key"
1204 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1205 #define KEY_FID2PATH "fid2path"
1206 #define KEY_CHECKSUM "checksum"
1207 #define KEY_CLEAR_FS "clear_fs"
1208 #define KEY_CONN_DATA "conn_data"
1209 #define KEY_EVICT_BY_NID "evict_by_nid"
1210 #define KEY_FIEMAP "fiemap"
1211 #define KEY_FLUSH_CTX "flush_ctx"
1212 #define KEY_GRANT_SHRINK "grant_shrink"
1213 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1214 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1215 #define KEY_INIT_RECOV "initial_recov"
1216 #define KEY_INTERMDS "inter_mds"
1217 #define KEY_LAST_ID "last_id"
1218 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1219 #define KEY_LOVDESC "lovdesc"
1220 #define KEY_LOV_IDX "lov_idx"
1221 #define KEY_MAX_EASIZE "max_easize"
1222 #define KEY_MDS_CONN "mds_conn"
1223 #define KEY_MGSSEC "mgssec"
1224 #define KEY_NEXT_ID "next_id"
1225 #define KEY_READ_ONLY "read-only"
1226 #define KEY_REGISTER_TARGET "register_target"
1227 #define KEY_REVIMP_UPD "revimp_update"
1228 #define KEY_SET_FS "set_fs"
1229 #define KEY_TGT_COUNT "tgt_count"
1230 /* KEY_SET_INFO in lustre_idl.h */
1231 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1232 #define KEY_CONNECT_FLAG "connect_flags"
1233 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1237 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1238 #define IT_OPEN (1 << 0)
1239 #define IT_CREAT (1 << 1)
1240 #define IT_READDIR (1 << 2)
1241 #define IT_GETATTR (1 << 3)
1242 #define IT_LOOKUP (1 << 4)
1243 #define IT_UNLINK (1 << 5)
1244 #define IT_TRUNC (1 << 6)
1245 #define IT_GETXATTR (1 << 7)
1246 #define IT_EXEC (1 << 8)
1247 #define IT_PIN (1 << 9)
1248 #define IT_LAYOUT (1 << 10)
1250 static inline int it_to_lock_mode(struct lookup_intent *it)
1252 /* CREAT needs to be tested before open (both could be set) */
1253 if (it->it_op & IT_CREAT)
1255 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1259 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1264 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1265 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1266 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1267 struct lu_fid op_fid4; /* to the operation locks. */
1268 mdsno_t op_mds; /* what mds server open will go to */
1269 struct lustre_handle op_handle;
1270 obd_time op_mod_time;
1271 const char *op_name;
1274 struct lmv_stripe_md *op_mea1;
1275 struct lmv_stripe_md *op_mea2;
1276 __u32 op_suppgids[2];
1282 /* iattr fields and blocks. */
1283 struct iattr op_attr;
1285 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
1286 unsigned int op_attr_flags;
1290 loff_t op_attr_blocks;
1292 /* Size-on-MDS epoch and flags. */
1297 struct obd_capa *op_capa1;
1298 struct obd_capa *op_capa2;
1300 /* Various operation flags. */
1303 /* Operation type */
1307 struct md_enqueue_info;
1308 /* metadata stat-ahead */
1309 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1310 struct md_enqueue_info *minfo,
1313 struct md_enqueue_info {
1314 struct md_op_data mi_data;
1315 struct lookup_intent mi_it;
1316 struct lustre_handle mi_lockh;
1317 struct inode *mi_dir;
1318 md_enqueue_cb_t mi_cb;
1320 unsigned int mi_generation;
1324 cfs_module_t *o_owner;
1325 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1326 void *karg, void *uarg);
1327 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
1328 __u32 *vallen, void *val, struct lov_stripe_md *lsm);
1329 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
1330 __u32 vallen, void *val,
1331 struct ptlrpc_request_set *set);
1332 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1333 int (*o_detach)(struct obd_device *dev);
1334 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1335 int (*o_precleanup)(struct obd_device *dev,
1336 enum obd_cleanup_stage cleanup_stage);
1337 int (*o_cleanup)(struct obd_device *dev);
1338 int (*o_process_config)(struct obd_device *dev, obd_count len,
1340 int (*o_postrecov)(struct obd_device *dev);
1341 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1343 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1344 /* connect to the target device with given connection
1345 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1346 * granted by the target, which are guaranteed to be a subset of flags
1347 * asked for. If @ocd == NULL, use default parameters. */
1348 int (*o_connect)(const struct lu_env *env,
1349 struct obd_export **exp, struct obd_device *src,
1350 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1352 int (*o_reconnect)(const struct lu_env *env,
1353 struct obd_export *exp, struct obd_device *src,
1354 struct obd_uuid *cluuid,
1355 struct obd_connect_data *ocd,
1357 int (*o_disconnect)(struct obd_export *exp);
1359 /* Initialize/finalize fids infrastructure. */
1360 int (*o_fid_init)(struct obd_export *exp);
1361 int (*o_fid_fini)(struct obd_export *exp);
1363 /* Allocate new fid according to passed @hint. */
1364 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1365 struct md_op_data *op_data);
1368 * Object with @fid is getting deleted, we may want to do something
1371 int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
1373 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
1374 __u64 max_age, __u32 flags);
1375 int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
1376 __u64 max_age, struct ptlrpc_request_set *set);
1377 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1378 struct lov_stripe_md *mem_src);
1379 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1380 struct lov_mds_md *disk_src, int disk_len);
1381 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1383 /* FIXME: add fid capability support for create & destroy! */
1384 int (*o_precreate)(struct obd_export *exp);
1385 int (*o_create)(struct obd_export *exp, struct obdo *oa,
1386 struct lov_stripe_md **ea, struct obd_trans_info *oti);
1387 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1388 struct lov_stripe_md **ea,
1389 struct obd_trans_info *oti);
1390 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
1391 struct lov_stripe_md *ea, struct obd_trans_info *oti,
1392 struct obd_export *md_exp, void *capa);
1393 int (*o_setattr)(struct obd_export *exp, struct obd_info *oinfo,
1394 struct obd_trans_info *oti);
1395 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1396 struct obd_trans_info *oti,
1397 struct ptlrpc_request_set *rqset);
1398 int (*o_getattr)(struct obd_export *exp, struct obd_info *oinfo);
1399 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1400 struct ptlrpc_request_set *set);
1401 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1402 obd_count oa_bufs, struct brw_page *pgarr,
1403 struct obd_trans_info *oti);
1404 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1405 struct ost_lvb *lvb, int kms_only);
1406 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1407 obd_off size, int shrink);
1408 int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
1409 struct obd_trans_info *oti,
1410 struct ptlrpc_request_set *rqset);
1411 int (*o_sync)(struct obd_export *exp, struct obd_info *oinfo,
1412 obd_size start, obd_size end,
1413 struct ptlrpc_request_set *set);
1414 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1415 struct lov_stripe_md *src, obd_size start,
1416 obd_size end, struct obd_trans_info *oti);
1417 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1418 struct lustre_handle *srconn, struct lov_stripe_md *src,
1419 obd_size start, obd_size end, struct obd_trans_info *);
1420 int (*o_iterate)(struct lustre_handle *conn,
1421 int (*)(obd_id, obd_seq, void *),
1422 obd_id *startid, obd_seq seq, void *data);
1423 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
1424 int objcount, struct obd_ioobj *obj,
1425 struct niobuf_remote *remote, int *nr_pages,
1426 struct niobuf_local *local,
1427 struct obd_trans_info *oti,
1428 struct lustre_capa *capa);
1429 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
1430 int objcount, struct obd_ioobj *obj,
1431 struct niobuf_remote *remote, int pages,
1432 struct niobuf_local *local,
1433 struct obd_trans_info *oti, int rc);
1434 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1435 struct ldlm_enqueue_info *einfo,
1436 struct ptlrpc_request_set *rqset);
1437 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1438 ldlm_iterator_t it, void *data);
1439 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1440 ldlm_iterator_t it, void *data);
1441 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1442 __u32 mode, struct lustre_handle *);
1443 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1444 ldlm_cancel_flags_t flags, void *opaque);
1445 int (*o_init_export)(struct obd_export *exp);
1446 int (*o_destroy_export)(struct obd_export *exp);
1447 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1448 int cmd, obd_off *);
1450 /* llog related obd_methods */
1451 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1452 struct obd_device *disk_obd, int *idx);
1453 int (*o_llog_finish)(struct obd_device *obd, int count);
1454 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1456 /* metadata-only methods */
1457 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1458 struct obd_capa *, struct obd_client_handle *, int flag);
1459 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1461 int (*o_import_event)(struct obd_device *, struct obd_import *,
1462 enum obd_import_event);
1464 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1465 enum obd_notify_event ev, void *data);
1467 int (*o_health_check)(struct obd_device *);
1468 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1471 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1472 struct obd_quotactl *);
1473 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1474 struct obd_quotactl *);
1475 int (*o_quota_adjust_qunit)(struct obd_export *exp,
1476 struct quota_adjust_qunit *oqaq,
1477 struct lustre_quota_ctxt *qctxt,
1478 struct ptlrpc_request_set *rqset);
1481 int (*o_ping)(struct obd_export *exp);
1484 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1485 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1486 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1488 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1490 void (*o_getref)(struct obd_device *obd);
1491 void (*o_putref)(struct obd_device *obd);
1493 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1494 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1495 * Also, add a wrapper function in include/linux/obd_class.h. */
1499 LUSTRE_OPC_MKDIR = (1 << 0),
1500 LUSTRE_OPC_SYMLINK = (1 << 1),
1501 LUSTRE_OPC_MKNOD = (1 << 2),
1502 LUSTRE_OPC_CREATE = (1 << 3),
1503 LUSTRE_OPC_ANY = (1 << 4)
1506 /* lmv structures */
1507 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1508 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1509 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1511 #define MAX_HASH_SIZE_32 0x7fffffffUL
1512 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1513 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1516 struct mdt_body *body;
1517 struct lov_stripe_md *lsm;
1518 struct lmv_stripe_md *mea;
1519 #ifdef CONFIG_FS_POSIX_ACL
1520 struct posix_acl *posix_acl;
1522 struct mdt_remote_perm *remote_perm;
1523 struct obd_capa *mds_capa;
1524 struct obd_capa *oss_capa;
1527 struct md_open_data {
1528 struct obd_client_handle *mod_och;
1529 struct ptlrpc_request *mod_open_req;
1530 struct ptlrpc_request *mod_close_req;
1531 cfs_atomic_t mod_refcount;
1534 struct lookup_intent;
1537 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1538 struct obd_capa **);
1539 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1540 ldlm_iterator_t, void *);
1541 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1542 ldlm_iterator_t, void *);
1543 int (*m_close)(struct obd_export *, struct md_op_data *,
1544 struct md_open_data *, struct ptlrpc_request **);
1545 int (*m_create)(struct obd_export *, struct md_op_data *,
1546 const void *, int, int, __u32, __u32, cfs_cap_t,
1547 __u64, struct ptlrpc_request **);
1548 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1549 struct md_open_data *);
1550 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1551 struct lookup_intent *, struct md_op_data *,
1552 struct lustre_handle *, void *, int,
1553 struct ptlrpc_request **, int);
1554 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1555 struct ptlrpc_request **);
1556 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1557 struct ptlrpc_request **);
1558 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1559 void *, int, struct lookup_intent *, int,
1560 struct ptlrpc_request **,
1561 ldlm_blocking_callback, int);
1562 int (*m_link)(struct obd_export *, struct md_op_data *,
1563 struct ptlrpc_request **);
1564 int (*m_rename)(struct obd_export *, struct md_op_data *,
1565 const char *, int, const char *, int,
1566 struct ptlrpc_request **);
1567 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1568 const struct lu_fid *,
1569 struct ptlrpc_request **);
1570 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1571 int , void *, int, struct ptlrpc_request **,
1572 struct md_open_data **mod);
1573 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1574 struct obd_capa *, struct ptlrpc_request **);
1575 int (*m_readpage)(struct obd_export *, const struct lu_fid *,
1576 struct obd_capa *, __u64, struct page **,
1577 unsigned, struct ptlrpc_request **);
1579 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1580 struct ptlrpc_request **);
1582 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1583 struct obd_capa *, obd_valid, const char *,
1584 const char *, int, int, int, __u32,
1585 struct ptlrpc_request **);
1587 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1588 struct obd_capa *, obd_valid, const char *,
1589 const char *, int, int, int,
1590 struct ptlrpc_request **);
1592 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1594 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1595 struct obd_export *, struct obd_export *,
1596 struct lustre_md *);
1598 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1600 int (*m_set_open_replay_data)(struct obd_export *,
1601 struct obd_client_handle *,
1602 struct ptlrpc_request *);
1603 int (*m_clear_open_replay_data)(struct obd_export *,
1604 struct obd_client_handle *);
1605 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1607 ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
1608 const struct lu_fid *, ldlm_type_t,
1609 ldlm_policy_data_t *, ldlm_mode_t,
1610 struct lustre_handle *);
1612 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1613 ldlm_policy_data_t *, ldlm_mode_t,
1614 ldlm_cancel_flags_t flags, void *opaque);
1615 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1616 renew_capa_cb_t cb);
1617 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1618 const struct req_msg_field *, struct obd_capa **);
1620 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1621 struct obd_capa *, __u32,
1622 struct ptlrpc_request **);
1624 int (*m_intent_getattr_async)(struct obd_export *,
1625 struct md_enqueue_info *,
1626 struct ldlm_enqueue_info *);
1628 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1629 struct lu_fid *, __u64 *bits);
1632 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1633 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1634 * wrapper function in include/linux/obd_class.h.
1638 struct lsm_operations {
1639 void (*lsm_free)(struct lov_stripe_md *);
1640 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1641 struct obd_export *md_exp);
1642 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1644 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1646 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1647 __u16 *stripe_count);
1648 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1649 struct lov_mds_md *lmm);
1652 extern const struct lsm_operations lsm_v1_ops;
1653 extern const struct lsm_operations lsm_v3_ops;
1654 static inline const struct lsm_operations *lsm_op_find(int magic)
1662 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1667 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1669 /* Requests for obd_extent_calc() */
1670 #define OBD_CALC_STRIPE_START 1
1671 #define OBD_CALC_STRIPE_END 2
1673 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1674 struct obd_export *exp, int error)
1677 CERROR("%s: transno "LPU64" commit error: %d\n",
1678 obd->obd_name, transno, error);
1681 if (exp && transno > exp->exp_last_committed) {
1682 CDEBUG(D_HA, "%s: transno "LPU64" committed\n",
1683 obd->obd_name, transno);
1684 exp->exp_last_committed = transno;
1685 ptlrpc_commit_replies(exp);
1687 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1688 obd->obd_name, transno);
1690 if (transno > obd->obd_last_committed)
1691 obd->obd_last_committed = transno;
1694 static inline void init_obd_quota_ops(quota_interface_t *interface,
1695 struct obd_ops *obd_ops)
1701 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1702 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1703 obd_ops->o_quota_adjust_qunit = QUOTA_OP(interface, adjust_qunit);
1706 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1708 return oinfo->oi_capa;
1711 static inline struct md_open_data *obd_mod_alloc(void)
1713 struct md_open_data *mod;
1717 cfs_atomic_set(&mod->mod_refcount, 1);
1721 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1722 #define obd_mod_put(mod) \
1724 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1725 if ((mod)->mod_open_req) \
1726 ptlrpc_req_finished((mod)->mod_open_req); \
1727 OBD_FREE_PTR(mod); \
1731 extern void obdo_from_inode(struct obdo *dst, struct inode *src,
1732 struct lu_fid *parent, obd_flag valid);
1734 /* return 1 if client should be resend request */
1735 static inline int client_should_resend(int resend, struct client_obd *cli)
1737 return cfs_atomic_read(&cli->cl_resends) ?
1738 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1742 * Return device name for this device
1744 * XXX: lu_device is declared before obd_device, while a pointer pointing
1745 * back to obd_device in lu_device, so this helper function defines here
1746 * instead of in lu_object.h
1748 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1750 return lu_dev->ld_obd->obd_name;
1753 #endif /* __OBD_H */