Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1 | /* |
| 2 | * GPL HEADER START |
| 3 | * |
| 4 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
| 5 | * |
| 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. |
| 9 | * |
| 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). |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * version 2 along with this program; If not, see |
Oleg Drokin | 6a5b99a | 2016-06-14 23:33:40 -0400 | [diff] [blame] | 18 | * http://www.gnu.org/licenses/gpl-2.0.html |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 19 | * |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 20 | * GPL HEADER END |
| 21 | */ |
| 22 | /* |
| 23 | * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. |
| 24 | * Use is subject to license terms. |
| 25 | * |
Andreas Dilger | 1dc563a | 2015-11-08 18:09:37 -0500 | [diff] [blame] | 26 | * Copyright (c) 2010, 2015, Intel Corporation. |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 27 | */ |
| 28 | /* |
| 29 | * This file is part of Lustre, http://www.lustre.org/ |
| 30 | * Lustre is a trademark of Sun Microsystems, Inc. |
| 31 | */ |
| 32 | |
| 33 | /** \defgroup LDLM Lustre Distributed Lock Manager |
| 34 | * |
| 35 | * Lustre DLM is based on VAX DLM. |
| 36 | * Its two main roles are: |
| 37 | * - To provide locking assuring consistency of data on all Lustre nodes. |
| 38 | * - To allow clients to cache state protected by a lock by holding the |
| 39 | * lock until a conflicting lock is requested or it is expired by the LRU. |
| 40 | * |
| 41 | * @{ |
| 42 | */ |
| 43 | |
| 44 | #ifndef _LUSTRE_DLM_H__ |
| 45 | #define _LUSTRE_DLM_H__ |
| 46 | |
Greg Kroah-Hartman | 1accaad | 2014-07-11 21:34:24 -0700 | [diff] [blame] | 47 | #include "lustre_lib.h" |
| 48 | #include "lustre_net.h" |
| 49 | #include "lustre_import.h" |
| 50 | #include "lustre_handles.h" |
| 51 | #include "interval_tree.h" /* for interval_node{}, ldlm_extent */ |
| 52 | #include "lu_ref.h" |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 53 | |
Bruce Korb | f2145ea | 2013-07-23 00:06:37 +0800 | [diff] [blame] | 54 | #include "lustre_dlm_flags.h" |
| 55 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 56 | struct obd_ops; |
| 57 | struct obd_device; |
| 58 | |
| 59 | #define OBD_LDLM_DEVICENAME "ldlm" |
| 60 | |
| 61 | #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus()) |
| 62 | #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000)) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 63 | #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024 |
| 64 | |
| 65 | /** |
| 66 | * LDLM non-error return states |
| 67 | */ |
Oleg Drokin | d1777aa | 2016-02-24 22:00:02 -0500 | [diff] [blame] | 68 | enum ldlm_error { |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 69 | ELDLM_OK = 0, |
Jinshan Xiong | 06563b5 | 2016-03-30 19:48:40 -0400 | [diff] [blame] | 70 | ELDLM_LOCK_MATCHED = 1, |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 71 | |
| 72 | ELDLM_LOCK_CHANGED = 300, |
| 73 | ELDLM_LOCK_ABORTED = 301, |
| 74 | ELDLM_LOCK_REPLACED = 302, |
| 75 | ELDLM_NO_LOCK_DATA = 303, |
| 76 | ELDLM_LOCK_WOULDBLOCK = 304, |
| 77 | |
| 78 | ELDLM_NAMESPACE_EXISTS = 400, |
| 79 | ELDLM_BAD_NAMESPACE = 401 |
Oleg Drokin | d1777aa | 2016-02-24 22:00:02 -0500 | [diff] [blame] | 80 | }; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 81 | |
| 82 | /** |
| 83 | * LDLM namespace type. |
| 84 | * The "client" type is actually an indication that this is a narrow local view |
| 85 | * into complete namespace on the server. Such namespaces cannot make any |
| 86 | * decisions about lack of conflicts or do any autonomous lock granting without |
| 87 | * first speaking to a server. |
| 88 | */ |
| 89 | typedef enum { |
| 90 | LDLM_NAMESPACE_SERVER = 1 << 0, |
| 91 | LDLM_NAMESPACE_CLIENT = 1 << 1 |
| 92 | } ldlm_side_t; |
| 93 | |
| 94 | /** |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 95 | * The blocking callback is overloaded to perform two functions. These flags |
| 96 | * indicate which operation should be performed. |
| 97 | */ |
| 98 | #define LDLM_CB_BLOCKING 1 |
| 99 | #define LDLM_CB_CANCELING 2 |
| 100 | |
| 101 | /** |
| 102 | * \name Lock Compatibility Matrix. |
| 103 | * |
| 104 | * A lock has both a type (extent, flock, inode bits, or plain) and a mode. |
| 105 | * Lock types are described in their respective implementation files: |
| 106 | * ldlm_{extent,flock,inodebits,plain}.c. |
| 107 | * |
| 108 | * There are six lock modes along with a compatibility matrix to indicate if |
| 109 | * two locks are compatible. |
| 110 | * |
| 111 | * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock |
| 112 | * on the parent. |
| 113 | * - PW: Protective Write (normal write) mode. When a client requests a write |
| 114 | * lock from an OST, a lock with PW mode will be issued. |
| 115 | * - PR: Protective Read (normal read) mode. When a client requests a read from |
| 116 | * an OST, a lock with PR mode will be issued. Also, if the client opens a |
| 117 | * file for execution, it is granted a lock with PR mode. |
| 118 | * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client |
| 119 | * requests a write lock during a file open operation. |
| 120 | * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants |
| 121 | * an inodebit lock with the CR mode on the intermediate path component. |
| 122 | * - NL Null mode. |
| 123 | * |
| 124 | * <PRE> |
| 125 | * NL CR CW PR PW EX |
| 126 | * NL 1 1 1 1 1 1 |
| 127 | * CR 1 1 1 1 1 0 |
| 128 | * CW 1 1 1 0 0 0 |
| 129 | * PR 1 1 0 1 0 0 |
| 130 | * PW 1 1 0 0 0 0 |
| 131 | * EX 1 0 0 0 0 0 |
| 132 | * </PRE> |
| 133 | */ |
| 134 | /** @{ */ |
| 135 | #define LCK_COMPAT_EX LCK_NL |
| 136 | #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR) |
| 137 | #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR) |
| 138 | #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW) |
| 139 | #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW) |
| 140 | #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP) |
| 141 | #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL) |
| 142 | #define LCK_COMPAT_COS (LCK_COS) |
| 143 | /** @} Lock Compatibility Matrix */ |
| 144 | |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 145 | extern enum ldlm_mode lck_compat_array[]; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 146 | |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 147 | static inline void lockmode_verify(enum ldlm_mode mode) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 148 | { |
Oleg Drokin | 22ea97f | 2016-02-26 01:50:10 -0500 | [diff] [blame] | 149 | LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 150 | } |
| 151 | |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 152 | static inline int lockmode_compat(enum ldlm_mode exist_mode, |
| 153 | enum ldlm_mode new_mode) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 154 | { |
Oleg Drokin | 22ea97f | 2016-02-26 01:50:10 -0500 | [diff] [blame] | 155 | return (lck_compat_array[exist_mode] & new_mode); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 156 | } |
| 157 | |
| 158 | /* |
| 159 | * |
| 160 | * cluster name spaces |
| 161 | * |
| 162 | */ |
| 163 | |
| 164 | #define DLM_OST_NAMESPACE 1 |
| 165 | #define DLM_MDS_NAMESPACE 2 |
| 166 | |
| 167 | /* XXX |
| 168 | - do we just separate this by security domains and use a prefix for |
| 169 | multiple namespaces in the same domain? |
| 170 | - |
| 171 | */ |
| 172 | |
| 173 | /** |
| 174 | * Locking rules for LDLM: |
| 175 | * |
| 176 | * lr_lock |
| 177 | * |
| 178 | * lr_lock |
| 179 | * waiting_locks_spinlock |
| 180 | * |
| 181 | * lr_lock |
| 182 | * led_lock |
| 183 | * |
| 184 | * lr_lock |
| 185 | * ns_lock |
| 186 | * |
| 187 | * lr_lvb_mutex |
| 188 | * lr_lock |
| 189 | * |
| 190 | */ |
| 191 | |
| 192 | struct ldlm_pool; |
| 193 | struct ldlm_lock; |
| 194 | struct ldlm_resource; |
| 195 | struct ldlm_namespace; |
| 196 | |
| 197 | /** |
| 198 | * Operations on LDLM pools. |
| 199 | * LDLM pool is a pool of locks in the namespace without any implicitly |
| 200 | * specified limits. |
| 201 | * Locks in the pool are organized in LRU. |
| 202 | * Local memory pressure or server instructions (e.g. mempressure on server) |
| 203 | * can trigger freeing of locks from the pool |
| 204 | */ |
| 205 | struct ldlm_pool_ops { |
| 206 | /** Recalculate pool \a pl usage */ |
| 207 | int (*po_recalc)(struct ldlm_pool *pl); |
| 208 | /** Cancel at least \a nr locks from pool \a pl */ |
| 209 | int (*po_shrink)(struct ldlm_pool *pl, int nr, |
Fabio Falzoi | 5802572 | 2014-04-25 11:41:43 +0200 | [diff] [blame] | 210 | gfp_t gfp_mask); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 211 | }; |
| 212 | |
| 213 | /** One second for pools thread check interval. Each pool has own period. */ |
| 214 | #define LDLM_POOLS_THREAD_PERIOD (1) |
| 215 | |
| 216 | /** ~6% margin for modest pools. See ldlm_pool.c for details. */ |
| 217 | #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4) |
| 218 | |
| 219 | /** Default recalc period for server side pools in sec. */ |
| 220 | #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1) |
| 221 | |
| 222 | /** Default recalc period for client side pools in sec. */ |
| 223 | #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10) |
| 224 | |
| 225 | /** |
| 226 | * LDLM pool structure to track granted locks. |
| 227 | * For purposes of determining when to release locks on e.g. memory pressure. |
| 228 | * This feature is commonly referred to as lru_resize. |
| 229 | */ |
| 230 | struct ldlm_pool { |
Dmitry Eremin | 700815d | 2015-05-21 15:32:11 -0400 | [diff] [blame] | 231 | /** Pool debugfs directory. */ |
| 232 | struct dentry *pl_debugfs_entry; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 233 | /** Pool name, must be long enough to hold compound proc entry name. */ |
| 234 | char pl_name[100]; |
| 235 | /** Lock for protecting SLV/CLV updates. */ |
| 236 | spinlock_t pl_lock; |
| 237 | /** Number of allowed locks in in pool, both, client and server side. */ |
| 238 | atomic_t pl_limit; |
| 239 | /** Number of granted locks in */ |
| 240 | atomic_t pl_granted; |
| 241 | /** Grant rate per T. */ |
| 242 | atomic_t pl_grant_rate; |
| 243 | /** Cancel rate per T. */ |
| 244 | atomic_t pl_cancel_rate; |
| 245 | /** Server lock volume (SLV). Protected by pl_lock. */ |
| 246 | __u64 pl_server_lock_volume; |
| 247 | /** Current biggest client lock volume. Protected by pl_lock. */ |
| 248 | __u64 pl_client_lock_volume; |
| 249 | /** Lock volume factor. SLV on client is calculated as following: |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 250 | * server_slv * lock_volume_factor. |
| 251 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 252 | atomic_t pl_lock_volume_factor; |
| 253 | /** Time when last SLV from server was obtained. */ |
Arnd Bergmann | 8f83409 | 2015-09-27 16:45:17 -0400 | [diff] [blame] | 254 | time64_t pl_recalc_time; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 255 | /** Recalculation period for pool. */ |
Arnd Bergmann | 8f83409 | 2015-09-27 16:45:17 -0400 | [diff] [blame] | 256 | time64_t pl_recalc_period; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 257 | /** Recalculation and shrink operations. */ |
Fabio Falzoi | b9c98cf | 2014-04-25 11:41:42 +0200 | [diff] [blame] | 258 | const struct ldlm_pool_ops *pl_ops; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 259 | /** Number of planned locks for next period. */ |
| 260 | int pl_grant_plan; |
| 261 | /** Pool statistics. */ |
| 262 | struct lprocfs_stats *pl_stats; |
Oleg Drokin | f2825e0 | 2015-05-21 15:32:04 -0400 | [diff] [blame] | 263 | |
| 264 | /* sysfs object */ |
| 265 | struct kobject pl_kobj; |
| 266 | struct completion pl_kobj_unregister; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 267 | }; |
| 268 | |
Jinshan Xiong | 7d44333 | 2016-03-30 19:49:02 -0400 | [diff] [blame] | 269 | typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 270 | |
| 271 | /** |
| 272 | * LVB operations. |
| 273 | * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could |
| 274 | * be associated with an LDLM lock and transferred from client to server and |
| 275 | * back. |
| 276 | * |
| 277 | * Currently LVBs are used by: |
| 278 | * - OSC-OST code to maintain current object size/times |
| 279 | * - layout lock code to return the layout when the layout lock is granted |
| 280 | */ |
| 281 | struct ldlm_valblock_ops { |
| 282 | int (*lvbo_init)(struct ldlm_resource *res); |
| 283 | int (*lvbo_update)(struct ldlm_resource *res, |
| 284 | struct ptlrpc_request *r, |
| 285 | int increase); |
| 286 | int (*lvbo_free)(struct ldlm_resource *res); |
| 287 | /* Return size of lvb data appropriate RPC size can be reserved */ |
| 288 | int (*lvbo_size)(struct ldlm_lock *lock); |
| 289 | /* Called to fill in lvb data to RPC buffer @buf */ |
| 290 | int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen); |
| 291 | }; |
| 292 | |
| 293 | /** |
| 294 | * LDLM pools related, type of lock pool in the namespace. |
| 295 | * Greedy means release cached locks aggressively |
| 296 | */ |
Oleg Drokin | 4d0d6b0 | 2016-02-24 22:00:03 -0500 | [diff] [blame] | 297 | enum ldlm_appetite { |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 298 | LDLM_NAMESPACE_GREEDY = 1 << 0, |
| 299 | LDLM_NAMESPACE_MODEST = 1 << 1 |
Oleg Drokin | 4d0d6b0 | 2016-02-24 22:00:03 -0500 | [diff] [blame] | 300 | }; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 301 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 302 | struct ldlm_ns_bucket { |
| 303 | /** back pointer to namespace */ |
| 304 | struct ldlm_namespace *nsb_namespace; |
| 305 | /** |
| 306 | * Estimated lock callback time. Used by adaptive timeout code to |
| 307 | * avoid spurious client evictions due to unresponsiveness when in |
| 308 | * fact the network or overall system load is at fault |
| 309 | */ |
| 310 | struct adaptive_timeout nsb_at_estimate; |
| 311 | }; |
| 312 | |
| 313 | enum { |
| 314 | /** LDLM namespace lock stats */ |
| 315 | LDLM_NSS_LOCKS = 0, |
| 316 | LDLM_NSS_LAST |
| 317 | }; |
| 318 | |
Oleg Drokin | 87ba6eb | 2016-02-24 22:00:04 -0500 | [diff] [blame] | 319 | enum ldlm_ns_type { |
Masanari Iida | bd9070c | 2014-03-08 22:58:34 +0900 | [diff] [blame] | 320 | /** invalid type */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 321 | LDLM_NS_TYPE_UNKNOWN = 0, |
| 322 | /** mdc namespace */ |
| 323 | LDLM_NS_TYPE_MDC, |
| 324 | /** mds namespace */ |
| 325 | LDLM_NS_TYPE_MDT, |
| 326 | /** osc namespace */ |
| 327 | LDLM_NS_TYPE_OSC, |
| 328 | /** ost namespace */ |
| 329 | LDLM_NS_TYPE_OST, |
| 330 | /** mgc namespace */ |
| 331 | LDLM_NS_TYPE_MGC, |
| 332 | /** mgs namespace */ |
| 333 | LDLM_NS_TYPE_MGT, |
Oleg Drokin | 87ba6eb | 2016-02-24 22:00:04 -0500 | [diff] [blame] | 334 | }; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 335 | |
| 336 | /** |
| 337 | * LDLM Namespace. |
| 338 | * |
| 339 | * Namespace serves to contain locks related to a particular service. |
| 340 | * There are two kinds of namespaces: |
| 341 | * - Server namespace has knowledge of all locks and is therefore authoritative |
| 342 | * to make decisions like what locks could be granted and what conflicts |
| 343 | * exist during new lock enqueue. |
| 344 | * - Client namespace only has limited knowledge about locks in the namespace, |
| 345 | * only seeing locks held by the client. |
| 346 | * |
| 347 | * Every Lustre service has one server namespace present on the server serving |
| 348 | * that service. Every client connected to the service has a client namespace |
| 349 | * for it. |
| 350 | * Every lock obtained by client in that namespace is actually represented by |
| 351 | * two in-memory locks. One on the server and one on the client. The locks are |
| 352 | * linked by a special cookie by which one node can tell to the other which lock |
| 353 | * it actually means during communications. Such locks are called remote locks. |
| 354 | * The locks held by server only without any reference to a client are called |
| 355 | * local locks. |
| 356 | */ |
| 357 | struct ldlm_namespace { |
| 358 | /** Backward link to OBD, required for LDLM pool to store new SLV. */ |
| 359 | struct obd_device *ns_obd; |
| 360 | |
| 361 | /** Flag indicating if namespace is on client instead of server */ |
| 362 | ldlm_side_t ns_client; |
| 363 | |
| 364 | /** Resource hash table for namespace. */ |
Lisa Nguyen | 6da6eab | 2013-10-21 18:16:26 -0700 | [diff] [blame] | 365 | struct cfs_hash *ns_rs_hash; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 366 | |
| 367 | /** serialize */ |
| 368 | spinlock_t ns_lock; |
| 369 | |
| 370 | /** big refcount (by bucket) */ |
| 371 | atomic_t ns_bref; |
| 372 | |
| 373 | /** |
| 374 | * Namespace connect flags supported by server (may be changed via |
Oleg Drokin | f6e42a4 | 2016-02-16 00:46:33 -0500 | [diff] [blame] | 375 | * sysfs, LRU resize may be disabled/enabled). |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 376 | */ |
| 377 | __u64 ns_connect_flags; |
| 378 | |
| 379 | /** Client side original connect flags supported by server. */ |
| 380 | __u64 ns_orig_connect_flags; |
| 381 | |
Dmitry Eremin | 700815d | 2015-05-21 15:32:11 -0400 | [diff] [blame] | 382 | /* namespace debugfs dir entry */ |
| 383 | struct dentry *ns_debugfs_entry; |
Peng Tao | 73bb1da | 2013-05-29 21:40:55 +0800 | [diff] [blame] | 384 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 385 | /** |
| 386 | * Position in global namespace list linking all namespaces on |
| 387 | * the node. |
| 388 | */ |
| 389 | struct list_head ns_list_chain; |
| 390 | |
| 391 | /** |
| 392 | * List of unused locks for this namespace. This list is also called |
| 393 | * LRU lock list. |
| 394 | * Unused locks are locks with zero reader/writer reference counts. |
| 395 | * This list is only used on clients for lock caching purposes. |
| 396 | * When we want to release some locks voluntarily or if server wants |
| 397 | * us to release some locks due to e.g. memory pressure, we take locks |
| 398 | * to release from the head of this list. |
| 399 | * Locks are linked via l_lru field in \see struct ldlm_lock. |
| 400 | */ |
| 401 | struct list_head ns_unused_list; |
| 402 | /** Number of locks in the LRU list above */ |
| 403 | int ns_nr_unused; |
| 404 | |
| 405 | /** |
| 406 | * Maximum number of locks permitted in the LRU. If 0, means locks |
| 407 | * are managed by pools and there is no preset limit, rather it is all |
| 408 | * controlled by available memory on this client and on server. |
| 409 | */ |
| 410 | unsigned int ns_max_unused; |
| 411 | /** Maximum allowed age (last used time) for locks in the LRU */ |
| 412 | unsigned int ns_max_age; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 413 | |
| 414 | /** |
| 415 | * Used to rate-limit ldlm_namespace_dump calls. |
| 416 | * \see ldlm_namespace_dump. Increased by 10 seconds every time |
| 417 | * it is called. |
| 418 | */ |
Greg Kroah-Hartman | a649ad1 | 2014-07-12 00:27:46 -0700 | [diff] [blame] | 419 | unsigned long ns_next_dump; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 420 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 421 | /** |
| 422 | * LVB operations for this namespace. |
| 423 | * \see struct ldlm_valblock_ops |
| 424 | */ |
| 425 | struct ldlm_valblock_ops *ns_lvbo; |
| 426 | |
| 427 | /** |
| 428 | * Used by filter code to store pointer to OBD of the service. |
| 429 | * Should be dropped in favor of \a ns_obd |
| 430 | */ |
| 431 | void *ns_lvbp; |
| 432 | |
| 433 | /** |
| 434 | * Wait queue used by __ldlm_namespace_free. Gets woken up every time |
| 435 | * a resource is removed. |
| 436 | */ |
| 437 | wait_queue_head_t ns_waitq; |
| 438 | /** LDLM pool structure for this namespace */ |
| 439 | struct ldlm_pool ns_pool; |
| 440 | /** Definition of how eagerly unused locks will be released from LRU */ |
Oleg Drokin | 4d0d6b0 | 2016-02-24 22:00:03 -0500 | [diff] [blame] | 441 | enum ldlm_appetite ns_appetite; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 442 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 443 | /** Limit of parallel AST RPC count. */ |
| 444 | unsigned ns_max_parallel_ast; |
| 445 | |
Jinshan Xiong | 7d44333 | 2016-03-30 19:49:02 -0400 | [diff] [blame] | 446 | /** |
| 447 | * Callback to check if a lock is good to be canceled by ELC or |
| 448 | * during recovery. |
| 449 | */ |
| 450 | ldlm_cancel_cbt ns_cancel; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 451 | |
| 452 | /** LDLM lock stats */ |
| 453 | struct lprocfs_stats *ns_stats; |
| 454 | |
| 455 | /** |
| 456 | * Flag to indicate namespace is being freed. Used to determine if |
| 457 | * recalculation of LDLM pool statistics should be skipped. |
| 458 | */ |
| 459 | unsigned ns_stopping:1; |
Oleg Drokin | 18fd885 | 2015-05-21 15:31:55 -0400 | [diff] [blame] | 460 | |
| 461 | struct kobject ns_kobj; /* sysfs object */ |
| 462 | struct completion ns_kobj_unregister; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 463 | }; |
| 464 | |
| 465 | /** |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 466 | * Returns 1 if namespace \a ns supports early lock cancel (ELC). |
| 467 | */ |
| 468 | static inline int ns_connect_cancelset(struct ldlm_namespace *ns) |
| 469 | { |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 470 | return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET); |
| 471 | } |
| 472 | |
| 473 | /** |
| 474 | * Returns 1 if this namespace supports lru_resize. |
| 475 | */ |
| 476 | static inline int ns_connect_lru_resize(struct ldlm_namespace *ns) |
| 477 | { |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 478 | return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE); |
| 479 | } |
| 480 | |
| 481 | static inline void ns_register_cancel(struct ldlm_namespace *ns, |
Jinshan Xiong | 7d44333 | 2016-03-30 19:49:02 -0400 | [diff] [blame] | 482 | ldlm_cancel_cbt arg) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 483 | { |
Jinshan Xiong | 7d44333 | 2016-03-30 19:49:02 -0400 | [diff] [blame] | 484 | ns->ns_cancel = arg; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 485 | } |
| 486 | |
| 487 | struct ldlm_lock; |
| 488 | |
| 489 | /** Type for blocking callback function of a lock. */ |
| 490 | typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock, |
| 491 | struct ldlm_lock_desc *new, void *data, |
| 492 | int flag); |
| 493 | /** Type for completion callback function of a lock. */ |
| 494 | typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags, |
| 495 | void *data); |
| 496 | /** Type for glimpse callback function of a lock. */ |
| 497 | typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 498 | |
| 499 | /** Work list for sending GL ASTs to multiple locks. */ |
| 500 | struct ldlm_glimpse_work { |
| 501 | struct ldlm_lock *gl_lock; /* lock to glimpse */ |
| 502 | struct list_head gl_list; /* linkage to other gl work structs */ |
| 503 | __u32 gl_flags;/* see LDLM_GL_WORK_* below */ |
| 504 | union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 505 | * glimpse callback request |
| 506 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 507 | }; |
| 508 | |
| 509 | /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */ |
| 510 | #define LDLM_GL_WORK_NOFREE 0x1 |
| 511 | |
| 512 | /** Interval node data for each LDLM_EXTENT lock. */ |
| 513 | struct ldlm_interval { |
| 514 | struct interval_node li_node; /* node for tree management */ |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 515 | struct list_head li_group; /* the locks which have the same |
| 516 | * policy - group of the policy |
| 517 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 518 | }; |
Mike Rapoport | c9f6bb9 | 2015-10-13 16:03:42 +0300 | [diff] [blame] | 519 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 520 | #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node) |
| 521 | |
| 522 | /** |
| 523 | * Interval tree for extent locks. |
| 524 | * The interval tree must be accessed under the resource lock. |
| 525 | * Interval trees are used for granted extent locks to speed up conflicts |
| 526 | * lookup. See ldlm/interval_tree.c for more details. |
| 527 | */ |
| 528 | struct ldlm_interval_tree { |
| 529 | /** Tree size. */ |
| 530 | int lit_size; |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 531 | enum ldlm_mode lit_mode; /* lock mode */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 532 | struct interval_node *lit_root; /* actual ldlm_interval */ |
| 533 | }; |
| 534 | |
| 535 | /** Whether to track references to exports by LDLM locks. */ |
| 536 | #define LUSTRE_TRACKS_LOCK_EXP_REFS (0) |
| 537 | |
| 538 | /** Cancel flags. */ |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 539 | enum ldlm_cancel_flags { |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 540 | LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */ |
| 541 | LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */ |
| 542 | LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 543 | * in the same RPC |
| 544 | */ |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 545 | }; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 546 | |
| 547 | struct ldlm_flock { |
| 548 | __u64 start; |
| 549 | __u64 end; |
| 550 | __u64 owner; |
| 551 | __u64 blocking_owner; |
| 552 | struct obd_export *blocking_export; |
| 553 | /* Protected by the hash lock */ |
| 554 | __u32 blocking_refs; |
| 555 | __u32 pid; |
| 556 | }; |
| 557 | |
| 558 | typedef union { |
| 559 | struct ldlm_extent l_extent; |
| 560 | struct ldlm_flock l_flock; |
| 561 | struct ldlm_inodebits l_inodebits; |
| 562 | } ldlm_policy_data_t; |
| 563 | |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 564 | void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type, |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 565 | const ldlm_wire_policy_data_t *wpolicy, |
| 566 | ldlm_policy_data_t *lpolicy); |
| 567 | |
| 568 | enum lvb_type { |
| 569 | LVB_T_NONE = 0, |
| 570 | LVB_T_OST = 1, |
| 571 | LVB_T_LQUOTA = 2, |
| 572 | LVB_T_LAYOUT = 3, |
| 573 | }; |
| 574 | |
| 575 | /** |
Jinshan Xiong | 8b915c1 | 2016-09-18 16:38:32 -0400 | [diff] [blame] | 576 | * LDLM_GID_ANY is used to match any group id in ldlm_lock_match(). |
| 577 | */ |
| 578 | #define LDLM_GID_ANY ((__u64)-1) |
| 579 | |
| 580 | /** |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 581 | * LDLM lock structure |
| 582 | * |
| 583 | * Represents a single LDLM lock and its state in memory. Each lock is |
| 584 | * associated with a single ldlm_resource, the object which is being |
| 585 | * locked. There may be multiple ldlm_locks on a single resource, |
| 586 | * depending on the lock type and whether the locks are conflicting or |
| 587 | * not. |
| 588 | */ |
| 589 | struct ldlm_lock { |
| 590 | /** |
| 591 | * Local lock handle. |
| 592 | * When remote side wants to tell us about a lock, they address |
| 593 | * it by this opaque handle. The handle does not hold a |
| 594 | * reference on the ldlm_lock, so it can be safely passed to |
| 595 | * other threads or nodes. When the lock needs to be accessed |
| 596 | * from the handle, it is looked up again in the lock table, and |
| 597 | * may no longer exist. |
| 598 | * |
| 599 | * Must be first in the structure. |
| 600 | */ |
| 601 | struct portals_handle l_handle; |
| 602 | /** |
| 603 | * Lock reference count. |
| 604 | * This is how many users have pointers to actual structure, so that |
| 605 | * we do not accidentally free lock structure that is in use. |
| 606 | */ |
| 607 | atomic_t l_refc; |
| 608 | /** |
| 609 | * Internal spinlock protects l_resource. We should hold this lock |
| 610 | * first before taking res_lock. |
| 611 | */ |
| 612 | spinlock_t l_lock; |
| 613 | /** |
| 614 | * Pointer to actual resource this lock is in. |
| 615 | * ldlm_lock_change_resource() can change this. |
| 616 | */ |
| 617 | struct ldlm_resource *l_resource; |
| 618 | /** |
| 619 | * List item for client side LRU list. |
| 620 | * Protected by ns_lock in struct ldlm_namespace. |
| 621 | */ |
| 622 | struct list_head l_lru; |
| 623 | /** |
| 624 | * Linkage to resource's lock queues according to current lock state. |
| 625 | * (could be granted, waiting or converting) |
| 626 | * Protected by lr_lock in struct ldlm_resource. |
| 627 | */ |
| 628 | struct list_head l_res_link; |
| 629 | /** |
| 630 | * Tree node for ldlm_extent. |
| 631 | */ |
| 632 | struct ldlm_interval *l_tree_node; |
| 633 | /** |
| 634 | * Per export hash of locks. |
| 635 | * Protected by per-bucket exp->exp_lock_hash locks. |
| 636 | */ |
| 637 | struct hlist_node l_exp_hash; |
| 638 | /** |
| 639 | * Per export hash of flock locks. |
| 640 | * Protected by per-bucket exp->exp_flock_hash locks. |
| 641 | */ |
| 642 | struct hlist_node l_exp_flock_hash; |
| 643 | /** |
| 644 | * Requested mode. |
| 645 | * Protected by lr_lock. |
| 646 | */ |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 647 | enum ldlm_mode l_req_mode; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 648 | /** |
| 649 | * Granted mode, also protected by lr_lock. |
| 650 | */ |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 651 | enum ldlm_mode l_granted_mode; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 652 | /** Lock completion handler pointer. Called when lock is granted. */ |
| 653 | ldlm_completion_callback l_completion_ast; |
| 654 | /** |
| 655 | * Lock blocking AST handler pointer. |
| 656 | * It plays two roles: |
| 657 | * - as a notification of an attempt to queue a conflicting lock (once) |
| 658 | * - as a notification when the lock is being cancelled. |
| 659 | * |
| 660 | * As such it's typically called twice: once for the initial conflict |
| 661 | * and then once more when the last user went away and the lock is |
| 662 | * cancelled (could happen recursively). |
| 663 | */ |
| 664 | ldlm_blocking_callback l_blocking_ast; |
| 665 | /** |
| 666 | * Lock glimpse handler. |
| 667 | * Glimpse handler is used to obtain LVB updates from a client by |
| 668 | * server |
| 669 | */ |
| 670 | ldlm_glimpse_callback l_glimpse_ast; |
| 671 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 672 | /** |
| 673 | * Lock export. |
| 674 | * This is a pointer to actual client export for locks that were granted |
| 675 | * to clients. Used server-side. |
| 676 | */ |
| 677 | struct obd_export *l_export; |
| 678 | /** |
| 679 | * Lock connection export. |
| 680 | * Pointer to server export on a client. |
| 681 | */ |
| 682 | struct obd_export *l_conn_export; |
| 683 | |
| 684 | /** |
| 685 | * Remote lock handle. |
| 686 | * If the lock is remote, this is the handle of the other side lock |
| 687 | * (l_handle) |
| 688 | */ |
| 689 | struct lustre_handle l_remote_handle; |
| 690 | |
| 691 | /** |
| 692 | * Representation of private data specific for a lock type. |
| 693 | * Examples are: extent range for extent lock or bitmask for ibits locks |
| 694 | */ |
| 695 | ldlm_policy_data_t l_policy_data; |
| 696 | |
| 697 | /** |
Bruce Korb | f2145ea | 2013-07-23 00:06:37 +0800 | [diff] [blame] | 698 | * Lock state flags. Protected by lr_lock. |
| 699 | * \see lustre_dlm_flags.h where the bits are defined. |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 700 | */ |
| 701 | __u64 l_flags; |
Bruce Korb | f2145ea | 2013-07-23 00:06:37 +0800 | [diff] [blame] | 702 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 703 | /** |
| 704 | * Lock r/w usage counters. |
| 705 | * Protected by lr_lock. |
| 706 | */ |
| 707 | __u32 l_readers; |
| 708 | __u32 l_writers; |
| 709 | /** |
| 710 | * If the lock is granted, a process sleeps on this waitq to learn when |
| 711 | * it's no longer in use. If the lock is not granted, a process sleeps |
| 712 | * on this waitq to learn when it becomes granted. |
| 713 | */ |
| 714 | wait_queue_head_t l_waitq; |
| 715 | |
| 716 | /** |
| 717 | * Seconds. It will be updated if there is any activity related to |
| 718 | * the lock, e.g. enqueue the lock or send blocking AST. |
| 719 | */ |
Arnd Bergmann | bf6d215 | 2015-09-27 16:45:06 -0400 | [diff] [blame] | 720 | time64_t l_last_activity; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 721 | |
| 722 | /** |
| 723 | * Time last used by e.g. being matched by lock match. |
| 724 | * Jiffies. Should be converted to time if needed. |
| 725 | */ |
Greg Kroah-Hartman | a649ad1 | 2014-07-12 00:27:46 -0700 | [diff] [blame] | 726 | unsigned long l_last_used; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 727 | |
| 728 | /** Originally requested extent for the extent lock. */ |
| 729 | struct ldlm_extent l_req_extent; |
| 730 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 731 | /* |
| 732 | * Client-side-only members. |
| 733 | */ |
| 734 | |
| 735 | enum lvb_type l_lvb_type; |
| 736 | |
| 737 | /** |
| 738 | * Temporary storage for a LVB received during an enqueue operation. |
| 739 | */ |
| 740 | __u32 l_lvb_len; |
| 741 | void *l_lvb_data; |
| 742 | |
| 743 | /** Private storage for lock user. Opaque to LDLM. */ |
| 744 | void *l_ast_data; |
| 745 | |
| 746 | /* |
| 747 | * Server-side-only members. |
| 748 | */ |
| 749 | |
| 750 | /** |
| 751 | * Connection cookie for the client originating the operation. |
| 752 | * Used by Commit on Share (COS) code. Currently only used for |
| 753 | * inodebits locks on MDS. |
| 754 | */ |
| 755 | __u64 l_client_cookie; |
| 756 | |
| 757 | /** |
| 758 | * List item for locks waiting for cancellation from clients. |
| 759 | * The lists this could be linked into are: |
| 760 | * waiting_locks_list (protected by waiting_locks_spinlock), |
| 761 | * then if the lock timed out, it is moved to |
| 762 | * expired_lock_thread.elt_expired_locks for further processing. |
| 763 | * Protected by elt_lock. |
| 764 | */ |
| 765 | struct list_head l_pending_chain; |
| 766 | |
| 767 | /** |
| 768 | * Set when lock is sent a blocking AST. Time in seconds when timeout |
| 769 | * is reached and client holding this lock could be evicted. |
| 770 | * This timeout could be further extended by e.g. certain IO activity |
| 771 | * under this lock. |
| 772 | * \see ost_rw_prolong_locks |
| 773 | */ |
Greg Kroah-Hartman | a649ad1 | 2014-07-12 00:27:46 -0700 | [diff] [blame] | 774 | unsigned long l_callback_timeout; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 775 | |
| 776 | /** Local PID of process which created this lock. */ |
| 777 | __u32 l_pid; |
| 778 | |
| 779 | /** |
| 780 | * Number of times blocking AST was sent for this lock. |
| 781 | * This is for debugging. Valid values are 0 and 1, if there is an |
| 782 | * attempt to send blocking AST more than once, an assertion would be |
| 783 | * hit. \see ldlm_work_bl_ast_lock |
| 784 | */ |
| 785 | int l_bl_ast_run; |
| 786 | /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */ |
| 787 | struct list_head l_bl_ast; |
| 788 | /** List item ldlm_add_ast_work_item() for case of completion ASTs. */ |
| 789 | struct list_head l_cp_ast; |
| 790 | /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */ |
| 791 | struct list_head l_rk_ast; |
| 792 | |
| 793 | /** |
| 794 | * Pointer to a conflicting lock that caused blocking AST to be sent |
| 795 | * for this lock |
| 796 | */ |
| 797 | struct ldlm_lock *l_blocking_lock; |
| 798 | |
| 799 | /** |
| 800 | * Protected by lr_lock, linkages to "skip lists". |
| 801 | * For more explanations of skip lists see ldlm/ldlm_inodebits.c |
| 802 | */ |
| 803 | struct list_head l_sl_mode; |
| 804 | struct list_head l_sl_policy; |
| 805 | |
| 806 | /** Reference tracking structure to debug leaked locks. */ |
| 807 | struct lu_ref l_reference; |
| 808 | #if LUSTRE_TRACKS_LOCK_EXP_REFS |
| 809 | /* Debugging stuff for bug 20498, for tracking export references. */ |
| 810 | /** number of export references taken */ |
| 811 | int l_exp_refs_nr; |
| 812 | /** link all locks referencing one export */ |
| 813 | struct list_head l_exp_refs_link; |
| 814 | /** referenced export object */ |
| 815 | struct obd_export *l_exp_refs_target; |
| 816 | #endif |
| 817 | /** |
| 818 | * export blocking dlm lock list, protected by |
| 819 | * l_export->exp_bl_list_lock. |
| 820 | * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock |
| 821 | * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock. |
| 822 | */ |
| 823 | struct list_head l_exp_list; |
| 824 | }; |
| 825 | |
| 826 | /** |
| 827 | * LDLM resource description. |
| 828 | * Basically, resource is a representation for a single object. |
| 829 | * Object has a name which is currently 4 64-bit integers. LDLM user is |
| 830 | * responsible for creation of a mapping between objects it wants to be |
| 831 | * protected and resource names. |
| 832 | * |
| 833 | * A resource can only hold locks of a single lock type, though there may be |
| 834 | * multiple ldlm_locks on a single resource, depending on the lock type and |
| 835 | * whether the locks are conflicting or not. |
| 836 | */ |
| 837 | struct ldlm_resource { |
| 838 | struct ldlm_ns_bucket *lr_ns_bucket; |
| 839 | |
| 840 | /** |
| 841 | * List item for list in namespace hash. |
| 842 | * protected by ns_lock |
| 843 | */ |
| 844 | struct hlist_node lr_hash; |
| 845 | |
| 846 | /** Spinlock to protect locks under this resource. */ |
| 847 | spinlock_t lr_lock; |
| 848 | |
| 849 | /** |
| 850 | * protected by lr_lock |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 851 | * @{ |
| 852 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 853 | /** List of locks in granted state */ |
| 854 | struct list_head lr_granted; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 855 | /** |
| 856 | * List of locks that could not be granted due to conflicts and |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 857 | * that are waiting for conflicts to go away |
| 858 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 859 | struct list_head lr_waiting; |
| 860 | /** @} */ |
| 861 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 862 | /** Type of locks this resource can hold. Only one type per resource. */ |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 863 | enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 864 | |
| 865 | /** Resource name */ |
| 866 | struct ldlm_res_id lr_name; |
| 867 | /** Reference count for this resource */ |
| 868 | atomic_t lr_refcount; |
| 869 | |
| 870 | /** |
| 871 | * Interval trees (only for extent locks) for all modes of this resource |
| 872 | */ |
| 873 | struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; |
| 874 | |
| 875 | /** |
| 876 | * Server-side-only lock value block elements. |
| 877 | * To serialize lvbo_init. |
| 878 | */ |
| 879 | struct mutex lr_lvb_mutex; |
| 880 | int lr_lvb_len; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 881 | |
| 882 | /** When the resource was considered as contended. */ |
Greg Kroah-Hartman | a649ad1 | 2014-07-12 00:27:46 -0700 | [diff] [blame] | 883 | unsigned long lr_contention_time; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 884 | /** List of references to this resource. For debugging. */ |
| 885 | struct lu_ref lr_reference; |
| 886 | |
| 887 | struct inode *lr_lvb_inode; |
| 888 | }; |
| 889 | |
| 890 | static inline bool ldlm_has_layout(struct ldlm_lock *lock) |
| 891 | { |
| 892 | return lock->l_resource->lr_type == LDLM_IBITS && |
| 893 | lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT; |
| 894 | } |
| 895 | |
| 896 | static inline char * |
| 897 | ldlm_ns_name(struct ldlm_namespace *ns) |
| 898 | { |
| 899 | return ns->ns_rs_hash->hs_name; |
| 900 | } |
| 901 | |
| 902 | static inline struct ldlm_namespace * |
| 903 | ldlm_res_to_ns(struct ldlm_resource *res) |
| 904 | { |
| 905 | return res->lr_ns_bucket->nsb_namespace; |
| 906 | } |
| 907 | |
| 908 | static inline struct ldlm_namespace * |
| 909 | ldlm_lock_to_ns(struct ldlm_lock *lock) |
| 910 | { |
| 911 | return ldlm_res_to_ns(lock->l_resource); |
| 912 | } |
| 913 | |
| 914 | static inline char * |
| 915 | ldlm_lock_to_ns_name(struct ldlm_lock *lock) |
| 916 | { |
| 917 | return ldlm_ns_name(ldlm_lock_to_ns(lock)); |
| 918 | } |
| 919 | |
| 920 | static inline struct adaptive_timeout * |
| 921 | ldlm_lock_to_ns_at(struct ldlm_lock *lock) |
| 922 | { |
| 923 | return &lock->l_resource->lr_ns_bucket->nsb_at_estimate; |
| 924 | } |
| 925 | |
| 926 | static inline int ldlm_lvbo_init(struct ldlm_resource *res) |
| 927 | { |
| 928 | struct ldlm_namespace *ns = ldlm_res_to_ns(res); |
| 929 | |
Oleg Drokin | d2a1398 | 2016-02-16 00:46:52 -0500 | [diff] [blame] | 930 | if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 931 | return ns->ns_lvbo->lvbo_init(res); |
| 932 | |
| 933 | return 0; |
| 934 | } |
| 935 | |
| 936 | static inline int ldlm_lvbo_size(struct ldlm_lock *lock) |
| 937 | { |
| 938 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); |
| 939 | |
Oleg Drokin | d2a1398 | 2016-02-16 00:46:52 -0500 | [diff] [blame] | 940 | if (ns->ns_lvbo && ns->ns_lvbo->lvbo_size) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 941 | return ns->ns_lvbo->lvbo_size(lock); |
| 942 | |
| 943 | return 0; |
| 944 | } |
| 945 | |
| 946 | static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len) |
| 947 | { |
| 948 | struct ldlm_namespace *ns = ldlm_lock_to_ns(lock); |
| 949 | |
Oleg Drokin | d2a1398 | 2016-02-16 00:46:52 -0500 | [diff] [blame] | 950 | if (ns->ns_lvbo) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 951 | return ns->ns_lvbo->lvbo_fill(lock, buf, len); |
Oleg Drokin | d2a1398 | 2016-02-16 00:46:52 -0500 | [diff] [blame] | 952 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 953 | return 0; |
| 954 | } |
| 955 | |
| 956 | struct ldlm_ast_work { |
| 957 | struct ldlm_lock *w_lock; |
| 958 | int w_blocking; |
| 959 | struct ldlm_lock_desc w_desc; |
| 960 | struct list_head w_list; |
| 961 | int w_flags; |
| 962 | void *w_data; |
| 963 | int w_datalen; |
| 964 | }; |
| 965 | |
| 966 | /** |
| 967 | * Common ldlm_enqueue parameters |
| 968 | */ |
| 969 | struct ldlm_enqueue_info { |
| 970 | __u32 ei_type; /** Type of the lock being enqueued. */ |
| 971 | __u32 ei_mode; /** Mode of the lock being enqueued. */ |
| 972 | void *ei_cb_bl; /** blocking lock callback */ |
| 973 | void *ei_cb_cp; /** lock completion callback */ |
| 974 | void *ei_cb_gl; /** lock glimpse callback */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 975 | void *ei_cbdata; /** Data to be passed into callbacks. */ |
wang di | 75ac62fc | 2016-08-16 16:18:27 -0400 | [diff] [blame] | 976 | unsigned int ei_enq_slave:1; /* whether enqueue slave stripes */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 977 | }; |
| 978 | |
| 979 | extern struct obd_ops ldlm_obd_ops; |
| 980 | |
| 981 | extern char *ldlm_lockname[]; |
Joe Perches | 8150a97 | 2015-08-10 14:51:21 -0700 | [diff] [blame] | 982 | char *ldlm_it2str(int it); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 983 | |
| 984 | /** |
| 985 | * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG. |
| 986 | * For the cases where we do not have actual lock to print along |
| 987 | * with a debugging message that is ldlm-related |
| 988 | */ |
| 989 | #define LDLM_DEBUG_NOLOCK(format, a...) \ |
Mike Rapoport | b2952d6 | 2015-09-03 11:49:13 +0300 | [diff] [blame] | 990 | CDEBUG(D_DLMTRACE, "### " format "\n", ##a) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 991 | |
| 992 | /** |
| 993 | * Support function for lock information printing into debug logs. |
| 994 | * \see LDLM_DEBUG |
| 995 | */ |
| 996 | #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \ |
| 997 | CFS_CHECK_STACK(msgdata, mask, cdls); \ |
| 998 | \ |
| 999 | if (((mask) & D_CANTMASK) != 0 || \ |
| 1000 | ((libcfs_debug & (mask)) != 0 && \ |
| 1001 | (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \ |
| 1002 | _ldlm_lock_debug(lock, msgdata, fmt, ##a); \ |
Greg Donald | a58a38a | 2014-08-21 12:40:35 -0500 | [diff] [blame] | 1003 | } while (0) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1004 | |
| 1005 | void _ldlm_lock_debug(struct ldlm_lock *lock, |
| 1006 | struct libcfs_debug_msg_data *data, |
| 1007 | const char *fmt, ...) |
Mario J. Rugiero | 70837c1 | 2015-03-10 12:02:03 -0300 | [diff] [blame] | 1008 | __printf(3, 4); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1009 | |
| 1010 | /** |
| 1011 | * Rate-limited version of lock printing function. |
| 1012 | */ |
| 1013 | #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \ |
Lisa Nguyen | a3ea59e | 2013-10-21 18:15:39 -0700 | [diff] [blame] | 1014 | static struct cfs_debug_limit_state _ldlm_cdls; \ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1015 | LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \ |
Mike Rapoport | b2952d6 | 2015-09-03 11:49:13 +0300 | [diff] [blame] | 1016 | ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a);\ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1017 | } while (0) |
| 1018 | |
| 1019 | #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a) |
| 1020 | #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a) |
| 1021 | |
| 1022 | /** Non-rate-limited lock printing function for debugging purposes. */ |
| 1023 | #define LDLM_DEBUG(lock, fmt, a...) do { \ |
Oleg Drokin | d2a1398 | 2016-02-16 00:46:52 -0500 | [diff] [blame] | 1024 | if (likely(lock)) { \ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1025 | LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \ |
| 1026 | ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \ |
Mike Rapoport | b2952d6 | 2015-09-03 11:49:13 +0300 | [diff] [blame] | 1027 | "### " fmt, ##a); \ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1028 | } else { \ |
| 1029 | LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \ |
| 1030 | } \ |
| 1031 | } while (0) |
| 1032 | |
| 1033 | typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags, |
Oleg Drokin | d1777aa | 2016-02-24 22:00:02 -0500 | [diff] [blame] | 1034 | int first_enq, enum ldlm_error *err, |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1035 | struct list_head *work_list); |
| 1036 | |
| 1037 | /** |
| 1038 | * Return values for lock iterators. |
| 1039 | * Also used during deciding of lock grants and cancellations. |
| 1040 | */ |
| 1041 | #define LDLM_ITER_CONTINUE 1 /* keep iterating */ |
| 1042 | #define LDLM_ITER_STOP 2 /* stop iterating */ |
| 1043 | |
| 1044 | typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *); |
| 1045 | typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *); |
| 1046 | |
| 1047 | /** \defgroup ldlm_iterator Lock iterators |
| 1048 | * |
| 1049 | * LDLM provides for a way to iterate through every lock on a resource or |
| 1050 | * namespace or every resource in a namespace. |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 1051 | * @{ |
| 1052 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1053 | int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *, |
| 1054 | ldlm_iterator_t iter, void *data); |
| 1055 | /** @} ldlm_iterator */ |
| 1056 | |
| 1057 | int ldlm_replay_locks(struct obd_import *imp); |
| 1058 | |
| 1059 | /* ldlm_flock.c */ |
| 1060 | int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data); |
| 1061 | |
| 1062 | /* ldlm_extent.c */ |
| 1063 | __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms); |
| 1064 | |
| 1065 | struct ldlm_callback_suite { |
| 1066 | ldlm_completion_callback lcs_completion; |
| 1067 | ldlm_blocking_callback lcs_blocking; |
| 1068 | ldlm_glimpse_callback lcs_glimpse; |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1069 | }; |
| 1070 | |
| 1071 | /* ldlm_lockd.c */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1072 | int ldlm_get_ref(void); |
| 1073 | void ldlm_put_ref(void); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1074 | struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req); |
| 1075 | |
| 1076 | /* ldlm_lock.c */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1077 | void ldlm_lock2handle(const struct ldlm_lock *lock, |
| 1078 | struct lustre_handle *lockh); |
| 1079 | struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags); |
| 1080 | void ldlm_cancel_callback(struct ldlm_lock *); |
| 1081 | int ldlm_lock_remove_from_lru(struct ldlm_lock *); |
John L. Hammond | e8beaf6 | 2016-06-20 16:55:41 -0400 | [diff] [blame] | 1082 | int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1083 | |
| 1084 | /** |
| 1085 | * Obtain a lock reference by its handle. |
| 1086 | */ |
| 1087 | static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h) |
| 1088 | { |
| 1089 | return __ldlm_handle2lock(h, 0); |
| 1090 | } |
| 1091 | |
| 1092 | #define LDLM_LOCK_REF_DEL(lock) \ |
| 1093 | lu_ref_del(&lock->l_reference, "handle", current) |
| 1094 | |
| 1095 | static inline struct ldlm_lock * |
| 1096 | ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags) |
| 1097 | { |
| 1098 | struct ldlm_lock *lock; |
| 1099 | |
| 1100 | lock = __ldlm_handle2lock(h, flags); |
Oleg Drokin | d2a1398 | 2016-02-16 00:46:52 -0500 | [diff] [blame] | 1101 | if (lock) |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1102 | LDLM_LOCK_REF_DEL(lock); |
| 1103 | return lock; |
| 1104 | } |
| 1105 | |
| 1106 | /** |
| 1107 | * Update Lock Value Block Operations (LVBO) on a resource taking into account |
Masanari Iida | 1789118 | 2014-03-08 22:58:33 +0900 | [diff] [blame] | 1108 | * data from request \a r |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1109 | */ |
| 1110 | static inline int ldlm_res_lvbo_update(struct ldlm_resource *res, |
| 1111 | struct ptlrpc_request *r, int increase) |
| 1112 | { |
| 1113 | if (ldlm_res_to_ns(res)->ns_lvbo && |
| 1114 | ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) { |
| 1115 | return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r, |
| 1116 | increase); |
| 1117 | } |
| 1118 | return 0; |
| 1119 | } |
| 1120 | |
Oleg Drokin | d1777aa | 2016-02-24 22:00:02 -0500 | [diff] [blame] | 1121 | int ldlm_error2errno(enum ldlm_error error); |
Oleg Drokin | 38c3f17 | 2015-10-01 00:12:25 -0400 | [diff] [blame] | 1122 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1123 | #if LUSTRE_TRACKS_LOCK_EXP_REFS |
| 1124 | void ldlm_dump_export_locks(struct obd_export *exp); |
| 1125 | #endif |
| 1126 | |
| 1127 | /** |
| 1128 | * Release a temporary lock reference obtained by ldlm_handle2lock() or |
| 1129 | * __ldlm_handle2lock(). |
| 1130 | */ |
| 1131 | #define LDLM_LOCK_PUT(lock) \ |
| 1132 | do { \ |
| 1133 | LDLM_LOCK_REF_DEL(lock); \ |
| 1134 | /*LDLM_DEBUG((lock), "put");*/ \ |
| 1135 | ldlm_lock_put(lock); \ |
| 1136 | } while (0) |
| 1137 | |
| 1138 | /** |
| 1139 | * Release a lock reference obtained by some other means (see |
| 1140 | * LDLM_LOCK_PUT()). |
| 1141 | */ |
| 1142 | #define LDLM_LOCK_RELEASE(lock) \ |
| 1143 | do { \ |
| 1144 | /*LDLM_DEBUG((lock), "put");*/ \ |
| 1145 | ldlm_lock_put(lock); \ |
| 1146 | } while (0) |
| 1147 | |
| 1148 | #define LDLM_LOCK_GET(lock) \ |
| 1149 | ({ \ |
| 1150 | ldlm_lock_get(lock); \ |
| 1151 | /*LDLM_DEBUG((lock), "get");*/ \ |
| 1152 | lock; \ |
| 1153 | }) |
| 1154 | |
| 1155 | #define ldlm_lock_list_put(head, member, count) \ |
| 1156 | ({ \ |
| 1157 | struct ldlm_lock *_lock, *_next; \ |
| 1158 | int c = count; \ |
| 1159 | list_for_each_entry_safe(_lock, _next, head, member) { \ |
| 1160 | if (c-- == 0) \ |
| 1161 | break; \ |
| 1162 | list_del_init(&_lock->member); \ |
| 1163 | LDLM_LOCK_RELEASE(_lock); \ |
| 1164 | } \ |
| 1165 | LASSERT(c <= 0); \ |
| 1166 | }) |
| 1167 | |
| 1168 | struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock); |
| 1169 | void ldlm_lock_put(struct ldlm_lock *lock); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1170 | void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc); |
John L. Hammond | e8beaf6 | 2016-06-20 16:55:41 -0400 | [diff] [blame] | 1171 | void ldlm_lock_addref(const struct lustre_handle *lockh, __u32 mode); |
| 1172 | int ldlm_lock_addref_try(const struct lustre_handle *lockh, __u32 mode); |
| 1173 | void ldlm_lock_decref(const struct lustre_handle *lockh, __u32 mode); |
| 1174 | void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh, __u32 mode); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1175 | void ldlm_lock_fail_match_locked(struct ldlm_lock *lock); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1176 | void ldlm_lock_allow_match(struct ldlm_lock *lock); |
| 1177 | void ldlm_lock_allow_match_locked(struct ldlm_lock *lock); |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 1178 | enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags, |
| 1179 | const struct ldlm_res_id *, |
| 1180 | enum ldlm_type type, ldlm_policy_data_t *, |
| 1181 | enum ldlm_mode mode, struct lustre_handle *, |
| 1182 | int unref); |
John L. Hammond | e8beaf6 | 2016-06-20 16:55:41 -0400 | [diff] [blame] | 1183 | enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh, |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 1184 | __u64 *bits); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1185 | void ldlm_lock_cancel(struct ldlm_lock *lock); |
John L. Hammond | e8beaf6 | 2016-06-20 16:55:41 -0400 | [diff] [blame] | 1186 | void ldlm_lock_dump_handle(int level, const struct lustre_handle *); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1187 | void ldlm_unlink_lock_skiplist(struct ldlm_lock *req); |
| 1188 | |
| 1189 | /* resource.c */ |
| 1190 | struct ldlm_namespace * |
| 1191 | ldlm_namespace_new(struct obd_device *obd, char *name, |
Oleg Drokin | 4d0d6b0 | 2016-02-24 22:00:03 -0500 | [diff] [blame] | 1192 | ldlm_side_t client, enum ldlm_appetite apt, |
Oleg Drokin | 87ba6eb | 2016-02-24 22:00:04 -0500 | [diff] [blame] | 1193 | enum ldlm_ns_type ns_type); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1194 | int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1195 | void ldlm_namespace_get(struct ldlm_namespace *ns); |
| 1196 | void ldlm_namespace_put(struct ldlm_namespace *ns); |
Dmitry Eremin | 700815d | 2015-05-21 15:32:11 -0400 | [diff] [blame] | 1197 | int ldlm_debugfs_setup(void); |
| 1198 | void ldlm_debugfs_cleanup(void); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1199 | |
| 1200 | /* resource.c - internal */ |
| 1201 | struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns, |
| 1202 | struct ldlm_resource *parent, |
| 1203 | const struct ldlm_res_id *, |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 1204 | enum ldlm_type type, int create); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1205 | int ldlm_resource_putref(struct ldlm_resource *res); |
| 1206 | void ldlm_resource_add_lock(struct ldlm_resource *res, |
| 1207 | struct list_head *head, |
| 1208 | struct ldlm_lock *lock); |
| 1209 | void ldlm_resource_unlink_lock(struct ldlm_lock *lock); |
| 1210 | void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc); |
| 1211 | void ldlm_dump_all_namespaces(ldlm_side_t client, int level); |
| 1212 | void ldlm_namespace_dump(int level, struct ldlm_namespace *); |
| 1213 | void ldlm_resource_dump(int level, struct ldlm_resource *); |
| 1214 | int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *, |
| 1215 | const struct ldlm_res_id *); |
| 1216 | |
| 1217 | #define LDLM_RESOURCE_ADDREF(res) do { \ |
Christoph Jaeger | f9bd9c1 | 2014-03-28 00:21:07 +0100 | [diff] [blame] | 1218 | lu_ref_add_atomic(&(res)->lr_reference, __func__, current); \ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1219 | } while (0) |
| 1220 | |
| 1221 | #define LDLM_RESOURCE_DELREF(res) do { \ |
Christoph Jaeger | f9bd9c1 | 2014-03-28 00:21:07 +0100 | [diff] [blame] | 1222 | lu_ref_del(&(res)->lr_reference, __func__, current); \ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1223 | } while (0) |
| 1224 | |
| 1225 | /* ldlm_request.c */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1226 | /** \defgroup ldlm_local_ast Default AST handlers for local locks |
| 1227 | * These AST handlers are typically used for server-side local locks and are |
| 1228 | * also used by client-side lock handlers to perform minimum level base |
| 1229 | * processing. |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 1230 | * @{ |
| 1231 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1232 | int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data); |
| 1233 | int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data); |
| 1234 | /** @} ldlm_local_ast */ |
| 1235 | |
| 1236 | /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users. |
| 1237 | * These are typically used by client and server (*_local versions) |
| 1238 | * to obtain and release locks. |
Oleg Drokin | c56e256 | 2016-02-24 22:00:25 -0500 | [diff] [blame] | 1239 | * @{ |
| 1240 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1241 | int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp, |
| 1242 | struct ldlm_enqueue_info *einfo, |
| 1243 | const struct ldlm_res_id *res_id, |
| 1244 | ldlm_policy_data_t const *policy, __u64 *flags, |
| 1245 | void *lvb, __u32 lvb_len, enum lvb_type lvb_type, |
| 1246 | struct lustre_handle *lockh, int async); |
| 1247 | int ldlm_prep_enqueue_req(struct obd_export *exp, |
| 1248 | struct ptlrpc_request *req, |
| 1249 | struct list_head *cancels, |
| 1250 | int count); |
| 1251 | int ldlm_prep_elc_req(struct obd_export *exp, |
| 1252 | struct ptlrpc_request *req, |
| 1253 | int version, int opc, int canceloff, |
| 1254 | struct list_head *cancels, int count); |
| 1255 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1256 | int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req, |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 1257 | enum ldlm_type type, __u8 with_policy, |
| 1258 | enum ldlm_mode mode, |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1259 | __u64 *flags, void *lvb, __u32 lvb_len, |
John L. Hammond | e8beaf6 | 2016-06-20 16:55:41 -0400 | [diff] [blame] | 1260 | const struct lustre_handle *lockh, int rc); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1261 | int ldlm_cli_update_pool(struct ptlrpc_request *req); |
John L. Hammond | e8beaf6 | 2016-06-20 16:55:41 -0400 | [diff] [blame] | 1262 | int ldlm_cli_cancel(const struct lustre_handle *lockh, |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 1263 | enum ldlm_cancel_flags cancel_flags); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1264 | int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *, |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 1265 | enum ldlm_cancel_flags flags, void *opaque); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1266 | int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns, |
| 1267 | const struct ldlm_res_id *res_id, |
| 1268 | ldlm_policy_data_t *policy, |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 1269 | enum ldlm_mode mode, |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 1270 | enum ldlm_cancel_flags flags, |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1271 | void *opaque); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1272 | int ldlm_cancel_resource_local(struct ldlm_resource *res, |
| 1273 | struct list_head *cancels, |
| 1274 | ldlm_policy_data_t *policy, |
Oleg Drokin | 52ee0d2 | 2016-02-24 21:59:54 -0500 | [diff] [blame] | 1275 | enum ldlm_mode mode, __u64 lock_flags, |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 1276 | enum ldlm_cancel_flags cancel_flags, |
| 1277 | void *opaque); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1278 | int ldlm_cli_cancel_list_local(struct list_head *cancels, int count, |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 1279 | enum ldlm_cancel_flags flags); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1280 | int ldlm_cli_cancel_list(struct list_head *head, int count, |
Oleg Drokin | f833ee4 | 2016-02-24 22:00:05 -0500 | [diff] [blame] | 1281 | struct ptlrpc_request *req, |
| 1282 | enum ldlm_cancel_flags flags); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1283 | /** @} ldlm_cli_api */ |
| 1284 | |
| 1285 | /* mds/handler.c */ |
| 1286 | /* This has to be here because recursive inclusion sucks. */ |
| 1287 | int intent_disposition(struct ldlm_reply *rep, int flag); |
| 1288 | void intent_set_disposition(struct ldlm_reply *rep, int flag); |
| 1289 | |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1290 | /** |
| 1291 | * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more |
| 1292 | * than one lock_res is dead-lock safe. |
| 1293 | */ |
| 1294 | enum lock_res_type { |
| 1295 | LRT_NORMAL, |
| 1296 | LRT_NEW |
| 1297 | }; |
| 1298 | |
| 1299 | /** Lock resource. */ |
| 1300 | static inline void lock_res(struct ldlm_resource *res) |
| 1301 | { |
| 1302 | spin_lock(&res->lr_lock); |
| 1303 | } |
| 1304 | |
| 1305 | /** Lock resource with a way to instruct lockdep code about nestedness-safe. */ |
| 1306 | static inline void lock_res_nested(struct ldlm_resource *res, |
| 1307 | enum lock_res_type mode) |
| 1308 | { |
| 1309 | spin_lock_nested(&res->lr_lock, mode); |
| 1310 | } |
| 1311 | |
| 1312 | /** Unlock resource. */ |
| 1313 | static inline void unlock_res(struct ldlm_resource *res) |
| 1314 | { |
| 1315 | spin_unlock(&res->lr_lock); |
| 1316 | } |
| 1317 | |
| 1318 | /** Check if resource is already locked, assert if not. */ |
| 1319 | static inline void check_res_locked(struct ldlm_resource *res) |
| 1320 | { |
Li Xi | 5e42bc9 | 2014-04-27 13:07:06 -0400 | [diff] [blame] | 1321 | assert_spin_locked(&res->lr_lock); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1322 | } |
| 1323 | |
Greg Donald | 5444261 | 2014-10-20 11:18:12 -0500 | [diff] [blame] | 1324 | struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1325 | void unlock_res_and_lock(struct ldlm_lock *lock); |
| 1326 | |
| 1327 | /* ldlm_pool.c */ |
| 1328 | /** \defgroup ldlm_pools Various LDLM pool related functions |
| 1329 | * There are not used outside of ldlm. |
| 1330 | * @{ |
| 1331 | */ |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1332 | int ldlm_pools_init(void); |
| 1333 | void ldlm_pools_fini(void); |
| 1334 | |
| 1335 | int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns, |
| 1336 | int idx, ldlm_side_t client); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1337 | void ldlm_pool_fini(struct ldlm_pool *pl); |
Peng Tao | d7e09d0 | 2013-05-02 16:46:55 +0800 | [diff] [blame] | 1338 | void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock); |
| 1339 | void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock); |
| 1340 | /** @} */ |
| 1341 | |
| 1342 | #endif |
| 1343 | /** @} LDLM */ |