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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/spinlock.h>
38 #include "lustre/lustre_idl.h"
39 #include "lustre_lib.h"
41 #include "lustre_export.h"
42 #include "lustre_fid.h"
43 #include "lustre_fld.h"
44 #include "lustre_handles.h"
45 #include "lustre_intent.h"
47 #define MAX_OBD_DEVICES 8192
55 struct lov_oinfo { /* per-stripe data structure */
56 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
57 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
58 int loi_ost_gen; /* generation of this loi_ost_idx */
60 unsigned long loi_kms_valid:1;
61 __u64 loi_kms; /* known minimum size */
62 struct ost_lvb loi_lvb;
63 struct osc_async_rc loi_ar;
66 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
69 oinfo->loi_kms_valid = 1;
72 static inline void loi_init(struct lov_oinfo *loi)
77 * If we are unable to get the maximum object size from the OST in
78 * ocd_maxbytes using OBD_CONNECT_MAXBYTES, then we fall back to using
79 * the old maximum object size from ext3.
81 #define LUSTRE_EXT3_STRIPE_MAXBYTES 0x1fffffff000ULL
83 struct lov_stripe_md {
86 pid_t lsm_lock_owner; /* debugging */
88 /* maximum possible file size, might change as OSTs status changes,
89 * e.g. disconnected, deactivated
94 __u32 lsm_stripe_size;
95 __u32 lsm_pattern; /* striping pattern (RAID0, RAID1) */
96 __u16 lsm_stripe_count;
98 char lsm_pool_name[LOV_MAXPOOLNAME + 1];
99 struct lov_oinfo *lsm_oinfo[0];
102 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
104 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
107 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
111 if (lsm_is_released(lsm))
116 static inline int lov_stripe_md_size(unsigned int stripe_count)
118 struct lov_stripe_md lsm;
120 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
125 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
127 /* obd info for a particular level (lov, osc). */
129 /* Flags used for set request specific flags:
130 - while lock handling, the flags obtained on the enqueue
131 request are set here.
132 - while stats, the flags used for control delay/resend.
133 - while setattr, the flags used for distinguish punch operation
136 /* lsm data specific for every OSC. */
137 struct lov_stripe_md *oi_md;
138 /* obdo data specific for every OSC, if needed at all. */
140 /* statfs data specific for every OSC, if needed at all. */
141 struct obd_statfs *oi_osfs;
142 /* An update callback which is called to update some data on upper
143 * level. E.g. it is used for update lsm->lsm_oinfo at every received
144 * request in osc level for enqueue requests. It is also possible to
145 * update some caller data from LOV layer if needed.
147 obd_enqueue_update_f oi_cb_up;
151 struct list_head typ_chain;
152 struct obd_ops *typ_dt_ops;
153 struct md_ops *typ_md_ops;
154 struct dentry *typ_debugfs_entry;
157 struct lu_device_type *typ_lu;
158 spinlock_t obd_type_lock;
159 struct kobject *typ_kobj;
169 struct timeout_item {
170 enum timeout_event ti_event;
171 unsigned long ti_timeout;
174 struct list_head ti_obd_list;
175 struct list_head ti_chain;
178 #define OBD_MAX_RIF_DEFAULT 8
179 #define OBD_MAX_RIF_MAX 512
180 #define OSC_MAX_RIF_MAX 256
181 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
182 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
183 #define OSC_DEFAULT_RESENDS 10
185 /* possible values for fo_sync_lock_cancel */
187 NEVER_SYNC_ON_CANCEL = 0,
188 BLOCKING_SYNC_ON_CANCEL = 1,
189 ALWAYS_SYNC_ON_CANCEL = 2,
190 NUM_SYNC_ON_CANCEL_STATES
193 enum obd_cl_sem_lock_class {
200 * Limit reply buffer size for striping data to one x86_64 page. This
201 * value is chosen to fit the striping data for common use cases while
202 * staying well below the limit at which the buffer must be backed by
203 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
206 #define OBD_MAX_DEFAULT_EA_SIZE 4096
207 #define OBD_MAX_DEFAULT_COOKIE_SIZE 4096
212 struct rw_semaphore cl_sem;
213 struct obd_uuid cl_target_uuid;
214 struct obd_import *cl_import; /* ptlrpc connection state */
215 size_t cl_conn_count;
217 * Cache maximum and default values for easize and cookiesize. This is
218 * strictly a performance optimization to minimize calls to
219 * obd_size_diskmd(). The default values are used to calculate the
220 * initial size of a request buffer. The ptlrpc layer will resize the
221 * buffer as needed to accommodate a larger reply from the
222 * server. The default values should be small enough to avoid wasted
223 * memory and excessive use of vmalloc(), yet large enough to avoid
224 * reallocating the buffer in the common use case.
227 * Default EA size for striping attributes. It is initialized at
228 * mount-time based on the default stripe width of the filesystem,
229 * then it tracks the largest observed EA size advertised by
230 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE.
232 u32 cl_default_mds_easize;
233 /* Maximum possible EA size computed at mount-time based on
234 * the number of OSTs in the filesystem. May be increased at
235 * run-time if a larger observed size is advertised by the MDT.
237 u32 cl_max_mds_easize;
238 /* Default cookie size for llog cookies (see struct llog_cookie). It is
239 * initialized to zero at mount-time, then it tracks the largest
240 * observed cookie size advertised by the MDT, up to a maximum value of
241 * OBD_MAX_DEFAULT_COOKIE_SIZE. Note that llog_cookies are not
242 * used by clients communicating with MDS versions 2.4.0 and later.
244 u32 cl_default_mds_cookiesize;
245 /* Maximum possible cookie size computed at mount-time based on
246 * the number of OSTs in the filesystem. May be increased at
247 * run-time if a larger observed size is advertised by the MDT.
249 u32 cl_max_mds_cookiesize;
251 enum lustre_sec_part cl_sp_me;
252 enum lustre_sec_part cl_sp_to;
253 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
255 /* the grant values are protected by loi_list_lock below */
256 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
257 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
258 unsigned long cl_dirty_transit; /* dirty synchronous */
259 unsigned long cl_avail_grant; /* bytes of credit for ost */
260 unsigned long cl_lost_grant; /* lost credits (trunc) */
262 /* since we allocate grant by blocks, we don't know how many grant will
263 * be used to add a page into cache. As a solution, we reserve maximum
264 * grant before trying to dirty a page and unreserve the rest.
265 * See osc_{reserve|unreserve}_grant for details.
267 long cl_reserved_grant;
268 struct list_head cl_cache_waiters; /* waiting for cache/grant */
269 unsigned long cl_next_shrink_grant; /* jiffies */
270 struct list_head cl_grant_shrink_list; /* Timeout event list */
271 int cl_grant_shrink_interval; /* seconds */
273 /* A chunk is an optimal size used by osc_extent to determine
274 * the extent size. A chunk is max(PAGE_SIZE, OST block size)
277 unsigned int cl_extent_tax; /* extent overhead, by bytes */
279 /* keep track of objects that have lois that contain pages which
280 * have been queued for async brw. this lock also protects the
281 * lists of osc_client_pages that hang off of the loi
284 * ->cl_loi_list_lock protects consistency of
285 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
286 * ->ap_completion() call-backs are executed under this lock. As we
287 * cannot guarantee that these call-backs never block on all platforms
288 * (as a matter of fact they do block on Mac OS X), type of
289 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
290 * and blocking mutex on Mac OS X. (Alternative is to make this lock
291 * blocking everywhere, but we don't want to slow down fast-path of
292 * our main platform.)
294 * NB by Jinshan: though field names are still _loi_, but actually
295 * osc_object{}s are in the list.
297 spinlock_t cl_loi_list_lock;
298 struct list_head cl_loi_ready_list;
299 struct list_head cl_loi_hp_ready_list;
300 struct list_head cl_loi_write_list;
301 struct list_head cl_loi_read_list;
302 __u32 cl_r_in_flight;
303 __u32 cl_w_in_flight;
304 /* just a sum of the loi/lop pending numbers to be exported by sysfs */
305 atomic_t cl_pending_w_pages;
306 atomic_t cl_pending_r_pages;
307 __u32 cl_max_pages_per_rpc;
308 __u32 cl_max_rpcs_in_flight;
309 struct obd_histogram cl_read_rpc_hist;
310 struct obd_histogram cl_write_rpc_hist;
311 struct obd_histogram cl_read_page_hist;
312 struct obd_histogram cl_write_page_hist;
313 struct obd_histogram cl_read_offset_hist;
314 struct obd_histogram cl_write_offset_hist;
316 /* lru for osc caching pages */
317 struct cl_client_cache *cl_cache;
318 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
319 atomic_long_t *cl_lru_left;
320 atomic_long_t cl_lru_busy;
321 atomic_long_t cl_lru_in_list;
322 atomic_t cl_lru_shrinkers;
323 struct list_head cl_lru_list; /* lru page list */
324 spinlock_t cl_lru_list_lock; /* page list protector */
325 atomic_long_t cl_unstable_count;
327 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
328 atomic_t cl_destroy_in_flight;
329 wait_queue_head_t cl_destroy_waitq;
331 struct mdc_rpc_lock *cl_rpc_lock;
332 struct mdc_rpc_lock *cl_close_lock;
335 atomic_t cl_mgc_refcount;
336 struct obd_export *cl_mgc_mgsexp;
338 /* checksumming for data sent over the network */
339 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
340 /* supported checksum types that are worked out at connect time */
341 __u32 cl_supp_cksum_types;
342 /* checksum algorithm to be used */
343 enum cksum_type cl_cksum_type;
345 /* also protected by the poorly named _loi_list_lock lock above */
346 struct osc_async_rc cl_ar;
348 /* used by quotacheck when the servers are older than 2.4 */
349 int cl_qchk_stat; /* quotacheck stat of the peer */
350 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
351 #if OBD_OCD_VERSION(2, 7, 53, 0) < LUSTRE_VERSION_CODE
352 #warning "please consider removing quotacheck compatibility code"
355 /* sequence manager */
356 struct lu_client_seq *cl_seq;
358 atomic_t cl_resends; /* resend count */
360 /* ptlrpc work for writeback in ptlrpcd context */
361 void *cl_writeback_work;
363 /* hash tables for osc_quota_info */
364 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
367 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
374 struct echo_client_obd {
375 struct obd_export *ec_exp; /* the local connection to osc/lov */
377 struct list_head ec_objects;
378 struct list_head ec_locks;
382 /* Generic subset of OSTs */
384 __u32 *op_array; /* array of index of lov_obd->lov_tgts */
385 unsigned int op_count; /* number of OSTs in the array */
386 unsigned int op_size; /* allocated size of lp_array */
387 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
390 /* allow statfs data caching for 1 second */
391 #define OBD_STATFS_CACHE_SECONDS 1
393 struct lov_tgt_desc {
394 struct list_head ltd_kill;
395 struct obd_uuid ltd_uuid;
396 struct obd_device *ltd_obd;
397 struct obd_export *ltd_exp;
399 __u32 ltd_index; /* index in lov_obd->tgts */
400 unsigned long ltd_active:1,/* is this target up for requests */
401 ltd_activate:1,/* should target be activated */
402 ltd_reap:1; /* should this target be deleted */
406 struct lov_desc desc;
407 struct lov_tgt_desc **lov_tgts; /* sparse array */
408 struct ost_pool lov_packed; /* all OSTs in a packed array */
409 struct mutex lov_lock;
410 struct obd_connect_data lov_ocd;
411 atomic_t lov_refcount;
412 __u32 lov_death_row;/* tgts scheduled to be deleted */
413 __u32 lov_tgt_size; /* size of tgts array */
416 struct cfs_hash *lov_pools_hash_body; /* used for key access */
417 struct list_head lov_pool_list; /* used for sequential access */
418 struct dentry *lov_pool_debugfs_entry;
419 enum lustre_sec_part lov_sp_me;
421 /* Cached LRU and unstable data from upper layer */
422 struct cl_client_cache *lov_cache;
424 struct rw_semaphore lov_notify_lock;
426 struct kobject *lov_tgts_kobj;
429 struct lmv_tgt_desc {
430 struct obd_uuid ltd_uuid;
431 struct obd_export *ltd_exp;
433 struct mutex ltd_fid_mutex;
434 unsigned long ltd_active:1; /* target up for requests */
437 enum placement_policy {
438 PLACEMENT_CHAR_POLICY = 0,
439 PLACEMENT_NID_POLICY = 1,
440 PLACEMENT_INVAL_POLICY = 2,
446 struct lu_client_fld lmv_fld;
448 enum placement_policy lmv_placement;
449 struct lmv_desc desc;
450 struct obd_uuid cluuid;
451 struct obd_export *exp;
453 struct mutex lmv_init_mutex;
458 int max_def_cookiesize;
460 u32 tgts_size; /* size of tgts array */
461 struct lmv_tgt_desc **tgts;
463 struct obd_connect_data conn_data;
464 struct kobject *lmv_tgts_kobj;
467 struct niobuf_local {
468 __u64 lnb_file_offset;
469 __u32 lnb_page_offset;
472 struct page *lnb_page;
477 #define LUSTRE_FLD_NAME "fld"
478 #define LUSTRE_SEQ_NAME "seq"
480 #define LUSTRE_MDD_NAME "mdd"
481 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
482 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
483 #define LUSTRE_VVP_NAME "vvp"
484 #define LUSTRE_LMV_NAME "lmv"
485 #define LUSTRE_SLP_NAME "slp"
486 #define LUSTRE_LOD_NAME "lod"
487 #define LUSTRE_OSP_NAME "osp"
488 #define LUSTRE_LWP_NAME "lwp"
490 /* obd device type names */
491 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
492 #define LUSTRE_MDS_NAME "mds"
493 #define LUSTRE_MDT_NAME "mdt"
494 #define LUSTRE_MDC_NAME "mdc"
495 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
496 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
497 #define LUSTRE_OSC_NAME "osc"
498 #define LUSTRE_LOV_NAME "lov"
499 #define LUSTRE_MGS_NAME "mgs"
500 #define LUSTRE_MGC_NAME "mgc"
502 #define LUSTRE_ECHO_NAME "obdecho"
503 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
504 #define LUSTRE_QMT_NAME "qmt"
506 /* Constant obd names (post-rename) */
507 #define LUSTRE_MDS_OBDNAME "MDS"
508 #define LUSTRE_OSS_OBDNAME "OSS"
509 #define LUSTRE_MGS_OBDNAME "MGS"
510 #define LUSTRE_MGC_OBDNAME "MGC"
512 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
513 #define N_LOCAL_TEMP_PAGE 0x10000000
515 struct obd_trans_info {
517 /* Only used on the server side for tracking acks. */
518 struct oti_req_ack_lock {
519 struct lustre_handle lock;
523 struct llog_cookie oti_onecookie;
524 struct llog_cookie *oti_logcookies;
527 __u64 oti_pre_version;
531 * Events signalled through obd_notify() upcall-chain.
533 enum obd_notify_event {
536 /* Device connect start */
538 /* Device activated */
540 /* Device deactivated */
542 /* Device disconnected */
544 /* Connect data for import were changed */
547 OBD_NOTIFY_SYNC_NONBLOCK,
549 /* Configuration event */
551 /* Administratively deactivate/activate event */
552 OBD_NOTIFY_DEACTIVATE,
557 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
558 * being main example).
560 struct obd_notify_upcall {
561 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
562 enum obd_notify_event ev, void *owner, void *data);
563 /* Opaque datum supplied by upper layer listener */
567 struct target_recovery_data {
568 svc_handler_t trd_recovery_handler;
569 pid_t trd_processing_task;
570 struct completion trd_starting;
571 struct completion trd_finishing;
574 struct obd_llog_group {
575 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
576 wait_queue_head_t olg_waitq;
578 struct mutex olg_cat_processing;
581 /* corresponds to one of the obd's */
582 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
584 struct lvfs_run_ctxt {
585 struct dt_device *dt;
589 struct obd_type *obd_type;
592 /* common and UUID name of this device */
593 char obd_name[MAX_OBD_NAME];
594 struct obd_uuid obd_uuid;
596 struct lu_device *obd_lu_dev;
599 /* bitfield modification is protected by obd_dev_lock */
600 unsigned long obd_attached:1, /* finished attach */
601 obd_set_up:1, /* finished setup */
602 obd_version_recov:1, /* obd uses version checking */
603 obd_replayable:1,/* recovery is enabled; inform clients */
604 obd_no_transno:1, /* no committed-transno notification */
605 obd_no_recov:1, /* fail instead of retry messages */
606 obd_stopping:1, /* started cleanup */
607 obd_starting:1, /* started setup */
608 obd_force:1, /* cleanup with > 0 obd refcount */
609 obd_fail:1, /* cleanup with failover */
610 obd_no_conn:1, /* deny new connections */
611 obd_inactive:1, /* device active/inactive
612 * (for sysfs status only!!)
614 obd_no_ir:1, /* no imperative recovery. */
615 obd_process_conf:1; /* device is processing mgs config */
616 /* use separate field as it is set in interrupt to don't mess with
617 * protection of other bits using _bh lock
619 unsigned long obd_recovery_expired:1;
620 /* uuid-export hash body */
621 struct cfs_hash *obd_uuid_hash;
622 atomic_t obd_refcount;
623 wait_queue_head_t obd_refcount_waitq;
624 struct list_head obd_exports;
625 struct list_head obd_unlinked_exports;
626 struct list_head obd_delayed_exports;
628 spinlock_t obd_nid_lock;
629 struct ldlm_namespace *obd_namespace;
630 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
631 /* a spinlock is OK for what we do now, may need a semaphore later */
632 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
633 struct mutex obd_dev_mutex;
634 __u64 obd_last_committed;
635 spinlock_t obd_osfs_lock;
636 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
638 struct lvfs_run_ctxt obd_lvfs_ctxt;
639 struct obd_llog_group obd_olg; /* default llog group */
640 struct obd_device *obd_observer;
641 struct rw_semaphore obd_observer_link_sem;
642 struct obd_notify_upcall obd_upcall;
643 struct obd_export *obd_self_export;
646 struct client_obd cli;
647 struct echo_client_obd echo_client;
651 /* Fields used by LProcFS */
652 unsigned int obd_cntr_base;
653 struct lprocfs_stats *obd_stats;
655 unsigned int md_cntr_base;
656 struct lprocfs_stats *md_stats;
658 struct dentry *obd_debugfs_entry;
659 struct dentry *obd_svc_debugfs_entry;
660 struct lprocfs_stats *obd_svc_stats;
661 atomic_t obd_evict_inprogress;
662 wait_queue_head_t obd_evict_inprogress_waitq;
663 struct list_head obd_evict_list; /* protected with pet_lock */
666 * Ldlm pool part. Save last calculated SLV and Limit.
668 rwlock_t obd_pool_lock;
673 * A list of outstanding class_incref()'s against this obd. For
676 struct lu_ref obd_reference;
678 int obd_conn_inprogress;
680 struct kobject obd_kobj; /* sysfs object */
681 struct completion obd_kobj_unregister;
684 enum obd_cleanup_stage {
685 /* Special case hack for MDS LOVs */
687 /* can be directly mapped to .ldto_device_fini() */
691 /* get/set_info keys */
692 #define KEY_ASYNC "async"
693 #define KEY_CHANGELOG_CLEAR "changelog_clear"
694 #define KEY_FID2PATH "fid2path"
695 #define KEY_CHECKSUM "checksum"
696 #define KEY_CLEAR_FS "clear_fs"
697 #define KEY_CONN_DATA "conn_data"
698 #define KEY_EVICT_BY_NID "evict_by_nid"
699 #define KEY_FIEMAP "fiemap"
700 #define KEY_FLUSH_CTX "flush_ctx"
701 #define KEY_GRANT_SHRINK "grant_shrink"
702 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
703 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
704 #define KEY_INTERMDS "inter_mds"
705 #define KEY_LAST_ID "last_id"
706 #define KEY_LAST_FID "last_fid"
707 #define KEY_LOVDESC "lovdesc"
708 #define KEY_MAX_EASIZE "max_easize"
709 #define KEY_DEFAULT_EASIZE "default_easize"
710 #define KEY_MGSSEC "mgssec"
711 #define KEY_READ_ONLY "read-only"
712 #define KEY_REGISTER_TARGET "register_target"
713 #define KEY_SET_FS "set_fs"
714 #define KEY_TGT_COUNT "tgt_count"
715 /* KEY_SET_INFO in lustre_idl.h */
716 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
718 #define KEY_CACHE_SET "cache_set"
719 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
723 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
724 #define IT_OPEN (1 << 0)
725 #define IT_CREAT (1 << 1)
726 #define IT_READDIR (1 << 2)
727 #define IT_GETATTR (1 << 3)
728 #define IT_LOOKUP (1 << 4)
729 #define IT_UNLINK (1 << 5)
730 #define IT_TRUNC (1 << 6)
731 #define IT_GETXATTR (1 << 7)
732 #define IT_EXEC (1 << 8)
733 #define IT_PIN (1 << 9)
734 #define IT_LAYOUT (1 << 10)
735 #define IT_QUOTA_DQACQ (1 << 11)
736 #define IT_QUOTA_CONN (1 << 12)
737 #define IT_SETXATTR (1 << 13)
739 static inline int it_to_lock_mode(struct lookup_intent *it)
741 /* CREAT needs to be tested before open (both could be set) */
742 if (it->it_op & IT_CREAT)
744 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
747 else if (it->it_op & IT_READDIR)
749 else if (it->it_op & IT_GETXATTR)
751 else if (it->it_op & IT_SETXATTR)
754 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
759 CLI_SET_MEA = BIT(0),
760 CLI_RM_ENTRY = BIT(1),
763 CLI_MIGRATE = BIT(4),
767 struct lu_fid op_fid1; /* operation fid1 (usually parent) */
768 struct lu_fid op_fid2; /* operation fid2 (usually child) */
769 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
770 struct lu_fid op_fid4; /* to the operation locks. */
771 u32 op_mds; /* what mds server open will go to */
772 struct lustre_handle op_handle;
777 struct lmv_stripe_md *op_mea1;
778 struct lmv_stripe_md *op_mea2;
779 __u32 op_suppgids[2];
786 /* iattr fields and blocks. */
787 struct iattr op_attr;
788 unsigned int op_attr_flags;
790 loff_t op_attr_blocks;
792 /* Size-on-MDS epoch and flags. */
796 /* Various operation flags. */
797 enum mds_op_bias op_bias;
799 /* Used by readdir */
802 /* Used by readdir */
805 /* used to transfer info between the stacks of MD client
806 * see enum op_cli_flags
808 enum md_cli_flags op_cli_flags;
810 /* File object data version for HSM release, on client */
811 __u64 op_data_version;
812 struct lustre_handle op_lease_handle;
814 /* default stripe offset */
815 __u32 op_default_stripe_offset;
819 int (*md_blocking_ast)(struct ldlm_lock *lock,
820 struct ldlm_lock_desc *desc,
821 void *data, int flag);
824 struct md_enqueue_info;
825 /* metadata stat-ahead */
827 struct md_enqueue_info {
828 struct md_op_data mi_data;
829 struct lookup_intent mi_it;
830 struct lustre_handle mi_lockh;
831 struct inode *mi_dir;
832 int (*mi_cb)(struct ptlrpc_request *req,
833 struct md_enqueue_info *minfo, int rc);
838 struct module *owner;
839 int (*iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
840 void *karg, void __user *uarg);
841 int (*get_info)(const struct lu_env *env, struct obd_export *,
842 __u32 keylen, void *key, __u32 *vallen, void *val,
843 struct lov_stripe_md *lsm);
844 int (*set_info_async)(const struct lu_env *, struct obd_export *,
845 __u32 keylen, void *key,
846 __u32 vallen, void *val,
847 struct ptlrpc_request_set *set);
848 int (*setup)(struct obd_device *dev, struct lustre_cfg *cfg);
849 int (*precleanup)(struct obd_device *dev,
850 enum obd_cleanup_stage cleanup_stage);
851 int (*cleanup)(struct obd_device *dev);
852 int (*process_config)(struct obd_device *dev, u32 len, void *data);
853 int (*postrecov)(struct obd_device *dev);
854 int (*add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
856 int (*del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
857 /* connect to the target device with given connection
858 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
859 * granted by the target, which are guaranteed to be a subset of flags
860 * asked for. If @ocd == NULL, use default parameters.
862 int (*connect)(const struct lu_env *env,
863 struct obd_export **exp, struct obd_device *src,
864 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
866 int (*reconnect)(const struct lu_env *env,
867 struct obd_export *exp, struct obd_device *src,
868 struct obd_uuid *cluuid,
869 struct obd_connect_data *ocd,
871 int (*disconnect)(struct obd_export *exp);
873 /* Initialize/finalize fids infrastructure. */
874 int (*fid_init)(struct obd_device *obd,
875 struct obd_export *exp, enum lu_cli_type type);
876 int (*fid_fini)(struct obd_device *obd);
878 /* Allocate new fid according to passed @hint. */
879 int (*fid_alloc)(const struct lu_env *env, struct obd_export *exp,
880 struct lu_fid *fid, struct md_op_data *op_data);
883 * Object with @fid is getting deleted, we may want to do something
886 int (*statfs)(const struct lu_env *, struct obd_export *exp,
887 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
888 int (*statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
889 __u64 max_age, struct ptlrpc_request_set *set);
890 int (*packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
891 struct lov_stripe_md *mem_src);
892 int (*unpackmd)(struct obd_export *exp,
893 struct lov_stripe_md **mem_tgt,
894 struct lov_mds_md *disk_src, int disk_len);
895 int (*create)(const struct lu_env *env, struct obd_export *exp,
896 struct obdo *oa, struct obd_trans_info *oti);
897 int (*destroy)(const struct lu_env *env, struct obd_export *exp,
898 struct obdo *oa, struct obd_trans_info *oti);
899 int (*setattr)(const struct lu_env *, struct obd_export *exp,
900 struct obd_info *oinfo, struct obd_trans_info *oti);
901 int (*setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
902 struct obd_trans_info *oti,
903 struct ptlrpc_request_set *rqset);
904 int (*getattr)(const struct lu_env *env, struct obd_export *exp,
905 struct obd_info *oinfo);
906 int (*getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
907 struct ptlrpc_request_set *set);
908 int (*preprw)(const struct lu_env *env, int cmd,
909 struct obd_export *exp, struct obdo *oa, int objcount,
910 struct obd_ioobj *obj, struct niobuf_remote *remote,
911 int *nr_pages, struct niobuf_local *local,
912 struct obd_trans_info *oti);
913 int (*commitrw)(const struct lu_env *env, int cmd,
914 struct obd_export *exp, struct obdo *oa,
915 int objcount, struct obd_ioobj *obj,
916 struct niobuf_remote *remote, int pages,
917 struct niobuf_local *local,
918 struct obd_trans_info *oti, int rc);
919 int (*init_export)(struct obd_export *exp);
920 int (*destroy_export)(struct obd_export *exp);
922 /* metadata-only methods */
923 int (*import_event)(struct obd_device *, struct obd_import *,
924 enum obd_import_event);
926 int (*notify)(struct obd_device *obd, struct obd_device *watched,
927 enum obd_notify_event ev, void *data);
929 int (*health_check)(const struct lu_env *env, struct obd_device *);
930 struct obd_uuid *(*get_uuid)(struct obd_export *exp);
933 int (*quotacheck)(struct obd_device *, struct obd_export *,
934 struct obd_quotactl *);
935 int (*quotactl)(struct obd_device *, struct obd_export *,
936 struct obd_quotactl *);
939 int (*pool_new)(struct obd_device *obd, char *poolname);
940 int (*pool_del)(struct obd_device *obd, char *poolname);
941 int (*pool_add)(struct obd_device *obd, char *poolname,
943 int (*pool_rem)(struct obd_device *obd, char *poolname,
945 void (*getref)(struct obd_device *obd);
946 void (*putref)(struct obd_device *obd);
948 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
949 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
950 * Also, add a wrapper function in include/linux/obd_class.h.
956 struct mdt_body *body;
957 struct lov_stripe_md *lsm;
958 struct lmv_stripe_md *lmv;
959 #ifdef CONFIG_FS_POSIX_ACL
960 struct posix_acl *posix_acl;
962 struct mdt_remote_perm *remote_perm;
965 struct md_open_data {
966 struct obd_client_handle *mod_och;
967 struct ptlrpc_request *mod_open_req;
968 struct ptlrpc_request *mod_close_req;
969 atomic_t mod_refcount;
973 struct obd_client_handle {
974 struct lustre_handle och_fh;
975 struct lu_fid och_fid;
976 struct md_open_data *och_mod;
977 struct lustre_handle och_lease_handle; /* open lock for lease */
982 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
984 struct lookup_intent;
988 int (*getstatus)(struct obd_export *, struct lu_fid *);
989 int (*null_inode)(struct obd_export *, const struct lu_fid *);
990 int (*close)(struct obd_export *, struct md_op_data *,
991 struct md_open_data *, struct ptlrpc_request **);
992 int (*create)(struct obd_export *, struct md_op_data *,
993 const void *, size_t, umode_t, uid_t, gid_t,
994 cfs_cap_t, __u64, struct ptlrpc_request **);
995 int (*done_writing)(struct obd_export *, struct md_op_data *,
996 struct md_open_data *);
997 int (*enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
998 const ldlm_policy_data_t *,
999 struct lookup_intent *, struct md_op_data *,
1000 struct lustre_handle *, __u64);
1001 int (*getattr)(struct obd_export *, struct md_op_data *,
1002 struct ptlrpc_request **);
1003 int (*getattr_name)(struct obd_export *, struct md_op_data *,
1004 struct ptlrpc_request **);
1005 int (*intent_lock)(struct obd_export *, struct md_op_data *,
1006 struct lookup_intent *,
1007 struct ptlrpc_request **,
1008 ldlm_blocking_callback, __u64);
1009 int (*link)(struct obd_export *, struct md_op_data *,
1010 struct ptlrpc_request **);
1011 int (*rename)(struct obd_export *, struct md_op_data *,
1012 const char *, size_t, const char *, size_t,
1013 struct ptlrpc_request **);
1014 int (*setattr)(struct obd_export *, struct md_op_data *, void *,
1015 size_t, void *, size_t, struct ptlrpc_request **,
1016 struct md_open_data **mod);
1017 int (*sync)(struct obd_export *, const struct lu_fid *,
1018 struct ptlrpc_request **);
1019 int (*read_page)(struct obd_export *, struct md_op_data *,
1020 struct md_callback *cb_op, __u64 hash_offset,
1021 struct page **ppage);
1022 int (*unlink)(struct obd_export *, struct md_op_data *,
1023 struct ptlrpc_request **);
1025 int (*setxattr)(struct obd_export *, const struct lu_fid *,
1026 u64, const char *, const char *, int, int, int, __u32,
1027 struct ptlrpc_request **);
1029 int (*getxattr)(struct obd_export *, const struct lu_fid *,
1030 u64, const char *, const char *, int, int, int,
1031 struct ptlrpc_request **);
1033 int (*init_ea_size)(struct obd_export *, u32, u32, u32, u32);
1035 int (*get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1036 struct obd_export *, struct obd_export *,
1037 struct lustre_md *);
1039 int (*free_lustre_md)(struct obd_export *, struct lustre_md *);
1041 int (*merge_attr)(struct obd_export *,
1042 const struct lmv_stripe_md *lsm,
1043 struct cl_attr *attr, ldlm_blocking_callback);
1045 int (*set_open_replay_data)(struct obd_export *,
1046 struct obd_client_handle *,
1047 struct lookup_intent *);
1048 int (*clear_open_replay_data)(struct obd_export *,
1049 struct obd_client_handle *);
1050 int (*set_lock_data)(struct obd_export *, const struct lustre_handle *,
1053 enum ldlm_mode (*lock_match)(struct obd_export *, __u64,
1054 const struct lu_fid *, enum ldlm_type,
1055 ldlm_policy_data_t *, enum ldlm_mode,
1056 struct lustre_handle *);
1058 int (*cancel_unused)(struct obd_export *, const struct lu_fid *,
1059 ldlm_policy_data_t *, enum ldlm_mode,
1060 enum ldlm_cancel_flags flags, void *opaque);
1062 int (*get_fid_from_lsm)(struct obd_export *,
1063 const struct lmv_stripe_md *,
1064 const char *name, int namelen,
1065 struct lu_fid *fid);
1067 int (*intent_getattr_async)(struct obd_export *,
1068 struct md_enqueue_info *,
1069 struct ldlm_enqueue_info *);
1071 int (*revalidate_lock)(struct obd_export *, struct lookup_intent *,
1072 struct lu_fid *, __u64 *bits);
1075 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1076 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1077 * wrapper function in include/linux/obd_class.h.
1081 struct lsm_operations {
1082 void (*lsm_free)(struct lov_stripe_md *);
1083 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, u64 *,
1085 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, u64 *,
1087 int (*lsm_lmm_verify)(struct lov_mds_md *lmm, int lmm_bytes,
1088 __u16 *stripe_count);
1089 int (*lsm_unpackmd)(struct lov_obd *lov, struct lov_stripe_md *lsm,
1090 struct lov_mds_md *lmm);
1093 extern const struct lsm_operations lsm_v1_ops;
1094 extern const struct lsm_operations lsm_v3_ops;
1095 static inline const struct lsm_operations *lsm_op_find(int magic)
1103 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1108 static inline struct md_open_data *obd_mod_alloc(void)
1110 struct md_open_data *mod;
1112 mod = kzalloc(sizeof(*mod), GFP_NOFS);
1115 atomic_set(&mod->mod_refcount, 1);
1119 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1120 #define obd_mod_put(mod) \
1122 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1123 if ((mod)->mod_open_req) \
1124 ptlrpc_req_finished((mod)->mod_open_req); \
1129 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1130 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1132 /* return 1 if client should be resend request */
1133 static inline int client_should_resend(int resend, struct client_obd *cli)
1135 return atomic_read(&cli->cl_resends) ?
1136 atomic_read(&cli->cl_resends) > resend : 1;
1140 * Return device name for this device
1142 * XXX: lu_device is declared before obd_device, while a pointer pointing
1143 * back to obd_device in lu_device, so this helper function defines here
1144 * instead of in lu_object.h
1146 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1148 return lu_dev->ld_obd->obd_name;
1151 static inline bool filename_is_volatile(const char *name, size_t namelen,
1157 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1160 /* caller does not care of idx */
1164 /* volatile file, the MDT can be set from name */
1165 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1166 /* if no MDT is specified, use std way */
1167 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1169 /* test for no MDT idx case */
1170 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1171 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1175 /* we have an idx, read it */
1176 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1177 *idx = simple_strtoul(start, &end, 0);
1179 * no digit, no trailing :, negative value
1181 if (((*idx == 0) && (end == start)) ||
1182 (*end != ':') || (*idx < 0))
1187 /* bad format of mdt idx, we cannot return an error
1188 * to caller so we use hash algo
1190 CERROR("Bad volatile file name format: %s\n",
1191 name + LUSTRE_VOLATILE_HDR_LEN);
1195 static inline int cli_brw_size(struct obd_device *obd)
1197 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1201 * when RPC size or the max RPCs in flight is increased, the max dirty pages
1202 * of the client should be increased accordingly to avoid sending fragmented
1203 * RPCs over the network when the client runs out of the maximum dirty space
1204 * when so many RPCs are being generated.
1206 static inline void client_adjust_max_dirty(struct client_obd *cli)
1209 if (cli->cl_dirty_max_pages <= 0)
1210 cli->cl_dirty_max_pages =
1211 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1213 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1214 cli->cl_max_pages_per_rpc;
1216 if (dirty_max > cli->cl_dirty_max_pages)
1217 cli->cl_dirty_max_pages = dirty_max;
1220 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1221 cli->cl_dirty_max_pages = totalram_pages / 8;
1224 #endif /* __OBD_H */