4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #include <linux/spinlock.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_lib.h>
44 #include <libcfs/bitmap.h>
45 #ifdef HAVE_SERVER_SUPPORT
46 # include <lu_target.h>
47 # include <obd_target.h>
50 #include <lustre_export.h>
51 #include <lustre_fid.h>
52 #include <lustre_fld.h>
53 #include <lustre_handles.h>
54 #include <lustre_intent.h>
55 #include <lustre_capa.h>
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
80 oinfo->loi_kms_valid = 1;
83 static inline void loi_init(struct lov_oinfo *loi)
90 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
92 /* obd info for a particular level (lov, osc). */
94 /* OBD_STATFS_* flags */
96 /* statfs data specific for every OSC, if needed at all. */
97 struct obd_statfs *oi_osfs;
98 /* An update callback which is called to update some data on upper
99 * level. E.g. it is used for update lsm->lsm_oinfo at every received
100 * request in osc level for enqueue requests. It is also possible to
101 * update some caller data from LOV layer if needed. */
102 obd_enqueue_update_f oi_cb_up;
106 struct list_head typ_chain;
107 struct obd_ops *typ_dt_ops;
108 struct md_ops *typ_md_ops;
109 struct proc_dir_entry *typ_procroot;
110 struct proc_dir_entry *typ_procsym;
111 __u32 typ_sym_filter;
114 struct lu_device_type *typ_lu;
115 spinlock_t obd_type_lock;
125 struct timeout_item {
126 enum timeout_event ti_event;
127 cfs_time_t ti_timeout;
130 struct list_head ti_obd_list;
131 struct list_head ti_chain;
134 #define OBD_MAX_RIF_DEFAULT 8
135 #define OBD_MAX_RIF_MAX 512
136 #define OSC_MAX_RIF_MAX 256
137 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
138 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
139 #define OSC_DEFAULT_RESENDS 10
141 /* possible values for fo_sync_lock_cancel */
143 NEVER_SYNC_ON_CANCEL = 0,
144 BLOCKING_SYNC_ON_CANCEL = 1,
145 ALWAYS_SYNC_ON_CANCEL = 2,
146 NUM_SYNC_ON_CANCEL_STATES
150 * Limit reply buffer size for striping data to one x86_64 page. This
151 * value is chosen to fit the striping data for common use cases while
152 * staying well below the limit at which the buffer must be backed by
153 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
156 #define OBD_MAX_DEFAULT_EA_SIZE 4096
161 struct rw_semaphore cl_sem;
162 struct obd_uuid cl_target_uuid;
163 struct obd_import *cl_import; /* ptlrpc connection state */
164 size_t cl_conn_count;
166 /* Cache maximum and default values for easize. This is
167 * strictly a performance optimization to minimize calls to
168 * obd_size_diskmd(). The default values are used to calculate the
169 * initial size of a request buffer. The ptlrpc layer will resize the
170 * buffer as needed to accommodate a larger reply from the
171 * server. The default values should be small enough to avoid wasted
172 * memory and excessive use of vmalloc(), yet large enough to avoid
173 * reallocating the buffer in the common use case. */
175 /* Default EA size for striping attributes. It is initialized at
176 * mount-time based on the default stripe width of the filesystem,
177 * then it tracks the largest observed EA size advertised by
178 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
179 __u32 cl_default_mds_easize;
181 /* Maximum possible EA size computed at mount-time based on
182 * the number of OSTs in the filesystem. May be increased at
183 * run-time if a larger observed size is advertised by the MDT. */
184 __u32 cl_max_mds_easize;
186 enum lustre_sec_part cl_sp_me;
187 enum lustre_sec_part cl_sp_to;
188 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
190 /* the grant values are protected by loi_list_lock below */
191 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
192 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
193 unsigned long cl_dirty_transit; /* dirty synchronous */
194 unsigned long cl_avail_grant; /* bytes of credit for ost */
195 unsigned long cl_lost_grant; /* lost credits (trunc) */
197 /* since we allocate grant by blocks, we don't know how many grant will
198 * be used to add a page into cache. As a solution, we reserve maximum
199 * grant before trying to dirty a page and unreserve the rest.
200 * See osc_{reserve|unreserve}_grant for details. */
201 long cl_reserved_grant;
202 struct list_head cl_cache_waiters; /* waiting for cache/grant */
203 cfs_time_t cl_next_shrink_grant; /* jiffies */
204 struct list_head cl_grant_shrink_list; /* Timeout event list */
205 int cl_grant_shrink_interval; /* seconds */
207 /* A chunk is an optimal size used by osc_extent to determine
208 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
210 unsigned int cl_extent_tax; /* extent overhead, by bytes */
212 /* keep track of objects that have lois that contain pages which
213 * have been queued for async brw. this lock also protects the
214 * lists of osc_client_pages that hang off of the loi */
216 * ->cl_loi_list_lock protects consistency of
217 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
218 * ->ap_completion() call-backs are executed under this lock. As we
219 * cannot guarantee that these call-backs never block on all platforms
220 * (as a matter of fact they do block on Mac OS X), type of
221 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
222 * and blocking mutex on Mac OS X. (Alternative is to make this lock
223 * blocking everywhere, but we don't want to slow down fast-path of
224 * our main platform.)
226 * NB by Jinshan: though field names are still _loi_, but actually
227 * osc_object{}s are in the list.
229 spinlock_t cl_loi_list_lock;
230 struct list_head cl_loi_ready_list;
231 struct list_head cl_loi_hp_ready_list;
232 struct list_head cl_loi_write_list;
233 struct list_head cl_loi_read_list;
234 __u32 cl_r_in_flight;
235 __u32 cl_w_in_flight;
236 /* just a sum of the loi/lop pending numbers to be exported by /proc */
237 atomic_t cl_pending_w_pages;
238 atomic_t cl_pending_r_pages;
239 __u32 cl_max_pages_per_rpc;
240 __u32 cl_max_rpcs_in_flight;
241 struct obd_histogram cl_read_rpc_hist;
242 struct obd_histogram cl_write_rpc_hist;
243 struct obd_histogram cl_read_page_hist;
244 struct obd_histogram cl_write_page_hist;
245 struct obd_histogram cl_read_offset_hist;
246 struct obd_histogram cl_write_offset_hist;
248 /* lru for osc caching pages */
249 struct cl_client_cache *cl_cache;
250 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
251 atomic_long_t *cl_lru_left;
252 atomic_long_t cl_lru_busy;
253 atomic_long_t cl_lru_in_list;
254 atomic_long_t cl_unstable_count;
255 struct list_head cl_lru_list; /* lru page list */
256 spinlock_t cl_lru_list_lock; /* page list protector */
257 atomic_t cl_lru_shrinkers;
259 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
260 atomic_t cl_destroy_in_flight;
261 wait_queue_head_t cl_destroy_waitq;
263 struct mdc_rpc_lock *cl_rpc_lock;
264 struct mdc_rpc_lock *cl_close_lock;
266 /* modify rpcs in flight
267 * currently used for metadata only */
268 spinlock_t cl_mod_rpcs_lock;
269 __u16 cl_max_mod_rpcs_in_flight;
272 struct mutex cl_mgc_mutex;
273 struct local_oid_storage *cl_mgc_los;
274 struct dt_object *cl_mgc_configs_dir;
275 atomic_t cl_mgc_refcount;
276 struct obd_export *cl_mgc_mgsexp;
278 /* checksumming for data sent over the network */
279 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
280 /* supported checksum types that are worked out at connect time */
281 __u32 cl_supp_cksum_types;
282 /* checksum algorithm to be used */
283 cksum_type_t cl_cksum_type;
285 /* also protected by the poorly named _loi_list_lock lock above */
286 struct osc_async_rc cl_ar;
288 /* sequence manager */
289 struct lu_client_seq *cl_seq;
291 atomic_t cl_resends; /* resend count */
293 /* ptlrpc work for writeback in ptlrpcd context */
294 void *cl_writeback_work;
296 /* hash tables for osc_quota_info */
297 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
299 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
306 struct echo_client_obd {
307 struct obd_export *ec_exp; /* the local connection to osc/lov */
309 struct list_head ec_objects;
310 struct list_head ec_locks;
314 /* Generic subset of OSTs */
316 __u32 *op_array; /* array of index of
318 unsigned int op_count; /* number of OSTs in the array */
319 unsigned int op_size; /* allocated size of lp_array */
320 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
323 /* allow statfs data caching for 1 second */
324 #define OBD_STATFS_CACHE_SECONDS 1
326 struct lov_tgt_desc {
327 struct list_head ltd_kill;
328 struct obd_uuid ltd_uuid;
329 struct obd_device *ltd_obd;
330 struct obd_export *ltd_exp;
332 __u32 ltd_index; /* index in lov_obd->tgts */
333 unsigned long ltd_active:1,/* is this target up for requests */
334 ltd_activate:1,/* should target be activated */
335 ltd_reap:1; /* should this target be deleted */
339 struct lov_desc desc;
340 struct lov_tgt_desc **lov_tgts; /* sparse array */
341 struct ost_pool lov_packed; /* all OSTs in a packed
343 struct mutex lov_lock;
344 struct obd_connect_data lov_ocd;
345 struct proc_dir_entry *targets_proc_entry;
346 atomic_t lov_refcount;
347 __u32 lov_death_row; /* tgts scheduled to be deleted */
348 __u32 lov_tgt_size; /* size of tgts array */
351 struct cfs_hash *lov_pools_hash_body; /* used for key access */
352 struct list_head lov_pool_list; /* used for sequential access */
353 struct proc_dir_entry *lov_pool_proc_entry;
354 enum lustre_sec_part lov_sp_me;
356 /* Cached LRU and unstable data from upper layer */
357 struct cl_client_cache *lov_cache;
359 struct rw_semaphore lov_notify_lock;
362 struct lmv_tgt_desc {
363 struct obd_uuid ltd_uuid;
364 struct obd_export *ltd_exp;
366 struct mutex ltd_fid_mutex;
367 unsigned long ltd_active:1; /* target up for requests */
370 enum placement_policy {
371 PLACEMENT_CHAR_POLICY = 0,
372 PLACEMENT_NID_POLICY = 1,
373 PLACEMENT_INVAL_POLICY = 2,
377 typedef enum placement_policy placement_policy_t;
381 struct lu_client_fld lmv_fld;
383 placement_policy_t lmv_placement;
384 struct lmv_desc desc;
385 struct obd_uuid cluuid;
386 struct obd_export *exp;
387 struct proc_dir_entry *targets_proc_entry;
389 struct mutex lmv_init_mutex;
394 __u32 tgts_size; /* size of tgts array */
395 struct lmv_tgt_desc **tgts;
397 struct obd_connect_data conn_data;
400 struct niobuf_local {
401 __u64 lnb_file_offset;
402 __u32 lnb_page_offset;
405 struct page *lnb_page;
410 #define LUSTRE_FLD_NAME "fld"
411 #define LUSTRE_SEQ_NAME "seq"
413 #define LUSTRE_MDD_NAME "mdd"
414 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
415 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
416 #define LUSTRE_VVP_NAME "vvp"
417 #define LUSTRE_LMV_NAME "lmv"
418 #define LUSTRE_SLP_NAME "slp"
419 #define LUSTRE_LOD_NAME "lod"
420 #define LUSTRE_OSP_NAME "osp"
421 #define LUSTRE_LWP_NAME "lwp"
423 /* obd device type names */
424 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
425 #define LUSTRE_MDS_NAME "mds"
426 #define LUSTRE_MDT_NAME "mdt"
427 #define LUSTRE_MDC_NAME "mdc"
428 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
429 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
430 #define LUSTRE_OSC_NAME "osc"
431 #define LUSTRE_LOV_NAME "lov"
432 #define LUSTRE_MGS_NAME "mgs"
433 #define LUSTRE_MGC_NAME "mgc"
435 #define LUSTRE_ECHO_NAME "obdecho"
436 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
437 #define LUSTRE_QMT_NAME "qmt"
439 /* Constant obd names (post-rename) */
440 #define LUSTRE_MDS_OBDNAME "MDS"
441 #define LUSTRE_OSS_OBDNAME "OSS"
442 #define LUSTRE_MGS_OBDNAME "MGS"
443 #define LUSTRE_MGC_OBDNAME "MGC"
446 * Events signalled through obd_notify() upcall-chain.
448 enum obd_notify_event {
451 /* Device connect start */
453 /* Device activated */
455 /* Device deactivated */
457 /* Device disconnected */
459 /* Connect data for import were changed */
462 OBD_NOTIFY_SYNC_NONBLOCK,
464 /* Configuration event */
466 /* Administratively deactivate/activate event */
467 OBD_NOTIFY_DEACTIVATE,
471 /* bit-mask flags for config events */
473 CONFIG_LOG = 0x1, /* finished processing config log */
474 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
475 CONFIG_TARGET = 0x4 /* one target is added */
479 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
480 * and liblustre being main examples).
482 struct obd_notify_upcall {
483 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
484 enum obd_notify_event ev, void *owner, void *data);
485 /* Opaque datum supplied by upper layer listener */
489 struct target_recovery_data {
490 svc_handler_t trd_recovery_handler;
491 pid_t trd_processing_task;
492 struct completion trd_starting;
493 struct completion trd_finishing;
496 struct obd_llog_group {
497 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
498 wait_queue_head_t olg_waitq;
500 struct mutex olg_cat_processing;
503 /* corresponds to one of the obd's */
504 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
507 struct obd_type *obd_type;
510 /* common and UUID name of this device */
511 char obd_name[MAX_OBD_NAME];
512 struct obd_uuid obd_uuid;
514 struct lu_device *obd_lu_dev;
516 /* bitfield modification is protected by obd_dev_lock */
518 obd_attached:1, /* finished attach */
519 obd_set_up:1, /* finished setup */
520 obd_recovering:1, /* there are recoverable clients */
521 obd_abort_recovery:1, /* recovery expired */
522 obd_version_recov:1, /* obd uses version checking */
523 obd_replayable:1, /* recovery is enabled;
525 obd_no_transno:1, /* no committed-transno notification */
526 obd_no_recov:1, /* fail instead of retry messages */
527 obd_stopping:1, /* started cleanup */
528 obd_starting:1, /* started setup */
529 obd_force:1, /* cleanup with > 0 obd refcount */
530 obd_fail:1, /* cleanup with failover */
531 obd_no_conn:1, /* deny new connections */
532 obd_inactive:1, /* device active/inactive
533 * (for /proc/status only!!) */
534 obd_no_ir:1, /* no imperative recovery. */
535 obd_process_conf:1, /* device is processing mgs config */
536 obd_uses_nid_stats:1, /* maintain per-client OBD stats */
537 obd_force_abort_recovery:1; /* abort recovery forcely */
539 /* use separate field as it is set in interrupt to don't mess with
540 * protection of other bits using _bh lock */
541 unsigned long obd_recovery_expired:1;
542 /* uuid-export hash body */
543 struct cfs_hash *obd_uuid_hash;
544 /* nid-export hash body */
545 struct cfs_hash *obd_nid_hash;
547 struct cfs_hash *obd_nid_stats_hash;
548 struct list_head obd_nid_stats;
549 atomic_t obd_refcount;
550 struct list_head obd_exports;
551 struct list_head obd_unlinked_exports;
552 struct list_head obd_delayed_exports;
553 struct list_head obd_lwp_list;
555 spinlock_t obd_nid_lock;
556 struct ldlm_namespace *obd_namespace;
557 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
558 /* a spinlock is OK for what we do now, may need a semaphore later */
559 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
560 struct mutex obd_dev_mutex;
561 __u64 obd_last_committed;
562 spinlock_t obd_osfs_lock;
563 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
565 struct lvfs_run_ctxt obd_lvfs_ctxt;
566 struct obd_llog_group obd_olg; /* default llog group */
567 struct obd_device *obd_observer;
568 struct rw_semaphore obd_observer_link_sem;
569 struct obd_notify_upcall obd_upcall;
570 struct obd_export *obd_self_export;
571 struct obd_export *obd_lwp_export;
572 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
573 struct list_head obd_exports_timed;
574 time_t obd_eviction_timer; /* for ping evictor */
576 int obd_max_recoverable_clients;
577 atomic_t obd_connected_clients;
578 int obd_stale_clients;
579 /* this lock protects all recovery list_heads, timer and
580 * obd_next_recovery_transno value */
581 spinlock_t obd_recovery_task_lock;
582 __u64 obd_next_recovery_transno;
583 int obd_replayed_requests;
584 int obd_requests_queued_for_recovery;
585 wait_queue_head_t obd_next_transno_waitq;
586 /* protected by obd_recovery_task_lock */
587 struct timer_list obd_recovery_timer;
589 time_t obd_recovery_start;
590 /* seconds, for lprocfs_status */
591 time_t obd_recovery_end;
592 int obd_recovery_time_hard;
593 int obd_recovery_timeout;
594 int obd_recovery_ir_factor;
596 /* new recovery stuff from CMD2 */
597 struct target_recovery_data obd_recovery_data;
598 int obd_replayed_locks;
599 atomic_t obd_req_replay_clients;
600 atomic_t obd_lock_replay_clients;
601 /* all lists are protected by obd_recovery_task_lock */
602 struct list_head obd_req_replay_queue;
603 struct list_head obd_lock_replay_queue;
604 struct list_head obd_final_req_queue;
607 #ifdef HAVE_SERVER_SUPPORT
608 struct obd_device_target obt;
609 struct filter_obd filter;
611 struct echo_obd echo;
613 struct client_obd cli;
614 struct echo_client_obd echo_client;
618 /* Fields used by LProcFS */
619 unsigned int obd_cntr_base;
620 struct lprocfs_stats *obd_stats;
622 unsigned int obd_md_cntr_base;
623 struct lprocfs_stats *obd_md_stats;
625 struct proc_dir_entry *obd_proc_entry;
626 struct proc_dir_entry *obd_proc_exports_entry;
627 struct proc_dir_entry *obd_svc_procroot;
628 struct lprocfs_stats *obd_svc_stats;
629 struct lprocfs_vars *obd_vars;
630 atomic_t obd_evict_inprogress;
631 wait_queue_head_t obd_evict_inprogress_waitq;
632 struct list_head obd_evict_list; /* protected with pet_lock */
635 * Ldlm pool part. Save last calculated SLV and Limit.
637 rwlock_t obd_pool_lock;
642 * A list of outstanding class_incref()'s against this obd. For
645 struct lu_ref obd_reference;
647 int obd_conn_inprogress;
650 enum obd_cleanup_stage {
651 /* Special case hack for MDS LOVs */
653 /* can be directly mapped to .ldto_device_fini() */
657 /* get/set_info keys */
658 #define KEY_ASYNC "async"
659 #define KEY_CAPA_KEY "capa_key"
660 #define KEY_CHANGELOG_CLEAR "changelog_clear"
661 #define KEY_FID2PATH "fid2path"
662 #define KEY_CHECKSUM "checksum"
663 #define KEY_CLEAR_FS "clear_fs"
664 #define KEY_CONN_DATA "conn_data"
665 #define KEY_EVICT_BY_NID "evict_by_nid"
666 #define KEY_FIEMAP "fiemap"
667 #define KEY_FLUSH_CTX "flush_ctx"
668 #define KEY_GRANT_SHRINK "grant_shrink"
669 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
670 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
671 #define KEY_INTERMDS "inter_mds"
672 #define KEY_LAST_ID "last_id"
673 #define KEY_LAST_FID "last_fid"
674 #define KEY_MAX_EASIZE "max_easize"
675 #define KEY_DEFAULT_EASIZE "default_easize"
676 #define KEY_MGSSEC "mgssec"
677 #define KEY_READ_ONLY "read-only"
678 #define KEY_REGISTER_TARGET "register_target"
679 #define KEY_SET_FS "set_fs"
680 #define KEY_TGT_COUNT "tgt_count"
681 /* KEY_SET_INFO in lustre_idl.h */
682 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
684 #define KEY_CACHE_SET "cache_set"
685 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
686 #define KEY_OSP_CONNECTED "osp_connected"
690 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
691 #define IT_OPEN (1 << 0)
692 #define IT_CREAT (1 << 1)
693 #define IT_READDIR (1 << 2)
694 #define IT_GETATTR (1 << 3)
695 #define IT_LOOKUP (1 << 4)
696 #define IT_UNLINK (1 << 5)
697 #define IT_TRUNC (1 << 6)
698 #define IT_GETXATTR (1 << 7)
699 #define IT_EXEC (1 << 8)
700 #define IT_PIN (1 << 9)
701 #define IT_LAYOUT (1 << 10)
702 #define IT_QUOTA_DQACQ (1 << 11)
703 #define IT_QUOTA_CONN (1 << 12)
704 #define IT_SETXATTR (1 << 13)
706 static inline int it_to_lock_mode(struct lookup_intent *it)
708 /* CREAT needs to be tested before open (both could be set) */
709 if (it->it_op & IT_CREAT)
711 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
714 else if (it->it_op & IT_READDIR)
716 else if (it->it_op & IT_GETXATTR)
718 else if (it->it_op & IT_SETXATTR)
721 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
726 MF_MDC_CANCEL_FID1 = 1 << 0,
727 MF_MDC_CANCEL_FID2 = 1 << 1,
728 MF_MDC_CANCEL_FID3 = 1 << 2,
729 MF_MDC_CANCEL_FID4 = 1 << 3,
730 MF_GET_MDT_IDX = 1 << 4,
734 CLI_SET_MEA = 1 << 0,
735 CLI_RM_ENTRY = 1 << 1,
738 CLI_MIGRATE = 1 << 4,
742 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
743 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
744 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
745 struct lu_fid op_fid4; /* to the operation locks. */
746 u32 op_mds; /* what mds server open will go to */
747 struct lustre_handle op_handle;
752 struct lmv_stripe_md *op_mea1;
753 struct lmv_stripe_md *op_mea2;
754 __u32 op_suppgids[2];
761 /* iattr fields and blocks. */
762 struct iattr op_attr;
763 loff_t op_attr_blocks;
764 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
765 __u64 op_valid; /* OBD_MD_* */
767 enum md_op_flags op_flags;
770 struct obd_capa *op_capa1;
771 struct obd_capa *op_capa2;
773 /* Various operation flags. */
774 enum mds_op_bias op_bias;
776 /* Used by readdir */
777 unsigned int op_max_pages;
779 /* used to transfer info between the stacks of MD client
780 * see enum op_cli_flags */
781 enum md_cli_flags op_cli_flags;
783 /* File object data version for HSM release, on client */
784 __u64 op_data_version;
785 struct lustre_handle op_lease_handle;
787 /* default stripe offset */
788 __u32 op_default_stripe_offset;
792 int (*md_blocking_ast)(struct ldlm_lock *lock,
793 struct ldlm_lock_desc *desc,
794 void *data, int flag);
797 struct md_enqueue_info;
798 /* metadata stat-ahead */
799 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
800 struct md_enqueue_info *minfo,
803 struct md_enqueue_info {
804 struct md_op_data mi_data;
805 struct lookup_intent mi_it;
806 struct lustre_handle mi_lockh;
807 struct inode *mi_dir;
808 md_enqueue_cb_t mi_cb;
813 struct module *o_owner;
814 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
815 void *karg, void __user *uarg);
816 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
817 __u32 keylen, void *key, __u32 *vallen, void *val);
818 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
819 __u32 keylen, void *key,
820 __u32 vallen, void *val,
821 struct ptlrpc_request_set *set);
822 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
823 int (*o_precleanup)(struct obd_device *dev,
824 enum obd_cleanup_stage cleanup_stage);
825 int (*o_cleanup)(struct obd_device *dev);
826 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
827 int (*o_postrecov)(struct obd_device *dev);
828 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
830 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
831 /* connect to the target device with given connection
832 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
833 * granted by the target, which are guaranteed to be a subset of flags
834 * asked for. If @ocd == NULL, use default parameters. */
835 int (*o_connect)(const struct lu_env *env,
836 struct obd_export **exp, struct obd_device *src,
837 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
839 int (*o_reconnect)(const struct lu_env *env,
840 struct obd_export *exp, struct obd_device *src,
841 struct obd_uuid *cluuid,
842 struct obd_connect_data *ocd,
844 int (*o_disconnect)(struct obd_export *exp);
846 /* Initialize/finalize fids infrastructure. */
847 int (*o_fid_init)(struct obd_device *obd,
848 struct obd_export *exp, enum lu_cli_type type);
849 int (*o_fid_fini)(struct obd_device *obd);
851 /* Allocate new fid according to passed @hint. */
852 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
853 struct lu_fid *fid, struct md_op_data *op_data);
856 * Object with @fid is getting deleted, we may want to do something
859 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
860 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
861 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
862 __u64 max_age, struct ptlrpc_request_set *set);
863 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
864 struct lov_mds_md *disk_src, int disk_len);
865 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
867 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
869 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
871 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
873 int (*o_preprw)(const struct lu_env *env, int cmd,
874 struct obd_export *exp, struct obdo *oa, int objcount,
875 struct obd_ioobj *obj, struct niobuf_remote *remote,
876 int *nr_pages, struct niobuf_local *local);
877 int (*o_commitrw)(const struct lu_env *env, int cmd,
878 struct obd_export *exp, struct obdo *oa,
879 int objcount, struct obd_ioobj *obj,
880 struct niobuf_remote *remote, int pages,
881 struct niobuf_local *local, int rc);
882 int (*o_init_export)(struct obd_export *exp);
883 int (*o_destroy_export)(struct obd_export *exp);
885 int (*o_import_event)(struct obd_device *, struct obd_import *,
886 enum obd_import_event);
888 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
889 enum obd_notify_event ev, void *data);
891 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
892 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
895 int (*o_quotactl)(struct obd_device *, struct obd_export *,
896 struct obd_quotactl *);
898 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
901 int (*o_pool_new)(struct obd_device *obd, char *poolname);
902 int (*o_pool_del)(struct obd_device *obd, char *poolname);
903 int (*o_pool_add)(struct obd_device *obd, char *poolname,
905 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
907 void (*o_getref)(struct obd_device *obd);
908 void (*o_putref)(struct obd_device *obd);
910 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
911 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
912 * Also, add a wrapper function in include/linux/obd_class.h. */
917 struct mdt_body *body;
918 struct lov_stripe_md *lsm;
919 struct lmv_stripe_md *lmv;
920 #ifdef CONFIG_FS_POSIX_ACL
921 struct posix_acl *posix_acl;
923 struct mdt_remote_perm *remote_perm;
924 struct obd_capa *mds_capa;
925 struct obd_capa *oss_capa;
928 struct md_open_data {
929 struct obd_client_handle *mod_och;
930 struct ptlrpc_request *mod_open_req;
931 struct ptlrpc_request *mod_close_req;
932 atomic_t mod_refcount;
936 struct obd_client_handle {
937 struct lustre_handle och_fh;
938 struct lu_fid och_fid;
939 struct md_open_data *och_mod;
940 struct lustre_handle och_lease_handle; /* open lock for lease */
945 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
947 struct lookup_intent;
951 /* Every operation from MD_STATS_FIRST_OP up to and including
952 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
953 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
954 * Operations after MD_STATS_LAST_OP are excluded from stats.
955 * There are a few reasons for doing this: we prune the 17
956 * counters which will be of minimal use in understanding
957 * metadata utilization, we save memory by allocating 15
958 * instead of 32 counters, we save cycles by not counting.
960 * MD_STATS_FIRST_OP must be the first member of md_ops.
962 #define MD_STATS_FIRST_OP m_close
963 int (*m_close)(struct obd_export *, struct md_op_data *,
964 struct md_open_data *, struct ptlrpc_request **);
966 int (*m_create)(struct obd_export *, struct md_op_data *,
967 const void *, size_t, umode_t, uid_t, gid_t,
968 cfs_cap_t, __u64, struct ptlrpc_request **);
970 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
971 const union ldlm_policy_data *,
972 struct lookup_intent *, struct md_op_data *,
973 struct lustre_handle *, __u64);
975 int (*m_getattr)(struct obd_export *, struct md_op_data *,
976 struct ptlrpc_request **);
978 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
979 struct lookup_intent *,
980 struct ptlrpc_request **,
981 ldlm_blocking_callback, __u64);
983 int (*m_link)(struct obd_export *, struct md_op_data *,
984 struct ptlrpc_request **);
986 int (*m_rename)(struct obd_export *, struct md_op_data *,
987 const char *, size_t, const char *, size_t,
988 struct ptlrpc_request **);
990 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
991 size_t , struct ptlrpc_request **);
993 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
994 struct obd_capa *, struct ptlrpc_request **);
996 int (*m_read_page)(struct obd_export *, struct md_op_data *,
997 struct md_callback *cb_op, __u64 hash_offset,
998 struct page **ppage);
1000 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1001 struct ptlrpc_request **);
1003 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1004 struct obd_capa *, u64, const char *,
1005 const char *, int, int, int, __u32,
1006 struct ptlrpc_request **);
1008 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1009 struct obd_capa *, u64, const char *,
1010 const char *, int, int, int,
1011 struct ptlrpc_request **);
1013 int (*m_intent_getattr_async)(struct obd_export *,
1014 struct md_enqueue_info *,
1015 struct ldlm_enqueue_info *);
1017 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1018 struct lu_fid *, __u64 *bits);
1020 #define MD_STATS_LAST_OP m_revalidate_lock
1022 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1023 struct obd_capa **);
1025 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1027 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1028 ldlm_iterator_t, void *);
1030 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1031 struct ptlrpc_request **);
1033 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1035 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1036 struct obd_export *, struct obd_export *,
1037 struct lustre_md *);
1039 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1041 int (*m_merge_attr)(struct obd_export *,
1042 const struct lmv_stripe_md *lsm,
1043 struct cl_attr *attr, ldlm_blocking_callback);
1045 int (*m_set_open_replay_data)(struct obd_export *,
1046 struct obd_client_handle *,
1047 struct lookup_intent *);
1049 int (*m_clear_open_replay_data)(struct obd_export *,
1050 struct obd_client_handle *);
1052 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1054 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1055 const struct lu_fid *, ldlm_type_t,
1056 ldlm_policy_data_t *, ldlm_mode_t,
1057 struct lustre_handle *);
1059 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1060 ldlm_policy_data_t *, ldlm_mode_t,
1061 ldlm_cancel_flags_t flags, void *opaque);
1063 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1064 renew_capa_cb_t cb);
1066 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1067 const struct req_msg_field *, struct obd_capa **);
1069 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1070 struct obd_capa *, __u32,
1071 struct ptlrpc_request **);
1073 int (*m_get_fid_from_lsm)(struct obd_export *,
1074 const struct lmv_stripe_md *,
1075 const char *name, int namelen,
1076 struct lu_fid *fid);
1079 static inline struct md_open_data *obd_mod_alloc(void)
1081 struct md_open_data *mod;
1085 atomic_set(&mod->mod_refcount, 1);
1089 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1090 #define obd_mod_put(mod) \
1092 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1093 if ((mod)->mod_open_req) \
1094 ptlrpc_req_finished((mod)->mod_open_req); \
1095 OBD_FREE_PTR(mod); \
1099 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1100 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1102 /* return 1 if client should be resend request */
1103 static inline int client_should_resend(int resend, struct client_obd *cli)
1105 return atomic_read(&cli->cl_resends) ?
1106 atomic_read(&cli->cl_resends) > resend : 1;
1110 * Return device name for this device
1112 * XXX: lu_device is declared before obd_device, while a pointer pointing
1113 * back to obd_device in lu_device, so this helper function defines here
1114 * instead of in lu_object.h
1116 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1118 return lu_dev->ld_obd->obd_name;
1121 static inline bool filename_is_volatile(const char *name, size_t namelen,
1127 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1130 /* caller does not care of idx */
1134 /* volatile file, the MDT can be set from name */
1135 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1136 /* if no MDT is specified, use std way */
1137 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1139 /* test for no MDT idx case */
1140 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1141 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1145 /* we have an idx, read it */
1146 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1147 *idx = simple_strtoul(start, &end, 16);
1149 * no digit, no trailing :, negative value
1151 if (((*idx == 0) && (end == start)) ||
1152 (*end != ':') || (*idx < 0))
1157 /* bad format of mdt idx, we cannot return an error
1158 * to caller so we use hash algo */
1159 CERROR("Bad volatile file name format: %s\n",
1160 name + LUSTRE_VOLATILE_HDR_LEN);
1164 static inline int cli_brw_size(struct obd_device *obd)
1166 LASSERT(obd != NULL);
1167 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1170 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1171 * of the client should be increased accordingly to avoid sending fragmented
1172 * RPCs over the network when the client runs out of the maximum dirty space
1173 * when so many RPCs are being generated.
1175 static inline void client_adjust_max_dirty(struct client_obd *cli)
1178 if (cli->cl_dirty_max_pages <= 0)
1179 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1180 >> PAGE_CACHE_SHIFT;
1182 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1183 cli->cl_max_pages_per_rpc;
1185 if (dirty_max > cli->cl_dirty_max_pages)
1186 cli->cl_dirty_max_pages = dirty_max;
1189 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1190 cli->cl_dirty_max_pages = totalram_pages / 8;
1193 #endif /* __OBD_H */