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)
79 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
81 /* obd info for a particular level (lov, osc). */
83 /* OBD_STATFS_* flags */
85 /* lsm data specific for every OSC. */
86 struct lov_stripe_md *oi_md;
87 /* statfs data specific for every OSC, if needed at all. */
88 struct obd_statfs *oi_osfs;
89 /* An update callback which is called to update some data on upper
90 * level. E.g. it is used for update lsm->lsm_oinfo at every received
91 * request in osc level for enqueue requests. It is also possible to
92 * update some caller data from LOV layer if needed.
94 obd_enqueue_update_f oi_cb_up;
98 struct list_head typ_chain;
99 struct obd_ops *typ_dt_ops;
100 struct md_ops *typ_md_ops;
101 struct dentry *typ_debugfs_entry;
104 struct lu_device_type *typ_lu;
105 spinlock_t obd_type_lock;
106 struct kobject *typ_kobj;
116 struct timeout_item {
117 enum timeout_event ti_event;
118 unsigned long ti_timeout;
121 struct list_head ti_obd_list;
122 struct list_head ti_chain;
125 #define OBD_MAX_RIF_DEFAULT 8
126 #define OBD_MAX_RIF_MAX 512
127 #define OSC_MAX_RIF_MAX 256
128 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
129 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
130 #define OSC_DEFAULT_RESENDS 10
132 /* possible values for fo_sync_lock_cancel */
134 NEVER_SYNC_ON_CANCEL = 0,
135 BLOCKING_SYNC_ON_CANCEL = 1,
136 ALWAYS_SYNC_ON_CANCEL = 2,
137 NUM_SYNC_ON_CANCEL_STATES
140 enum obd_cl_sem_lock_class {
147 * Limit reply buffer size for striping data to one x86_64 page. This
148 * value is chosen to fit the striping data for common use cases while
149 * staying well below the limit at which the buffer must be backed by
150 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
153 #define OBD_MAX_DEFAULT_EA_SIZE 4096
158 struct rw_semaphore cl_sem;
159 struct obd_uuid cl_target_uuid;
160 struct obd_import *cl_import; /* ptlrpc connection state */
161 size_t cl_conn_count;
163 * Cache maximum and default values for easize. This is
164 * strictly a performance optimization to minimize calls to
165 * obd_size_diskmd(). The default values are used to calculate the
166 * initial size of a request buffer. The ptlrpc layer will resize the
167 * buffer as needed to accommodate a larger reply from the
168 * server. The default values should be small enough to avoid wasted
169 * memory and excessive use of vmalloc(), yet large enough to avoid
170 * reallocating the buffer in the common use case.
173 * Default EA size for striping attributes. It is initialized at
174 * mount-time based on the default stripe width of the filesystem,
175 * then it tracks the largest observed EA size advertised by
176 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE.
178 u32 cl_default_mds_easize;
179 /* Maximum possible EA size computed at mount-time based on
180 * the number of OSTs in the filesystem. May be increased at
181 * run-time if a larger observed size is advertised by the MDT.
183 u32 cl_max_mds_easize;
185 enum lustre_sec_part cl_sp_me;
186 enum lustre_sec_part cl_sp_to;
187 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
189 /* the grant values are protected by loi_list_lock below */
190 unsigned long cl_dirty_pages; /* all _dirty_ in pahges */
191 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
192 unsigned long cl_dirty_transit; /* dirty synchronous */
193 unsigned long cl_avail_grant; /* bytes of credit for ost */
194 unsigned long cl_lost_grant; /* lost credits (trunc) */
196 /* since we allocate grant by blocks, we don't know how many grant will
197 * be used to add a page into cache. As a solution, we reserve maximum
198 * grant before trying to dirty a page and unreserve the rest.
199 * See osc_{reserve|unreserve}_grant for details.
201 long cl_reserved_grant;
202 struct list_head cl_cache_waiters; /* waiting for cache/grant */
203 unsigned long 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_SIZE, OST block size)
211 unsigned int cl_extent_tax; /* extent overhead, by bytes */
213 /* keep track of objects that have lois that contain pages which
214 * have been queued for async brw. this lock also protects the
215 * lists of osc_client_pages that hang off of the loi
218 * ->cl_loi_list_lock protects consistency of
219 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
220 * ->ap_completion() call-backs are executed under this lock. As we
221 * cannot guarantee that these call-backs never block on all platforms
222 * (as a matter of fact they do block on Mac OS X), type of
223 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
224 * and blocking mutex on Mac OS X. (Alternative is to make this lock
225 * blocking everywhere, but we don't want to slow down fast-path of
226 * our main platform.)
228 * NB by Jinshan: though field names are still _loi_, but actually
229 * osc_object{}s are in the list.
231 spinlock_t cl_loi_list_lock;
232 struct list_head cl_loi_ready_list;
233 struct list_head cl_loi_hp_ready_list;
234 struct list_head cl_loi_write_list;
235 struct list_head cl_loi_read_list;
236 __u32 cl_r_in_flight;
237 __u32 cl_w_in_flight;
238 /* just a sum of the loi/lop pending numbers to be exported by sysfs */
239 atomic_t cl_pending_w_pages;
240 atomic_t cl_pending_r_pages;
241 __u32 cl_max_pages_per_rpc;
242 __u32 cl_max_rpcs_in_flight;
243 struct obd_histogram cl_read_rpc_hist;
244 struct obd_histogram cl_write_rpc_hist;
245 struct obd_histogram cl_read_page_hist;
246 struct obd_histogram cl_write_page_hist;
247 struct obd_histogram cl_read_offset_hist;
248 struct obd_histogram cl_write_offset_hist;
250 /* LRU for osc caching pages */
251 struct cl_client_cache *cl_cache;
252 /** member of cl_cache->ccc_lru */
253 struct list_head cl_lru_osc;
254 /** # of available LRU slots left in the per-OSC cache.
255 * Available LRU slots are shared by all OSCs of the same file system,
256 * therefore this is a pointer to cl_client_cache::ccc_lru_left.
258 atomic_long_t *cl_lru_left;
259 /** # of busy LRU pages. A page is considered busy if it's in writeback
260 * queue, or in transfer. Busy pages can't be discarded so they are not
263 atomic_long_t cl_lru_busy;
264 /** # of LRU pages in the cache for this client_obd */
265 atomic_long_t cl_lru_in_list;
266 /** # of threads are shrinking LRU cache. To avoid contention, it's not
267 * allowed to have multiple threads shrinking LRU cache.
269 atomic_t cl_lru_shrinkers;
270 /** The time when this LRU cache was last used. */
271 time64_t cl_lru_last_used;
272 /** stats: how many reclaims have happened for this client_obd.
273 * reclaim and shrink - shrink is async, voluntarily rebalancing;
274 * reclaim is sync, initiated by IO thread when the LRU slots are
278 /** List of LRU pages for this client_obd */
279 struct list_head cl_lru_list;
280 /** Lock for LRU page list */
281 spinlock_t cl_lru_list_lock;
282 /** # of unstable pages in this client_obd.
283 * An unstable page is a page state that WRITE RPC has finished but
284 * the transaction has NOT yet committed.
286 atomic_long_t cl_unstable_count;
288 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
289 atomic_t cl_destroy_in_flight;
290 wait_queue_head_t cl_destroy_waitq;
292 struct mdc_rpc_lock *cl_rpc_lock;
294 /* modify rpcs in flight
295 * currently used for metadata only
297 spinlock_t cl_mod_rpcs_lock;
298 u16 cl_max_mod_rpcs_in_flight;
299 u16 cl_mod_rpcs_in_flight;
300 u16 cl_close_rpcs_in_flight;
301 wait_queue_head_t cl_mod_rpcs_waitq;
302 unsigned long *cl_mod_tag_bitmap;
303 struct obd_histogram cl_mod_rpcs_hist;
306 atomic_t cl_mgc_refcount;
307 struct obd_export *cl_mgc_mgsexp;
309 /* checksumming for data sent over the network */
310 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
311 /* supported checksum types that are worked out at connect time */
312 __u32 cl_supp_cksum_types;
313 /* checksum algorithm to be used */
314 enum cksum_type cl_cksum_type;
316 /* also protected by the poorly named _loi_list_lock lock above */
317 struct osc_async_rc cl_ar;
319 /* sequence manager */
320 struct lu_client_seq *cl_seq;
322 atomic_t cl_resends; /* resend count */
324 /* ptlrpc work for writeback in ptlrpcd context */
325 void *cl_writeback_work;
327 /* hash tables for osc_quota_info */
328 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
331 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
338 struct echo_client_obd {
339 struct obd_export *ec_exp; /* the local connection to osc/lov */
341 struct list_head ec_objects;
342 struct list_head ec_locks;
346 /* Generic subset of OSTs */
348 __u32 *op_array; /* array of index of lov_obd->lov_tgts */
349 unsigned int op_count; /* number of OSTs in the array */
350 unsigned int op_size; /* allocated size of lp_array */
351 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
354 /* allow statfs data caching for 1 second */
355 #define OBD_STATFS_CACHE_SECONDS 1
357 struct lov_tgt_desc {
358 struct list_head ltd_kill;
359 struct obd_uuid ltd_uuid;
360 struct obd_device *ltd_obd;
361 struct obd_export *ltd_exp;
363 __u32 ltd_index; /* index in lov_obd->tgts */
364 unsigned long ltd_active:1,/* is this target up for requests */
365 ltd_activate:1,/* should target be activated */
366 ltd_reap:1; /* should this target be deleted */
370 struct lov_desc desc;
371 struct lov_tgt_desc **lov_tgts; /* sparse array */
372 struct ost_pool lov_packed; /* all OSTs in a packed array */
373 struct mutex lov_lock;
374 struct obd_connect_data lov_ocd;
375 atomic_t lov_refcount;
376 __u32 lov_death_row;/* tgts scheduled to be deleted */
377 __u32 lov_tgt_size; /* size of tgts array */
380 struct cfs_hash *lov_pools_hash_body; /* used for key access */
381 struct list_head lov_pool_list; /* used for sequential access */
382 struct dentry *lov_pool_debugfs_entry;
383 enum lustre_sec_part lov_sp_me;
385 /* Cached LRU and unstable data from upper layer */
386 struct cl_client_cache *lov_cache;
388 struct rw_semaphore lov_notify_lock;
390 struct kobject *lov_tgts_kobj;
393 struct lmv_tgt_desc {
394 struct obd_uuid ltd_uuid;
395 struct obd_export *ltd_exp;
397 struct mutex ltd_fid_mutex;
398 unsigned long ltd_active:1; /* target up for requests */
401 enum placement_policy {
402 PLACEMENT_CHAR_POLICY = 0,
403 PLACEMENT_NID_POLICY = 1,
404 PLACEMENT_INVAL_POLICY = 2,
410 struct lu_client_fld lmv_fld;
412 enum placement_policy lmv_placement;
413 struct lmv_desc desc;
414 struct obd_uuid cluuid;
415 struct obd_export *exp;
417 struct mutex lmv_init_mutex;
422 u32 tgts_size; /* size of tgts array */
423 struct lmv_tgt_desc **tgts;
425 struct obd_connect_data conn_data;
426 struct kobject *lmv_tgts_kobj;
429 struct niobuf_local {
430 __u64 lnb_file_offset;
431 __u32 lnb_page_offset;
435 struct page *lnb_page;
439 #define LUSTRE_FLD_NAME "fld"
440 #define LUSTRE_SEQ_NAME "seq"
442 #define LUSTRE_MDD_NAME "mdd"
443 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
444 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
445 #define LUSTRE_VVP_NAME "vvp"
446 #define LUSTRE_LMV_NAME "lmv"
447 #define LUSTRE_SLP_NAME "slp"
448 #define LUSTRE_LOD_NAME "lod"
449 #define LUSTRE_OSP_NAME "osp"
450 #define LUSTRE_LWP_NAME "lwp"
452 /* obd device type names */
453 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
454 #define LUSTRE_MDS_NAME "mds"
455 #define LUSTRE_MDT_NAME "mdt"
456 #define LUSTRE_MDC_NAME "mdc"
457 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
458 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
459 #define LUSTRE_OSC_NAME "osc"
460 #define LUSTRE_LOV_NAME "lov"
461 #define LUSTRE_MGS_NAME "mgs"
462 #define LUSTRE_MGC_NAME "mgc"
464 #define LUSTRE_ECHO_NAME "obdecho"
465 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
466 #define LUSTRE_QMT_NAME "qmt"
468 /* Constant obd names (post-rename) */
469 #define LUSTRE_MDS_OBDNAME "MDS"
470 #define LUSTRE_OSS_OBDNAME "OSS"
471 #define LUSTRE_MGS_OBDNAME "MGS"
472 #define LUSTRE_MGC_OBDNAME "MGC"
474 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
475 #define N_LOCAL_TEMP_PAGE 0x10000000
478 * Events signalled through obd_notify() upcall-chain.
480 enum obd_notify_event {
483 /* Device connect start */
485 /* Device activated */
487 /* Device deactivated */
489 /* Device disconnected */
491 /* Connect data for import were changed */
494 OBD_NOTIFY_SYNC_NONBLOCK,
496 /* Configuration event */
498 /* Administratively deactivate/activate event */
499 OBD_NOTIFY_DEACTIVATE,
504 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
505 * being main example).
507 struct obd_notify_upcall {
508 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
509 enum obd_notify_event ev, void *owner, void *data);
510 /* Opaque datum supplied by upper layer listener */
514 struct target_recovery_data {
515 svc_handler_t trd_recovery_handler;
516 pid_t trd_processing_task;
517 struct completion trd_starting;
518 struct completion trd_finishing;
521 struct obd_llog_group {
522 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
523 wait_queue_head_t olg_waitq;
525 struct mutex olg_cat_processing;
528 /* corresponds to one of the obd's */
529 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
531 struct lvfs_run_ctxt {
532 struct dt_device *dt;
536 struct obd_type *obd_type;
537 u32 obd_magic; /* OBD_DEVICE_MAGIC */
538 int obd_minor; /* device number: lctl dl */
539 struct lu_device *obd_lu_dev;
541 /* common and UUID name of this device */
542 struct obd_uuid obd_uuid;
543 char obd_name[MAX_OBD_NAME];
545 /* bitfield modification is protected by obd_dev_lock */
546 unsigned long obd_attached:1, /* finished attach */
547 obd_set_up:1, /* finished setup */
548 obd_version_recov:1, /* obd uses version checking */
549 obd_replayable:1,/* recovery is enabled; inform clients */
550 obd_no_transno:1, /* no committed-transno notification */
551 obd_no_recov:1, /* fail instead of retry messages */
552 obd_stopping:1, /* started cleanup */
553 obd_starting:1, /* started setup */
554 obd_force:1, /* cleanup with > 0 obd refcount */
555 obd_fail:1, /* cleanup with failover */
556 obd_no_conn:1, /* deny new connections */
557 obd_inactive:1, /* device active/inactive
558 * (for sysfs status only!!)
560 obd_no_ir:1, /* no imperative recovery. */
561 obd_process_conf:1; /* device is processing mgs config */
562 /* use separate field as it is set in interrupt to don't mess with
563 * protection of other bits using _bh lock
565 unsigned long obd_recovery_expired:1;
566 /* uuid-export hash body */
567 struct cfs_hash *obd_uuid_hash;
568 wait_queue_head_t obd_refcount_waitq;
569 struct list_head obd_exports;
570 struct list_head obd_unlinked_exports;
571 struct list_head obd_delayed_exports;
572 atomic_t obd_refcount;
574 spinlock_t obd_nid_lock;
575 struct ldlm_namespace *obd_namespace;
576 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
577 /* a spinlock is OK for what we do now, may need a semaphore later */
578 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
579 spinlock_t obd_osfs_lock;
580 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
582 u64 obd_last_committed;
583 struct mutex obd_dev_mutex;
584 struct lvfs_run_ctxt obd_lvfs_ctxt;
585 struct obd_llog_group obd_olg; /* default llog group */
586 struct obd_device *obd_observer;
587 struct rw_semaphore obd_observer_link_sem;
588 struct obd_notify_upcall obd_upcall;
589 struct obd_export *obd_self_export;
592 struct client_obd cli;
593 struct echo_client_obd echo_client;
598 /* Fields used by LProcFS */
599 struct lprocfs_stats *obd_stats;
600 unsigned int obd_cntr_base;
602 struct lprocfs_stats *md_stats;
603 unsigned int md_cntr_base;
605 struct dentry *obd_debugfs_entry;
606 struct dentry *obd_svc_debugfs_entry;
607 struct lprocfs_stats *obd_svc_stats;
608 atomic_t obd_evict_inprogress;
609 wait_queue_head_t obd_evict_inprogress_waitq;
610 struct list_head obd_evict_list; /* protected with pet_lock */
613 * Ldlm pool part. Save last calculated SLV and Limit.
615 rwlock_t obd_pool_lock;
619 int obd_conn_inprogress;
622 * A list of outstanding class_incref()'s against this obd. For
625 struct lu_ref obd_reference;
627 struct kobject obd_kobj; /* sysfs object */
628 struct completion obd_kobj_unregister;
631 /* get/set_info keys */
632 #define KEY_ASYNC "async"
633 #define KEY_CHANGELOG_CLEAR "changelog_clear"
634 #define KEY_FID2PATH "fid2path"
635 #define KEY_CHECKSUM "checksum"
636 #define KEY_CLEAR_FS "clear_fs"
637 #define KEY_CONN_DATA "conn_data"
638 #define KEY_EVICT_BY_NID "evict_by_nid"
639 #define KEY_FIEMAP "fiemap"
640 #define KEY_FLUSH_CTX "flush_ctx"
641 #define KEY_GRANT_SHRINK "grant_shrink"
642 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
643 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
644 #define KEY_INTERMDS "inter_mds"
645 #define KEY_LAST_ID "last_id"
646 #define KEY_LAST_FID "last_fid"
647 #define KEY_MAX_EASIZE "max_easize"
648 #define KEY_DEFAULT_EASIZE "default_easize"
649 #define KEY_MGSSEC "mgssec"
650 #define KEY_READ_ONLY "read-only"
651 #define KEY_REGISTER_TARGET "register_target"
652 #define KEY_SET_FS "set_fs"
653 #define KEY_TGT_COUNT "tgt_count"
654 /* KEY_SET_INFO in lustre_idl.h */
655 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
657 #define KEY_CACHE_SET "cache_set"
658 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
662 static inline int it_to_lock_mode(struct lookup_intent *it)
664 /* CREAT needs to be tested before open (both could be set) */
665 if (it->it_op & IT_CREAT)
667 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
670 else if (it->it_op & IT_READDIR)
672 else if (it->it_op & IT_GETXATTR)
674 else if (it->it_op & IT_SETXATTR)
677 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
682 MF_MDC_CANCEL_FID1 = BIT(0),
683 MF_MDC_CANCEL_FID2 = BIT(1),
684 MF_MDC_CANCEL_FID3 = BIT(2),
685 MF_MDC_CANCEL_FID4 = BIT(3),
686 MF_GET_MDT_IDX = BIT(4),
690 CLI_SET_MEA = BIT(0),
691 CLI_RM_ENTRY = BIT(1),
694 CLI_MIGRATE = BIT(4),
698 struct lu_fid op_fid1; /* operation fid1 (usually parent) */
699 struct lu_fid op_fid2; /* operation fid2 (usually child) */
700 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
701 struct lu_fid op_fid4; /* to the operation locks. */
702 u32 op_mds; /* what mds server open will go to */
703 struct lustre_handle op_handle;
708 struct lmv_stripe_md *op_mea1;
709 struct lmv_stripe_md *op_mea2;
710 __u32 op_suppgids[2];
717 /* iattr fields and blocks. */
718 struct iattr op_attr;
719 unsigned int op_attr_flags;
721 loff_t op_attr_blocks;
725 /* Various operation flags. */
726 enum mds_op_bias op_bias;
728 /* Used by readdir */
731 /* Used by readdir */
734 /* used to transfer info between the stacks of MD client
735 * see enum op_cli_flags
737 enum md_cli_flags op_cli_flags;
739 /* File object data version for HSM release, on client */
740 __u64 op_data_version;
741 struct lustre_handle op_lease_handle;
743 /* default stripe offset */
744 __u32 op_default_stripe_offset;
748 int (*md_blocking_ast)(struct ldlm_lock *lock,
749 struct ldlm_lock_desc *desc,
750 void *data, int flag);
753 struct md_enqueue_info;
754 /* metadata stat-ahead */
756 struct md_enqueue_info {
757 struct md_op_data mi_data;
758 struct lookup_intent mi_it;
759 struct lustre_handle mi_lockh;
760 struct inode *mi_dir;
761 int (*mi_cb)(struct ptlrpc_request *req,
762 struct md_enqueue_info *minfo, int rc);
767 struct module *owner;
768 int (*iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
769 void *karg, void __user *uarg);
770 int (*get_info)(const struct lu_env *env, struct obd_export *,
771 __u32 keylen, void *key, __u32 *vallen, void *val);
772 int (*set_info_async)(const struct lu_env *, struct obd_export *,
773 __u32 keylen, void *key,
774 __u32 vallen, void *val,
775 struct ptlrpc_request_set *set);
776 int (*setup)(struct obd_device *dev, struct lustre_cfg *cfg);
777 int (*precleanup)(struct obd_device *dev);
778 int (*cleanup)(struct obd_device *dev);
779 int (*process_config)(struct obd_device *dev, u32 len, void *data);
780 int (*postrecov)(struct obd_device *dev);
781 int (*add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
783 int (*del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
784 /* connect to the target device with given connection
785 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
786 * granted by the target, which are guaranteed to be a subset of flags
787 * asked for. If @ocd == NULL, use default parameters.
789 int (*connect)(const struct lu_env *env,
790 struct obd_export **exp, struct obd_device *src,
791 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
793 int (*reconnect)(const struct lu_env *env,
794 struct obd_export *exp, struct obd_device *src,
795 struct obd_uuid *cluuid,
796 struct obd_connect_data *ocd,
798 int (*disconnect)(struct obd_export *exp);
800 /* Initialize/finalize fids infrastructure. */
801 int (*fid_init)(struct obd_device *obd,
802 struct obd_export *exp, enum lu_cli_type type);
803 int (*fid_fini)(struct obd_device *obd);
805 /* Allocate new fid according to passed @hint. */
806 int (*fid_alloc)(const struct lu_env *env, struct obd_export *exp,
807 struct lu_fid *fid, struct md_op_data *op_data);
810 * Object with @fid is getting deleted, we may want to do something
813 int (*statfs)(const struct lu_env *, struct obd_export *exp,
814 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
815 int (*statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
816 __u64 max_age, struct ptlrpc_request_set *set);
817 int (*create)(const struct lu_env *env, struct obd_export *exp,
819 int (*destroy)(const struct lu_env *env, struct obd_export *exp,
821 int (*setattr)(const struct lu_env *, struct obd_export *exp,
823 int (*getattr)(const struct lu_env *env, struct obd_export *exp,
825 int (*preprw)(const struct lu_env *env, int cmd,
826 struct obd_export *exp, struct obdo *oa, int objcount,
827 struct obd_ioobj *obj, struct niobuf_remote *remote,
828 int *nr_pages, struct niobuf_local *local);
829 int (*commitrw)(const struct lu_env *env, int cmd,
830 struct obd_export *exp, struct obdo *oa,
831 int objcount, struct obd_ioobj *obj,
832 struct niobuf_remote *remote, int pages,
833 struct niobuf_local *local, int rc);
834 int (*init_export)(struct obd_export *exp);
835 int (*destroy_export)(struct obd_export *exp);
837 /* metadata-only methods */
838 int (*import_event)(struct obd_device *, struct obd_import *,
839 enum obd_import_event);
841 int (*notify)(struct obd_device *obd, struct obd_device *watched,
842 enum obd_notify_event ev, void *data);
844 int (*health_check)(const struct lu_env *env, struct obd_device *);
845 struct obd_uuid *(*get_uuid)(struct obd_export *exp);
848 int (*quotactl)(struct obd_device *, struct obd_export *,
849 struct obd_quotactl *);
852 int (*pool_new)(struct obd_device *obd, char *poolname);
853 int (*pool_del)(struct obd_device *obd, char *poolname);
854 int (*pool_add)(struct obd_device *obd, char *poolname,
856 int (*pool_rem)(struct obd_device *obd, char *poolname,
858 void (*getref)(struct obd_device *obd);
859 void (*putref)(struct obd_device *obd);
861 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
862 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
863 * Also, add a wrapper function in include/linux/obd_class.h.
869 struct mdt_body *body;
870 struct lu_buf layout;
871 struct lmv_stripe_md *lmv;
872 #ifdef CONFIG_FS_POSIX_ACL
873 struct posix_acl *posix_acl;
875 struct mdt_remote_perm *remote_perm;
878 struct md_open_data {
879 struct obd_client_handle *mod_och;
880 struct ptlrpc_request *mod_open_req;
881 struct ptlrpc_request *mod_close_req;
882 atomic_t mod_refcount;
886 struct obd_client_handle {
887 struct lustre_handle och_fh;
888 struct lu_fid och_fid;
889 struct md_open_data *och_mod;
890 struct lustre_handle och_lease_handle; /* open lock for lease */
895 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
897 struct lookup_intent;
901 int (*getstatus)(struct obd_export *, struct lu_fid *);
902 int (*null_inode)(struct obd_export *, const struct lu_fid *);
903 int (*close)(struct obd_export *, struct md_op_data *,
904 struct md_open_data *, struct ptlrpc_request **);
905 int (*create)(struct obd_export *, struct md_op_data *,
906 const void *, size_t, umode_t, uid_t, gid_t,
907 cfs_cap_t, __u64, struct ptlrpc_request **);
908 int (*enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
909 const union ldlm_policy_data *,
910 struct lookup_intent *, struct md_op_data *,
911 struct lustre_handle *, __u64);
912 int (*getattr)(struct obd_export *, struct md_op_data *,
913 struct ptlrpc_request **);
914 int (*getattr_name)(struct obd_export *, struct md_op_data *,
915 struct ptlrpc_request **);
916 int (*intent_lock)(struct obd_export *, struct md_op_data *,
917 struct lookup_intent *,
918 struct ptlrpc_request **,
919 ldlm_blocking_callback, __u64);
920 int (*link)(struct obd_export *, struct md_op_data *,
921 struct ptlrpc_request **);
922 int (*rename)(struct obd_export *, struct md_op_data *,
923 const char *, size_t, const char *, size_t,
924 struct ptlrpc_request **);
925 int (*setattr)(struct obd_export *, struct md_op_data *, void *,
926 size_t, struct ptlrpc_request **);
927 int (*sync)(struct obd_export *, const struct lu_fid *,
928 struct ptlrpc_request **);
929 int (*read_page)(struct obd_export *, struct md_op_data *,
930 struct md_callback *cb_op, __u64 hash_offset,
931 struct page **ppage);
932 int (*unlink)(struct obd_export *, struct md_op_data *,
933 struct ptlrpc_request **);
935 int (*setxattr)(struct obd_export *, const struct lu_fid *,
936 u64, const char *, const char *, int, int, int, __u32,
937 struct ptlrpc_request **);
939 int (*getxattr)(struct obd_export *, const struct lu_fid *,
940 u64, const char *, const char *, int, int, int,
941 struct ptlrpc_request **);
943 int (*init_ea_size)(struct obd_export *, u32, u32);
945 int (*get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
946 struct obd_export *, struct obd_export *,
949 int (*free_lustre_md)(struct obd_export *, struct lustre_md *);
951 int (*merge_attr)(struct obd_export *,
952 const struct lmv_stripe_md *lsm,
953 struct cl_attr *attr, ldlm_blocking_callback);
955 int (*set_open_replay_data)(struct obd_export *,
956 struct obd_client_handle *,
957 struct lookup_intent *);
958 int (*clear_open_replay_data)(struct obd_export *,
959 struct obd_client_handle *);
960 int (*set_lock_data)(struct obd_export *, const struct lustre_handle *,
963 enum ldlm_mode (*lock_match)(struct obd_export *, __u64,
964 const struct lu_fid *, enum ldlm_type,
965 union ldlm_policy_data *, enum ldlm_mode,
966 struct lustre_handle *);
968 int (*cancel_unused)(struct obd_export *, const struct lu_fid *,
969 union ldlm_policy_data *, enum ldlm_mode,
970 enum ldlm_cancel_flags flags, void *opaque);
972 int (*get_fid_from_lsm)(struct obd_export *,
973 const struct lmv_stripe_md *,
974 const char *name, int namelen,
977 int (*intent_getattr_async)(struct obd_export *,
978 struct md_enqueue_info *,
979 struct ldlm_enqueue_info *);
981 int (*revalidate_lock)(struct obd_export *, struct lookup_intent *,
982 struct lu_fid *, __u64 *bits);
984 int (*unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
985 const union lmv_mds_md *lmv, size_t lmv_size);
987 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
988 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
989 * wrapper function in include/linux/obd_class.h.
993 static inline struct md_open_data *obd_mod_alloc(void)
995 struct md_open_data *mod;
997 mod = kzalloc(sizeof(*mod), GFP_NOFS);
1000 atomic_set(&mod->mod_refcount, 1);
1004 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1005 #define obd_mod_put(mod) \
1007 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1008 if ((mod)->mod_open_req) \
1009 ptlrpc_req_finished((mod)->mod_open_req); \
1014 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1015 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1017 /* return 1 if client should be resend request */
1018 static inline int client_should_resend(int resend, struct client_obd *cli)
1020 return atomic_read(&cli->cl_resends) ?
1021 atomic_read(&cli->cl_resends) > resend : 1;
1025 * Return device name for this device
1027 * XXX: lu_device is declared before obd_device, while a pointer pointing
1028 * back to obd_device in lu_device, so this helper function defines here
1029 * instead of in lu_object.h
1031 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1033 return lu_dev->ld_obd->obd_name;
1036 static inline bool filename_is_volatile(const char *name, size_t namelen,
1042 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1045 /* caller does not care of idx */
1049 /* volatile file, the MDT can be set from name */
1050 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1051 /* if no MDT is specified, use std way */
1052 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1054 /* test for no MDT idx case */
1055 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1056 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1060 /* we have an idx, read it */
1061 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1062 *idx = simple_strtoul(start, &end, 0);
1064 * no digit, no trailing :, negative value
1066 if (((*idx == 0) && (end == start)) ||
1067 (*end != ':') || (*idx < 0))
1072 /* bad format of mdt idx, we cannot return an error
1073 * to caller so we use hash algo
1075 CERROR("Bad volatile file name format: %s\n",
1076 name + LUSTRE_VOLATILE_HDR_LEN);
1080 static inline int cli_brw_size(struct obd_device *obd)
1082 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1086 * when RPC size or the max RPCs in flight is increased, the max dirty pages
1087 * of the client should be increased accordingly to avoid sending fragmented
1088 * RPCs over the network when the client runs out of the maximum dirty space
1089 * when so many RPCs are being generated.
1091 static inline void client_adjust_max_dirty(struct client_obd *cli)
1094 if (cli->cl_dirty_max_pages <= 0)
1095 cli->cl_dirty_max_pages =
1096 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1098 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1099 cli->cl_max_pages_per_rpc;
1101 if (dirty_max > cli->cl_dirty_max_pages)
1102 cli->cl_dirty_max_pages = dirty_max;
1105 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1106 cli->cl_dirty_max_pages = totalram_pages / 8;
1109 #endif /* __OBD_H */