1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
6 * This code is issued under the GNU General Public License.
7 * See the file COPYING in this distribution
13 #define IOC_OSC_TYPE 'h'
14 #define IOC_OSC_MIN_NR 20
15 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
16 #define IOC_OSC_MAX_NR 50
18 #define IOC_MDC_TYPE 'i'
19 #define IOC_MDC_MIN_NR 20
20 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_device *)
21 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *)
22 #define IOC_MDC_MAX_NR 50
25 # include <linux/fs.h>
26 # include <linux/list.h>
27 # include <linux/sched.h> /* for struct task_struct, for current.h */
28 # include <asm/current.h> /* for smp_lock.h */
29 # include <linux/smp_lock.h>
30 # include <linux/proc_fs.h>
31 # include <linux/mount.h>
34 #include <linux/lustre_lib.h>
35 #include <linux/lustre_idl.h>
36 #include <linux/lustre_export.h>
38 /* this is really local to the OSC */
39 struct loi_oap_pages {
40 struct list_head lop_pending;
42 struct list_head lop_urgent;
43 struct list_head lop_pending_sync;
46 struct lov_oinfo { /* per-stripe data structure */
47 __u64 loi_id; /* object ID on the target OST */
48 __u64 loi_gr; /* object group on the target OST */
49 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
50 int loi_ost_gen; /* generation of this loi_ost_idx */
52 /* used by the osc to keep track of what objects to build into rpcs */
53 struct loi_oap_pages loi_read_lop;
54 struct loi_oap_pages loi_write_lop;
55 /* _cli_ is poorly named, it should be _ready_ */
56 struct list_head loi_cli_item;
57 struct list_head loi_write_item;
60 static inline void loi_init(struct lov_oinfo *loi)
62 INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
63 INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
64 INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_sync);
65 INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
66 INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
67 INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_sync);
68 INIT_LIST_HEAD(&loi->loi_cli_item);
69 INIT_LIST_HEAD(&loi->loi_write_item);
72 struct lov_stripe_md {
74 __u64 lsm_object_id; /* lov object id */
75 __u64 lsm_object_gr; /* lov object id */
78 /* LOV-private members start here -- only for use in lov/. */
80 __u32 lsm_stripe_size; /* size of the stripe */
81 __u32 lsm_pattern; /* striping pattern (RAID0, RAID1) */
82 unsigned lsm_stripe_count; /* number of objects being striped over */
83 struct lov_oinfo lsm_oinfo[0];
87 struct list_head typ_chain;
88 struct obd_ops *typ_ops;
89 struct proc_dir_entry *typ_procroot;
104 ASYNC_COUNT_STABLE = 0x4,
107 struct obd_async_page_ops {
108 int (*ap_make_ready)(void *data, int cmd);
109 int (*ap_refresh_count)(void *data, int cmd);
110 void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
111 void (*ap_completion)(void *data, int cmd, int rc);
114 struct obd_sync_io_container {
115 spinlock_t osic_lock;
116 atomic_t osic_refcount;
119 wait_queue_head_t osic_waitq;
122 /* if we find more consumers this could be generalized */
123 #define OBD_HIST_MAX 32
124 struct obd_histogram {
126 unsigned long oh_buckets[OBD_HIST_MAX];
129 /* Individual type definitions */
131 struct ost_server_data;
134 const char *fo_fstype;
135 struct super_block *fo_sb;
136 struct vfsmount *fo_vfsmnt;
137 struct dentry *fo_dentry_O;
138 struct dentry **fo_dentry_O_groups;
139 struct dentry **fo_dentry_O_sub;
140 spinlock_t fo_objidlock; /* protect fo_lastobjid increment */
141 spinlock_t fo_translock; /* protect fsd_last_rcvd increment */
142 struct file *fo_rcvd_filp;
143 struct filter_server_data *fo_fsd;
144 unsigned long *fo_last_rcvd_slots;
145 __u64 fo_mount_count;
147 struct file_operations *fo_fop;
148 struct inode_operations *fo_iop;
149 struct address_space_operations *fo_aops;
151 struct list_head fo_export_list;
153 spinlock_t fo_grant_lock; /* protects tot_granted */
154 obd_size fo_tot_granted;
155 obd_size fo_tot_cached;
157 struct obd_import *fo_mdc_imp;
158 struct obd_uuid fo_mdc_uuid;
159 struct lustre_handle fo_mdc_conn;
161 struct ptlrpc_client fo_mdc_client;
163 struct file **fo_last_objid_files;
164 __u64 *fo_last_objids; //last created object ID for groups
166 struct semaphore fo_alloc_lock;
168 struct obd_histogram fo_r_pages;
169 struct obd_histogram fo_w_pages;
170 struct obd_histogram fo_r_discont_pages;
171 struct obd_histogram fo_w_discont_pages;
172 struct obd_histogram fo_r_discont_blocks;
173 struct obd_histogram fo_w_discont_blocks;
176 struct mds_server_data;
178 #define OSC_MAX_RIF_DEFAULT 4
179 #define OSC_MAX_RIF_MAX 32
180 #define OSC_MAX_DIRTY_DEFAULT 4
181 #define OSC_MAX_DIRTY_MB_MAX 256 /* totally arbitrary */
185 struct obd_import *cl_import;
186 struct semaphore cl_sem;
188 /* max_mds_easize is purely a performance thing so we don't have to
189 * call obd_size_wiremd() all the time. */
190 int cl_max_mds_easize;
191 int cl_max_mds_cookiesize;
194 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
195 void *cl_llcd_offset;
197 struct semaphore cl_dirty_sem;
198 obd_size cl_dirty; /* all _dirty_ in bytes */
199 obd_size cl_dirty_granted; /* from ost */
200 obd_size cl_dirty_max; /* allowed w/o rpc */
201 struct list_head cl_cache_waiters;
203 struct obd_device *cl_mgmtcli_obd;
205 /* this is just to keep existing infinitely caching behaviour between
206 * clients and OSTs that don't have the grant code in yet.. it can
207 * be yanked once everything speaks grants */
208 char cl_ost_can_grant;
210 /* keep track of objects that have lois that contain pages which
211 * have been queued for async brw. this lock also protects the
212 * lists of osc_client_pages that hang off of the loi */
213 spinlock_t cl_loi_list_lock;
214 struct list_head cl_loi_ready_list;
215 struct list_head cl_loi_write_list;
216 int cl_brw_in_flight;
217 /* just a sum of the loi/lop pending numbers to be exported by /proc */
218 int cl_pending_w_pages;
219 int cl_pending_r_pages;
220 int cl_max_pages_per_rpc;
221 int cl_max_rpcs_in_flight;
222 struct obd_histogram cl_read_rpc_hist;
223 struct obd_histogram cl_write_rpc_hist;
224 struct obd_histogram cl_read_page_hist;
225 struct obd_histogram cl_write_page_hist;
227 struct mdc_rpc_lock *cl_rpc_lock;
228 struct mdc_rpc_lock *cl_setattr_lock;
231 /* Like a client, with some hangers-on. Keep mc_client_obd first so that we
232 * can reuse the various client setup/connect functions. */
234 struct client_obd mc_client_obd; /* nested */
235 struct ptlrpc_thread *mc_ping_thread;
236 struct obd_export *mc_ping_exp; /* XXX single-target */
237 struct list_head mc_registered;
241 #define mc_import mc_client_obd.cl_import
244 struct ptlrpc_service *mds_service;
245 struct ptlrpc_service *mds_setattr_service;
246 struct ptlrpc_service *mds_readpage_service;
247 struct super_block *mds_sb;
248 struct vfsmount *mds_vfsmnt;
249 struct dentry *mds_fid_de;
251 int mds_max_cookiesize;
252 struct file *mds_rcvd_filp;
253 spinlock_t mds_transno_lock;
254 __u64 mds_last_transno;
255 __u64 mds_mount_count;
257 struct semaphore mds_epoch_sem;
258 struct ll_fid mds_rootfid;
259 struct mds_server_data *mds_server_data;
260 struct dentry *mds_pending_dir;
261 struct dentry *mds_logs_dir;
262 struct dentry *mds_objects_dir;
263 struct llog_handle *mds_cfg_llh;
264 // struct llog_handle *mds_catalog;
265 struct obd_device *mds_osc_obd; /* XXX lov_obd */
266 struct obd_uuid mds_lov_uuid;
268 struct obd_export *mds_osc_exp; /* XXX lov_exp */
269 int mds_has_lov_desc;
270 struct lov_desc mds_lov_desc;
271 obd_id *mds_lov_objids;
272 int mds_lov_objids_valid;
273 int mds_lov_nextid_set;
274 struct file *mds_lov_objid_filp;
275 unsigned long *mds_client_bitmap;
276 struct semaphore mds_orphan_recovery_sem;
278 atomic_t mds_open_count;
295 * this struct does double-duty acting as either a client or
296 * server instance .. maybe not wise.
300 struct ptlrpc_service *ptlbd_service;
303 struct ptlrpc_client bd_client;
304 struct obd_import *bd_import;
305 struct obd_uuid bd_server_uuid;
306 struct obd_export *bd_exp;
307 int refcount; /* XXX sigh */
311 spinlock_t recovd_lock;
312 struct list_head recovd_managed_items; /* items managed */
313 struct list_head recovd_troubled_items; /* items in recovery */
315 wait_queue_head_t recovd_recovery_waitq;
316 wait_queue_head_t recovd_ctl_waitq;
317 wait_queue_head_t recovd_waitq;
318 struct task_struct *recovd_thread;
323 struct ptlrpc_service *ost_service;
324 struct ptlrpc_service *ost_create_service;
327 struct echo_client_obd {
328 struct obd_export *ec_exp; /* the local connection to osc/lov */
330 struct list_head ec_objects;
336 struct obd_export *cobd_target_exp;/* local connection to target obd */
337 struct obd_export *cobd_cache_exp; /* local connection to cache obd */
340 struct lov_tgt_desc {
341 struct obd_uuid uuid;
342 struct obd_export *ltd_exp;
343 int active; /* is this target up for requests */
348 struct lov_desc desc;
351 int lo_catalog_loaded:1;
352 struct lov_tgt_desc *tgts;
355 struct niobuf_local {
361 struct dentry *dentry;
366 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
367 #define N_LOCAL_TEMP_PAGE 0x10000000
369 struct obd_trans_info {
372 /* Only used on the server side for tracking acks. */
373 struct oti_req_ack_lock {
374 struct lustre_handle lock;
378 struct llog_cookie oti_onecookie;
379 struct llog_cookie *oti_logcookies;
383 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
388 if (num_cookies == 1)
389 oti->oti_logcookies = &oti->oti_onecookie;
391 OBD_ALLOC(oti->oti_logcookies,
392 num_cookies * sizeof(oti->oti_onecookie));
394 oti->oti_numcookies = num_cookies;
397 static inline void oti_free_cookies(struct obd_trans_info *oti)
399 if (!oti || !oti->oti_logcookies)
402 if (oti->oti_logcookies == &oti->oti_onecookie)
403 LASSERT(oti->oti_numcookies == 1);
405 OBD_FREE(oti->oti_logcookies,
406 oti->oti_numcookies * sizeof(oti->oti_onecookie));
407 oti->oti_logcookies = NULL;
408 oti->oti_numcookies = 0;
413 LLOG_CONFIG_ORIG_CTXT = 0,
414 LLOG_CONFIG_REPL_CTXT = 1,
415 LLOG_UNLINK_ORIG_CTXT = 2,
416 LLOG_UNLINK_REPL_CTXT = 3,
417 LLOG_SIZE_ORIG_CTXT = 4,
418 LLOG_SIZE_REPL_CTXT = 5,
419 LLOG_MD_ORIG_CTXT = 6,
420 LLOG_MD_REPL_CTXT = 7,
421 LLOG_RD1_ORIG_CTXT = 8,
422 LLOG_RD1_REPL_CTXT = 9,
423 LLOG_TEST_ORIG_CTXT = 10,
424 LLOG_TEST_REPL_CTXT = 11,
429 /* corresponds to one of the obd's */
431 struct obd_type *obd_type;
433 /* common and UUID name of this device */
435 struct obd_uuid obd_uuid;
438 int obd_attached:1, obd_set_up:1, obd_recovering:1,
439 obd_abort_recovery:1, obd_replayable:1, obd_no_transno:1,
440 obd_no_recov:1, obd_stopping:1;
441 atomic_t obd_refcount;
442 wait_queue_head_t obd_refcount_waitq;
443 struct proc_dir_entry *obd_proc_entry;
444 struct list_head obd_exports;
446 struct ldlm_namespace *obd_namespace;
447 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
448 /* a spinlock is OK for what we do now, may need a semaphore later */
449 spinlock_t obd_dev_lock;
450 __u64 obd_last_committed;
451 struct fsfilt_operations *obd_fsops;
452 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
453 struct obd_statfs obd_osfs;
454 unsigned long obd_osfs_age;
455 struct obd_run_ctxt obd_ctxt;
456 struct obd_device *obd_observer;
457 struct obd_export *obd_self_export;
459 /* XXX encapsulate all this recovery data into one struct */
460 svc_handler_t obd_recovery_handler;
461 int obd_max_recoverable_clients;
462 int obd_connected_clients;
463 int obd_recoverable_clients;
464 spinlock_t obd_processing_task_lock;
465 pid_t obd_processing_task;
466 __u64 obd_next_recovery_transno;
467 int obd_replayed_requests;
468 int obd_requests_queued_for_recovery;
469 wait_queue_head_t obd_next_transno_waitq;
470 wait_queue_head_t obd_commit_waitq;
471 struct timer_list obd_recovery_timer;
472 struct list_head obd_recovery_queue;
473 struct list_head obd_delayed_reply_queue;
476 struct filter_obd filter;
478 struct client_obd cli;
480 struct echo_client_obd echo_client;
481 struct echo_obd echo;
482 struct recovd_obd recovd;
484 struct cache_obd cobd;
485 struct ptlbd_obd ptlbd;
486 struct mgmtcli_obd mgmtcli;
488 /* Fields used by LProcFS */
489 unsigned int obd_cntr_base;
490 struct lprocfs_stats *obd_stats;
491 struct proc_dir_entry *obd_svc_procroot;
492 struct lprocfs_stats *obd_svc_stats;
495 #define OBD_OPT_FORCE 0x0001
496 #define OBD_OPT_FAILOVER 0x0002
498 #define OBD_LLOG_FL_SENDNOW 0x0001
501 struct module *o_owner;
502 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
503 void *karg, void *uarg);
504 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
505 __u32 *vallen, void *val);
506 int (*o_set_info)(struct obd_export *, __u32 keylen, void *key,
507 __u32 vallen, void *val);
508 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
509 int (*o_detach)(struct obd_device *dev);
510 int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
511 int (*o_postsetup) (struct obd_device *dev);
512 int (*o_precleanup)(struct obd_device *dev, int flags);
513 int (*o_cleanup)(struct obd_device *dev, int flags);
514 int (*o_postrecov)(struct obd_device *dev);
515 int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
516 struct obd_uuid *cluuid);
517 int (*o_disconnect)(struct obd_export *exp, int flags);
519 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
520 unsigned long max_age);
521 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
522 struct lov_stripe_md *mem_src);
523 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
524 struct lov_mds_md *disk_src, int disk_len);
525 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
527 int (*o_create)(struct obd_export *exp, struct obdo *oa,
528 struct lov_stripe_md **ea, struct obd_trans_info *oti);
529 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
530 struct lov_stripe_md *ea, struct obd_trans_info *oti);
531 int (*o_setattr)(struct obd_export *exp, struct obdo *oa,
532 struct lov_stripe_md *ea, struct obd_trans_info *oti);
533 int (*o_getattr)(struct obd_export *exp, struct obdo *oa,
534 struct lov_stripe_md *ea);
535 int (*o_getattr_async)(struct obd_export *exp, struct obdo *oa,
536 struct lov_stripe_md *ea,
537 struct ptlrpc_request_set *set);
538 int (*o_brw)(int rw, struct obd_export *exp, struct obdo *oa,
539 struct lov_stripe_md *ea, obd_count oa_bufs,
540 struct brw_page *pgarr, struct obd_trans_info *oti);
541 int (*o_brw_async)(int rw, struct obd_export *exp, struct obdo *oa,
542 struct lov_stripe_md *ea, obd_count oa_bufs,
543 struct brw_page *pgarr, struct ptlrpc_request_set *,
544 struct obd_trans_info *oti);
545 int (*o_prep_async_page)(struct obd_export *exp,
546 struct lov_stripe_md *lsm,
547 struct lov_oinfo *loi,
548 struct page *page, obd_off offset,
549 struct obd_async_page_ops *ops, void *data,
551 int (*o_queue_async_io)(struct obd_export *exp,
552 struct lov_stripe_md *lsm,
553 struct lov_oinfo *loi, void *cookie,
554 int cmd, obd_off off, int count,
555 obd_flag brw_flags, obd_flag async_flags);
556 int (*o_queue_sync_io)(struct obd_export *exp,
557 struct lov_stripe_md *lsm,
558 struct lov_oinfo *loi,
559 struct obd_sync_io_container *osic,
560 void *cookie, int cmd, obd_off off, int count,
562 int (*o_trigger_sync_io)(struct obd_export *exp,
563 struct lov_stripe_md *lsm,
564 struct lov_oinfo *loi,
565 struct obd_sync_io_container *osic);
566 int (*o_set_async_flags)(struct obd_export *exp,
567 struct lov_stripe_md *lsm,
568 struct lov_oinfo *loi, void *cookie,
569 obd_flag async_flags);
570 int (*o_teardown_async_page)(struct obd_export *exp,
571 struct lov_stripe_md *lsm,
572 struct lov_oinfo *loi, void *cookie);
573 int (*o_punch)(struct obd_export *exp, struct obdo *oa,
574 struct lov_stripe_md *ea, obd_size start,
575 obd_size end, struct obd_trans_info *oti);
576 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
577 struct lov_stripe_md *ea, obd_size start, obd_size end);
578 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
579 struct lov_stripe_md *src, obd_size start,
580 obd_size end, struct obd_trans_info *oti);
581 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
582 struct lustre_handle *srconn, struct lov_stripe_md *src,
583 obd_size start, obd_size end, struct obd_trans_info *);
584 int (*o_iterate)(struct lustre_handle *conn,
585 int (*)(obd_id, obd_gr, void *),
586 obd_id *startid, obd_gr group, void *data);
587 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
588 int objcount, struct obd_ioobj *obj,
589 int niocount, struct niobuf_remote *remote,
590 struct niobuf_local *local, struct obd_trans_info *oti);
591 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
592 int objcount, struct obd_ioobj *obj,
593 int niocount, struct niobuf_local *local,
594 struct obd_trans_info *oti);
595 int (*o_enqueue)(struct obd_export *exp, struct lov_stripe_md *md,
596 struct lustre_handle *parent_lock,
597 __u32 type, void *cookie, int cookielen, __u32 mode,
598 int *flags, void *cb, void *data,
599 struct lustre_handle *lockh);
600 int (*o_match)(struct obd_export *exp, struct lov_stripe_md *md,
601 __u32 type, void *cookie, int cookielen, __u32 mode,
602 int *flags, void *data, struct lustre_handle *lockh);
603 int (*o_change_cbdata)(struct obd_export *exp,
604 struct lov_stripe_md *lsm, ldlm_iterator_t it,
606 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
607 __u32 mode, struct lustre_handle *);
608 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
609 int flags, void *opaque);
610 int (*o_san_preprw)(int cmd, struct obd_export *exp,
611 struct obdo *oa, int objcount,
612 struct obd_ioobj *obj, int niocount,
613 struct niobuf_remote *remote);
614 int (*o_init_export)(struct obd_export *exp);
615 int (*o_destroy_export)(struct obd_export *exp);
617 /* llog related obd_methods */
618 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
619 int count, struct llog_logid *logid);
620 int (*o_llog_finish)(struct obd_device *obd, int count);
622 /* only until proper file size mechanics arrive */
623 int (*o_lock_contains)(struct obd_export *exp,
624 struct lov_stripe_md *lsm,
625 struct ldlm_lock *lock, obd_off offset);
627 /* metadata-only methods */
628 int (*o_pin)(struct obd_export *, obd_id ino, __u32 gen, int type,
629 struct obd_client_handle *, int flag);
630 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
632 int (*o_invalidate_import)(struct obd_device *, struct obd_import *);
634 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
637 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
638 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
639 * Also, add a wrapper function in include/linux/obd_class.h.
645 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
649 CERROR("%s: transno "LPD64" commit error: %d\n",
650 obd->obd_name, transno, error);
653 CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
654 obd->obd_name, transno);
655 if (transno > obd->obd_last_committed) {
656 obd->obd_last_committed = transno;
657 wake_up(&obd->obd_commit_waitq);