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_ioctl_data *)
27 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
28 #define IOC_MDC_MAX_NR 50
30 #include <lustre_lib.h>
31 #include <lustre/lustre_idl.h>
32 #include <lustre_export.h>
33 #include <lustre_quota.h>
34 #include <class_hash.h>
36 #define MAX_OBD_DEVICES 8192
38 /* this is really local to the OSC */
39 struct loi_oap_pages {
40 struct list_head lop_pending;
41 struct list_head lop_urgent;
42 struct list_head lop_pending_group;
52 struct lov_oinfo { /* per-stripe data structure */
53 __u64 loi_id; /* object ID on the target OST */
54 __u64 loi_gr; /* object group on the target OST */
55 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
56 int loi_ost_gen; /* generation of this loi_ost_idx */
58 /* used by the osc to keep track of what objects to build into rpcs */
59 struct loi_oap_pages loi_read_lop;
60 struct loi_oap_pages loi_write_lop;
61 /* _cli_ is poorly named, it should be _ready_ */
62 struct list_head loi_cli_item;
63 struct list_head loi_write_item;
64 struct list_head loi_read_item;
66 unsigned loi_kms_valid:1;
67 __u64 loi_kms; /* known minimum size */
68 struct ost_lvb loi_lvb;
69 struct osc_async_rc loi_ar;
72 static inline void loi_init(struct lov_oinfo *loi)
74 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
75 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
76 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
77 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
78 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
79 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
80 CFS_INIT_LIST_HEAD(&loi->loi_cli_item);
81 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
82 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
85 /*extent array item for describing the joined file extent info*/
87 __u64 le_start; /* extent start */
88 __u64 le_len; /* extent length */
89 int le_loi_idx; /* extent #1 loi's index in lsm loi array */
90 int le_stripe_count; /* extent stripe count*/
93 /*Lov array info for describing joined file array EA info*/
94 struct lov_array_info {
95 struct llog_logid lai_array_id; /* MDS med llog object id */
96 unsigned lai_ext_count; /* number of extent count */
97 struct lov_extent *lai_ext_array; /* extent desc array */
100 struct lov_stripe_md {
102 void *lsm_lock_owner; /* debugging */
105 /* Public members. */
106 __u64 lw_object_id; /* lov object id */
107 __u64 lw_object_gr; /* lov object group */
108 __u64 lw_maxbytes; /* maximum possible file 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_magic lsm_wire.lw_magic
125 #define lsm_stripe_size lsm_wire.lw_stripe_size
126 #define lsm_pattern lsm_wire.lw_pattern
127 #define lsm_stripe_count lsm_wire.lw_stripe_count
131 typedef int (*obd_enqueue_update_f)(struct obd_info *oinfo, int rc);
133 /* obd info for a particular level (lov, osc). */
135 /* Lock policy. It keeps an extent which is specific for a particular
136 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
137 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
138 ldlm_policy_data_t oi_policy;
139 /* Flags used while lock handling. The flags obtained on the enqueue
140 * request are set here, therefore they are request specific. */
142 /* Lock handle specific for every OSC lock. */
143 struct lustre_handle *oi_lockh;
144 /* lsm data specific for every OSC. */
145 struct lov_stripe_md *oi_md;
146 /* obdo data specific for every OSC, if needed at all. */
148 /* statfs data specific for every OSC, if needed at all. */
149 struct obd_statfs *oi_osfs;
150 /* An update callback which is called to update some data on upper
151 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
152 * request in osc level for enqueue requests. It is also possible to
153 * update some caller data from LOV layer if needed. */
154 obd_enqueue_update_f oi_cb_up;
157 /* compare all relevant fields. */
158 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
159 struct lov_stripe_md *m2)
162 * ->lsm_wire contains padding, but it should be zeroed out during
165 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
168 void lov_stripe_lock(struct lov_stripe_md *md);
169 void lov_stripe_unlock(struct lov_stripe_md *md);
172 struct list_head typ_chain;
173 struct obd_ops *typ_ops;
174 cfs_proc_dir_entry_t *typ_procroot;
177 spinlock_t obd_type_lock;
188 ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
189 page is added to an rpc */
190 ASYNC_URGENT = 0x2, /* page must be put into an RPC before return */
191 ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
192 to give the caller a chance to update
193 or cancel the size of the io */
194 ASYNC_GROUP_SYNC = 0x8, /* ap_completion will not be called, instead
195 the page is accounted for in the
196 obd_io_group given to
197 obd_queue_group_io */
200 struct obd_async_page_ops {
201 int (*ap_make_ready)(void *data, int cmd);
202 int (*ap_refresh_count)(void *data, int cmd);
203 void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
204 void (*ap_update_obdo)(void *data, int cmd, struct obdo *oa,
206 int (*ap_completion)(void *data, int cmd, struct obdo *oa, int rc);
209 /* the `oig' is passed down from a caller of obd rw methods. the callee
210 * records enough state such that the caller can sleep on the oig and
211 * be woken when all the callees have finished their work */
212 struct obd_io_group {
214 atomic_t oig_refcount;
217 struct list_head oig_occ_list;
218 cfs_waitq_t oig_waitq;
221 /* the oig callback context lets the callee of obd rw methods register
222 * for callbacks from the caller. */
223 struct oig_callback_context {
224 struct list_head occ_oig_item;
225 /* called when the caller has received a signal while sleeping.
226 * callees of this method are encouraged to abort their state
227 * in the oig. This may be called multiple times. */
228 void (*occ_interrupted)(struct oig_callback_context *occ);
229 unsigned int interrupted:1;
232 /* Individual type definitions */
234 struct ost_server_data;
236 /* hold common fields for "target" device */
237 struct obd_device_target {
238 struct super_block *obt_sb;
239 atomic_t obt_quotachecking;
240 struct lustre_quota_ctxt obt_qctxt;
243 #define FILTER_GROUP_LLOG 1
244 #define FILTER_GROUP_ECHO 2
252 /* NB this field MUST be first */
253 struct obd_device_target fo_obt;
254 const char *fo_fstype;
255 struct vfsmount *fo_vfsmnt;
256 cfs_dentry_t *fo_dentry_O;
257 cfs_dentry_t **fo_dentry_O_groups;
258 cfs_dentry_t **fo_dentry_O_sub;
259 spinlock_t fo_objidlock; /* protect fo_lastobjid */
260 spinlock_t fo_translock; /* protect fsd_last_transno */
261 struct file *fo_rcvd_filp;
262 struct file *fo_health_check_filp;
263 struct lr_server_data *fo_fsd;
264 unsigned long *fo_last_rcvd_slots;
265 __u64 fo_mount_count;
267 int fo_destroy_in_progress;
268 struct semaphore fo_create_lock;
270 struct list_head fo_export_list;
273 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
274 obd_size fo_tot_granted; /* all values in bytes */
275 obd_size fo_tot_pending;
277 obd_size fo_readcache_max_filesize;
279 struct obd_import *fo_mdc_imp;
280 struct obd_uuid fo_mdc_uuid;
281 struct lustre_handle fo_mdc_conn;
282 struct file **fo_last_objid_files;
283 __u64 *fo_last_objids; /* last created objid for groups,
284 * protected by fo_objidlock */
286 struct semaphore fo_alloc_lock;
288 atomic_t fo_r_in_flight;
289 atomic_t fo_w_in_flight;
292 * per-filter pool of kiobuf's allocated by filter_common_setup() and
293 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
294 * which ->fo_iobuf_count were allocated.
296 * This pool contains kiobuf used by
297 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
300 * Locking: none, each OST thread uses only one element, determined by
301 * its "ordinal number", ->t_id.
303 struct filter_iobuf **fo_iobuf_pool;
306 struct brw_stats fo_filter_stats;
307 struct lustre_quota_ctxt fo_quota_ctxt;
308 spinlock_t fo_quotacheck_lock;
309 atomic_t fo_quotachecking;
311 int fo_fmd_max_num; /* per exp filter_mod_data */
312 int fo_fmd_max_age; /* jiffies to fmd expiry */
316 #define OSC_MAX_RIF_DEFAULT 8
317 #define OSC_MAX_RIF_MAX 256
318 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
319 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
321 #define MDC_MAX_RIF_DEFAULT 8
322 #define MDC_MAX_RIF_MAX 512
327 struct semaphore cl_sem;
328 struct obd_uuid cl_target_uuid;
329 struct obd_import *cl_import; /* ptlrpc connection state */
331 /* max_mds_easize is purely a performance thing so we don't have to
332 * call obd_size_diskmd() all the time. */
333 int cl_default_mds_easize;
334 int cl_max_mds_easize;
335 int cl_max_mds_cookiesize;
337 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
338 void *cl_llcd_offset;
340 /* the grant values are protected by loi_list_lock below */
341 long cl_dirty; /* all _dirty_ in bytes */
342 long cl_dirty_max; /* allowed w/o rpc */
343 long cl_avail_grant; /* bytes of credit for ost */
344 long cl_lost_grant; /* lost credits (trunc) */
345 struct list_head cl_cache_waiters; /* waiting for cache/grant */
347 /* keep track of objects that have lois that contain pages which
348 * have been queued for async brw. this lock also protects the
349 * lists of osc_client_pages that hang off of the loi */
351 * ->cl_loi_list_lock protects consistency of
352 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
353 * ->ap_completion() call-backs are executed under this lock. As we
354 * cannot guarantee that these call-backs never block on all platforms
355 * (as a matter of fact they do block on Mac OS X), type of
356 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
357 * and blocking mutex on Mac OS X. (Alternative is to make this lock
358 * blocking everywhere, but we don't want to slow down fast-path of
359 * our main platform.)
361 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
362 * with client_obd_list_{un,}lock() and
363 * client_obd_list_lock_{init,done}() functions.
365 client_obd_lock_t cl_loi_list_lock;
366 struct list_head cl_loi_ready_list;
367 struct list_head cl_loi_write_list;
368 struct list_head cl_loi_read_list;
371 /* just a sum of the loi/lop pending numbers to be exported by /proc */
372 int cl_pending_w_pages;
373 int cl_pending_r_pages;
374 int cl_max_pages_per_rpc;
375 int cl_max_rpcs_in_flight;
376 struct obd_histogram cl_read_rpc_hist;
377 struct obd_histogram cl_write_rpc_hist;
378 struct obd_histogram cl_read_page_hist;
379 struct obd_histogram cl_write_page_hist;
380 struct obd_histogram cl_read_offset_hist;
381 struct obd_histogram cl_write_offset_hist;
383 struct mdc_rpc_lock *cl_rpc_lock;
384 struct mdc_rpc_lock *cl_setattr_lock;
385 struct mdc_rpc_lock *cl_close_lock;
386 struct osc_creator cl_oscc;
389 struct semaphore cl_mgc_sem;
390 struct vfsmount *cl_mgc_vfsmnt;
391 struct dentry *cl_mgc_configs_dir;
392 atomic_t cl_mgc_refcount;
393 struct obd_export *cl_mgc_mgsexp;
396 unsigned int cl_checksum:1; /* debug checksums */
398 /* also protected by the poorly named _loi_list_lock lock above */
399 struct osc_async_rc cl_ar;
401 /* used by quotacheck */
402 int cl_qchk_stat; /* quotacheck stat of the peer */
404 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
406 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
409 struct ptlrpc_service *mgs_service;
410 struct vfsmount *mgs_vfsmnt;
411 struct super_block *mgs_sb;
412 struct dentry *mgs_configs_dir;
413 struct dentry *mgs_fid_de;
414 struct list_head mgs_fs_db_list;
415 struct semaphore mgs_sem;
416 cfs_proc_dir_entry_t *mgs_proc_live;
420 /* NB this field MUST be first */
421 struct obd_device_target mds_obt;
422 struct ptlrpc_service *mds_service;
423 struct ptlrpc_service *mds_setattr_service;
424 struct ptlrpc_service *mds_readpage_service;
425 struct vfsmount *mds_vfsmnt;
426 cfs_dentry_t *mds_fid_de;
428 int mds_max_cookiesize;
429 struct file *mds_rcvd_filp;
430 spinlock_t mds_transno_lock;
431 __u64 mds_last_transno;
432 __u64 mds_mount_count;
434 unsigned long mds_atime_diff;
435 struct semaphore mds_epoch_sem;
436 struct ll_fid mds_rootfid;
437 struct lr_server_data *mds_server_data;
438 cfs_dentry_t *mds_pending_dir;
439 cfs_dentry_t *mds_logs_dir;
440 cfs_dentry_t *mds_objects_dir;
441 struct llog_handle *mds_cfg_llh;
442 // struct llog_handle *mds_catalog;
443 struct obd_device *mds_osc_obd; /* XXX lov_obd */
444 struct obd_uuid mds_lov_uuid;
446 struct obd_export *mds_osc_exp; /* XXX lov_exp */
447 struct lov_desc mds_lov_desc;
448 obd_id *mds_lov_objids;
449 int mds_lov_objids_size;
450 __u32 mds_lov_objids_in_file;
451 int mds_lov_nextid_set;
452 struct file *mds_lov_objid_filp;
453 struct file *mds_health_check_filp;
454 unsigned long *mds_client_bitmap;
455 struct upcall_cache *mds_group_hash;
457 struct lustre_quota_info mds_quota_info;
458 struct semaphore mds_qonoff_sem;
459 struct semaphore mds_health_sem;
460 unsigned long mds_lov_objids_valid:1,
461 mds_lov_objids_dirty:1,
472 struct lustre_handle eo_nl_lock;
477 struct ptlrpc_service *ost_service;
478 struct ptlrpc_service *ost_create_service;
479 struct ptlrpc_service *ost_io_service;
480 struct semaphore ost_health_sem;
483 struct echo_client_obd {
484 struct obd_export *ec_exp; /* the local connection to osc/lov */
486 struct list_head ec_objects;
492 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
493 struct list_head lqo_oss_list; /* link to lov_qos */
494 __u32 lqo_ost_count; /* number of osts on this oss */
495 __u64 lqo_bavail; /* total bytes avail on OSS */
496 __u64 lqo_penalty; /* current penalty */
497 __u64 lqo_penalty_per_obj; /* penalty decrease every obj*/
501 struct lov_qos_oss *ltq_oss; /* oss info */
502 __u64 ltq_penalty; /* current penalty */
503 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
504 __u64 ltq_weight; /* net weighting */
505 unsigned int ltq_usable:1; /* usable for striping */
509 struct list_head lq_oss_list; /* list of OSSs that targets use */
510 struct rw_semaphore lq_rw_sem;
511 __u32 lq_active_oss_count;
512 __u32 *lq_rr_array; /* round-robin optimized list */
513 unsigned int lq_rr_size; /* rr array size */
514 unsigned int lq_prio_free; /* priority for free space */
515 unsigned int lq_dirty:1, /* recalc qos data */
516 lq_dirty_rr:1, /* recalc round-robin list */
517 lq_same_space:1,/* the ost's all have approx.
518 the same space avail */
519 lq_reset:1; /* zero current penalties */
522 struct lov_tgt_desc {
523 struct obd_uuid ltd_uuid;
524 struct obd_export *ltd_exp;
525 struct ltd_qos ltd_qos; /* qos info per target */
527 __u32 ltd_index; /* index in lov_obd->tgts */
528 unsigned int ltd_active:1,/* is this target up for requests */
529 ltd_activate:1,/* should this target be activated */
530 ltd_reap:1; /* should this target be deleted */
534 struct lov_desc desc;
535 struct lov_tgt_desc **lov_tgts;
536 struct semaphore lov_lock;
537 struct obd_connect_data lov_ocd;
538 struct lov_qos lov_qos; /* qos info per lov */
539 atomic_t lov_refcount;
540 __u32 lov_tgt_count; /* how many OBD's */
541 __u32 lov_active_tgt_count; /* how many active */
542 __u32 lov_death_row;/* tgts scheduled to be deleted */
543 __u32 lov_tgt_size; /* size of tgts array */
544 __u32 lov_start_idx; /* start index of new inode */
545 __u32 lov_offset_idx; /* aliasing for start_idx */
546 int lov_start_count;/* reseed counter */
550 struct niobuf_local {
555 cfs_dentry_t *dentry;
560 /* obd device type names */
561 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
562 #define LUSTRE_MDS_NAME "mds"
563 #define LUSTRE_MDT_NAME "mdt"
564 #define LUSTRE_MDC_NAME "mdc"
565 #define LUSTRE_OSS_NAME "ost" /*FIXME change name to oss*/
566 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost*/
567 #define LUSTRE_OSC_NAME "osc"
568 #define LUSTRE_LOV_NAME "lov"
569 #define LUSTRE_MGS_NAME "mgs"
570 #define LUSTRE_MGC_NAME "mgc"
572 #define LUSTRE_CACHEOBD_NAME "cobd"
573 #define LUSTRE_ECHO_NAME "obdecho"
574 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
576 /* Constant obd names (post-rename) */
577 #define LUSTRE_MDS_OBDNAME "MDS"
578 #define LUSTRE_OSS_OBDNAME "OSS"
579 #define LUSTRE_MGS_OBDNAME "MGS"
580 #define LUSTRE_MGC_OBDNAME "MGC"
582 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
583 #define N_LOCAL_TEMP_PAGE 0x10000000
585 struct obd_trans_info {
589 /* Only used on the server side for tracking acks. */
590 struct oti_req_ack_lock {
591 struct lustre_handle lock;
595 struct llog_cookie oti_onecookie;
596 struct llog_cookie *oti_logcookies;
599 /* initial thread handling transaction */
604 static inline void oti_init(struct obd_trans_info *oti,
605 struct ptlrpc_request *req)
609 memset(oti, 0, sizeof *oti);
614 oti->oti_xid = req->rq_xid;
616 if (req->rq_repmsg && req->rq_reqmsg != 0)
617 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
618 oti->oti_thread_id = req->rq_svc_thread ? req->rq_svc_thread->t_id : -1;
619 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
622 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
627 if (num_cookies == 1)
628 oti->oti_logcookies = &oti->oti_onecookie;
630 OBD_ALLOC(oti->oti_logcookies,
631 num_cookies * sizeof(oti->oti_onecookie));
633 oti->oti_numcookies = num_cookies;
636 static inline void oti_free_cookies(struct obd_trans_info *oti)
638 if (!oti || !oti->oti_logcookies)
641 if (oti->oti_logcookies == &oti->oti_onecookie)
642 LASSERT(oti->oti_numcookies == 1);
644 OBD_FREE(oti->oti_logcookies,
645 oti->oti_numcookies * sizeof(oti->oti_onecookie));
646 oti->oti_logcookies = NULL;
647 oti->oti_numcookies = 0;
652 LLOG_CONFIG_ORIG_CTXT = 0,
653 LLOG_CONFIG_REPL_CTXT = 1,
654 LLOG_MDS_OST_ORIG_CTXT = 2,
655 LLOG_MDS_OST_REPL_CTXT = 3,
656 LLOG_SIZE_ORIG_CTXT = 4,
657 LLOG_SIZE_REPL_CTXT = 5,
658 LLOG_MD_ORIG_CTXT = 6,
659 LLOG_MD_REPL_CTXT = 7,
660 LLOG_RD1_ORIG_CTXT = 8,
661 LLOG_RD1_REPL_CTXT = 9,
662 LLOG_TEST_ORIG_CTXT = 10,
663 LLOG_TEST_REPL_CTXT = 11,
664 LLOG_LOVEA_ORIG_CTXT = 12,
665 LLOG_LOVEA_REPL_CTXT = 13,
670 * Events signalled through obd_notify() upcall-chain.
672 enum obd_notify_event {
673 /* Device activated */
675 /* Device deactivated */
677 /* Connect data for import were changed */
680 OBD_NOTIFY_SYNC_NONBLOCK,
682 /* Configuration event */
686 #define CONFIG_LOG 0x1 /* finished processing config log */
687 #define CONFIG_SYNC 0x2 /* mdt synced 1 ost */
690 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
691 * and liblustre being main examples).
693 struct obd_notify_upcall {
694 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
695 enum obd_notify_event ev, void *owner);
696 /* Opaque datum supplied by upper layer listener */
700 /* corresponds to one of the obd's */
701 #define MAX_OBD_NAME 128
702 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
704 struct obd_type *obd_type;
707 /* common and UUID name of this device */
708 char obd_name[MAX_OBD_NAME];
709 struct obd_uuid obd_uuid;
712 unsigned int obd_attached:1, /* finished attach */
713 obd_set_up:1, /* finished setup */
714 obd_recovering:1, /* there are recoverable clients */
715 obd_abort_recovery:1,/* somebody ioctl'ed us to abort */
716 obd_replayable:1, /* recovery is enabled; inform clients */
717 obd_no_transno:1, /* no committed-transno notification */
718 obd_no_recov:1, /* fail instead of retry messages */
719 obd_stopping:1, /* started cleanup */
720 obd_starting:1, /* started setup */
721 obd_force:1, /* cleanup with > 0 obd refcount */
722 obd_fail:1, /* cleanup with failover */
723 obd_async_recov:1, /* allow asyncronous orphan cleanup */
724 obd_no_conn:1; /* deny new connections */
725 /* uuid-export hash body */
726 struct lustre_class_hash_body *obd_uuid_hash_body;
727 /* nid-export hash body */
728 struct lustre_class_hash_body *obd_nid_hash_body;
729 atomic_t obd_refcount;
730 cfs_waitq_t obd_refcount_waitq;
731 cfs_waitq_t obd_llog_waitq;
732 struct list_head obd_exports;
734 struct ldlm_namespace *obd_namespace;
735 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
736 /* a spinlock is OK for what we do now, may need a semaphore later */
737 spinlock_t obd_dev_lock;
738 struct semaphore obd_dev_sem;
739 __u64 obd_last_committed;
740 struct fsfilt_operations *obd_fsops;
741 spinlock_t obd_osfs_lock;
742 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
744 struct lvfs_run_ctxt obd_lvfs_ctxt;
745 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
746 struct obd_device *obd_observer;
747 struct obd_notify_upcall obd_upcall;
748 struct obd_export *obd_self_export;
749 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
750 struct list_head obd_exports_timed;
751 time_t obd_eviction_timer; /* for ping evictor */
753 /* XXX encapsulate all this recovery data into one struct */
754 svc_handler_t obd_recovery_handler;
755 int obd_max_recoverable_clients;
756 int obd_connected_clients;
757 int obd_recoverable_clients;
758 spinlock_t obd_processing_task_lock; /* BH lock (timer) */
759 pid_t obd_processing_task;
760 __u64 obd_next_recovery_transno;
761 int obd_replayed_requests;
762 int obd_requests_queued_for_recovery;
763 cfs_waitq_t obd_next_transno_waitq;
764 struct list_head obd_uncommitted_replies;
765 spinlock_t obd_uncommitted_replies_lock;
766 cfs_timer_t obd_recovery_timer;
767 struct list_head obd_recovery_queue;
768 struct list_head obd_delayed_reply_queue;
769 time_t obd_recovery_start;
770 time_t obd_recovery_end;
773 struct obd_device_target obt;
774 struct filter_obd filter;
776 struct client_obd cli;
778 struct echo_client_obd echo_client;
779 struct echo_obd echo;
783 /* Fields used by LProcFS */
784 cfs_proc_dir_entry_t *obd_proc_entry;
785 cfs_proc_dir_entry_t *obd_proc_exports;
786 cfs_proc_dir_entry_t *obd_svc_procroot;
787 struct lprocfs_stats *obd_stats;
788 struct lprocfs_stats *obd_svc_stats;
789 unsigned int obd_cntr_base;
790 struct semaphore obd_proc_exp_sem;
793 #define OBD_OPT_FORCE 0x0001
794 #define OBD_OPT_FAILOVER 0x0002
796 #define OBD_LLOG_FL_SENDNOW 0x0001
798 enum obd_cleanup_stage {
799 /* Special case hack for MDS LOVs */
801 /* Precleanup stage 1, we must make sure all exports (other than the
802 self-export) get destroyed. */
804 /* Precleanup stage 2, do other type-specific cleanup requiring the
806 OBD_CLEANUP_SELF_EXP,
807 /* FIXME we should eliminate the "precleanup" function and make them stages
808 of the "cleanup" function. */
812 /* get/set_info keys */
813 #define KEY_MDS_CONN "mds_conn"
814 #define KEY_NEXT_ID "next_id"
815 #define KEY_LOVDESC "lovdesc"
816 #define KEY_INIT_RECOV "initial_recov"
817 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
820 struct module *o_owner;
821 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
822 void *karg, void *uarg);
823 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
824 __u32 *vallen, void *val);
825 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
826 __u32 vallen, void *val,
827 struct ptlrpc_request_set *set);
828 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
829 int (*o_detach)(struct obd_device *dev);
830 int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
831 int (*o_precleanup)(struct obd_device *dev,
832 enum obd_cleanup_stage cleanup_stage);
833 int (*o_cleanup)(struct obd_device *dev);
834 int (*o_process_config)(struct obd_device *dev, obd_count len,
836 int (*o_postrecov)(struct obd_device *dev);
837 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
839 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
840 /* connect to the target device with given connection
841 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
842 * granted by the target, which are guaranteed to be a subset of flags
843 * asked for. If @ocd == NULL, use default parameters. */
844 int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
845 struct obd_uuid *cluuid, struct obd_connect_data *ocd);
846 int (*o_reconnect)(struct obd_export *exp, struct obd_device *src,
847 struct obd_uuid *cluuid,
848 struct obd_connect_data *ocd);
849 int (*o_disconnect)(struct obd_export *exp);
851 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
853 int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
854 __u64 max_age, struct ptlrpc_request_set *set);
855 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
856 struct lov_stripe_md *mem_src);
857 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
858 struct lov_mds_md *disk_src, int disk_len);
859 int (*o_checkmd)(struct obd_export *exp, struct obd_export *md_exp,
860 struct lov_stripe_md *mem_tgt);
861 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
863 int (*o_precreate)(struct obd_export *exp, int need_create);
864 int (*o_create)(struct obd_export *exp, struct obdo *oa,
865 struct lov_stripe_md **ea, struct obd_trans_info *oti);
866 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
867 struct lov_stripe_md *ea, struct obd_trans_info *oti,
868 struct obd_export *md_exp);
869 int (*o_setattr)(struct obd_export *exp, struct obd_info *oinfo,
870 struct obd_trans_info *oti);
871 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
872 struct obd_trans_info *oti,
873 struct ptlrpc_request_set *rqset);
874 int (*o_getattr)(struct obd_export *exp, struct obd_info *oinfo);
875 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
876 struct ptlrpc_request_set *set);
877 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
878 obd_count oa_bufs, struct brw_page *pgarr,
879 struct obd_trans_info *oti);
880 int (*o_brw_async)(int rw, struct obd_export *exp,
881 struct obd_info *oinfo, obd_count oa_bufs,
882 struct brw_page *pgarr, struct obd_trans_info *oti,
883 struct ptlrpc_request_set *);
884 int (*o_prep_async_page)(struct obd_export *exp,
885 struct lov_stripe_md *lsm,
886 struct lov_oinfo *loi,
887 cfs_page_t *page, obd_off offset,
888 struct obd_async_page_ops *ops, void *data,
890 int (*o_queue_async_io)(struct obd_export *exp,
891 struct lov_stripe_md *lsm,
892 struct lov_oinfo *loi, void *cookie,
893 int cmd, obd_off off, int count,
894 obd_flag brw_flags, obd_flag async_flags);
895 int (*o_queue_group_io)(struct obd_export *exp,
896 struct lov_stripe_md *lsm,
897 struct lov_oinfo *loi,
898 struct obd_io_group *oig,
899 void *cookie, int cmd, obd_off off, int count,
900 obd_flag brw_flags, obd_flag async_flags);
901 int (*o_trigger_group_io)(struct obd_export *exp,
902 struct lov_stripe_md *lsm,
903 struct lov_oinfo *loi,
904 struct obd_io_group *oig);
905 int (*o_set_async_flags)(struct obd_export *exp,
906 struct lov_stripe_md *lsm,
907 struct lov_oinfo *loi, void *cookie,
908 obd_flag async_flags);
909 int (*o_teardown_async_page)(struct obd_export *exp,
910 struct lov_stripe_md *lsm,
911 struct lov_oinfo *loi, void *cookie);
912 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
913 struct ost_lvb *lvb, int kms_only);
914 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
915 obd_off size, int shrink);
916 int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
917 struct obd_trans_info *oti,
918 struct ptlrpc_request_set *rqset);
919 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
920 struct lov_stripe_md *ea, obd_size start, obd_size end);
921 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
922 struct lov_stripe_md *src, obd_size start,
923 obd_size end, struct obd_trans_info *oti);
924 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
925 struct lustre_handle *srconn, struct lov_stripe_md *src,
926 obd_size start, obd_size end, struct obd_trans_info *);
927 int (*o_iterate)(struct lustre_handle *conn,
928 int (*)(obd_id, obd_gr, void *),
929 obd_id *startid, obd_gr group, void *data);
930 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
931 int objcount, struct obd_ioobj *obj,
932 int niocount, struct niobuf_remote *remote,
933 struct niobuf_local *local, struct obd_trans_info *oti);
934 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
935 int objcount, struct obd_ioobj *obj,
936 int niocount, struct niobuf_local *local,
937 struct obd_trans_info *oti, int rc);
938 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
939 struct ldlm_enqueue_info *einfo,
940 struct ptlrpc_request_set *rqset);
941 int (*o_match)(struct obd_export *, struct lov_stripe_md *, __u32 type,
942 ldlm_policy_data_t *, __u32 mode, int *flags, void *data,
943 struct lustre_handle *lockh);
944 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
945 ldlm_iterator_t it, void *data);
946 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
947 __u32 mode, struct lustre_handle *);
948 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
949 int flags, void *opaque);
950 int (*o_join_lru)(struct obd_export *, struct lov_stripe_md *,
952 int (*o_init_export)(struct obd_export *exp);
953 int (*o_destroy_export)(struct obd_export *exp);
954 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
957 /* llog related obd_methods */
958 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
959 int count, struct llog_catid *logid,
960 struct obd_uuid *uuid);
961 int (*o_llog_finish)(struct obd_device *obd, int count);
963 /* metadata-only methods */
964 int (*o_pin)(struct obd_export *, obd_id ino, __u32 gen, int type,
965 struct obd_client_handle *, int flag);
966 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
968 int (*o_import_event)(struct obd_device *, struct obd_import *,
969 enum obd_import_event);
971 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
972 enum obd_notify_event ev, void *data);
974 int (*o_health_check)(struct obd_device *);
977 int (*o_quotacheck)(struct obd_export *, struct obd_quotactl *);
978 int (*o_quotactl)(struct obd_export *, struct obd_quotactl *);
980 int (*o_ping)(struct obd_export *exp);
982 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
983 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
984 * Also, add a wrapper function in include/linux/obd_class.h. */
987 struct lsm_operations {
988 void (*lsm_free)(struct lov_stripe_md *);
989 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
990 struct obd_export *md_exp);
991 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
993 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
995 obd_off (*lsm_stripe_offset_by_index)(struct lov_stripe_md *, int);
996 obd_off (*lsm_stripe_offset_by_offset)(struct lov_stripe_md *, obd_off);
997 int (*lsm_stripe_index_by_offset)(struct lov_stripe_md *, obd_off);
998 int (*lsm_revalidate) (struct lov_stripe_md *, struct obd_device *obd);
999 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1001 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1002 struct lov_mds_md *lmm);
1005 extern struct lsm_operations lsm_plain_ops;
1006 extern struct lsm_operations lsm_join_ops;
1007 static inline struct lsm_operations *lsm_op_find(int magic)
1011 return &lsm_plain_ops;
1012 case LOV_MAGIC_JOIN:
1013 return &lsm_join_ops;
1015 CERROR("Cannot recognize lsm_magic %d", magic);
1020 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1022 /* Requests for obd_extent_calc() */
1023 #define OBD_CALC_STRIPE_START 1
1024 #define OBD_CALC_STRIPE_END 2
1026 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1030 CERROR("%s: transno "LPD64" commit error: %d\n",
1031 obd->obd_name, transno, error);
1034 CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
1035 obd->obd_name, transno);
1036 if (transno > obd->obd_last_committed) {
1037 obd->obd_last_committed = transno;
1038 ptlrpc_commit_replies (obd);
1042 static inline void init_obd_quota_ops(quota_interface_t *interface,
1043 struct obd_ops *obd_ops)
1049 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1050 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1053 #endif /* __OBD_H */