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"
46 #include "cl_object.h"
48 #define MAX_OBD_DEVICES 8192
56 struct lov_oinfo { /* per-stripe data structure */
57 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
58 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
59 int loi_ost_gen; /* generation of this loi_ost_idx */
61 unsigned long loi_kms_valid:1;
62 __u64 loi_kms; /* known minimum size */
63 struct ost_lvb loi_lvb;
64 struct osc_async_rc loi_ar;
67 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
70 oinfo->loi_kms_valid = 1;
73 static inline void loi_init(struct lov_oinfo *loi)
80 int lov_read_and_clear_async_rc(struct cl_object *clob);
82 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
84 /* obd info for a particular level (lov, osc). */
86 /* OBD_STATFS_* flags */
88 /* lsm data specific for every OSC. */
89 struct lov_stripe_md *oi_md;
90 /* statfs data specific for every OSC, if needed at all. */
91 struct obd_statfs *oi_osfs;
92 /* An update callback which is called to update some data on upper
93 * level. E.g. it is used for update lsm->lsm_oinfo at every received
94 * request in osc level for enqueue requests. It is also possible to
95 * update some caller data from LOV layer if needed.
97 obd_enqueue_update_f oi_cb_up;
101 struct list_head typ_chain;
102 struct obd_ops *typ_dt_ops;
103 struct md_ops *typ_md_ops;
104 struct dentry *typ_debugfs_entry;
107 struct lu_device_type *typ_lu;
108 spinlock_t obd_type_lock;
109 struct kobject *typ_kobj;
119 struct timeout_item {
120 enum timeout_event ti_event;
121 unsigned long ti_timeout;
124 struct list_head ti_obd_list;
125 struct list_head ti_chain;
128 #define OBD_MAX_RIF_DEFAULT 8
129 #define OBD_MAX_RIF_MAX 512
130 #define OSC_MAX_RIF_MAX 256
131 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
132 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
133 #define OSC_DEFAULT_RESENDS 10
135 /* possible values for fo_sync_lock_cancel */
137 NEVER_SYNC_ON_CANCEL = 0,
138 BLOCKING_SYNC_ON_CANCEL = 1,
139 ALWAYS_SYNC_ON_CANCEL = 2,
140 NUM_SYNC_ON_CANCEL_STATES
143 enum obd_cl_sem_lock_class {
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.
176 * Default EA size for striping attributes. It is initialized at
177 * mount-time based on the default stripe width of the filesystem,
178 * then it tracks the largest observed EA size advertised by
179 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE.
181 u32 cl_default_mds_easize;
182 /* Maximum possible EA size computed at mount-time based on
183 * the number of OSTs in the filesystem. May be increased at
184 * run-time if a larger observed size is advertised by the MDT.
186 u32 cl_max_mds_easize;
188 enum lustre_sec_part cl_sp_me;
189 enum lustre_sec_part cl_sp_to;
190 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
192 /* the grant values are protected by loi_list_lock below */
193 unsigned long cl_dirty_pages; /* all _dirty_ in pahges */
194 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
195 unsigned long cl_dirty_transit; /* dirty synchronous */
196 unsigned long cl_avail_grant; /* bytes of credit for ost */
197 unsigned long cl_lost_grant; /* lost credits (trunc) */
199 /* since we allocate grant by blocks, we don't know how many grant will
200 * be used to add a page into cache. As a solution, we reserve maximum
201 * grant before trying to dirty a page and unreserve the rest.
202 * See osc_{reserve|unreserve}_grant for details.
204 long cl_reserved_grant;
205 struct list_head cl_cache_waiters; /* waiting for cache/grant */
206 unsigned long cl_next_shrink_grant; /* jiffies */
207 struct list_head cl_grant_shrink_list; /* Timeout event list */
208 int cl_grant_shrink_interval; /* seconds */
210 /* A chunk is an optimal size used by osc_extent to determine
211 * the extent size. A chunk is max(PAGE_SIZE, OST block size)
214 unsigned int cl_extent_tax; /* extent overhead, by bytes */
216 /* keep track of objects that have lois that contain pages which
217 * have been queued for async brw. this lock also protects the
218 * lists of osc_client_pages that hang off of the loi
221 * ->cl_loi_list_lock protects consistency of
222 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
223 * ->ap_completion() call-backs are executed under this lock. As we
224 * cannot guarantee that these call-backs never block on all platforms
225 * (as a matter of fact they do block on Mac OS X), type of
226 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
227 * and blocking mutex on Mac OS X. (Alternative is to make this lock
228 * blocking everywhere, but we don't want to slow down fast-path of
229 * our main platform.)
231 * NB by Jinshan: though field names are still _loi_, but actually
232 * osc_object{}s are in the list.
234 spinlock_t cl_loi_list_lock;
235 struct list_head cl_loi_ready_list;
236 struct list_head cl_loi_hp_ready_list;
237 struct list_head cl_loi_write_list;
238 struct list_head cl_loi_read_list;
239 __u32 cl_r_in_flight;
240 __u32 cl_w_in_flight;
241 /* just a sum of the loi/lop pending numbers to be exported by sysfs */
242 atomic_t cl_pending_w_pages;
243 atomic_t cl_pending_r_pages;
244 __u32 cl_max_pages_per_rpc;
245 __u32 cl_max_rpcs_in_flight;
246 struct obd_histogram cl_read_rpc_hist;
247 struct obd_histogram cl_write_rpc_hist;
248 struct obd_histogram cl_read_page_hist;
249 struct obd_histogram cl_write_page_hist;
250 struct obd_histogram cl_read_offset_hist;
251 struct obd_histogram cl_write_offset_hist;
253 /* LRU for osc caching pages */
254 struct cl_client_cache *cl_cache;
255 /** member of cl_cache->ccc_lru */
256 struct list_head cl_lru_osc;
257 /** # of available LRU slots left in the per-OSC cache.
258 * Available LRU slots are shared by all OSCs of the same file system,
259 * therefore this is a pointer to cl_client_cache::ccc_lru_left.
261 atomic_long_t *cl_lru_left;
262 /** # of busy LRU pages. A page is considered busy if it's in writeback
263 * queue, or in transfer. Busy pages can't be discarded so they are not
266 atomic_long_t cl_lru_busy;
267 /** # of LRU pages in the cache for this client_obd */
268 atomic_long_t cl_lru_in_list;
269 /** # of threads are shrinking LRU cache. To avoid contention, it's not
270 * allowed to have multiple threads shrinking LRU cache.
272 atomic_t cl_lru_shrinkers;
273 /** The time when this LRU cache was last used. */
274 time64_t cl_lru_last_used;
275 /** stats: how many reclaims have happened for this client_obd.
276 * reclaim and shrink - shrink is async, voluntarily rebalancing;
277 * reclaim is sync, initiated by IO thread when the LRU slots are
281 /** List of LRU pages for this client_obd */
282 struct list_head cl_lru_list;
283 /** Lock for LRU page list */
284 spinlock_t cl_lru_list_lock;
285 /** # of unstable pages in this client_obd.
286 * An unstable page is a page state that WRITE RPC has finished but
287 * the transaction has NOT yet committed.
289 atomic_long_t cl_unstable_count;
290 /** Link to osc_shrinker_list */
291 struct list_head cl_shrink_list;
293 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
294 atomic_t cl_destroy_in_flight;
295 wait_queue_head_t cl_destroy_waitq;
297 struct mdc_rpc_lock *cl_rpc_lock;
299 /* modify rpcs in flight
300 * currently used for metadata only
302 spinlock_t cl_mod_rpcs_lock;
303 u16 cl_max_mod_rpcs_in_flight;
304 u16 cl_mod_rpcs_in_flight;
305 u16 cl_close_rpcs_in_flight;
306 wait_queue_head_t cl_mod_rpcs_waitq;
307 unsigned long *cl_mod_tag_bitmap;
308 struct obd_histogram cl_mod_rpcs_hist;
311 atomic_t cl_mgc_refcount;
312 struct obd_export *cl_mgc_mgsexp;
314 /* checksumming for data sent over the network */
315 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
316 /* supported checksum types that are worked out at connect time */
317 __u32 cl_supp_cksum_types;
318 /* checksum algorithm to be used */
319 enum cksum_type cl_cksum_type;
321 /* also protected by the poorly named _loi_list_lock lock above */
322 struct osc_async_rc cl_ar;
324 /* sequence manager */
325 struct lu_client_seq *cl_seq;
327 atomic_t cl_resends; /* resend count */
329 /* ptlrpc work for writeback in ptlrpcd context */
330 void *cl_writeback_work;
332 /* hash tables for osc_quota_info */
333 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
336 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
343 struct echo_client_obd {
344 struct obd_export *ec_exp; /* the local connection to osc/lov */
346 struct list_head ec_objects;
347 struct list_head ec_locks;
351 /* Generic subset of OSTs */
353 __u32 *op_array; /* array of index of lov_obd->lov_tgts */
354 unsigned int op_count; /* number of OSTs in the array */
355 unsigned int op_size; /* allocated size of lp_array */
356 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
359 /* allow statfs data caching for 1 second */
360 #define OBD_STATFS_CACHE_SECONDS 1
362 struct lov_tgt_desc {
363 struct list_head ltd_kill;
364 struct obd_uuid ltd_uuid;
365 struct obd_device *ltd_obd;
366 struct obd_export *ltd_exp;
368 __u32 ltd_index; /* index in lov_obd->tgts */
369 unsigned long ltd_active:1,/* is this target up for requests */
370 ltd_activate:1,/* should target be activated */
371 ltd_reap:1; /* should this target be deleted */
375 struct lov_desc desc;
376 struct lov_tgt_desc **lov_tgts; /* sparse array */
377 struct ost_pool lov_packed; /* all OSTs in a packed array */
378 struct mutex lov_lock;
379 struct obd_connect_data lov_ocd;
380 atomic_t lov_refcount;
381 __u32 lov_death_row;/* tgts scheduled to be deleted */
382 __u32 lov_tgt_size; /* size of tgts array */
385 struct cfs_hash *lov_pools_hash_body; /* used for key access */
386 struct list_head lov_pool_list; /* used for sequential access */
387 struct dentry *lov_pool_debugfs_entry;
388 enum lustre_sec_part lov_sp_me;
390 /* Cached LRU and unstable data from upper layer */
391 struct cl_client_cache *lov_cache;
393 struct rw_semaphore lov_notify_lock;
395 struct kobject *lov_tgts_kobj;
398 struct lmv_tgt_desc {
399 struct obd_uuid ltd_uuid;
400 struct obd_export *ltd_exp;
402 struct mutex ltd_fid_mutex;
403 unsigned long ltd_active:1; /* target up for requests */
408 struct lu_client_fld lmv_fld;
410 struct lmv_desc desc;
411 struct obd_uuid cluuid;
412 struct obd_export *exp;
414 struct mutex lmv_init_mutex;
419 u32 tgts_size; /* size of tgts array */
420 struct lmv_tgt_desc **tgts;
422 struct obd_connect_data conn_data;
423 struct kobject *lmv_tgts_kobj;
426 struct niobuf_local {
427 __u64 lnb_file_offset;
428 __u32 lnb_page_offset;
432 struct page *lnb_page;
436 #define LUSTRE_FLD_NAME "fld"
437 #define LUSTRE_SEQ_NAME "seq"
439 #define LUSTRE_MDD_NAME "mdd"
440 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
441 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
442 #define LUSTRE_VVP_NAME "vvp"
443 #define LUSTRE_LMV_NAME "lmv"
444 #define LUSTRE_SLP_NAME "slp"
445 #define LUSTRE_LOD_NAME "lod"
446 #define LUSTRE_OSP_NAME "osp"
447 #define LUSTRE_LWP_NAME "lwp"
449 /* obd device type names */
450 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
451 #define LUSTRE_MDS_NAME "mds"
452 #define LUSTRE_MDT_NAME "mdt"
453 #define LUSTRE_MDC_NAME "mdc"
454 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
455 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
456 #define LUSTRE_OSC_NAME "osc"
457 #define LUSTRE_LOV_NAME "lov"
458 #define LUSTRE_MGS_NAME "mgs"
459 #define LUSTRE_MGC_NAME "mgc"
461 #define LUSTRE_ECHO_NAME "obdecho"
462 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
463 #define LUSTRE_QMT_NAME "qmt"
465 /* Constant obd names (post-rename) */
466 #define LUSTRE_MDS_OBDNAME "MDS"
467 #define LUSTRE_OSS_OBDNAME "OSS"
468 #define LUSTRE_MGS_OBDNAME "MGS"
469 #define LUSTRE_MGC_OBDNAME "MGC"
471 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
472 #define N_LOCAL_TEMP_PAGE 0x10000000
475 * Events signalled through obd_notify() upcall-chain.
477 enum obd_notify_event {
478 /* Device connect start */
480 /* Device activated */
482 /* Device deactivated */
484 /* Connect data for import were changed */
487 OBD_NOTIFY_SYNC_NONBLOCK,
489 /* Configuration event */
491 /* Administratively deactivate/activate event */
492 OBD_NOTIFY_DEACTIVATE,
497 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
498 * being main example).
500 struct obd_notify_upcall {
501 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
502 enum obd_notify_event ev, void *owner, void *data);
503 /* Opaque datum supplied by upper layer listener */
507 struct target_recovery_data {
508 svc_handler_t trd_recovery_handler;
509 pid_t trd_processing_task;
510 struct completion trd_starting;
511 struct completion trd_finishing;
514 struct obd_llog_group {
515 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
516 wait_queue_head_t olg_waitq;
518 struct mutex olg_cat_processing;
521 /* corresponds to one of the obd's */
522 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
524 struct lvfs_run_ctxt {
525 struct dt_device *dt;
529 struct obd_type *obd_type;
530 u32 obd_magic; /* OBD_DEVICE_MAGIC */
531 int obd_minor; /* device number: lctl dl */
532 struct lu_device *obd_lu_dev;
534 /* common and UUID name of this device */
535 struct obd_uuid obd_uuid;
536 char obd_name[MAX_OBD_NAME];
538 /* bitfield modification is protected by obd_dev_lock */
539 unsigned long obd_attached:1, /* finished attach */
540 obd_set_up:1, /* finished setup */
541 obd_version_recov:1, /* obd uses version checking */
542 obd_replayable:1,/* recovery is enabled; inform clients */
543 obd_no_transno:1, /* no committed-transno notification */
544 obd_no_recov:1, /* fail instead of retry messages */
545 obd_stopping:1, /* started cleanup */
546 obd_starting:1, /* started setup */
547 obd_force:1, /* cleanup with > 0 obd refcount */
548 obd_fail:1, /* cleanup with failover */
549 obd_no_conn:1, /* deny new connections */
550 obd_inactive:1, /* device active/inactive
551 * (for sysfs status only!!)
553 obd_no_ir:1, /* no imperative recovery. */
554 obd_process_conf:1; /* device is processing mgs config */
555 /* use separate field as it is set in interrupt to don't mess with
556 * protection of other bits using _bh lock
558 unsigned long obd_recovery_expired:1;
559 /* uuid-export hash body */
560 struct cfs_hash *obd_uuid_hash;
561 wait_queue_head_t obd_refcount_waitq;
562 struct list_head obd_exports;
563 struct list_head obd_unlinked_exports;
564 struct list_head obd_delayed_exports;
565 atomic_t obd_refcount;
567 spinlock_t obd_nid_lock;
568 struct ldlm_namespace *obd_namespace;
569 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
570 /* a spinlock is OK for what we do now, may need a semaphore later */
571 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
572 spinlock_t obd_osfs_lock;
573 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
575 u64 obd_last_committed;
576 struct mutex obd_dev_mutex;
577 struct lvfs_run_ctxt obd_lvfs_ctxt;
578 struct obd_llog_group obd_olg; /* default llog group */
579 struct obd_device *obd_observer;
580 struct rw_semaphore obd_observer_link_sem;
581 struct obd_notify_upcall obd_upcall;
582 struct obd_export *obd_self_export;
585 struct client_obd cli;
586 struct echo_client_obd echo_client;
591 /* Fields used by LProcFS */
592 struct lprocfs_stats *obd_stats;
593 unsigned int obd_cntr_base;
595 struct lprocfs_stats *md_stats;
596 unsigned int md_cntr_base;
598 struct dentry *obd_debugfs_entry;
599 struct dentry *obd_svc_debugfs_entry;
600 struct lprocfs_stats *obd_svc_stats;
601 atomic_t obd_evict_inprogress;
602 wait_queue_head_t obd_evict_inprogress_waitq;
603 struct list_head obd_evict_list; /* protected with pet_lock */
606 * Ldlm pool part. Save last calculated SLV and Limit.
608 rwlock_t obd_pool_lock;
612 int obd_conn_inprogress;
615 * A list of outstanding class_incref()'s against this obd. For
618 struct lu_ref obd_reference;
620 struct kobject obd_kobj; /* sysfs object */
621 struct completion obd_kobj_unregister;
624 /* get/set_info keys */
625 #define KEY_ASYNC "async"
626 #define KEY_CHANGELOG_CLEAR "changelog_clear"
627 #define KEY_FID2PATH "fid2path"
628 #define KEY_CHECKSUM "checksum"
629 #define KEY_CLEAR_FS "clear_fs"
630 #define KEY_CONN_DATA "conn_data"
631 #define KEY_EVICT_BY_NID "evict_by_nid"
632 #define KEY_FIEMAP "fiemap"
633 #define KEY_FLUSH_CTX "flush_ctx"
634 #define KEY_GRANT_SHRINK "grant_shrink"
635 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
636 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
637 #define KEY_INTERMDS "inter_mds"
638 #define KEY_LAST_ID "last_id"
639 #define KEY_LAST_FID "last_fid"
640 #define KEY_MAX_EASIZE "max_easize"
641 #define KEY_DEFAULT_EASIZE "default_easize"
642 #define KEY_MGSSEC "mgssec"
643 #define KEY_READ_ONLY "read-only"
644 #define KEY_REGISTER_TARGET "register_target"
645 #define KEY_SET_FS "set_fs"
646 #define KEY_TGT_COUNT "tgt_count"
647 /* KEY_SET_INFO in lustre_idl.h */
648 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
650 #define KEY_CACHE_SET "cache_set"
651 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
655 static inline int it_to_lock_mode(struct lookup_intent *it)
657 /* CREAT needs to be tested before open (both could be set) */
658 if (it->it_op & IT_CREAT)
660 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
663 else if (it->it_op & IT_READDIR)
665 else if (it->it_op & IT_GETXATTR)
667 else if (it->it_op & IT_SETXATTR)
670 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
675 MF_MDC_CANCEL_FID1 = BIT(0),
676 MF_MDC_CANCEL_FID2 = BIT(1),
677 MF_MDC_CANCEL_FID3 = BIT(2),
678 MF_MDC_CANCEL_FID4 = BIT(3),
679 MF_GET_MDT_IDX = BIT(4),
683 CLI_SET_MEA = BIT(0),
684 CLI_RM_ENTRY = BIT(1),
687 CLI_MIGRATE = BIT(4),
691 struct lu_fid op_fid1; /* operation fid1 (usually parent) */
692 struct lu_fid op_fid2; /* operation fid2 (usually child) */
693 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
694 struct lu_fid op_fid4; /* to the operation locks. */
695 u32 op_mds; /* what mds server open will go to */
696 struct lustre_handle op_handle;
701 struct lmv_stripe_md *op_mea1;
702 struct lmv_stripe_md *op_mea2;
703 __u32 op_suppgids[2];
710 /* iattr fields and blocks. */
711 struct iattr op_attr;
712 unsigned int op_attr_flags;
714 loff_t op_attr_blocks;
718 /* Various operation flags. */
719 enum mds_op_bias op_bias;
721 /* Used by readdir */
724 /* Used by readdir */
727 /* used to transfer info between the stacks of MD client
728 * see enum op_cli_flags
730 enum md_cli_flags op_cli_flags;
732 /* File object data version for HSM release, on client */
733 __u64 op_data_version;
734 struct lustre_handle op_lease_handle;
736 /* default stripe offset */
737 __u32 op_default_stripe_offset;
741 int (*md_blocking_ast)(struct ldlm_lock *lock,
742 struct ldlm_lock_desc *desc,
743 void *data, int flag);
746 struct md_enqueue_info;
747 /* metadata stat-ahead */
749 struct md_enqueue_info {
750 struct md_op_data mi_data;
751 struct lookup_intent mi_it;
752 struct lustre_handle mi_lockh;
753 struct inode *mi_dir;
754 struct ldlm_enqueue_info mi_einfo;
755 int (*mi_cb)(struct ptlrpc_request *req,
756 struct md_enqueue_info *minfo, int rc);
761 struct module *owner;
762 int (*iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
763 void *karg, void __user *uarg);
764 int (*get_info)(const struct lu_env *env, struct obd_export *,
765 __u32 keylen, void *key, __u32 *vallen, void *val);
766 int (*set_info_async)(const struct lu_env *, struct obd_export *,
767 __u32 keylen, void *key,
768 __u32 vallen, void *val,
769 struct ptlrpc_request_set *set);
770 int (*setup)(struct obd_device *dev, struct lustre_cfg *cfg);
771 int (*precleanup)(struct obd_device *dev);
772 int (*cleanup)(struct obd_device *dev);
773 int (*process_config)(struct obd_device *dev, u32 len, void *data);
774 int (*postrecov)(struct obd_device *dev);
775 int (*add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
777 int (*del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
778 /* connect to the target device with given connection
779 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
780 * granted by the target, which are guaranteed to be a subset of flags
781 * asked for. If @ocd == NULL, use default parameters.
783 int (*connect)(const struct lu_env *env,
784 struct obd_export **exp, struct obd_device *src,
785 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
787 int (*reconnect)(const struct lu_env *env,
788 struct obd_export *exp, struct obd_device *src,
789 struct obd_uuid *cluuid,
790 struct obd_connect_data *ocd,
792 int (*disconnect)(struct obd_export *exp);
794 /* Initialize/finalize fids infrastructure. */
795 int (*fid_init)(struct obd_device *obd,
796 struct obd_export *exp, enum lu_cli_type type);
797 int (*fid_fini)(struct obd_device *obd);
799 /* Allocate new fid according to passed @hint. */
800 int (*fid_alloc)(const struct lu_env *env, struct obd_export *exp,
801 struct lu_fid *fid, struct md_op_data *op_data);
804 * Object with @fid is getting deleted, we may want to do something
807 int (*statfs)(const struct lu_env *, struct obd_export *exp,
808 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
809 int (*statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
810 __u64 max_age, struct ptlrpc_request_set *set);
811 int (*create)(const struct lu_env *env, struct obd_export *exp,
813 int (*destroy)(const struct lu_env *env, struct obd_export *exp,
815 int (*setattr)(const struct lu_env *, struct obd_export *exp,
817 int (*getattr)(const struct lu_env *env, struct obd_export *exp,
819 int (*preprw)(const struct lu_env *env, int cmd,
820 struct obd_export *exp, struct obdo *oa, int objcount,
821 struct obd_ioobj *obj, struct niobuf_remote *remote,
822 int *nr_pages, struct niobuf_local *local);
823 int (*commitrw)(const struct lu_env *env, int cmd,
824 struct obd_export *exp, struct obdo *oa,
825 int objcount, struct obd_ioobj *obj,
826 struct niobuf_remote *remote, int pages,
827 struct niobuf_local *local, int rc);
828 int (*init_export)(struct obd_export *exp);
829 int (*destroy_export)(struct obd_export *exp);
831 /* metadata-only methods */
832 int (*import_event)(struct obd_device *, struct obd_import *,
833 enum obd_import_event);
835 int (*notify)(struct obd_device *obd, struct obd_device *watched,
836 enum obd_notify_event ev, void *data);
838 int (*health_check)(const struct lu_env *env, struct obd_device *);
839 struct obd_uuid *(*get_uuid)(struct obd_export *exp);
842 int (*quotactl)(struct obd_device *, struct obd_export *,
843 struct obd_quotactl *);
846 int (*pool_new)(struct obd_device *obd, char *poolname);
847 int (*pool_del)(struct obd_device *obd, char *poolname);
848 int (*pool_add)(struct obd_device *obd, char *poolname,
850 int (*pool_rem)(struct obd_device *obd, char *poolname,
852 void (*getref)(struct obd_device *obd);
853 void (*putref)(struct obd_device *obd);
855 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
856 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
857 * Also, add a wrapper function in include/linux/obd_class.h.
863 struct mdt_body *body;
864 struct lu_buf layout;
865 struct lmv_stripe_md *lmv;
866 #ifdef CONFIG_FS_POSIX_ACL
867 struct posix_acl *posix_acl;
869 struct mdt_remote_perm *remote_perm;
872 struct md_open_data {
873 struct obd_client_handle *mod_och;
874 struct ptlrpc_request *mod_open_req;
875 struct ptlrpc_request *mod_close_req;
876 atomic_t mod_refcount;
880 struct obd_client_handle {
881 struct lustre_handle och_fh;
882 struct lu_fid och_fid;
883 struct md_open_data *och_mod;
884 struct lustre_handle och_lease_handle; /* open lock for lease */
889 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
891 struct lookup_intent;
895 int (*getstatus)(struct obd_export *, struct lu_fid *);
896 int (*null_inode)(struct obd_export *, const struct lu_fid *);
897 int (*close)(struct obd_export *, struct md_op_data *,
898 struct md_open_data *, struct ptlrpc_request **);
899 int (*create)(struct obd_export *, struct md_op_data *,
900 const void *, size_t, umode_t, uid_t, gid_t,
901 cfs_cap_t, __u64, struct ptlrpc_request **);
902 int (*enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
903 const union ldlm_policy_data *,
904 struct lookup_intent *, struct md_op_data *,
905 struct lustre_handle *, __u64);
906 int (*getattr)(struct obd_export *, struct md_op_data *,
907 struct ptlrpc_request **);
908 int (*getattr_name)(struct obd_export *, struct md_op_data *,
909 struct ptlrpc_request **);
910 int (*intent_lock)(struct obd_export *, struct md_op_data *,
911 struct lookup_intent *,
912 struct ptlrpc_request **,
913 ldlm_blocking_callback, __u64);
914 int (*link)(struct obd_export *, struct md_op_data *,
915 struct ptlrpc_request **);
916 int (*rename)(struct obd_export *, struct md_op_data *,
917 const char *, size_t, const char *, size_t,
918 struct ptlrpc_request **);
919 int (*setattr)(struct obd_export *, struct md_op_data *, void *,
920 size_t, struct ptlrpc_request **);
921 int (*sync)(struct obd_export *, const struct lu_fid *,
922 struct ptlrpc_request **);
923 int (*read_page)(struct obd_export *, struct md_op_data *,
924 struct md_callback *cb_op, __u64 hash_offset,
925 struct page **ppage);
926 int (*unlink)(struct obd_export *, struct md_op_data *,
927 struct ptlrpc_request **);
929 int (*setxattr)(struct obd_export *, const struct lu_fid *,
930 u64, const char *, const char *, int, int, int, __u32,
931 struct ptlrpc_request **);
933 int (*getxattr)(struct obd_export *, const struct lu_fid *,
934 u64, const char *, const char *, int, int, int,
935 struct ptlrpc_request **);
937 int (*init_ea_size)(struct obd_export *, u32, u32);
939 int (*get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
940 struct obd_export *, struct obd_export *,
943 int (*free_lustre_md)(struct obd_export *, struct lustre_md *);
945 int (*merge_attr)(struct obd_export *,
946 const struct lmv_stripe_md *lsm,
947 struct cl_attr *attr, ldlm_blocking_callback);
949 int (*set_open_replay_data)(struct obd_export *,
950 struct obd_client_handle *,
951 struct lookup_intent *);
952 int (*clear_open_replay_data)(struct obd_export *,
953 struct obd_client_handle *);
954 int (*set_lock_data)(struct obd_export *, const struct lustre_handle *,
957 enum ldlm_mode (*lock_match)(struct obd_export *, __u64,
958 const struct lu_fid *, enum ldlm_type,
959 union ldlm_policy_data *, enum ldlm_mode,
960 struct lustre_handle *);
962 int (*cancel_unused)(struct obd_export *, const struct lu_fid *,
963 union ldlm_policy_data *, enum ldlm_mode,
964 enum ldlm_cancel_flags flags, void *opaque);
966 int (*get_fid_from_lsm)(struct obd_export *,
967 const struct lmv_stripe_md *,
968 const char *name, int namelen,
971 int (*intent_getattr_async)(struct obd_export *,
972 struct md_enqueue_info *);
974 int (*revalidate_lock)(struct obd_export *, struct lookup_intent *,
975 struct lu_fid *, __u64 *bits);
977 int (*unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
978 const union lmv_mds_md *lmv, size_t lmv_size);
980 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
981 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
982 * wrapper function in include/linux/obd_class.h.
986 static inline struct md_open_data *obd_mod_alloc(void)
988 struct md_open_data *mod;
990 mod = kzalloc(sizeof(*mod), GFP_NOFS);
993 atomic_set(&mod->mod_refcount, 1);
997 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
998 #define obd_mod_put(mod) \
1000 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1001 if ((mod)->mod_open_req) \
1002 ptlrpc_req_finished((mod)->mod_open_req); \
1007 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1008 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1010 /* return 1 if client should be resend request */
1011 static inline int client_should_resend(int resend, struct client_obd *cli)
1013 return atomic_read(&cli->cl_resends) ?
1014 atomic_read(&cli->cl_resends) > resend : 1;
1018 * Return device name for this device
1020 * XXX: lu_device is declared before obd_device, while a pointer pointing
1021 * back to obd_device in lu_device, so this helper function defines here
1022 * instead of in lu_object.h
1024 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1026 return lu_dev->ld_obd->obd_name;
1029 static inline bool filename_is_volatile(const char *name, size_t namelen,
1035 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1038 /* caller does not care of idx */
1042 /* volatile file, the MDT can be set from name */
1043 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1044 /* if no MDT is specified, use std way */
1045 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1047 /* test for no MDT idx case */
1048 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1049 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1053 /* we have an idx, read it */
1054 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1055 *idx = simple_strtoul(start, &end, 0);
1057 * no digit, no trailing :, negative value
1059 if (((*idx == 0) && (end == start)) ||
1060 (*end != ':') || (*idx < 0))
1065 /* bad format of mdt idx, we cannot return an error
1066 * to caller so we use hash algo
1068 CERROR("Bad volatile file name format: %s\n",
1069 name + LUSTRE_VOLATILE_HDR_LEN);
1073 static inline int cli_brw_size(struct obd_device *obd)
1075 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1079 * when RPC size or the max RPCs in flight is increased, the max dirty pages
1080 * of the client should be increased accordingly to avoid sending fragmented
1081 * RPCs over the network when the client runs out of the maximum dirty space
1082 * when so many RPCs are being generated.
1084 static inline void client_adjust_max_dirty(struct client_obd *cli)
1087 if (cli->cl_dirty_max_pages <= 0)
1088 cli->cl_dirty_max_pages =
1089 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1091 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1092 cli->cl_max_pages_per_rpc;
1094 if (dirty_max > cli->cl_dirty_max_pages)
1095 cli->cl_dirty_max_pages = dirty_max;
1098 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1099 cli->cl_dirty_max_pages = totalram_pages / 8;
1102 #endif /* __OBD_H */