1 /* 2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.129 2008/10/15 22:38:37 dillon Exp $ 35 */ 36 /* 37 * This header file contains structures used internally by the HAMMERFS 38 * implementation. See hammer_disk.h for on-disk structures. 39 */ 40 41 #include <sys/param.h> 42 #include <sys/types.h> 43 #include <sys/kernel.h> 44 #include <sys/conf.h> 45 #include <sys/systm.h> 46 #include <sys/tree.h> 47 #include <sys/malloc.h> 48 #include <sys/mount.h> 49 #include <sys/mountctl.h> 50 #include <sys/vnode.h> 51 #include <sys/proc.h> 52 #include <sys/stat.h> 53 #include <sys/globaldata.h> 54 #include <sys/lockf.h> 55 #include <sys/buf.h> 56 #include <sys/queue.h> 57 #include <sys/ktr.h> 58 #include <sys/globaldata.h> 59 60 #include <sys/buf2.h> 61 #include <sys/signal2.h> 62 #include "hammer_disk.h" 63 #include "hammer_mount.h" 64 #include "hammer_ioctl.h" 65 66 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES) 67 68 MALLOC_DECLARE(M_HAMMER); 69 MALLOC_DECLARE(M_HAMMER_INO); 70 71 /* 72 * Kernel trace 73 */ 74 #if !defined(KTR_HAMMER) 75 #define KTR_HAMMER KTR_ALL 76 #endif 77 KTR_INFO_MASTER_EXTERN(hammer); 78 79 /* 80 * Misc structures 81 */ 82 struct hammer_mount; 83 84 /* 85 * Key structure used for custom RB tree inode lookups. This prototypes 86 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info). 87 */ 88 typedef struct hammer_inode_info { 89 int64_t obj_id; /* (key) object identifier */ 90 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */ 91 u_int32_t obj_localization; /* (key) pseudo-fs */ 92 union { 93 struct hammer_btree_leaf_elm *leaf; 94 } u; 95 } *hammer_inode_info_t; 96 97 typedef enum hammer_transaction_type { 98 HAMMER_TRANS_RO, 99 HAMMER_TRANS_STD, 100 HAMMER_TRANS_FLS 101 } hammer_transaction_type_t; 102 103 /* 104 * HAMMER Transaction tracking 105 */ 106 struct hammer_transaction { 107 hammer_transaction_type_t type; 108 struct hammer_mount *hmp; 109 hammer_tid_t tid; 110 u_int64_t time; 111 u_int32_t time32; 112 int sync_lock_refs; 113 int flags; 114 struct hammer_volume *rootvol; 115 }; 116 117 typedef struct hammer_transaction *hammer_transaction_t; 118 119 #define HAMMER_TRANSF_NEWINODE 0x0001 120 121 /* 122 * HAMMER locks 123 */ 124 struct hammer_lock { 125 int refs; /* active references delay writes */ 126 int lockcount; /* lock count for exclusive/shared access */ 127 int wanted; 128 int exwanted; /* number of threads waiting for ex lock */ 129 struct thread *locktd; 130 }; 131 132 static __inline int 133 hammer_islocked(struct hammer_lock *lock) 134 { 135 return(lock->lockcount != 0); 136 } 137 138 static __inline int 139 hammer_isactive(struct hammer_lock *lock) 140 { 141 return(lock->refs != 0); 142 } 143 144 static __inline int 145 hammer_islastref(struct hammer_lock *lock) 146 { 147 return(lock->refs == 1); 148 } 149 150 /* 151 * Return if we specifically own the lock exclusively. 152 */ 153 static __inline int 154 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td) 155 { 156 if (lock->lockcount > 0 && lock->locktd == td) 157 return(1); 158 return(0); 159 } 160 161 /* 162 * Flush state, used by various structures 163 */ 164 typedef enum hammer_inode_state { 165 HAMMER_FST_IDLE, 166 HAMMER_FST_SETUP, 167 HAMMER_FST_FLUSH 168 } hammer_inode_state_t; 169 170 TAILQ_HEAD(hammer_record_list, hammer_record); 171 172 /* 173 * Pseudo-filesystem extended data tracking 174 */ 175 struct hammer_pfs_rb_tree; 176 struct hammer_pseudofs_inmem; 177 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem); 178 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node, 179 hammer_pfs_rb_compare, u_int32_t); 180 181 struct hammer_pseudofs_inmem { 182 RB_ENTRY(hammer_pseudofs_inmem) rb_node; 183 struct hammer_lock lock; 184 u_int32_t localization; 185 hammer_tid_t create_tid; 186 int flags; 187 udev_t fsid_udev; 188 struct hammer_pseudofs_data pfsd; 189 }; 190 191 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t; 192 193 #define HAMMER_PFSM_DELETED 0x0001 194 195 /* 196 * Cache object ids. A fixed number of objid cache structures are 197 * created to reserve object id's for newly created files in multiples 198 * of 100,000, localized to a particular directory, and recycled as 199 * needed. This allows parallel create operations in different 200 * directories to retain fairly localized object ids which in turn 201 * improves reblocking performance and layout. 202 */ 203 #define OBJID_CACHE_SIZE 1024 204 #define OBJID_CACHE_BULK 100000 205 206 typedef struct hammer_objid_cache { 207 TAILQ_ENTRY(hammer_objid_cache) entry; 208 struct hammer_inode *dip; 209 hammer_tid_t next_tid; 210 int count; 211 } *hammer_objid_cache_t; 212 213 /* 214 * Associate an inode with a B-Tree node to cache search start positions 215 */ 216 typedef struct hammer_node_cache { 217 TAILQ_ENTRY(hammer_node_cache) entry; 218 struct hammer_node *node; 219 struct hammer_inode *ip; 220 } *hammer_node_cache_t; 221 222 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache); 223 224 /* 225 * Structure used to organize flush groups. Flush groups must be 226 * organized into chunks in order to avoid blowing out the UNDO FIFO. 227 * Without this a 'sync' could end up flushing 50,000 inodes in a single 228 * transaction. 229 */ 230 struct hammer_flush_group { 231 TAILQ_ENTRY(hammer_flush_group) flush_entry; 232 TAILQ_HEAD(, hammer_inode) flush_list; 233 int inode_count; /* inode load */ 234 int total_count; /* record load */ 235 int running; /* group is running */ 236 int closed; 237 int refs; 238 }; 239 240 typedef struct hammer_flush_group *hammer_flush_group_t; 241 242 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group); 243 244 /* 245 * Structure used to represent an inode in-memory. 246 * 247 * The record and data associated with an inode may be out of sync with 248 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag 249 * clear). 250 * 251 * An inode may also hold a cache of unsynchronized records, used for 252 * database and directories only. Unsynchronized regular file data is 253 * stored in the buffer cache. 254 * 255 * NOTE: A file which is created and destroyed within the initial 256 * synchronization period can wind up not doing any disk I/O at all. 257 * 258 * Finally, an inode may cache numerous disk-referencing B-Tree cursors. 259 */ 260 struct hammer_ino_rb_tree; 261 struct hammer_inode; 262 RB_HEAD(hammer_ino_rb_tree, hammer_inode); 263 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node, 264 hammer_ino_rb_compare, hammer_inode_info_t); 265 266 struct hammer_rec_rb_tree; 267 struct hammer_record; 268 RB_HEAD(hammer_rec_rb_tree, hammer_record); 269 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node, 270 hammer_rec_rb_compare, hammer_btree_leaf_elm_t); 271 272 TAILQ_HEAD(hammer_node_list, hammer_node); 273 274 struct hammer_inode { 275 RB_ENTRY(hammer_inode) rb_node; 276 hammer_inode_state_t flush_state; 277 hammer_flush_group_t flush_group; 278 TAILQ_ENTRY(hammer_inode) flush_entry; 279 struct hammer_record_list target_list; /* target of dependant recs */ 280 int64_t obj_id; /* (key) object identifier */ 281 hammer_tid_t obj_asof; /* (key) snapshot or 0 */ 282 u_int32_t obj_localization; /* (key) pseudo-fs */ 283 struct hammer_mount *hmp; 284 hammer_objid_cache_t objid_cache; 285 int flags; 286 int error; /* flush error */ 287 int cursor_ip_refs; /* sanity */ 288 int rsv_recs; 289 struct vnode *vp; 290 hammer_pseudofs_inmem_t pfsm; 291 struct lockf advlock; 292 struct hammer_lock lock; /* sync copy interlock */ 293 off_t trunc_off; 294 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */ 295 struct hammer_inode_data ino_data; /* in-memory cache */ 296 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */ 297 struct hammer_node_cache cache[2]; /* search initiate cache */ 298 299 /* 300 * When a demark is created to synchronize an inode to 301 * disk, certain fields are copied so the front-end VOPs 302 * can continue to run in parallel with the synchronization 303 * occuring in the background. 304 */ 305 int sync_flags; /* to-sync flags cache */ 306 off_t sync_trunc_off; /* to-sync truncation */ 307 off_t save_trunc_off; /* write optimization */ 308 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */ 309 struct hammer_inode_data sync_ino_data; /* to-sync cache */ 310 }; 311 312 typedef struct hammer_inode *hammer_inode_t; 313 314 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data) 315 316 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */ 317 /* (not including atime/mtime) */ 318 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */ 319 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */ 320 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */ 321 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */ 322 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */ 323 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */ 324 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */ 325 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */ 326 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */ 327 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */ 328 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */ 329 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */ 330 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */ 331 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */ 332 333 #define HAMMER_INODE_TRUNCATED 0x00010000 334 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/ 335 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */ 336 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */ 337 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */ 338 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */ 339 340 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \ 341 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \ 342 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \ 343 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING) 344 345 #define HAMMER_INODE_MODMASK_NOXDIRTY \ 346 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY) 347 348 #define HAMMER_FLUSH_GROUP_SIZE 64 349 350 #define HAMMER_FLUSH_SIGNAL 0x0001 351 #define HAMMER_FLUSH_RECURSION 0x0002 352 353 /* 354 * Used by the inode reclaim code to pipeline reclaims and avoid 355 * blowing out kernel memory or letting the flusher get too far 356 * behind. 357 */ 358 struct hammer_reclaim { 359 TAILQ_ENTRY(hammer_reclaim) entry; 360 int okydoky; 361 }; 362 363 #define HAMMER_RECLAIM_FLUSH 2000 364 #define HAMMER_RECLAIM_WAIT 4000 365 366 /* 367 * Structure used to represent an unsynchronized record in-memory. These 368 * records typically represent directory entries. Only non-historical 369 * records are kept in-memory. 370 * 371 * Records are organized as a per-inode RB-Tree. If the inode is not 372 * on disk then neither are any records and the in-memory record tree 373 * represents the entire contents of the inode. If the inode is on disk 374 * then the on-disk B-Tree is scanned in parallel with the in-memory 375 * RB-Tree to synthesize the current state of the file. 376 * 377 * Records are also used to enforce the ordering of directory create/delete 378 * operations. A new inode will not be flushed to disk unless its related 379 * directory entry is also being flushed at the same time. A directory entry 380 * will not be removed unless its related inode is also being removed at the 381 * same time. 382 */ 383 typedef enum hammer_record_type { 384 HAMMER_MEM_RECORD_GENERAL, /* misc record */ 385 HAMMER_MEM_RECORD_INODE, /* inode record */ 386 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */ 387 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */ 388 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */ 389 } hammer_record_type_t; 390 391 struct hammer_record { 392 RB_ENTRY(hammer_record) rb_node; 393 TAILQ_ENTRY(hammer_record) target_entry; 394 hammer_inode_state_t flush_state; 395 hammer_flush_group_t flush_group; 396 hammer_record_type_t type; 397 struct hammer_lock lock; 398 struct hammer_reserve *resv; 399 struct hammer_inode *ip; 400 struct hammer_inode *target_ip; 401 struct hammer_btree_leaf_elm leaf; 402 union hammer_data_ondisk *data; 403 int flags; 404 hammer_off_t zone2_offset; /* direct-write only */ 405 }; 406 407 typedef struct hammer_record *hammer_record_t; 408 409 /* 410 * Record flags. Note that FE can only be set by the frontend if the 411 * record has not been interlocked by the backend w/ BE. 412 */ 413 #define HAMMER_RECF_ALLOCDATA 0x0001 414 #define HAMMER_RECF_ONRBTREE 0x0002 415 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */ 416 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */ 417 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */ 418 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */ 419 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */ 420 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */ 421 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/ 422 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/ 423 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */ 424 425 /* 426 * hammer_delete_at_cursor() flags 427 */ 428 #define HAMMER_DELETE_ADJUST 0x0001 429 #define HAMMER_DELETE_DESTROY 0x0002 430 431 /* 432 * In-memory structures representing on-disk structures. 433 */ 434 struct hammer_volume; 435 struct hammer_buffer; 436 struct hammer_node; 437 struct hammer_undo; 438 struct hammer_reserve; 439 440 RB_HEAD(hammer_vol_rb_tree, hammer_volume); 441 RB_HEAD(hammer_buf_rb_tree, hammer_buffer); 442 RB_HEAD(hammer_nod_rb_tree, hammer_node); 443 RB_HEAD(hammer_und_rb_tree, hammer_undo); 444 RB_HEAD(hammer_res_rb_tree, hammer_reserve); 445 446 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node, 447 hammer_vol_rb_compare, int32_t); 448 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node, 449 hammer_buf_rb_compare, hammer_off_t); 450 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node, 451 hammer_nod_rb_compare, hammer_off_t); 452 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node, 453 hammer_und_rb_compare, hammer_off_t); 454 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node, 455 hammer_res_rb_compare, hammer_off_t); 456 457 /* 458 * IO management - embedded at the head of various in-memory structures 459 * 460 * VOLUME - hammer_volume containing meta-data 461 * META_BUFFER - hammer_buffer containing meta-data 462 * DATA_BUFFER - hammer_buffer containing pure-data 463 * 464 * Dirty volume headers and dirty meta-data buffers are locked until the 465 * flusher can sequence them out. Dirty pure-data buffers can be written. 466 * Clean buffers can be passively released. 467 */ 468 typedef enum hammer_io_type { 469 HAMMER_STRUCTURE_VOLUME, 470 HAMMER_STRUCTURE_META_BUFFER, 471 HAMMER_STRUCTURE_UNDO_BUFFER, 472 HAMMER_STRUCTURE_DATA_BUFFER 473 } hammer_io_type_t; 474 475 union hammer_io_structure; 476 struct hammer_io; 477 478 struct worklist { 479 LIST_ENTRY(worklist) node; 480 }; 481 482 TAILQ_HEAD(hammer_io_list, hammer_io); 483 typedef struct hammer_io_list *hammer_io_list_t; 484 485 struct hammer_io { 486 struct worklist worklist; 487 struct hammer_lock lock; 488 enum hammer_io_type type; 489 struct hammer_mount *hmp; 490 struct hammer_volume *volume; 491 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */ 492 hammer_io_list_t mod_list; 493 struct buf *bp; 494 int64_t offset; /* zone-2 offset */ 495 int bytes; /* buffer cache buffer size */ 496 int loading; /* loading/unloading interlock */ 497 int modify_refs; 498 499 u_int modified : 1; /* bp's data was modified */ 500 u_int released : 1; /* bp released (w/ B_LOCKED set) */ 501 u_int running : 1; /* bp write IO in progress */ 502 u_int waiting : 1; /* someone is waiting on us */ 503 u_int validated : 1; /* ondisk has been validated */ 504 u_int waitdep : 1; /* flush waits for dependancies */ 505 u_int recovered : 1; /* has recovery ref */ 506 u_int waitmod : 1; /* waiting for modify_refs */ 507 u_int reclaim : 1; /* reclaim requested */ 508 u_int gencrc : 1; /* crc needs to be generated */ 509 u_int ioerror : 1; /* abort on io-error */ 510 }; 511 512 typedef struct hammer_io *hammer_io_t; 513 514 #define HAMMER_CLUSTER_SIZE (64 * 1024) 515 #if HAMMER_CLUSTER_SIZE > MAXBSIZE 516 #undef HAMMER_CLUSTER_SIZE 517 #define HAMMER_CLUSTER_SIZE MAXBSIZE 518 #endif 519 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE) 520 521 /* 522 * In-memory volume representing on-disk buffer 523 */ 524 struct hammer_volume { 525 struct hammer_io io; 526 RB_ENTRY(hammer_volume) rb_node; 527 struct hammer_volume_ondisk *ondisk; 528 int32_t vol_no; 529 int64_t nblocks; /* note: special calculation for statfs */ 530 int64_t buffer_base; /* base offset of buffer 0 */ 531 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */ 532 hammer_off_t maxraw_off; /* Maximum raw offset for device */ 533 char *vol_name; 534 struct vnode *devvp; 535 int vol_flags; 536 }; 537 538 typedef struct hammer_volume *hammer_volume_t; 539 540 /* 541 * In-memory buffer (other then volume, super-cluster, or cluster), 542 * representing an on-disk buffer. 543 */ 544 struct hammer_buffer { 545 struct hammer_io io; 546 RB_ENTRY(hammer_buffer) rb_node; 547 void *ondisk; 548 hammer_off_t zoneX_offset; 549 hammer_off_t zone2_offset; 550 struct hammer_reserve *resv; 551 struct hammer_node_list clist; 552 }; 553 554 typedef struct hammer_buffer *hammer_buffer_t; 555 556 /* 557 * In-memory B-Tree node, representing an on-disk B-Tree node. 558 * 559 * This is a hang-on structure which is backed by a hammer_buffer, 560 * indexed by a hammer_cluster, and used for fine-grained locking of 561 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer 562 * can contain multiple nodes representing wildly disassociated portions 563 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis. 564 * 565 * This structure uses a cluster-relative index to reduce the number 566 * of layers required to access it, and also because all on-disk B-Tree 567 * references are cluster-relative offsets. 568 */ 569 struct hammer_node { 570 struct hammer_lock lock; /* node-by-node lock */ 571 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */ 572 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */ 573 hammer_off_t node_offset; /* full offset spec */ 574 struct hammer_mount *hmp; 575 struct hammer_buffer *buffer; /* backing buffer */ 576 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */ 577 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */ 578 struct hammer_node_cache_list cache_list; /* passive caches */ 579 int flags; 580 int loading; /* load interlock */ 581 }; 582 583 #define HAMMER_NODE_DELETED 0x0001 584 #define HAMMER_NODE_FLUSH 0x0002 585 #define HAMMER_NODE_CRCGOOD 0x0004 586 #define HAMMER_NODE_NEEDSCRC 0x0008 587 #define HAMMER_NODE_NEEDSMIRROR 0x0010 588 589 typedef struct hammer_node *hammer_node_t; 590 591 /* 592 * List of locked nodes. 593 */ 594 struct hammer_node_locklist { 595 struct hammer_node_locklist *next; 596 hammer_node_t node; 597 }; 598 599 typedef struct hammer_node_locklist *hammer_node_locklist_t; 600 601 602 /* 603 * Common I/O management structure - embedded in in-memory structures 604 * which are backed by filesystem buffers. 605 */ 606 union hammer_io_structure { 607 struct hammer_io io; 608 struct hammer_volume volume; 609 struct hammer_buffer buffer; 610 }; 611 612 typedef union hammer_io_structure *hammer_io_structure_t; 613 614 /* 615 * The reserve structure prevents the blockmap from allocating 616 * out of a reserved bigblock. Such reservations are used by 617 * the direct-write mechanism. 618 * 619 * The structure is also used to hold off on reallocations of 620 * big blocks from the freemap until flush dependancies have 621 * been dealt with. 622 */ 623 struct hammer_reserve { 624 RB_ENTRY(hammer_reserve) rb_node; 625 TAILQ_ENTRY(hammer_reserve) delay_entry; 626 int flush_group; 627 int flags; 628 int refs; 629 int zone; 630 int bytes_freed; 631 int append_off; 632 hammer_off_t zone_offset; 633 }; 634 635 typedef struct hammer_reserve *hammer_reserve_t; 636 637 #define HAMMER_RESF_ONDELAY 0x0001 638 639 #include "hammer_cursor.h" 640 641 /* 642 * The undo structure tracks recent undos to avoid laying down duplicate 643 * undos within a flush group, saving us a significant amount of overhead. 644 * 645 * This is strictly a heuristic. 646 */ 647 #define HAMMER_MAX_UNDOS 1024 648 #define HAMMER_MAX_FLUSHERS 4 649 650 struct hammer_undo { 651 RB_ENTRY(hammer_undo) rb_node; 652 TAILQ_ENTRY(hammer_undo) lru_entry; 653 hammer_off_t offset; 654 int bytes; 655 }; 656 657 typedef struct hammer_undo *hammer_undo_t; 658 659 struct hammer_flusher_info; 660 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info); 661 662 struct hammer_flusher { 663 int signal; /* flusher thread sequencer */ 664 int act; /* currently active flush group */ 665 int done; /* set to act when complete */ 666 int next; /* next flush group */ 667 int group_lock; /* lock sequencing of the next flush */ 668 int exiting; /* request master exit */ 669 thread_t td; /* master flusher thread */ 670 hammer_tid_t tid; /* last flushed transaction id */ 671 int finalize_want; /* serialize finalization */ 672 struct hammer_lock finalize_lock; /* serialize finalization */ 673 struct hammer_transaction trans; /* shared transaction */ 674 struct hammer_flusher_info_list run_list; 675 struct hammer_flusher_info_list ready_list; 676 }; 677 678 /* 679 * Internal hammer mount data structure 680 */ 681 struct hammer_mount { 682 struct mount *mp; 683 /*struct vnode *rootvp;*/ 684 struct hammer_ino_rb_tree rb_inos_root; 685 struct hammer_vol_rb_tree rb_vols_root; 686 struct hammer_nod_rb_tree rb_nods_root; 687 struct hammer_und_rb_tree rb_undo_root; 688 struct hammer_res_rb_tree rb_resv_root; 689 struct hammer_buf_rb_tree rb_bufs_root; 690 struct hammer_pfs_rb_tree rb_pfsm_root; 691 struct hammer_volume *rootvol; 692 struct hammer_base_elm root_btree_beg; 693 struct hammer_base_elm root_btree_end; 694 int flags; /* HAMMER_MOUNT_xxx flags */ 695 int hflags; 696 int ronly; 697 int nvolumes; 698 int volume_iterator; 699 int master_id; /* -1 or 0-15 - clustering and mirroring */ 700 int rsv_inodes; /* reserved space due to dirty inodes */ 701 int64_t rsv_databytes; /* reserved space due to record data */ 702 int rsv_recs; /* reserved space due to dirty records */ 703 int rsv_fromdelay; /* bigblocks reserved due to flush delay */ 704 int undo_rec_limit; /* based on size of undo area */ 705 int last_newrecords; 706 int count_newrecords; 707 708 int inode_reclaims; /* inodes pending reclaim by flusher */ 709 int count_inodes; /* total number of inodes */ 710 int count_iqueued; /* inodes queued to flusher */ 711 712 struct hammer_flusher flusher; 713 714 u_int check_interrupt; 715 uuid_t fsid; 716 struct hammer_io_list volu_list; /* dirty undo buffers */ 717 struct hammer_io_list undo_list; /* dirty undo buffers */ 718 struct hammer_io_list data_list; /* dirty data buffers */ 719 struct hammer_io_list alt_data_list; /* dirty data buffers */ 720 struct hammer_io_list meta_list; /* dirty meta bufs */ 721 struct hammer_io_list lose_list; /* loose buffers */ 722 int locked_dirty_space; /* meta/volu count */ 723 int io_running_space; 724 int objid_cache_count; 725 int error; /* critical I/O error */ 726 struct krate krate; /* rate limited kprintf */ 727 hammer_tid_t asof; /* snapshot mount */ 728 hammer_tid_t next_tid; 729 hammer_tid_t flush_tid1; /* flusher tid sequencing */ 730 hammer_tid_t flush_tid2; /* flusher tid sequencing */ 731 int64_t copy_stat_freebigblocks; /* number of free bigblocks */ 732 733 u_int32_t namekey_iterator; 734 struct netexport export; 735 struct hammer_lock sync_lock; 736 struct hammer_lock free_lock; 737 struct hammer_lock undo_lock; 738 struct hammer_lock blkmap_lock; 739 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES]; 740 struct hammer_undo undos[HAMMER_MAX_UNDOS]; 741 int undo_alloc; 742 TAILQ_HEAD(, hammer_undo) undo_lru_list; 743 TAILQ_HEAD(, hammer_reserve) delay_list; 744 struct hammer_flush_group_list flush_group_list; 745 hammer_flush_group_t next_flush_group; 746 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list; 747 TAILQ_HEAD(, hammer_reclaim) reclaim_list; 748 }; 749 750 typedef struct hammer_mount *hammer_mount_t; 751 752 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001 753 754 struct hammer_sync_info { 755 int error; 756 int waitfor; 757 }; 758 759 #endif 760 761 /* 762 * checkspace slop (8MB chunks), higher numbers are more conservative. 763 */ 764 #define HAMMER_CHKSPC_REBLOCK 25 765 #define HAMMER_CHKSPC_MIRROR 20 766 #define HAMMER_CHKSPC_WRITE 20 767 #define HAMMER_CHKSPC_CREATE 20 768 #define HAMMER_CHKSPC_REMOVE 10 769 #define HAMMER_CHKSPC_EMERGENCY 0 770 771 #if defined(_KERNEL) 772 773 extern struct vop_ops hammer_vnode_vops; 774 extern struct vop_ops hammer_spec_vops; 775 extern struct vop_ops hammer_fifo_vops; 776 extern struct bio_ops hammer_bioops; 777 778 extern int hammer_debug_io; 779 extern int hammer_debug_general; 780 extern int hammer_debug_debug; 781 extern int hammer_debug_inode; 782 extern int hammer_debug_locks; 783 extern int hammer_debug_btree; 784 extern int hammer_debug_tid; 785 extern int hammer_debug_recover; 786 extern int hammer_debug_recover_faults; 787 extern int hammer_cluster_enable; 788 extern int hammer_count_fsyncs; 789 extern int hammer_count_inodes; 790 extern int hammer_count_iqueued; 791 extern int hammer_count_reclaiming; 792 extern int hammer_count_records; 793 extern int hammer_count_record_datas; 794 extern int hammer_count_volumes; 795 extern int hammer_count_buffers; 796 extern int hammer_count_nodes; 797 extern int64_t hammer_count_extra_space_used; 798 extern int64_t hammer_stats_btree_lookups; 799 extern int64_t hammer_stats_btree_searches; 800 extern int64_t hammer_stats_btree_inserts; 801 extern int64_t hammer_stats_btree_deletes; 802 extern int64_t hammer_stats_btree_elements; 803 extern int64_t hammer_stats_btree_splits; 804 extern int64_t hammer_stats_btree_iterations; 805 extern int64_t hammer_stats_record_iterations; 806 extern int64_t hammer_stats_file_read; 807 extern int64_t hammer_stats_file_write; 808 extern int64_t hammer_stats_file_iopsr; 809 extern int64_t hammer_stats_file_iopsw; 810 extern int64_t hammer_stats_disk_read; 811 extern int64_t hammer_stats_disk_write; 812 extern int64_t hammer_stats_inode_flushes; 813 extern int64_t hammer_stats_commits; 814 extern int hammer_count_dirtybufspace; 815 extern int hammer_count_refedbufs; 816 extern int hammer_count_reservations; 817 extern int hammer_count_io_running_read; 818 extern int hammer_count_io_running_write; 819 extern int hammer_count_io_locked; 820 extern int hammer_limit_dirtybufspace; 821 extern int hammer_limit_iqueued; 822 extern int hammer_limit_recs; 823 extern int hammer_bio_count; 824 extern int hammer_verify_zone; 825 extern int hammer_verify_data; 826 extern int hammer_write_mode; 827 extern int hammer_autoflush; 828 extern int64_t hammer_contention_count; 829 830 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 831 int error, const char *msg); 832 int hammer_vop_inactive(struct vop_inactive_args *); 833 int hammer_vop_reclaim(struct vop_reclaim_args *); 834 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp); 835 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans, 836 hammer_inode_t dip, int64_t obj_id, 837 hammer_tid_t asof, u_int32_t localization, 838 int flags, int *errorp); 839 void hammer_scan_inode_snapshots(hammer_mount_t hmp, 840 hammer_inode_info_t iinfo, 841 int (*callback)(hammer_inode_t ip, void *data), 842 void *data); 843 void hammer_put_inode(struct hammer_inode *ip); 844 void hammer_put_inode_ref(struct hammer_inode *ip); 845 void hammer_inode_waitreclaims(hammer_mount_t hmp); 846 847 int hammer_unload_volume(hammer_volume_t volume, void *data __unused); 848 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused); 849 850 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused); 851 int hammer_install_volume(hammer_mount_t hmp, const char *volname, 852 struct vnode *devvp); 853 int hammer_mountcheck_volumes(hammer_mount_t hmp); 854 855 int hammer_mem_add(hammer_record_t record); 856 int hammer_ip_lookup(hammer_cursor_t cursor); 857 int hammer_ip_first(hammer_cursor_t cursor); 858 int hammer_ip_next(hammer_cursor_t cursor); 859 int hammer_ip_resolve_data(hammer_cursor_t cursor); 860 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip, 861 hammer_tid_t tid); 862 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags, 863 hammer_tid_t delete_tid, u_int32_t delete_ts, 864 int track, int64_t *stat_bytes); 865 int hammer_ip_check_directory_empty(hammer_transaction_t trans, 866 hammer_inode_t ip); 867 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor); 868 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor); 869 870 hammer_record_t 871 hammer_alloc_mem_record(hammer_inode_t ip, int data_len); 872 void hammer_flush_record_done(hammer_record_t record, int error); 873 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident); 874 void hammer_rel_mem_record(hammer_record_t record); 875 876 int hammer_cursor_up(hammer_cursor_t cursor); 877 int hammer_cursor_up_locked(hammer_cursor_t cursor); 878 int hammer_cursor_down(hammer_cursor_t cursor); 879 int hammer_cursor_upgrade(hammer_cursor_t cursor); 880 int hammer_cursor_upgrade_node(hammer_cursor_t cursor); 881 void hammer_cursor_downgrade(hammer_cursor_t cursor); 882 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, 883 int index); 884 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident); 885 int hammer_lock_ex_try(struct hammer_lock *lock); 886 void hammer_lock_sh(struct hammer_lock *lock); 887 int hammer_lock_sh_try(struct hammer_lock *lock); 888 int hammer_lock_upgrade(struct hammer_lock *lock); 889 void hammer_lock_downgrade(struct hammer_lock *lock); 890 int hammer_lock_status(struct hammer_lock *lock); 891 void hammer_unlock(struct hammer_lock *lock); 892 void hammer_ref(struct hammer_lock *lock); 893 void hammer_unref(struct hammer_lock *lock); 894 895 void hammer_sync_lock_ex(hammer_transaction_t trans); 896 void hammer_sync_lock_sh(hammer_transaction_t trans); 897 int hammer_sync_lock_sh_try(hammer_transaction_t trans); 898 void hammer_sync_unlock(hammer_transaction_t trans); 899 900 u_int32_t hammer_to_unix_xid(uuid_t *uuid); 901 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid); 902 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts); 903 u_int64_t hammer_timespec_to_time(struct timespec *ts); 904 int hammer_str_to_tid(const char *str, int *ispfsp, 905 hammer_tid_t *tidp, u_int32_t *localizationp); 906 int hammer_is_atatext(const char *name, int len); 907 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip); 908 void hammer_clear_objid(hammer_inode_t dip); 909 void hammer_destroy_objid_cache(hammer_mount_t hmp); 910 911 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset, 912 int bytes); 913 void hammer_clear_undo_history(hammer_mount_t hmp); 914 enum vtype hammer_get_vnode_type(u_int8_t obj_type); 915 int hammer_get_dtype(u_int8_t obj_type); 916 u_int8_t hammer_get_obj_type(enum vtype vtype); 917 int64_t hammer_directory_namekey(const void *name, int len); 918 int hammer_nohistory(hammer_inode_t ip); 919 920 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor, 921 hammer_node_cache_t cache, hammer_inode_t ip); 922 void hammer_normalize_cursor(hammer_cursor_t cursor); 923 void hammer_done_cursor(hammer_cursor_t cursor); 924 int hammer_recover_cursor(hammer_cursor_t cursor); 925 void hammer_unlock_cursor(hammer_cursor_t cursor, int also_ip); 926 int hammer_lock_cursor(hammer_cursor_t cursor, int also_ip); 927 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor); 928 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor); 929 930 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode); 931 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent, 932 int index); 933 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode, 934 int index); 935 void hammer_cursor_inserted_element(hammer_node_t node, int index); 936 void hammer_cursor_deleted_element(hammer_node_t node, int index); 937 938 int hammer_btree_lookup(hammer_cursor_t cursor); 939 int hammer_btree_first(hammer_cursor_t cursor); 940 int hammer_btree_last(hammer_cursor_t cursor); 941 int hammer_btree_extract(hammer_cursor_t cursor, int flags); 942 int hammer_btree_iterate(hammer_cursor_t cursor); 943 int hammer_btree_iterate_reverse(hammer_cursor_t cursor); 944 int hammer_btree_insert(hammer_cursor_t cursor, 945 hammer_btree_leaf_elm_t elm, int *doprop); 946 int hammer_btree_delete(hammer_cursor_t cursor); 947 void hammer_btree_do_propagation(hammer_cursor_t cursor, 948 hammer_pseudofs_inmem_t pfsm, 949 hammer_btree_leaf_elm_t leaf); 950 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2); 951 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key); 952 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid); 953 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid); 954 955 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node, 956 hammer_btree_elm_t elm); 957 int hammer_btree_lock_children(hammer_cursor_t cursor, 958 struct hammer_node_locklist **locklistp); 959 void hammer_btree_unlock_children(struct hammer_node_locklist **locklistp); 960 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node); 961 hammer_node_t hammer_btree_get_parent(hammer_node_t node, int *parent_indexp, 962 int *errorp, int try_exclusive); 963 964 void hammer_print_btree_node(hammer_node_ondisk_t ondisk); 965 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i); 966 967 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off, 968 int *errorp, struct hammer_buffer **bufferp); 969 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off, 970 int *errorp, struct hammer_buffer **bufferp); 971 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes, 972 int *errorp, struct hammer_buffer **bufferp); 973 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes, 974 int *errorp, struct hammer_buffer **bufferp); 975 976 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp); 977 978 hammer_volume_t hammer_get_volume(hammer_mount_t hmp, 979 int32_t vol_no, int *errorp); 980 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset, 981 int bytes, int isnew, int *errorp); 982 void hammer_sync_buffers(hammer_mount_t hmp, 983 hammer_off_t base_offset, int bytes); 984 void hammer_del_buffers(hammer_mount_t hmp, 985 hammer_off_t base_offset, 986 hammer_off_t zone2_offset, int bytes); 987 988 int hammer_ref_volume(hammer_volume_t volume); 989 int hammer_ref_buffer(hammer_buffer_t buffer); 990 void hammer_flush_buffer_nodes(hammer_buffer_t buffer); 991 992 void hammer_rel_volume(hammer_volume_t volume, int flush); 993 void hammer_rel_buffer(hammer_buffer_t buffer, int flush); 994 995 int hammer_vfs_export(struct mount *mp, int op, 996 const struct export_args *export); 997 hammer_node_t hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset, 998 int isnew, int *errorp); 999 void hammer_ref_node(hammer_node_t node); 1000 hammer_node_t hammer_ref_node_safe(struct hammer_mount *hmp, 1001 hammer_node_cache_t cache, int *errorp); 1002 void hammer_rel_node(hammer_node_t node); 1003 void hammer_delete_node(hammer_transaction_t trans, 1004 hammer_node_t node); 1005 void hammer_cache_node(hammer_node_cache_t cache, 1006 hammer_node_t node); 1007 void hammer_uncache_node(hammer_node_cache_t cache); 1008 void hammer_flush_node(hammer_node_t node); 1009 1010 void hammer_dup_buffer(struct hammer_buffer **bufferp, 1011 struct hammer_buffer *buffer); 1012 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp); 1013 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len, 1014 u_int16_t rec_type, hammer_off_t *data_offsetp, 1015 struct hammer_buffer **data_bufferp, int *errorp); 1016 1017 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io, 1018 hammer_off_t zone1_offset, void *base, int len); 1019 1020 void hammer_put_volume(struct hammer_volume *volume, int flush); 1021 void hammer_put_buffer(struct hammer_buffer *buffer, int flush); 1022 1023 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans, 1024 hammer_off_t owner, int *errorp); 1025 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset, 1026 hammer_off_t owner, int *errorp); 1027 int hammer_checkspace(hammer_mount_t hmp, int slop); 1028 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone, 1029 int bytes, int *errorp); 1030 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone, 1031 int bytes, hammer_off_t *zone_offp, int *errorp); 1032 void hammer_blockmap_reserve_undo(hammer_reserve_t resv, 1033 hammer_off_t zone_offset, int bytes); 1034 void hammer_blockmap_reserve_complete(hammer_mount_t hmp, 1035 hammer_reserve_t resv); 1036 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv); 1037 void hammer_blockmap_free(hammer_transaction_t trans, 1038 hammer_off_t bmap_off, int bytes); 1039 int hammer_blockmap_finalize(hammer_transaction_t trans, 1040 hammer_off_t bmap_off, int bytes); 1041 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off, 1042 int *curp, int *errorp); 1043 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, 1044 int *errorp); 1045 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, 1046 int *errorp); 1047 int64_t hammer_undo_used(hammer_transaction_t trans); 1048 int64_t hammer_undo_space(hammer_transaction_t trans); 1049 int64_t hammer_undo_max(hammer_mount_t hmp); 1050 1051 void hammer_start_transaction(struct hammer_transaction *trans, 1052 struct hammer_mount *hmp); 1053 void hammer_simple_transaction(struct hammer_transaction *trans, 1054 struct hammer_mount *hmp); 1055 void hammer_start_transaction_fls(struct hammer_transaction *trans, 1056 struct hammer_mount *hmp); 1057 void hammer_done_transaction(struct hammer_transaction *trans); 1058 1059 void hammer_modify_inode(hammer_inode_t ip, int flags); 1060 void hammer_flush_inode(hammer_inode_t ip, int flags); 1061 void hammer_flush_inode_done(hammer_inode_t ip, int error); 1062 void hammer_wait_inode(hammer_inode_t ip); 1063 1064 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap, 1065 struct ucred *cred, struct hammer_inode *dip, 1066 hammer_pseudofs_inmem_t pfsm, 1067 struct hammer_inode **ipp); 1068 void hammer_rel_inode(hammer_inode_t ip, int flush); 1069 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused); 1070 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2); 1071 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused); 1072 1073 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip); 1074 void hammer_test_inode(hammer_inode_t dip); 1075 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp); 1076 1077 int hammer_ip_add_directory(struct hammer_transaction *trans, 1078 hammer_inode_t dip, const char *name, int bytes, 1079 hammer_inode_t nip); 1080 int hammer_ip_del_directory(struct hammer_transaction *trans, 1081 hammer_cursor_t cursor, hammer_inode_t dip, 1082 hammer_inode_t ip); 1083 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset, 1084 void *data, int bytes, int *errorp); 1085 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size); 1086 int hammer_ip_add_record(struct hammer_transaction *trans, 1087 hammer_record_t record); 1088 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip, 1089 int64_t ran_beg, int64_t ran_end, int truncating); 1090 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip, 1091 int *countp); 1092 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip, 1093 int64_t offset, void *data, int bytes); 1094 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec); 1095 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec); 1096 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans, 1097 u_int32_t localization, int *errorp); 1098 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred, 1099 hammer_pseudofs_inmem_t pfsm); 1100 int hammer_save_pseudofs(hammer_transaction_t trans, 1101 hammer_pseudofs_inmem_t pfsm); 1102 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization); 1103 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm); 1104 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag, 1105 struct ucred *cred); 1106 1107 void hammer_io_init(hammer_io_t io, hammer_volume_t volume, 1108 enum hammer_io_type type); 1109 int hammer_io_read(struct vnode *devvp, struct hammer_io *io, 1110 hammer_off_t limit); 1111 int hammer_io_new(struct vnode *devvp, struct hammer_io *io); 1112 void hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset); 1113 struct buf *hammer_io_release(struct hammer_io *io, int flush); 1114 void hammer_io_flush(struct hammer_io *io); 1115 void hammer_io_wait(struct hammer_io *io); 1116 void hammer_io_waitdep(struct hammer_io *io); 1117 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident); 1118 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio, 1119 hammer_btree_leaf_elm_t leaf); 1120 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record, 1121 struct bio *bio); 1122 void hammer_io_direct_wait(hammer_record_t record); 1123 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf); 1124 void hammer_io_write_interlock(hammer_io_t io); 1125 void hammer_io_done_interlock(hammer_io_t io); 1126 void hammer_io_clear_modify(struct hammer_io *io, int inval); 1127 void hammer_io_clear_modlist(struct hammer_io *io); 1128 void hammer_io_flush_sync(hammer_mount_t hmp); 1129 1130 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume, 1131 void *base, int len); 1132 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer, 1133 void *base, int len); 1134 void hammer_modify_volume_done(hammer_volume_t volume); 1135 void hammer_modify_buffer_done(hammer_buffer_t buffer); 1136 1137 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip, 1138 struct hammer_ioc_reblock *reblock); 1139 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip, 1140 struct hammer_ioc_prune *prune); 1141 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip, 1142 struct hammer_ioc_mirror_rw *mirror); 1143 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip, 1144 struct hammer_ioc_mirror_rw *mirror); 1145 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1146 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs); 1147 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1148 struct hammer_ioc_pseudofs_rw *pfs); 1149 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1150 struct hammer_ioc_pseudofs_rw *pfs); 1151 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1152 struct hammer_ioc_pseudofs_rw *pfs); 1153 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1154 struct hammer_ioc_pseudofs_rw *pfs); 1155 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1156 struct hammer_ioc_pseudofs_rw *pfs); 1157 1158 int hammer_signal_check(hammer_mount_t hmp); 1159 1160 void hammer_flusher_create(hammer_mount_t hmp); 1161 void hammer_flusher_destroy(hammer_mount_t hmp); 1162 void hammer_flusher_sync(hammer_mount_t hmp); 1163 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg); 1164 int hammer_flusher_async_one(hammer_mount_t hmp); 1165 void hammer_flusher_wait(hammer_mount_t hmp, int seq); 1166 int hammer_flusher_meta_limit(hammer_mount_t hmp); 1167 int hammer_flusher_meta_halflimit(hammer_mount_t hmp); 1168 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter); 1169 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp); 1170 void hammer_flusher_finalize(hammer_transaction_t trans, int final); 1171 int hammer_flusher_haswork(hammer_mount_t hmp); 1172 1173 1174 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol); 1175 void hammer_recover_flush_buffers(hammer_mount_t hmp, 1176 hammer_volume_t root_volume, int final); 1177 1178 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap); 1179 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk); 1180 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf); 1181 1182 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap); 1183 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk); 1184 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk); 1185 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf); 1186 void hkprintf(const char *ctl, ...); 1187 udev_t hammer_fsid_to_udev(uuid_t *uuid); 1188 1189 1190 int hammer_blocksize(int64_t file_offset); 1191 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2); 1192 1193 #endif 1194 1195 static __inline void 1196 hammer_wait_mem_record(hammer_record_t record) 1197 { 1198 hammer_wait_mem_record_ident(record, "hmmwai"); 1199 } 1200 1201 static __inline void 1202 hammer_lock_ex(struct hammer_lock *lock) 1203 { 1204 hammer_lock_ex_ident(lock, "hmrlck"); 1205 } 1206 1207 /* 1208 * Indicate that a B-Tree node is being modified. 1209 */ 1210 static __inline void 1211 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node) 1212 { 1213 hammer_modify_buffer(trans, node->buffer, NULL, 0); 1214 } 1215 1216 static __inline void 1217 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node) 1218 { 1219 hammer_modify_buffer(trans, node->buffer, 1220 node->ondisk, sizeof(*node->ondisk)); 1221 } 1222 1223 static __inline void 1224 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node, 1225 void *base, int len) 1226 { 1227 hammer_crc_t *crcptr; 1228 1229 KKASSERT((char *)base >= (char *)node->ondisk && 1230 (char *)base + len <= 1231 (char *)node->ondisk + sizeof(*node->ondisk)); 1232 hammer_modify_buffer(trans, node->buffer, base, len); 1233 crcptr = &node->ondisk->crc; 1234 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t)); 1235 --node->buffer->io.modify_refs; /* only want one ref */ 1236 } 1237 1238 /* 1239 * Indicate that the specified modifications have been completed. 1240 * 1241 * Do not try to generate the crc here, it's very expensive to do and a 1242 * sequence of insertions or deletions can result in many calls to this 1243 * function on the same node. 1244 */ 1245 static __inline void 1246 hammer_modify_node_done(hammer_node_t node) 1247 { 1248 node->flags |= HAMMER_NODE_CRCGOOD; 1249 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) { 1250 node->flags |= HAMMER_NODE_NEEDSCRC; 1251 node->buffer->io.gencrc = 1; 1252 hammer_ref_node(node); 1253 } 1254 hammer_modify_buffer_done(node->buffer); 1255 } 1256 1257 #define hammer_modify_volume_field(trans, vol, field) \ 1258 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \ 1259 sizeof((vol)->ondisk->field)) 1260 1261 #define hammer_modify_node_field(trans, node, field) \ 1262 hammer_modify_node(trans, node, &(node)->ondisk->field, \ 1263 sizeof((node)->ondisk->field)) 1264 1265