1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
10 #elif defined(__APPLE__)
11 #include <darwin/obd.h>
12 #elif defined(__WINNT__)
13 #include <winnt/obd.h>
15 #error Unsupported operating system.
18 #define IOC_OSC_TYPE 'h'
19 #define IOC_OSC_MIN_NR 20
20 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
21 #define IOC_OSC_MAX_NR 50
23 #define IOC_MDC_TYPE 'i'
24 #define IOC_MDC_MIN_NR 20
25 /* Moved to lustre_user.h
26 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_device *)
27 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
28 #define IOC_MDC_MAX_NR 50
30 #include <lustre/lustre_idl.h>
31 #include <lu_object.h>
32 #include <lustre_lib.h>
33 #include <lustre_export.h>
34 #include <lustre_quota.h>
35 #include <lustre_fld.h>
37 /* this is really local to the OSC */
38 struct loi_oap_pages {
39 struct list_head lop_pending;
41 struct list_head lop_urgent;
42 struct list_head lop_pending_group;
51 struct lov_oinfo { /* per-stripe data structure */
52 __u64 loi_id; /* object ID on the target OST */
53 __u64 loi_gr; /* object group on the target OST */
54 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
55 int loi_ost_gen; /* generation of this loi_ost_idx */
57 /* used by the osc to keep track of what objects to build into rpcs */
58 struct loi_oap_pages loi_read_lop;
59 struct loi_oap_pages loi_write_lop;
60 /* _cli_ is poorly named, it should be _ready_ */
61 struct list_head loi_cli_item;
62 struct list_head loi_write_item;
63 struct list_head loi_read_item;
65 unsigned loi_kms_valid:1;
66 __u64 loi_kms; /* known minimum size */
67 struct ost_lvb loi_lvb;
68 struct osc_async_rc loi_ar;
71 static inline void loi_init(struct lov_oinfo *loi)
73 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
74 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
75 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
76 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
77 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
78 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
79 CFS_INIT_LIST_HEAD(&loi->loi_cli_item);
80 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
81 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
84 /*extent array item for describing the joined file extent info*/
86 __u64 le_start; /* extent start */
87 __u64 le_len; /* extent length */
88 int le_loi_idx; /* extent #1 loi's index in lsm loi array */
89 int le_stripe_count; /* extent stripe count*/
92 /*Lov array info for describing joined file array EA info*/
93 struct lov_array_info {
94 struct llog_logid lai_array_id; /* MDS med llog object id */
95 unsigned lai_ext_count; /* number of extent count */
96 struct lov_extent *lai_ext_array; /* extent desc array */
99 struct lov_stripe_md {
101 void *lsm_lock_owner; /* debugging */
104 /* Public members. */
105 __u64 lw_object_id; /* lov object id */
106 __u64 lw_object_gr; /* lov object group */
107 __u64 lw_maxbytes; /* maximum possible file size */
108 unsigned long lw_xfersize; /* optimal transfer size */
110 /* LOV-private members start here -- only for use in lov/. */
112 __u32 lw_stripe_size; /* size of the stripe */
113 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
114 unsigned lw_stripe_count; /* number of objects being striped over */
117 struct lov_array_info *lsm_array; /*Only for joined file array info*/
118 struct lov_oinfo lsm_oinfo[0];
121 #define lsm_object_id lsm_wire.lw_object_id
122 #define lsm_object_gr lsm_wire.lw_object_gr
123 #define lsm_maxbytes lsm_wire.lw_maxbytes
124 #define lsm_xfersize lsm_wire.lw_xfersize
125 #define lsm_magic lsm_wire.lw_magic
126 #define lsm_stripe_size lsm_wire.lw_stripe_size
127 #define lsm_pattern lsm_wire.lw_pattern
128 #define lsm_stripe_count lsm_wire.lw_stripe_count
130 /* compare all relevant fields. */
131 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
132 struct lov_stripe_md *m2)
135 * ->lsm_wire contains padding, but it should be zeroed out during
138 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
141 void lov_stripe_lock(struct lov_stripe_md *md);
142 void lov_stripe_unlock(struct lov_stripe_md *md);
145 struct list_head typ_chain;
146 struct obd_ops *typ_dt_ops;
147 struct md_ops *typ_md_ops;
148 struct proc_dir_entry *typ_procroot;
151 struct lu_device_type *typ_lu;
162 ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
163 page is added to an rpc */
164 ASYNC_URGENT = 0x2, /* page must be put into an RPC before return */
165 ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
166 to give the caller a chance to update
167 or cancel the size of the io */
168 ASYNC_GROUP_SYNC = 0x8, /* ap_completion will not be called, instead
169 the page is accounted for in the
170 obd_io_group given to
171 obd_queue_group_io */
174 struct obd_async_page_ops {
175 int (*ap_make_ready)(void *data, int cmd);
176 int (*ap_refresh_count)(void *data, int cmd);
177 void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
178 void (*ap_completion)(void *data, int cmd, struct obdo *oa, int rc);
181 /* the `oig' is passed down from a caller of obd rw methods. the callee
182 * records enough state such that the caller can sleep on the oig and
183 * be woken when all the callees have finished their work */
184 struct obd_io_group {
186 atomic_t oig_refcount;
189 struct list_head oig_occ_list;
190 cfs_waitq_t oig_waitq;
193 /* the oig callback context lets the callee of obd rw methods register
194 * for callbacks from the caller. */
195 struct oig_callback_context {
196 struct list_head occ_oig_item;
197 /* called when the caller has received a signal while sleeping.
198 * callees of this method are encouraged to abort their state
199 * in the oig. This may be called multiple times. */
200 void (*occ_interrupted)(struct oig_callback_context *occ);
201 unsigned int interrupted:1;
204 /* if we find more consumers this could be generalized */
205 #define OBD_HIST_MAX 32
206 struct obd_histogram {
208 unsigned long oh_buckets[OBD_HIST_MAX];
211 /* Individual type definitions */
213 struct ost_server_data;
215 /* hold common fields for "target" device */
216 struct obd_device_target {
217 struct super_block *obt_sb;
218 atomic_t obt_quotachecking;
219 struct lustre_quota_ctxt obt_qctxt;
222 #define FILTER_GROUP_LLOG 1
223 #define FILTER_GROUP_ECHO 2
231 /* NB this field MUST be first */
232 struct obd_device_target fo_obt;
233 const char *fo_fstype;
234 struct vfsmount *fo_vfsmnt;
235 cfs_dentry_t *fo_dentry_O;
236 cfs_dentry_t **fo_dentry_O_groups;
237 cfs_dentry_t **fo_dentry_O_sub;
238 spinlock_t fo_objidlock; /* protect fo_lastobjid */
239 spinlock_t fo_translock; /* protect fsd_last_transno */
240 struct file *fo_rcvd_filp;
241 struct file *fo_health_check_filp;
242 struct lr_server_data *fo_fsd;
243 unsigned long *fo_last_rcvd_slots;
244 __u64 fo_mount_count;
246 int fo_destroy_in_progress;
247 struct semaphore fo_create_lock;
249 struct list_head fo_export_list;
252 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
253 obd_size fo_tot_granted; /* all values in bytes */
254 obd_size fo_tot_pending;
256 obd_size fo_readcache_max_filesize;
258 struct obd_import *fo_mdc_imp;
259 struct obd_uuid fo_mdc_uuid;
260 struct lustre_handle fo_mdc_conn;
261 struct file **fo_last_objid_files;
262 __u64 *fo_last_objids; /* last created objid for groups,
263 * protected by fo_objidlock */
265 struct semaphore fo_alloc_lock;
267 spinlock_t fo_stats_lock;
268 int fo_r_in_flight; /* protected by fo_stats_lock */
269 int fo_w_in_flight; /* protected by fo_stats_lock */
272 * per-filter pool of kiobuf's allocated by filter_common_setup() and
273 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
274 * which ->fo_iobuf_count were allocated.
276 * This pool contains kiobuf used by
277 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
280 * Locking: none, each OST thread uses only one element, determined by
281 * its "ordinal number", ->t_id.
283 struct filter_iobuf **fo_iobuf_pool;
286 struct obd_histogram fo_r_pages;
287 struct obd_histogram fo_w_pages;
288 struct obd_histogram fo_read_rpc_hist;
289 struct obd_histogram fo_write_rpc_hist;
290 struct obd_histogram fo_r_io_time;
291 struct obd_histogram fo_w_io_time;
292 struct obd_histogram fo_r_discont_pages;
293 struct obd_histogram fo_w_discont_pages;
294 struct obd_histogram fo_r_discont_blocks;
295 struct obd_histogram fo_w_discont_blocks;
296 struct obd_histogram fo_r_disk_iosize;
297 struct obd_histogram fo_w_disk_iosize;
299 struct lustre_quota_ctxt fo_quota_ctxt;
300 spinlock_t fo_quotacheck_lock;
301 atomic_t fo_quotachecking;
304 #define OSC_MAX_RIF_DEFAULT 8
305 #define OSC_MAX_RIF_MAX 256
306 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
307 #define OSC_MAX_DIRTY_MB_MAX 2048 /* totally arbitrary */
312 struct semaphore cl_sem;
313 struct obd_uuid cl_target_uuid;
314 struct obd_import *cl_import; /* ptlrpc connection state */
316 /* max_mds_easize is purely a performance thing so we don't have to
317 * call obd_size_diskmd() all the time. */
318 int cl_default_mds_easize;
319 int cl_max_mds_easize;
320 int cl_max_mds_cookiesize;
323 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
324 void *cl_llcd_offset;
326 /* the grant values are protected by loi_list_lock below */
327 long cl_dirty; /* all _dirty_ in bytes */
328 long cl_dirty_max; /* allowed w/o rpc */
329 long cl_avail_grant; /* bytes of credit for ost */
330 long cl_lost_grant; /* lost credits (trunc) */
331 struct list_head cl_cache_waiters; /* waiting for cache/grant */
333 /* keep track of objects that have lois that contain pages which
334 * have been queued for async brw. this lock also protects the
335 * lists of osc_client_pages that hang off of the loi */
337 * ->cl_loi_list_lock protects consistency of
338 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
339 * ->ap_completion() call-backs are executed under this lock. As we
340 * cannot guarantee that these call-backs never block on all platforms
341 * (as a matter of fact they do block on Mac OS X), type of
342 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
343 * and blocking mutex on Mac OS X. (Alternative is to make this lock
344 * blocking everywhere, but we don't want to slow down fast-path of
345 * our main platform.)
347 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
348 * with client_obd_list_{un,}lock() and
349 * client_obd_list_lock_{init,done}() functions.
351 client_obd_lock_t cl_loi_list_lock;
352 struct list_head cl_loi_ready_list;
353 struct list_head cl_loi_write_list;
354 struct list_head cl_loi_read_list;
357 /* just a sum of the loi/lop pending numbers to be exported by /proc */
358 int cl_pending_w_pages;
359 int cl_pending_r_pages;
360 int cl_max_pages_per_rpc;
361 int cl_max_rpcs_in_flight;
362 struct obd_histogram cl_read_rpc_hist;
363 struct obd_histogram cl_write_rpc_hist;
364 struct obd_histogram cl_read_page_hist;
365 struct obd_histogram cl_write_page_hist;
366 struct obd_histogram cl_read_offset_hist;
367 struct obd_histogram cl_write_offset_hist;
369 struct mdc_rpc_lock *cl_rpc_lock;
370 struct mdc_rpc_lock *cl_setattr_lock;
371 struct osc_creator cl_oscc;
374 struct semaphore cl_mgc_sem;
375 struct vfsmount *cl_mgc_vfsmnt;
376 struct dentry *cl_mgc_configs_dir;
377 atomic_t cl_mgc_refcount;
378 struct obd_export *cl_mgc_mgsexp;
381 unsigned int cl_checksum:1; /* debug checksums */
383 /* also protected by the poorly named _loi_list_lock lock above */
384 struct osc_async_rc cl_ar;
386 /* used by quotacheck */
387 int cl_qchk_stat; /* quotacheck stat of the peer */
389 /* sequence manager */
390 struct lu_client_seq *cl_seq;
392 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
394 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
397 struct ptlrpc_service *mgs_service;
398 struct vfsmount *mgs_vfsmnt;
399 struct super_block *mgs_sb;
400 struct dentry *mgs_configs_dir;
401 struct dentry *mgs_fid_de;
402 struct list_head mgs_fs_db_list;
403 struct semaphore mgs_sem;
409 int (*ml_read_objids)(struct obd_device *obd, struct md_lov_info *mli,
411 int (*ml_write_objids)(struct obd_device *obd, struct md_lov_info *mli,
415 struct obd_device *md_lov_obd; /* XXX lov_obd */
416 struct obd_uuid md_lov_uuid;
417 struct obd_export *md_lov_exp; /* XXX lov_exp */
418 struct lov_desc md_lov_desc;
419 obd_id *md_lov_objids;
420 int md_lov_objids_size;
421 __u32 md_lov_objids_in_file;
422 unsigned int md_lov_objids_dirty:1;
423 int md_lov_nextid_set;
424 void *md_lov_objid_obj;
425 struct lu_fid md_lov_objid_fid;
426 unsigned long md_lov_objids_valid:1;
427 int md_lov_max_mdsize;
428 int md_lov_max_cookiesize;
429 struct semaphore md_lov_orphan_recovery_sem;
430 struct md_lov_ops *md_lov_ops;
433 #define mds_osc_obd mds_lov_info.md_lov_obd
434 #define mds_lov_uuid mds_lov_info.md_lov_uuid
435 #define mds_osc_exp mds_lov_info.md_lov_exp
436 #define mds_lov_desc mds_lov_info.md_lov_desc
437 #define mds_lov_objids mds_lov_info.md_lov_objids
438 #define mds_lov_objids_size mds_lov_info.md_lov_objids_size
439 #define mds_lov_objids_in_file mds_lov_info.md_lov_objids_in_file
440 #define mds_lov_objids_dirty mds_lov_info.md_lov_objids_dirty
441 #define mds_lov_nextid_set mds_lov_info.md_lov_nextid_set
442 #define mds_lov_objid_filp mds_lov_info.md_lov_objid_obj
443 #define mds_lov_objids_valid mds_lov_info.md_lov_objids_valid
444 #define mds_max_mdsize mds_lov_info.md_lov_max_mdsize
445 #define mds_max_cookiesize mds_lov_info.md_lov_max_cookiesize
446 #define mds_orphan_recovery_sem mds_lov_info.md_lov_orphan_recovery_sem
449 /* NB this field MUST be first */
450 struct obd_device_target mds_obt;
451 struct ptlrpc_service *mds_service;
452 struct ptlrpc_service *mds_setattr_service;
453 struct ptlrpc_service *mds_readpage_service;
454 struct vfsmount *mds_vfsmnt;
455 cfs_dentry_t *mds_fid_de;
456 struct file *mds_rcvd_filp;
457 spinlock_t mds_transno_lock;
458 __u64 mds_last_transno;
459 __u64 mds_mount_count;
461 unsigned long mds_atime_diff;
462 struct semaphore mds_epoch_sem;
463 struct ll_fid mds_rootfid;
464 struct lr_server_data *mds_server_data;
465 cfs_dentry_t *mds_pending_dir;
466 cfs_dentry_t *mds_logs_dir;
467 cfs_dentry_t *mds_objects_dir;
468 struct llog_handle *mds_cfg_llh;
469 // struct llog_handle *mds_catalog;
470 struct md_lov_info mds_lov_info;
472 struct file *mds_health_check_filp;
473 unsigned long *mds_client_bitmap;
474 struct upcall_cache *mds_group_hash;
476 struct lustre_quota_info mds_quota_info;
477 struct semaphore mds_qonoff_sem;
478 struct semaphore mds_health_sem;
479 unsigned long mds_fl_user_xattr:1,
487 struct lustre_handle eo_nl_lock;
492 struct ptlrpc_service *ost_service;
493 struct ptlrpc_service *ost_create_service;
494 struct ptlrpc_service *ost_io_service;
495 struct semaphore ost_health_sem;
498 struct echo_client_obd {
499 struct obd_export *ec_exp; /* the local connection to osc/lov */
501 struct list_head ec_objects;
506 struct lov_tgt_desc {
507 struct obd_uuid uuid;
509 struct obd_export *ltd_exp;
510 unsigned int active:1, /* is this target up for requests */
511 reap:1; /* should this target be deleted */
512 int index; /* index of target array in lov_obd */
513 struct list_head qos_bavail_list; /* link entry to lov_obd */
517 struct semaphore lov_lock;
519 struct lov_desc desc;
520 struct obd_connect_data ocd;
523 int death_row; /* Do we have tgts scheduled to be deleted?
524 (Make this a linked list?) */
525 struct list_head qos_bavail_list; /* tgts list, sorted by available
526 space, protected by lov_lock */
527 struct lov_tgt_desc *tgts;
530 struct lmv_tgt_desc {
531 struct obd_uuid uuid;
532 struct obd_export *ltd_exp;
533 int active; /* is this target up for requests */
539 struct lu_client_fld lmv_fld;
541 struct lmv_desc desc;
542 struct obd_uuid cluuid;
543 struct obd_export *exp;
550 struct semaphore init_sem;
552 struct lmv_tgt_desc *tgts;
555 struct obd_connect_data *datas;
558 struct obd_connect_data conn_data;
561 struct niobuf_local {
566 cfs_dentry_t *dentry;
571 #define LUSTRE_OPC_MKDIR (1 << 0)
572 #define LUSTRE_OPC_SYMLINK (1 << 1)
573 #define LUSTRE_OPC_MKNOD (1 << 2)
574 #define LUSTRE_OPC_CREATE (1 << 3)
576 struct lu_placement_hint {
577 struct qstr *ph_pname;
578 struct qstr *ph_cname;
582 #define LUSTRE_FLD_NAME "fld"
583 #define LUSTRE_SEQ_NAME "seq"
585 /* device types (not names--FIXME) */
586 /* FIXME all the references to these defines need to be updated */
587 #define LUSTRE_MDS_NAME "mds"
588 #define LUSTRE_MDT_NAME "mdt"
590 /* new MDS layers. Prototype */
591 #define LUSTRE_MDT0_NAME "mdt0"
592 #define LUSTRE_CMM0_NAME "cmm0"
593 #define LUSTRE_MDD0_NAME "mdd0"
594 #define LUSTRE_OSD0_NAME "osd0"
595 #define LUSTRE_MDC0_NAME "mdc0"
597 #define LUSTRE_MDC_NAME "mdc"
598 #define LUSTRE_LOV_NAME "lov"
599 #define LUSTRE_LMV_NAME "lmv"
601 /* FIXME just the names need to be changed */
602 #define LUSTRE_OSS_NAME "ost" /* FIXME oss */
603 #define LUSTRE_OST_NAME "obdfilter" /* FIXME ost */
604 #define LUSTRE_OSTSAN_NAME "sanobdfilter"
606 #define LUSTRE_OSC_NAME "osc"
607 #define LUSTRE_FILTER_NAME "filter"
608 #define LUSTRE_SANOSC_NAME "sanosc"
609 #define LUSTRE_SANOST_NAME "sanost"
610 #define LUSTRE_MGS_NAME "mgs"
611 #define LUSTRE_MGC_NAME "mgc"
613 #define LUSTRE_ECHO_NAME "obdecho"
614 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
616 #define LUSTRE_MGS_OBDNAME "MGS"
617 #define LUSTRE_MGC_OBDNAME "MGC"
619 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
620 #define N_LOCAL_TEMP_PAGE 0x10000000
622 struct obd_trans_info {
625 /* Only used on the server side for tracking acks. */
626 struct oti_req_ack_lock {
627 struct lustre_handle lock;
631 struct llog_cookie oti_onecookie;
632 struct llog_cookie *oti_logcookies;
635 /* initial thread handling transaction */
639 static inline void oti_init(struct obd_trans_info *oti,
640 struct ptlrpc_request *req)
644 memset(oti, 0, sizeof *oti);
649 if (req->rq_repmsg && req->rq_reqmsg != 0)
650 oti->oti_transno = req->rq_repmsg->transno;
651 oti->oti_thread_id = req->rq_svc_thread ? req->rq_svc_thread->t_id : -1;
654 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
659 if (num_cookies == 1)
660 oti->oti_logcookies = &oti->oti_onecookie;
662 OBD_ALLOC(oti->oti_logcookies,
663 num_cookies * sizeof(oti->oti_onecookie));
665 oti->oti_numcookies = num_cookies;
668 static inline void oti_free_cookies(struct obd_trans_info *oti)
670 if (!oti || !oti->oti_logcookies)
673 if (oti->oti_logcookies == &oti->oti_onecookie)
674 LASSERT(oti->oti_numcookies == 1);
676 OBD_FREE(oti->oti_logcookies,
677 oti->oti_numcookies * sizeof(oti->oti_onecookie));
678 oti->oti_logcookies = NULL;
679 oti->oti_numcookies = 0;
684 LLOG_CONFIG_ORIG_CTXT = 0,
685 LLOG_CONFIG_REPL_CTXT = 1,
686 LLOG_MDS_OST_ORIG_CTXT = 2,
687 LLOG_MDS_OST_REPL_CTXT = 3,
688 LLOG_SIZE_ORIG_CTXT = 4,
689 LLOG_SIZE_REPL_CTXT = 5,
690 LLOG_MD_ORIG_CTXT = 6,
691 LLOG_MD_REPL_CTXT = 7,
692 LLOG_RD1_ORIG_CTXT = 8,
693 LLOG_RD1_REPL_CTXT = 9,
694 LLOG_TEST_ORIG_CTXT = 10,
695 LLOG_TEST_REPL_CTXT = 11,
696 LLOG_LOVEA_ORIG_CTXT = 12,
697 LLOG_LOVEA_REPL_CTXT = 13,
702 * Events signalled through obd_notify() upcall-chain.
704 enum obd_notify_event {
705 /* Device activated */
707 /* Device deactivated */
709 /* Connect data for import were changed */
712 OBD_NOTIFY_SYNC_NONBLOCK,
716 #include <lu_object.h>
718 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
719 * and liblustre being main examples).
721 struct obd_notify_upcall {
722 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
723 enum obd_notify_event ev, void *owner);
724 /* Opaque datum supplied by upper layer listener */
728 /* corresponds to one of the obd's */
730 struct obd_type *obd_type;
731 /* common and UUID name of this device */
733 struct obd_uuid obd_uuid;
735 struct lu_device *obd_lu_dev;
738 unsigned int obd_attached:1, obd_set_up:1, obd_recovering:1,
739 obd_abort_recovery:1, obd_replayable:1, obd_no_transno:1,
740 obd_no_recov:1, obd_stopping:1, obd_starting:1,
741 obd_force:1, obd_fail:1, obd_async_recov:1;
742 atomic_t obd_refcount;
743 cfs_waitq_t obd_refcount_waitq;
744 cfs_proc_dir_entry_t *obd_proc_entry;
745 struct list_head obd_exports;
747 struct ldlm_namespace *obd_namespace;
748 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
749 /* a spinlock is OK for what we do now, may need a semaphore later */
750 spinlock_t obd_dev_lock;
751 __u64 obd_last_committed;
752 struct fsfilt_operations *obd_fsops;
753 spinlock_t obd_osfs_lock;
754 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
755 cfs_time_t obd_osfs_age;
756 struct lvfs_run_ctxt obd_lvfs_ctxt;
757 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
758 struct obd_device *obd_observer;
759 struct obd_notify_upcall obd_upcall;
760 struct obd_export *obd_self_export;
761 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
762 struct list_head obd_exports_timed;
763 time_t obd_eviction_timer; /* for ping evictor */
765 /* XXX encapsulate all this recovery data into one struct */
766 svc_handler_t obd_recovery_handler;
767 int obd_max_recoverable_clients;
768 int obd_connected_clients;
769 int obd_recoverable_clients;
770 spinlock_t obd_processing_task_lock;
771 pid_t obd_processing_task;
772 __u64 obd_next_recovery_transno;
773 int obd_replayed_requests;
774 int obd_requests_queued_for_recovery;
775 cfs_waitq_t obd_next_transno_waitq;
776 struct list_head obd_uncommitted_replies;
777 spinlock_t obd_uncommitted_replies_lock;
778 cfs_timer_t obd_recovery_timer;
779 struct list_head obd_recovery_queue;
780 struct list_head obd_delayed_reply_queue;
781 time_t obd_recovery_start;
782 time_t obd_recovery_end;
785 struct obd_device_target obt;
786 struct filter_obd filter;
788 struct client_obd cli;
790 struct echo_client_obd echo_client;
791 struct echo_obd echo;
796 /* Fields used by LProcFS */
797 unsigned int obd_cntr_base;
798 struct lprocfs_stats *obd_stats;
800 unsigned int md_cntr_base;
801 struct lprocfs_stats *md_stats;
803 cfs_proc_dir_entry_t *obd_svc_procroot;
804 struct lprocfs_stats *obd_svc_stats;
807 #define OBD_OPT_FORCE 0x0001
808 #define OBD_OPT_FAILOVER 0x0002
810 #define OBD_LLOG_FL_SENDNOW 0x0001
812 enum obd_cleanup_stage {
813 /* Special case hack for MDS LOVs */
815 /* Precleanup stage 1, we must make sure all exports (other than the
816 self-export) get destroyed. */
818 /* Precleanup stage 2, do other type-specific cleanup requiring the
820 OBD_CLEANUP_SELF_EXP,
821 /* FIXME we should eliminate the "precleanup" function and make them stages
822 of the "cleanup" function. */
826 /* get/set_info keys */
827 #define KEY_MDS_CONN "mds_conn"
828 #define KEY_NEXT_ID "next_id"
829 #define KEY_LOVDESC "lovdesc"
830 #define KEY_INIT_RECOV "initial_recov"
831 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
836 struct module *o_owner;
837 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
838 void *karg, void *uarg);
839 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
840 __u32 *vallen, void *val);
841 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
842 __u32 vallen, void *val,
843 struct ptlrpc_request_set *set);
844 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
845 int (*o_detach)(struct obd_device *dev);
846 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
847 int (*o_precleanup)(struct obd_device *dev,
848 enum obd_cleanup_stage cleanup_stage);
849 int (*o_cleanup)(struct obd_device *dev);
850 int (*o_process_config)(struct obd_device *dev, obd_count len,
852 int (*o_postrecov)(struct obd_device *dev);
853 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
855 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
856 /* connect to the target device with given connection
857 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
858 * granted by the target, which are guaranteed to be a subset of flags
859 * asked for. If @ocd == NULL, use default parameters. */
860 int (*o_connect)(const struct lu_context *ctx,
861 struct lustre_handle *conn, struct obd_device *src,
862 struct obd_uuid *cluuid, struct obd_connect_data *ocd);
863 int (*o_reconnect)(struct obd_export *exp, struct obd_device *src,
864 struct obd_uuid *cluuid,
865 struct obd_connect_data *ocd);
866 int (*o_disconnect)(struct obd_export *exp);
868 /* maybe later these should be moved into separate fid_ops */
869 int (*o_fid_init)(struct obd_export *exp);
870 int (*o_fid_fini)(struct obd_export *exp);
872 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
873 struct lu_placement_hint *hint);
875 int (*o_fid_delete)(struct obd_export *exp, struct lu_fid *fid);
877 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
878 unsigned long max_age);
879 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
880 struct lov_stripe_md *mem_src);
881 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
882 struct lov_mds_md *disk_src, int disk_len);
883 int (*o_checkmd)(struct obd_export *exp, struct obd_export *md_exp,
884 struct lov_stripe_md *mem_tgt);
885 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
887 int (*o_create)(struct obd_export *exp, struct obdo *oa,
888 struct lov_stripe_md **ea, struct obd_trans_info *oti);
889 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
890 struct lov_stripe_md *ea, struct obd_trans_info *oti,
891 struct obd_export *md_exp);
892 int (*o_setattr)(struct obd_export *exp, struct obdo *oa,
893 struct lov_stripe_md *ea, struct obd_trans_info *oti);
894 int (*o_setattr_async)(struct obd_export *exp, struct obdo *oa,
895 struct lov_stripe_md *ea, struct obd_trans_info *oti);
896 int (*o_getattr)(struct obd_export *exp, struct obdo *oa,
897 struct lov_stripe_md *ea);
898 int (*o_getattr_async)(struct obd_export *exp, struct obdo *oa,
899 struct lov_stripe_md *ea,
900 struct ptlrpc_request_set *set);
901 int (*o_brw)(int rw, struct obd_export *exp, struct obdo *oa,
902 struct lov_stripe_md *ea, obd_count oa_bufs,
903 struct brw_page *pgarr, struct obd_trans_info *oti);
904 int (*o_brw_async)(int rw, struct obd_export *exp, struct obdo *oa,
905 struct lov_stripe_md *ea, obd_count oa_bufs,
906 struct brw_page *pgarr, struct ptlrpc_request_set *,
907 struct obd_trans_info *oti);
908 int (*o_prep_async_page)(struct obd_export *exp,
909 struct lov_stripe_md *lsm,
910 struct lov_oinfo *loi,
911 struct page *page, obd_off offset,
912 struct obd_async_page_ops *ops, void *data,
914 int (*o_queue_async_io)(struct obd_export *exp,
915 struct lov_stripe_md *lsm,
916 struct lov_oinfo *loi, void *cookie,
917 int cmd, obd_off off, int count,
918 obd_flag brw_flags, obd_flag async_flags);
919 int (*o_queue_group_io)(struct obd_export *exp,
920 struct lov_stripe_md *lsm,
921 struct lov_oinfo *loi,
922 struct obd_io_group *oig,
923 void *cookie, int cmd, obd_off off, int count,
924 obd_flag brw_flags, obd_flag async_flags);
925 int (*o_trigger_group_io)(struct obd_export *exp,
926 struct lov_stripe_md *lsm,
927 struct lov_oinfo *loi,
928 struct obd_io_group *oig);
929 int (*o_set_async_flags)(struct obd_export *exp,
930 struct lov_stripe_md *lsm,
931 struct lov_oinfo *loi, void *cookie,
932 obd_flag async_flags);
933 int (*o_teardown_async_page)(struct obd_export *exp,
934 struct lov_stripe_md *lsm,
935 struct lov_oinfo *loi, void *cookie);
936 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
937 struct ost_lvb *lvb, int kms_only);
938 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
939 obd_off size, int shrink);
940 int (*o_punch)(struct obd_export *exp, struct obdo *oa,
941 struct lov_stripe_md *ea, obd_size start,
942 obd_size end, struct obd_trans_info *oti);
943 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
944 struct lov_stripe_md *ea, obd_size start, obd_size end);
945 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
946 struct lov_stripe_md *src, obd_size start,
947 obd_size end, struct obd_trans_info *oti);
948 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
949 struct lustre_handle *srconn, struct lov_stripe_md *src,
950 obd_size start, obd_size end, struct obd_trans_info *);
951 int (*o_iterate)(struct lustre_handle *conn,
952 int (*)(obd_id, obd_gr, void *),
953 obd_id *startid, obd_gr group, void *data);
954 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
955 int objcount, struct obd_ioobj *obj,
956 int niocount, struct niobuf_remote *remote,
957 struct niobuf_local *local, struct obd_trans_info *oti);
958 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
959 int objcount, struct obd_ioobj *obj,
960 int niocount, struct niobuf_local *local,
961 struct obd_trans_info *oti, int rc);
962 int (*o_enqueue)(struct obd_export *, struct lov_stripe_md *,
963 __u32 type, ldlm_policy_data_t *, __u32 mode,
964 int *flags, void *bl_cb, void *cp_cb, void *gl_cb,
965 void *data, __u32 lvb_len, void *lvb_swabber,
966 struct lustre_handle *lockh);
967 int (*o_match)(struct obd_export *, struct lov_stripe_md *, __u32 type,
968 ldlm_policy_data_t *, __u32 mode, int *flags, void *data,
969 struct lustre_handle *lockh);
970 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
971 ldlm_iterator_t it, void *data);
972 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
973 __u32 mode, struct lustre_handle *);
974 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
975 int flags, void *opaque);
976 int (*o_join_lru)(struct obd_export *, struct lov_stripe_md *,
978 int (*o_san_preprw)(int cmd, struct obd_export *exp,
979 struct obdo *oa, int objcount,
980 struct obd_ioobj *obj, int niocount,
981 struct niobuf_remote *remote);
982 int (*o_init_export)(struct obd_export *exp);
983 int (*o_destroy_export)(struct obd_export *exp);
985 /* llog related obd_methods */
986 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
987 int count, struct llog_catid *logid);
988 int (*o_llog_finish)(struct obd_device *obd, int count);
990 /* metadata-only methods */
991 int (*o_pin)(struct obd_export *, struct lu_fid *fid,
992 struct obd_client_handle *, int flag);
993 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
995 int (*o_import_event)(struct obd_device *, struct obd_import *,
996 enum obd_import_event);
998 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
999 enum obd_notify_event ev, void *data);
1001 int (*o_health_check)(struct obd_device *);
1004 int (*o_quotacheck)(struct obd_export *, struct obd_quotactl *);
1005 int (*o_quotactl)(struct obd_export *, struct obd_quotactl *);
1008 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1009 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1010 * Also, add a wrapper function in include/linux/obd_class.h.
1012 * Also note that if you add it to the END, you also have to change
1013 * the num_stats calculation.
1019 int (*m_getstatus)(struct obd_export *, struct lu_fid *);
1020 int (*m_change_cbdata)(struct obd_export *, struct lu_fid *,
1021 ldlm_iterator_t, void *);
1022 int (*m_close)(struct obd_export *, struct md_op_data *,
1023 struct obd_client_handle *, struct ptlrpc_request **);
1024 int (*m_create)(struct obd_export *, struct md_op_data *,
1025 const void *, int, int, __u32, __u32, __u32,
1026 __u64, struct ptlrpc_request **);
1027 int (*m_done_writing)(struct obd_export *, struct md_op_data *);
1028 int (*m_enqueue)(struct obd_export *, int, struct lookup_intent *,
1029 int, struct md_op_data *, struct lustre_handle *,
1030 void *, int, ldlm_completion_callback,
1031 ldlm_blocking_callback, void *, int);
1032 int (*m_getattr)(struct obd_export *, struct lu_fid *,
1033 obd_valid, int, struct ptlrpc_request **);
1034 int (*m_getattr_name)(struct obd_export *, struct lu_fid *,
1035 const char *, int, obd_valid,
1036 int, struct ptlrpc_request **);
1037 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1038 void *, int, struct lookup_intent *, int,
1039 struct ptlrpc_request **,
1040 ldlm_blocking_callback, int);
1041 int (*m_link)(struct obd_export *, struct md_op_data *,
1042 struct ptlrpc_request **);
1043 int (*m_rename)(struct obd_export *, struct md_op_data *,
1044 const char *, int, const char *, int,
1045 struct ptlrpc_request **);
1046 int (*m_setattr)(struct obd_export *, struct md_op_data *,
1047 struct iattr *, void *, int , void *, int,
1048 struct ptlrpc_request **);
1049 int (*m_sync)(struct obd_export *, struct lu_fid *,
1050 struct ptlrpc_request **);
1051 int (*m_readpage)(struct obd_export *, struct lu_fid *,
1052 __u64, struct page *, struct ptlrpc_request **);
1053 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1054 struct ptlrpc_request **);
1056 int (*m_setxattr)(struct obd_export *, struct lu_fid *,
1057 obd_valid, const char *, const char *,
1058 int, int, int, struct ptlrpc_request **);
1060 int (*m_getxattr)(struct obd_export *, struct lu_fid *,
1061 obd_valid, const char *, const char *,
1062 int, int, int, struct ptlrpc_request **);
1064 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1066 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1067 int, struct obd_export *, struct lustre_md *);
1069 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1071 int (*m_set_open_replay_data)(struct obd_export *,
1072 struct obd_client_handle *,
1073 struct ptlrpc_request *);
1074 int (*m_clear_open_replay_data)(struct obd_export *,
1075 struct obd_client_handle *);
1076 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *);
1078 int (*m_lock_match)(struct obd_export *, int, struct lu_fid *,
1079 ldlm_type_t, ldlm_policy_data_t *, ldlm_mode_t,
1080 struct lustre_handle *);
1082 int (*m_cancel_unused)(struct obd_export *, struct lu_fid *,
1083 int flags, void *opaque);
1086 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1087 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1088 * wrapper function in include/linux/obd_class.h.
1092 struct lsm_operations {
1093 void (*lsm_free)(struct lov_stripe_md *);
1094 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1095 struct obd_export *md_exp);
1096 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1098 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1100 obd_off (*lsm_stripe_offset_by_index)(struct lov_stripe_md *, int);
1101 int (*lsm_stripe_index_by_offset)(struct lov_stripe_md *, obd_off);
1102 int (*lsm_revalidate) (struct lov_stripe_md *, struct obd_device *obd);
1103 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1105 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1106 struct lov_mds_md *lmm);
1109 extern struct lsm_operations lsm_plain_ops;
1110 extern struct lsm_operations lsm_join_ops;
1111 static inline struct lsm_operations *lsm_op_find(int magic)
1115 return &lsm_plain_ops;
1116 case LOV_MAGIC_JOIN:
1117 return &lsm_join_ops;
1119 CERROR("Cannot recognize lsm_magic %d", magic);
1124 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1126 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1130 CERROR("%s: transno "LPD64" commit error: %d\n",
1131 obd->obd_name, transno, error);
1134 CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
1135 obd->obd_name, transno);
1136 if (transno > obd->obd_last_committed) {
1137 obd->obd_last_committed = transno;
1138 ptlrpc_commit_replies (obd);
1142 static inline void init_obd_quota_ops(quota_interface_t *interface,
1143 struct obd_ops *obd_ops)
1149 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1150 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1153 #endif /* __OBD_H */