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 1.130 2008/11/13 02:18:43 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/priv.h> 53 #include <sys/stat.h> 54 #include <sys/globaldata.h> 55 #include <sys/lockf.h> 56 #include <sys/buf.h> 57 #include <sys/queue.h> 58 #include <sys/ktr.h> 59 #include <sys/globaldata.h> 60 61 #include <sys/buf2.h> 62 #include <sys/signal2.h> 63 #include "hammer_disk.h" 64 #include "hammer_mount.h" 65 #include "hammer_ioctl.h" 66 67 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES) 68 69 MALLOC_DECLARE(M_HAMMER); 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 #define HAMMER_TRANSF_DIDIO 0x0002 121 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */ 122 123 /* 124 * HAMMER locks 125 */ 126 struct hammer_lock { 127 int refs; /* active references delay writes */ 128 volatile u_int lockval; /* lock count and control bits */ 129 struct thread *owner; /* owner if exclusively held */ 130 }; 131 132 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000 133 #define HAMMER_LOCKF_WANTED 0x80000000 134 135 static __inline int 136 hammer_notlocked(struct hammer_lock *lock) 137 { 138 return(lock->lockval == 0); 139 } 140 141 static __inline int 142 hammer_islocked(struct hammer_lock *lock) 143 { 144 return(lock->lockval != 0); 145 } 146 147 static __inline int 148 hammer_isactive(struct hammer_lock *lock) 149 { 150 return(lock->refs != 0); 151 } 152 153 static __inline int 154 hammer_islastref(struct hammer_lock *lock) 155 { 156 return(lock->refs == 1); 157 } 158 159 /* 160 * Return if we specifically own the lock exclusively. 161 */ 162 static __inline int 163 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td) 164 { 165 if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) && 166 lock->owner == td) { 167 return(1); 168 } 169 return(0); 170 } 171 172 /* 173 * Flush state, used by various structures 174 */ 175 typedef enum hammer_inode_state { 176 HAMMER_FST_IDLE, 177 HAMMER_FST_SETUP, 178 HAMMER_FST_FLUSH 179 } hammer_inode_state_t; 180 181 TAILQ_HEAD(hammer_record_list, hammer_record); 182 183 /* 184 * Pseudo-filesystem extended data tracking 185 */ 186 struct hammer_pfs_rb_tree; 187 struct hammer_pseudofs_inmem; 188 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem); 189 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node, 190 hammer_pfs_rb_compare, u_int32_t); 191 192 struct hammer_pseudofs_inmem { 193 RB_ENTRY(hammer_pseudofs_inmem) rb_node; 194 struct hammer_lock lock; 195 u_int32_t localization; 196 hammer_tid_t create_tid; 197 int flags; 198 udev_t fsid_udev; 199 struct hammer_pseudofs_data pfsd; 200 }; 201 202 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t; 203 204 #define HAMMER_PFSM_DELETED 0x0001 205 206 /* 207 * Cache object ids. A fixed number of objid cache structures are 208 * created to reserve object id's for newly created files in multiples 209 * of 100,000, localized to a particular directory, and recycled as 210 * needed. This allows parallel create operations in different 211 * directories to retain fairly localized object ids which in turn 212 * improves reblocking performance and layout. 213 */ 214 #define OBJID_CACHE_SIZE 1024 215 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */ 216 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */ 217 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1) 218 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1)) 219 220 typedef struct hammer_objid_cache { 221 TAILQ_ENTRY(hammer_objid_cache) entry; 222 struct hammer_inode *dip; 223 hammer_tid_t base_tid; 224 int count; 225 u_int32_t bm0; 226 u_int32_t bm1[32]; 227 } *hammer_objid_cache_t; 228 229 /* 230 * Associate an inode with a B-Tree node to cache search start positions 231 */ 232 typedef struct hammer_node_cache { 233 TAILQ_ENTRY(hammer_node_cache) entry; 234 struct hammer_node *node; 235 struct hammer_inode *ip; 236 } *hammer_node_cache_t; 237 238 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache); 239 240 /* 241 * Structure used to organize flush groups. Flush groups must be 242 * organized into chunks in order to avoid blowing out the UNDO FIFO. 243 * Without this a 'sync' could end up flushing 50,000 inodes in a single 244 * transaction. 245 */ 246 struct hammer_flush_group { 247 TAILQ_ENTRY(hammer_flush_group) flush_entry; 248 TAILQ_HEAD(, hammer_inode) flush_list; 249 int unused01; /* inode load */ 250 int total_count; /* record load */ 251 int running; /* group is running */ 252 int closed; 253 int refs; 254 }; 255 256 typedef struct hammer_flush_group *hammer_flush_group_t; 257 258 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group); 259 260 /* 261 * Structure used to represent an inode in-memory. 262 * 263 * The record and data associated with an inode may be out of sync with 264 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag 265 * clear). 266 * 267 * An inode may also hold a cache of unsynchronized records, used for 268 * database and directories only. Unsynchronized regular file data is 269 * stored in the buffer cache. 270 * 271 * NOTE: A file which is created and destroyed within the initial 272 * synchronization period can wind up not doing any disk I/O at all. 273 * 274 * Finally, an inode may cache numerous disk-referencing B-Tree cursors. 275 */ 276 struct hammer_ino_rb_tree; 277 struct hammer_inode; 278 RB_HEAD(hammer_ino_rb_tree, hammer_inode); 279 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node, 280 hammer_ino_rb_compare, hammer_inode_info_t); 281 282 struct hammer_rec_rb_tree; 283 struct hammer_record; 284 RB_HEAD(hammer_rec_rb_tree, hammer_record); 285 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node, 286 hammer_rec_rb_compare, hammer_btree_leaf_elm_t); 287 288 TAILQ_HEAD(hammer_node_list, hammer_node); 289 290 struct hammer_inode { 291 RB_ENTRY(hammer_inode) rb_node; 292 hammer_inode_state_t flush_state; 293 hammer_flush_group_t flush_group; 294 TAILQ_ENTRY(hammer_inode) flush_entry; 295 struct hammer_record_list target_list; /* target of dependant recs */ 296 int64_t obj_id; /* (key) object identifier */ 297 hammer_tid_t obj_asof; /* (key) snapshot or 0 */ 298 u_int32_t obj_localization; /* (key) pseudo-fs */ 299 struct hammer_mount *hmp; 300 hammer_objid_cache_t objid_cache; 301 int flags; 302 int error; /* flush error */ 303 int cursor_ip_refs; /* sanity */ 304 int rsv_recs; 305 struct vnode *vp; 306 hammer_pseudofs_inmem_t pfsm; 307 struct lockf advlock; 308 struct hammer_lock lock; /* sync copy interlock */ 309 off_t trunc_off; 310 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */ 311 struct hammer_inode_data ino_data; /* in-memory cache */ 312 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */ 313 int rec_generation; 314 struct hammer_node_cache cache[4]; /* search initiate cache */ 315 316 /* 317 * When a demark is created to synchronize an inode to 318 * disk, certain fields are copied so the front-end VOPs 319 * can continue to run in parallel with the synchronization 320 * occuring in the background. 321 */ 322 int sync_flags; /* to-sync flags cache */ 323 off_t sync_trunc_off; /* to-sync truncation */ 324 off_t save_trunc_off; /* write optimization */ 325 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */ 326 struct hammer_inode_data sync_ino_data; /* to-sync cache */ 327 }; 328 329 typedef struct hammer_inode *hammer_inode_t; 330 331 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data) 332 333 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */ 334 /* (not including atime/mtime) */ 335 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */ 336 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */ 337 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */ 338 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */ 339 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */ 340 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */ 341 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */ 342 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */ 343 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */ 344 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */ 345 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */ 346 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */ 347 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */ 348 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */ 349 350 #define HAMMER_INODE_TRUNCATED 0x00010000 351 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/ 352 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */ 353 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */ 354 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */ 355 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */ 356 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */ 357 #define HAMMER_INODE_CLOSESYNC 0x01000000 /* synchronously fsync on close */ 358 #define HAMMER_INODE_CLOSEASYNC 0x02000000 /* asynchronously fsync on close */ 359 360 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \ 361 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \ 362 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \ 363 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING) 364 365 #define HAMMER_INODE_MODMASK_NOXDIRTY \ 366 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY) 367 368 #define HAMMER_FLUSH_GROUP_SIZE 64 369 370 #define HAMMER_FLUSH_SIGNAL 0x0001 371 #define HAMMER_FLUSH_RECURSION 0x0002 372 373 /* 374 * Used by the inode reclaim code to pipeline reclaims and avoid 375 * blowing out kernel memory or letting the flusher get too far 376 * behind. The reclaim wakes up when count reaches 0 or the 377 * timer expires. 378 */ 379 struct hammer_reclaim { 380 TAILQ_ENTRY(hammer_reclaim) entry; 381 int count; 382 }; 383 384 #define HAMMER_RECLAIM_FLUSH 2000 385 #define HAMMER_RECLAIM_WAIT 4000 386 387 /* 388 * Structure used to represent an unsynchronized record in-memory. These 389 * records typically represent directory entries. Only non-historical 390 * records are kept in-memory. 391 * 392 * Records are organized as a per-inode RB-Tree. If the inode is not 393 * on disk then neither are any records and the in-memory record tree 394 * represents the entire contents of the inode. If the inode is on disk 395 * then the on-disk B-Tree is scanned in parallel with the in-memory 396 * RB-Tree to synthesize the current state of the file. 397 * 398 * Records are also used to enforce the ordering of directory create/delete 399 * operations. A new inode will not be flushed to disk unless its related 400 * directory entry is also being flushed at the same time. A directory entry 401 * will not be removed unless its related inode is also being removed at the 402 * same time. 403 */ 404 typedef enum hammer_record_type { 405 HAMMER_MEM_RECORD_GENERAL, /* misc record */ 406 HAMMER_MEM_RECORD_INODE, /* inode record */ 407 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */ 408 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */ 409 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */ 410 } hammer_record_type_t; 411 412 struct hammer_record { 413 RB_ENTRY(hammer_record) rb_node; 414 TAILQ_ENTRY(hammer_record) target_entry; 415 hammer_inode_state_t flush_state; 416 hammer_flush_group_t flush_group; 417 hammer_record_type_t type; 418 struct hammer_lock lock; 419 struct hammer_reserve *resv; 420 struct hammer_inode *ip; 421 struct hammer_inode *target_ip; 422 struct hammer_btree_leaf_elm leaf; 423 union hammer_data_ondisk *data; 424 int flags; 425 hammer_off_t zone2_offset; /* direct-write only */ 426 }; 427 428 typedef struct hammer_record *hammer_record_t; 429 430 /* 431 * Record flags. Note that FE can only be set by the frontend if the 432 * record has not been interlocked by the backend w/ BE. 433 */ 434 #define HAMMER_RECF_ALLOCDATA 0x0001 435 #define HAMMER_RECF_ONRBTREE 0x0002 436 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */ 437 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */ 438 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */ 439 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */ 440 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */ 441 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */ 442 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/ 443 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/ 444 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */ 445 446 /* 447 * hammer_delete_at_cursor() flags 448 */ 449 #define HAMMER_DELETE_ADJUST 0x0001 450 #define HAMMER_DELETE_DESTROY 0x0002 451 452 /* 453 * In-memory structures representing on-disk structures. 454 */ 455 struct hammer_volume; 456 struct hammer_buffer; 457 struct hammer_node; 458 struct hammer_undo; 459 struct hammer_reserve; 460 461 RB_HEAD(hammer_vol_rb_tree, hammer_volume); 462 RB_HEAD(hammer_buf_rb_tree, hammer_buffer); 463 RB_HEAD(hammer_nod_rb_tree, hammer_node); 464 RB_HEAD(hammer_und_rb_tree, hammer_undo); 465 RB_HEAD(hammer_res_rb_tree, hammer_reserve); 466 467 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node, 468 hammer_vol_rb_compare, int32_t); 469 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node, 470 hammer_buf_rb_compare, hammer_off_t); 471 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node, 472 hammer_nod_rb_compare, hammer_off_t); 473 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node, 474 hammer_und_rb_compare, hammer_off_t); 475 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node, 476 hammer_res_rb_compare, hammer_off_t); 477 478 /* 479 * IO management - embedded at the head of various in-memory structures 480 * 481 * VOLUME - hammer_volume containing meta-data 482 * META_BUFFER - hammer_buffer containing meta-data 483 * DATA_BUFFER - hammer_buffer containing pure-data 484 * 485 * Dirty volume headers and dirty meta-data buffers are locked until the 486 * flusher can sequence them out. Dirty pure-data buffers can be written. 487 * Clean buffers can be passively released. 488 */ 489 typedef enum hammer_io_type { 490 HAMMER_STRUCTURE_VOLUME, 491 HAMMER_STRUCTURE_META_BUFFER, 492 HAMMER_STRUCTURE_UNDO_BUFFER, 493 HAMMER_STRUCTURE_DATA_BUFFER 494 } hammer_io_type_t; 495 496 union hammer_io_structure; 497 struct hammer_io; 498 499 struct worklist { 500 LIST_ENTRY(worklist) node; 501 }; 502 503 TAILQ_HEAD(hammer_io_list, hammer_io); 504 typedef struct hammer_io_list *hammer_io_list_t; 505 506 struct hammer_io { 507 struct worklist worklist; 508 struct hammer_lock lock; 509 enum hammer_io_type type; 510 struct hammer_mount *hmp; 511 struct hammer_volume *volume; 512 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */ 513 hammer_io_list_t mod_list; 514 struct buf *bp; 515 int64_t offset; /* zone-2 offset */ 516 int bytes; /* buffer cache buffer size */ 517 int loading; /* loading/unloading interlock */ 518 int modify_refs; 519 520 u_int modified : 1; /* bp's data was modified */ 521 u_int released : 1; /* bp released (w/ B_LOCKED set) */ 522 u_int running : 1; /* bp write IO in progress */ 523 u_int waiting : 1; /* someone is waiting on us */ 524 u_int validated : 1; /* ondisk has been validated */ 525 u_int waitdep : 1; /* flush waits for dependancies */ 526 u_int recovered : 1; /* has recovery ref */ 527 u_int waitmod : 1; /* waiting for modify_refs */ 528 u_int reclaim : 1; /* reclaim requested */ 529 u_int gencrc : 1; /* crc needs to be generated */ 530 u_int ioerror : 1; /* abort on io-error */ 531 }; 532 533 typedef struct hammer_io *hammer_io_t; 534 535 #define HAMMER_CLUSTER_SIZE (64 * 1024) 536 #if HAMMER_CLUSTER_SIZE > MAXBSIZE 537 #undef HAMMER_CLUSTER_SIZE 538 #define HAMMER_CLUSTER_SIZE MAXBSIZE 539 #endif 540 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE) 541 542 /* 543 * In-memory volume representing on-disk buffer 544 */ 545 struct hammer_volume { 546 struct hammer_io io; 547 RB_ENTRY(hammer_volume) rb_node; 548 struct hammer_volume_ondisk *ondisk; 549 int32_t vol_no; 550 int64_t nblocks; /* note: special calculation for statfs */ 551 int64_t buffer_base; /* base offset of buffer 0 */ 552 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */ 553 hammer_off_t maxraw_off; /* Maximum raw offset for device */ 554 char *vol_name; 555 struct vnode *devvp; 556 int vol_flags; 557 }; 558 559 typedef struct hammer_volume *hammer_volume_t; 560 561 /* 562 * In-memory buffer (other then volume, super-cluster, or cluster), 563 * representing an on-disk buffer. 564 */ 565 struct hammer_buffer { 566 struct hammer_io io; 567 RB_ENTRY(hammer_buffer) rb_node; 568 void *ondisk; 569 hammer_off_t zoneX_offset; 570 hammer_off_t zone2_offset; 571 struct hammer_reserve *resv; 572 struct hammer_node_list clist; 573 }; 574 575 typedef struct hammer_buffer *hammer_buffer_t; 576 577 /* 578 * In-memory B-Tree node, representing an on-disk B-Tree node. 579 * 580 * This is a hang-on structure which is backed by a hammer_buffer, 581 * indexed by a hammer_cluster, and used for fine-grained locking of 582 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer 583 * can contain multiple nodes representing wildly disassociated portions 584 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis. 585 * 586 * This structure uses a cluster-relative index to reduce the number 587 * of layers required to access it, and also because all on-disk B-Tree 588 * references are cluster-relative offsets. 589 */ 590 struct hammer_node { 591 struct hammer_lock lock; /* node-by-node lock */ 592 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */ 593 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */ 594 hammer_off_t node_offset; /* full offset spec */ 595 struct hammer_mount *hmp; 596 struct hammer_buffer *buffer; /* backing buffer */ 597 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */ 598 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */ 599 struct hammer_node_cache_list cache_list; /* passive caches */ 600 int flags; 601 int loading; /* load interlock */ 602 }; 603 604 #define HAMMER_NODE_DELETED 0x0001 605 #define HAMMER_NODE_FLUSH 0x0002 606 #define HAMMER_NODE_CRCGOOD 0x0004 607 #define HAMMER_NODE_NEEDSCRC 0x0008 608 #define HAMMER_NODE_NEEDSMIRROR 0x0010 609 #define HAMMER_NODE_CRCBAD 0x0020 610 611 #define HAMMER_NODE_CRCANY (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD) 612 613 typedef struct hammer_node *hammer_node_t; 614 615 /* 616 * List of locked nodes. This structure is used to lock potentially large 617 * numbers of nodes as an aid for complex B-Tree operations. 618 */ 619 struct hammer_node_lock; 620 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock); 621 622 struct hammer_node_lock { 623 TAILQ_ENTRY(hammer_node_lock) entry; 624 struct hammer_node_lock_list list; 625 struct hammer_node_lock *parent; 626 hammer_node_t node; 627 hammer_node_ondisk_t copy; /* copy of on-disk data */ 628 int index; /* index of this node in parent */ 629 int count; /* count children */ 630 int flags; 631 }; 632 633 typedef struct hammer_node_lock *hammer_node_lock_t; 634 635 #define HAMMER_NODE_LOCK_UPDATED 0x0001 636 637 /* 638 * Common I/O management structure - embedded in in-memory structures 639 * which are backed by filesystem buffers. 640 */ 641 union hammer_io_structure { 642 struct hammer_io io; 643 struct hammer_volume volume; 644 struct hammer_buffer buffer; 645 }; 646 647 typedef union hammer_io_structure *hammer_io_structure_t; 648 649 /* 650 * The reserve structure prevents the blockmap from allocating 651 * out of a reserved bigblock. Such reservations are used by 652 * the direct-write mechanism. 653 * 654 * The structure is also used to hold off on reallocations of 655 * big blocks from the freemap until flush dependancies have 656 * been dealt with. 657 */ 658 struct hammer_reserve { 659 RB_ENTRY(hammer_reserve) rb_node; 660 TAILQ_ENTRY(hammer_reserve) delay_entry; 661 int flush_group; 662 int flags; 663 int refs; 664 int zone; 665 int append_off; 666 hammer_off_t zone_offset; 667 }; 668 669 typedef struct hammer_reserve *hammer_reserve_t; 670 671 #define HAMMER_RESF_ONDELAY 0x0001 672 #define HAMMER_RESF_LAYER2FREE 0x0002 673 674 #include "hammer_cursor.h" 675 676 /* 677 * The undo structure tracks recent undos to avoid laying down duplicate 678 * undos within a flush group, saving us a significant amount of overhead. 679 * 680 * This is strictly a heuristic. 681 */ 682 #define HAMMER_MAX_UNDOS 1024 683 #define HAMMER_MAX_FLUSHERS 4 684 685 struct hammer_undo { 686 RB_ENTRY(hammer_undo) rb_node; 687 TAILQ_ENTRY(hammer_undo) lru_entry; 688 hammer_off_t offset; 689 int bytes; 690 }; 691 692 typedef struct hammer_undo *hammer_undo_t; 693 694 struct hammer_flusher_info; 695 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info); 696 697 struct hammer_flusher { 698 int signal; /* flusher thread sequencer */ 699 int act; /* currently active flush group */ 700 int done; /* set to act when complete */ 701 int next; /* next flush group */ 702 int group_lock; /* lock sequencing of the next flush */ 703 int exiting; /* request master exit */ 704 thread_t td; /* master flusher thread */ 705 hammer_tid_t tid; /* last flushed transaction id */ 706 int finalize_want; /* serialize finalization */ 707 struct hammer_lock finalize_lock; /* serialize finalization */ 708 struct hammer_transaction trans; /* shared transaction */ 709 struct hammer_flusher_info_list run_list; 710 struct hammer_flusher_info_list ready_list; 711 }; 712 713 /* 714 * Internal hammer mount data structure 715 */ 716 struct hammer_mount { 717 struct mount *mp; 718 /*struct vnode *rootvp;*/ 719 struct hammer_ino_rb_tree rb_inos_root; 720 struct hammer_vol_rb_tree rb_vols_root; 721 struct hammer_nod_rb_tree rb_nods_root; 722 struct hammer_und_rb_tree rb_undo_root; 723 struct hammer_res_rb_tree rb_resv_root; 724 struct hammer_buf_rb_tree rb_bufs_root; 725 struct hammer_pfs_rb_tree rb_pfsm_root; 726 struct hammer_volume *rootvol; 727 struct hammer_base_elm root_btree_beg; 728 struct hammer_base_elm root_btree_end; 729 730 struct malloc_type *m_misc; 731 struct malloc_type *m_inodes; 732 733 int flags; /* HAMMER_MOUNT_xxx flags */ 734 int hflags; 735 int ronly; 736 int nvolumes; 737 int volume_iterator; 738 int master_id; /* -1 or 0-15 - clustering and mirroring */ 739 int version; /* hammer filesystem version to use */ 740 int rsv_inodes; /* reserved space due to dirty inodes */ 741 int64_t rsv_databytes; /* reserved space due to record data */ 742 int rsv_recs; /* reserved space due to dirty records */ 743 int rsv_fromdelay; /* bigblocks reserved due to flush delay */ 744 int undo_rec_limit; /* based on size of undo area */ 745 int last_newrecords; 746 int count_newrecords; 747 748 int inode_reclaims; /* inodes pending reclaim by flusher */ 749 int count_inodes; /* total number of inodes */ 750 int count_iqueued; /* inodes queued to flusher */ 751 752 struct hammer_flusher flusher; 753 754 u_int check_interrupt; 755 u_int check_yield; 756 uuid_t fsid; 757 struct hammer_io_list volu_list; /* dirty undo buffers */ 758 struct hammer_io_list undo_list; /* dirty undo buffers */ 759 struct hammer_io_list data_list; /* dirty data buffers */ 760 struct hammer_io_list alt_data_list; /* dirty data buffers */ 761 struct hammer_io_list meta_list; /* dirty meta bufs */ 762 struct hammer_io_list lose_list; /* loose buffers */ 763 int locked_dirty_space; /* meta/volu count */ 764 int io_running_space; 765 int objid_cache_count; 766 int error; /* critical I/O error */ 767 struct krate krate; /* rate limited kprintf */ 768 hammer_tid_t asof; /* snapshot mount */ 769 hammer_tid_t next_tid; 770 hammer_tid_t flush_tid1; /* flusher tid sequencing */ 771 hammer_tid_t flush_tid2; /* flusher tid sequencing */ 772 int64_t copy_stat_freebigblocks; /* number of free bigblocks */ 773 774 struct netexport export; 775 struct hammer_lock sync_lock; 776 struct hammer_lock free_lock; 777 struct hammer_lock undo_lock; 778 struct hammer_lock blkmap_lock; 779 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES]; 780 struct hammer_undo undos[HAMMER_MAX_UNDOS]; 781 int undo_alloc; 782 TAILQ_HEAD(, hammer_undo) undo_lru_list; 783 TAILQ_HEAD(, hammer_reserve) delay_list; 784 struct hammer_flush_group_list flush_group_list; 785 hammer_flush_group_t next_flush_group; 786 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list; 787 TAILQ_HEAD(, hammer_reclaim) reclaim_list; 788 }; 789 790 typedef struct hammer_mount *hammer_mount_t; 791 792 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001 793 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002 794 795 struct hammer_sync_info { 796 int error; 797 int waitfor; 798 }; 799 800 #endif 801 802 /* 803 * checkspace slop (8MB chunks), higher numbers are more conservative. 804 */ 805 #define HAMMER_CHKSPC_REBLOCK 25 806 #define HAMMER_CHKSPC_MIRROR 20 807 #define HAMMER_CHKSPC_WRITE 20 808 #define HAMMER_CHKSPC_CREATE 20 809 #define HAMMER_CHKSPC_REMOVE 10 810 #define HAMMER_CHKSPC_EMERGENCY 0 811 812 #if defined(_KERNEL) 813 814 extern struct vop_ops hammer_vnode_vops; 815 extern struct vop_ops hammer_spec_vops; 816 extern struct vop_ops hammer_fifo_vops; 817 extern struct bio_ops hammer_bioops; 818 819 extern int hammer_debug_io; 820 extern int hammer_debug_general; 821 extern int hammer_debug_debug; 822 extern int hammer_debug_inode; 823 extern int hammer_debug_locks; 824 extern int hammer_debug_btree; 825 extern int hammer_debug_tid; 826 extern int hammer_debug_recover; 827 extern int hammer_debug_recover_faults; 828 extern int hammer_cluster_enable; 829 extern int hammer_count_fsyncs; 830 extern int hammer_count_inodes; 831 extern int hammer_count_iqueued; 832 extern int hammer_count_reclaiming; 833 extern int hammer_count_records; 834 extern int hammer_count_record_datas; 835 extern int hammer_count_volumes; 836 extern int hammer_count_buffers; 837 extern int hammer_count_nodes; 838 extern int64_t hammer_count_extra_space_used; 839 extern int64_t hammer_stats_btree_lookups; 840 extern int64_t hammer_stats_btree_searches; 841 extern int64_t hammer_stats_btree_inserts; 842 extern int64_t hammer_stats_btree_deletes; 843 extern int64_t hammer_stats_btree_elements; 844 extern int64_t hammer_stats_btree_splits; 845 extern int64_t hammer_stats_btree_iterations; 846 extern int64_t hammer_stats_btree_root_iterations; 847 extern int64_t hammer_stats_record_iterations; 848 extern int64_t hammer_stats_file_read; 849 extern int64_t hammer_stats_file_write; 850 extern int64_t hammer_stats_file_iopsr; 851 extern int64_t hammer_stats_file_iopsw; 852 extern int64_t hammer_stats_disk_read; 853 extern int64_t hammer_stats_disk_write; 854 extern int64_t hammer_stats_inode_flushes; 855 extern int64_t hammer_stats_commits; 856 extern int64_t hammer_stats_undo; 857 extern int hammer_count_dirtybufspace; 858 extern int hammer_count_refedbufs; 859 extern int hammer_count_reservations; 860 extern int hammer_count_io_running_read; 861 extern int hammer_count_io_running_write; 862 extern int hammer_count_io_locked; 863 extern int hammer_limit_dirtybufspace; 864 extern int hammer_limit_recs; 865 extern int hammer_limit_inode_recs; 866 extern int hammer_bio_count; 867 extern int hammer_verify_zone; 868 extern int hammer_verify_data; 869 extern int hammer_write_mode; 870 extern int hammer_yield_check; 871 extern int hammer_fsync_mode; 872 extern int hammer_autoflush; 873 extern int64_t hammer_contention_count; 874 875 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 876 int error, const char *msg); 877 int hammer_vop_inactive(struct vop_inactive_args *); 878 int hammer_vop_reclaim(struct vop_reclaim_args *); 879 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp); 880 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans, 881 hammer_inode_t dip, int64_t obj_id, 882 hammer_tid_t asof, u_int32_t localization, 883 int flags, int *errorp); 884 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans, 885 hammer_inode_t dip, int64_t obj_id, 886 hammer_tid_t asof, u_int32_t localization, 887 int flags, int *errorp); 888 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans, 889 int64_t obj_id, hammer_tid_t asof, 890 u_int32_t localization); 891 void hammer_scan_inode_snapshots(hammer_mount_t hmp, 892 hammer_inode_info_t iinfo, 893 int (*callback)(hammer_inode_t ip, void *data), 894 void *data); 895 void hammer_put_inode(struct hammer_inode *ip); 896 void hammer_put_inode_ref(struct hammer_inode *ip); 897 void hammer_inode_waitreclaims(hammer_mount_t hmp); 898 void hammer_inode_waithard(hammer_mount_t hmp); 899 900 int hammer_unload_volume(hammer_volume_t volume, void *data __unused); 901 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused); 902 903 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused); 904 int hammer_install_volume(hammer_mount_t hmp, const char *volname, 905 struct vnode *devvp); 906 int hammer_mountcheck_volumes(hammer_mount_t hmp); 907 908 int hammer_mem_add(hammer_record_t record); 909 int hammer_ip_lookup(hammer_cursor_t cursor); 910 int hammer_ip_first(hammer_cursor_t cursor); 911 int hammer_ip_next(hammer_cursor_t cursor); 912 int hammer_ip_resolve_data(hammer_cursor_t cursor); 913 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip, 914 hammer_tid_t tid); 915 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags, 916 hammer_tid_t delete_tid, u_int32_t delete_ts, 917 int track, int64_t *stat_bytes); 918 int hammer_ip_check_directory_empty(hammer_transaction_t trans, 919 hammer_inode_t ip); 920 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor); 921 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor); 922 923 hammer_record_t 924 hammer_alloc_mem_record(hammer_inode_t ip, int data_len); 925 void hammer_flush_record_done(hammer_record_t record, int error); 926 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident); 927 void hammer_rel_mem_record(hammer_record_t record); 928 929 int hammer_cursor_up(hammer_cursor_t cursor); 930 int hammer_cursor_up_locked(hammer_cursor_t cursor); 931 int hammer_cursor_down(hammer_cursor_t cursor); 932 int hammer_cursor_upgrade(hammer_cursor_t cursor); 933 int hammer_cursor_upgrade_node(hammer_cursor_t cursor); 934 void hammer_cursor_downgrade(hammer_cursor_t cursor); 935 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, 936 int index); 937 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident); 938 int hammer_lock_ex_try(struct hammer_lock *lock); 939 void hammer_lock_sh(struct hammer_lock *lock); 940 int hammer_lock_sh_try(struct hammer_lock *lock); 941 int hammer_lock_upgrade(struct hammer_lock *lock); 942 void hammer_lock_downgrade(struct hammer_lock *lock); 943 int hammer_lock_status(struct hammer_lock *lock); 944 void hammer_unlock(struct hammer_lock *lock); 945 void hammer_ref(struct hammer_lock *lock); 946 void hammer_unref(struct hammer_lock *lock); 947 948 void hammer_sync_lock_ex(hammer_transaction_t trans); 949 void hammer_sync_lock_sh(hammer_transaction_t trans); 950 int hammer_sync_lock_sh_try(hammer_transaction_t trans); 951 void hammer_sync_unlock(hammer_transaction_t trans); 952 953 u_int32_t hammer_to_unix_xid(uuid_t *uuid); 954 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid); 955 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts); 956 u_int64_t hammer_timespec_to_time(struct timespec *ts); 957 int hammer_str_to_tid(const char *str, int *ispfsp, 958 hammer_tid_t *tidp, u_int32_t *localizationp); 959 int hammer_is_atatext(const char *name, int len); 960 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip, 961 int64_t namekey); 962 void hammer_clear_objid(hammer_inode_t dip); 963 void hammer_destroy_objid_cache(hammer_mount_t hmp); 964 965 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset, 966 int bytes); 967 void hammer_clear_undo_history(hammer_mount_t hmp); 968 enum vtype hammer_get_vnode_type(u_int8_t obj_type); 969 int hammer_get_dtype(u_int8_t obj_type); 970 u_int8_t hammer_get_obj_type(enum vtype vtype); 971 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len, 972 u_int32_t *max_iterationsp); 973 int hammer_nohistory(hammer_inode_t ip); 974 975 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor, 976 hammer_node_cache_t cache, hammer_inode_t ip); 977 void hammer_normalize_cursor(hammer_cursor_t cursor); 978 void hammer_done_cursor(hammer_cursor_t cursor); 979 int hammer_recover_cursor(hammer_cursor_t cursor); 980 void hammer_unlock_cursor(hammer_cursor_t cursor); 981 int hammer_lock_cursor(hammer_cursor_t cursor); 982 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor); 983 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor); 984 985 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode); 986 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent, 987 int index); 988 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode, 989 int index); 990 void hammer_cursor_moved_element(hammer_node_t onode, hammer_node_t nnode, 991 int oindex, int nindex); 992 void hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent, 993 hammer_node_t nparent, int nindex); 994 void hammer_cursor_inserted_element(hammer_node_t node, int index); 995 void hammer_cursor_deleted_element(hammer_node_t node, int index); 996 997 int hammer_btree_lookup(hammer_cursor_t cursor); 998 int hammer_btree_first(hammer_cursor_t cursor); 999 int hammer_btree_last(hammer_cursor_t cursor); 1000 int hammer_btree_extract(hammer_cursor_t cursor, int flags); 1001 int hammer_btree_iterate(hammer_cursor_t cursor); 1002 int hammer_btree_iterate_reverse(hammer_cursor_t cursor); 1003 int hammer_btree_insert(hammer_cursor_t cursor, 1004 hammer_btree_leaf_elm_t elm, int *doprop); 1005 int hammer_btree_delete(hammer_cursor_t cursor); 1006 void hammer_btree_do_propagation(hammer_cursor_t cursor, 1007 hammer_pseudofs_inmem_t pfsm, 1008 hammer_btree_leaf_elm_t leaf); 1009 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2); 1010 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key); 1011 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid); 1012 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid); 1013 1014 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node, 1015 hammer_btree_elm_t elm); 1016 void hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node); 1017 int hammer_btree_lock_children(hammer_cursor_t cursor, int depth, 1018 hammer_node_lock_t parent); 1019 void hammer_btree_lock_copy(hammer_cursor_t cursor, 1020 hammer_node_lock_t parent); 1021 int hammer_btree_sync_copy(hammer_cursor_t cursor, 1022 hammer_node_lock_t parent); 1023 void hammer_btree_unlock_children(hammer_cursor_t cursor, 1024 hammer_node_lock_t parent); 1025 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node); 1026 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans, 1027 hammer_node_t node, int *parent_indexp, 1028 int *errorp, int try_exclusive); 1029 1030 void hammer_print_btree_node(hammer_node_ondisk_t ondisk); 1031 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i); 1032 1033 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off, 1034 int *errorp, struct hammer_buffer **bufferp); 1035 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off, 1036 int *errorp, struct hammer_buffer **bufferp); 1037 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes, 1038 int *errorp, struct hammer_buffer **bufferp); 1039 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes, 1040 int *errorp, struct hammer_buffer **bufferp); 1041 1042 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp); 1043 1044 hammer_volume_t hammer_get_volume(hammer_mount_t hmp, 1045 int32_t vol_no, int *errorp); 1046 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset, 1047 int bytes, int isnew, int *errorp); 1048 void hammer_sync_buffers(hammer_mount_t hmp, 1049 hammer_off_t base_offset, int bytes); 1050 int hammer_del_buffers(hammer_mount_t hmp, 1051 hammer_off_t base_offset, 1052 hammer_off_t zone2_offset, int bytes, 1053 int report_conflicts); 1054 1055 int hammer_ref_volume(hammer_volume_t volume); 1056 int hammer_ref_buffer(hammer_buffer_t buffer); 1057 void hammer_flush_buffer_nodes(hammer_buffer_t buffer); 1058 1059 void hammer_rel_volume(hammer_volume_t volume, int flush); 1060 void hammer_rel_buffer(hammer_buffer_t buffer, int flush); 1061 1062 int hammer_vfs_export(struct mount *mp, int op, 1063 const struct export_args *export); 1064 hammer_node_t hammer_get_node(hammer_transaction_t trans, 1065 hammer_off_t node_offset, int isnew, int *errorp); 1066 void hammer_ref_node(hammer_node_t node); 1067 hammer_node_t hammer_ref_node_safe(hammer_transaction_t trans, 1068 hammer_node_cache_t cache, int *errorp); 1069 void hammer_rel_node(hammer_node_t node); 1070 void hammer_delete_node(hammer_transaction_t trans, 1071 hammer_node_t node); 1072 void hammer_cache_node(hammer_node_cache_t cache, 1073 hammer_node_t node); 1074 void hammer_uncache_node(hammer_node_cache_t cache); 1075 void hammer_flush_node(hammer_node_t node); 1076 1077 void hammer_dup_buffer(struct hammer_buffer **bufferp, 1078 struct hammer_buffer *buffer); 1079 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, 1080 hammer_off_t hint, int *errorp); 1081 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len, 1082 u_int16_t rec_type, hammer_off_t *data_offsetp, 1083 struct hammer_buffer **data_bufferp, 1084 hammer_off_t hint, int *errorp); 1085 1086 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io, 1087 hammer_off_t zone1_offset, void *base, int len); 1088 1089 void hammer_put_volume(struct hammer_volume *volume, int flush); 1090 void hammer_put_buffer(struct hammer_buffer *buffer, int flush); 1091 1092 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans, 1093 hammer_off_t owner, int *errorp); 1094 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset, 1095 hammer_off_t owner, int *errorp); 1096 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp); 1097 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone, 1098 int bytes, hammer_off_t hint, int *errorp); 1099 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone, 1100 int bytes, hammer_off_t *zone_offp, int *errorp); 1101 void hammer_blockmap_reserve_complete(hammer_mount_t hmp, 1102 hammer_reserve_t resv); 1103 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv); 1104 void hammer_blockmap_free(hammer_transaction_t trans, 1105 hammer_off_t bmap_off, int bytes); 1106 int hammer_blockmap_finalize(hammer_transaction_t trans, 1107 hammer_reserve_t resv, 1108 hammer_off_t bmap_off, int bytes); 1109 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off, 1110 int *curp, int *errorp); 1111 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, 1112 int *errorp); 1113 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, 1114 int *errorp); 1115 int64_t hammer_undo_used(hammer_transaction_t trans); 1116 int64_t hammer_undo_space(hammer_transaction_t trans); 1117 int64_t hammer_undo_max(hammer_mount_t hmp); 1118 int hammer_undo_reclaim(hammer_io_t io); 1119 1120 void hammer_start_transaction(struct hammer_transaction *trans, 1121 struct hammer_mount *hmp); 1122 void hammer_simple_transaction(struct hammer_transaction *trans, 1123 struct hammer_mount *hmp); 1124 void hammer_start_transaction_fls(struct hammer_transaction *trans, 1125 struct hammer_mount *hmp); 1126 void hammer_done_transaction(struct hammer_transaction *trans); 1127 1128 void hammer_modify_inode(hammer_inode_t ip, int flags); 1129 void hammer_flush_inode(hammer_inode_t ip, int flags); 1130 void hammer_flush_inode_done(hammer_inode_t ip, int error); 1131 void hammer_wait_inode(hammer_inode_t ip); 1132 1133 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap, 1134 struct ucred *cred, struct hammer_inode *dip, 1135 const char *name, int namelen, 1136 hammer_pseudofs_inmem_t pfsm, 1137 struct hammer_inode **ipp); 1138 void hammer_rel_inode(hammer_inode_t ip, int flush); 1139 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused); 1140 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2); 1141 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused); 1142 1143 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip); 1144 void hammer_test_inode(hammer_inode_t dip); 1145 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp); 1146 1147 int hammer_ip_add_directory(struct hammer_transaction *trans, 1148 hammer_inode_t dip, const char *name, int bytes, 1149 hammer_inode_t nip); 1150 int hammer_ip_del_directory(struct hammer_transaction *trans, 1151 hammer_cursor_t cursor, hammer_inode_t dip, 1152 hammer_inode_t ip); 1153 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset, 1154 void *data, int bytes, int *errorp); 1155 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size); 1156 int hammer_ip_add_record(struct hammer_transaction *trans, 1157 hammer_record_t record); 1158 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip, 1159 int64_t ran_beg, int64_t ran_end, int truncating); 1160 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip, 1161 int *countp); 1162 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip, 1163 int64_t offset, void *data, int bytes); 1164 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec); 1165 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec); 1166 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans, 1167 u_int32_t localization, int *errorp); 1168 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred, 1169 hammer_pseudofs_inmem_t pfsm); 1170 int hammer_save_pseudofs(hammer_transaction_t trans, 1171 hammer_pseudofs_inmem_t pfsm); 1172 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization); 1173 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm); 1174 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag, 1175 struct ucred *cred); 1176 1177 void hammer_io_init(hammer_io_t io, hammer_volume_t volume, 1178 enum hammer_io_type type); 1179 int hammer_io_read(struct vnode *devvp, struct hammer_io *io, 1180 hammer_off_t limit); 1181 int hammer_io_new(struct vnode *devvp, struct hammer_io *io); 1182 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset); 1183 struct buf *hammer_io_release(struct hammer_io *io, int flush); 1184 void hammer_io_flush(struct hammer_io *io, int reclaim); 1185 void hammer_io_wait(struct hammer_io *io); 1186 void hammer_io_waitdep(struct hammer_io *io); 1187 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident); 1188 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio, 1189 hammer_btree_leaf_elm_t leaf); 1190 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record, 1191 struct bio *bio); 1192 void hammer_io_direct_wait(hammer_record_t record); 1193 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf); 1194 void hammer_io_write_interlock(hammer_io_t io); 1195 void hammer_io_done_interlock(hammer_io_t io); 1196 void hammer_io_clear_modify(struct hammer_io *io, int inval); 1197 void hammer_io_clear_modlist(struct hammer_io *io); 1198 void hammer_io_flush_sync(hammer_mount_t hmp); 1199 1200 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume, 1201 void *base, int len); 1202 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer, 1203 void *base, int len); 1204 void hammer_modify_volume_done(hammer_volume_t volume); 1205 void hammer_modify_buffer_done(hammer_buffer_t buffer); 1206 1207 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip, 1208 struct hammer_ioc_reblock *reblock); 1209 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip, 1210 struct hammer_ioc_rebalance *rebal); 1211 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip, 1212 struct hammer_ioc_prune *prune); 1213 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip, 1214 struct hammer_ioc_mirror_rw *mirror); 1215 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip, 1216 struct hammer_ioc_mirror_rw *mirror); 1217 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1218 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs); 1219 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1220 struct hammer_ioc_pseudofs_rw *pfs); 1221 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1222 struct hammer_ioc_pseudofs_rw *pfs); 1223 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1224 struct hammer_ioc_pseudofs_rw *pfs); 1225 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1226 struct hammer_ioc_pseudofs_rw *pfs); 1227 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1228 struct hammer_ioc_pseudofs_rw *pfs); 1229 int hammer_ioc_expand(hammer_transaction_t trans, hammer_inode_t ip, 1230 struct hammer_ioc_expand *expand); 1231 1232 int hammer_signal_check(hammer_mount_t hmp); 1233 1234 void hammer_flusher_create(hammer_mount_t hmp); 1235 void hammer_flusher_destroy(hammer_mount_t hmp); 1236 void hammer_flusher_sync(hammer_mount_t hmp); 1237 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg); 1238 int hammer_flusher_async_one(hammer_mount_t hmp); 1239 void hammer_flusher_wait(hammer_mount_t hmp, int seq); 1240 void hammer_flusher_wait_next(hammer_mount_t hmp); 1241 int hammer_flusher_meta_limit(hammer_mount_t hmp); 1242 int hammer_flusher_meta_halflimit(hammer_mount_t hmp); 1243 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter); 1244 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp); 1245 void hammer_flusher_finalize(hammer_transaction_t trans, int final); 1246 int hammer_flusher_haswork(hammer_mount_t hmp); 1247 1248 1249 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol); 1250 void hammer_recover_flush_buffers(hammer_mount_t hmp, 1251 hammer_volume_t root_volume, int final); 1252 1253 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap); 1254 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk); 1255 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf); 1256 1257 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap); 1258 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk); 1259 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk); 1260 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf); 1261 void hkprintf(const char *ctl, ...); 1262 udev_t hammer_fsid_to_udev(uuid_t *uuid); 1263 1264 1265 int hammer_blocksize(int64_t file_offset); 1266 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2); 1267 1268 /* 1269 * Shortcut for _hammer_checkspace(), used all over the code. 1270 */ 1271 static __inline int 1272 hammer_checkspace(hammer_mount_t hmp, int slop) 1273 { 1274 return(_hammer_checkspace(hmp, slop, NULL)); 1275 } 1276 1277 #endif 1278 1279 static __inline void 1280 hammer_wait_mem_record(hammer_record_t record) 1281 { 1282 hammer_wait_mem_record_ident(record, "hmmwai"); 1283 } 1284 1285 static __inline void 1286 hammer_lock_ex(struct hammer_lock *lock) 1287 { 1288 hammer_lock_ex_ident(lock, "hmrlck"); 1289 } 1290 1291 /* 1292 * Indicate that a B-Tree node is being modified. 1293 */ 1294 static __inline void 1295 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node) 1296 { 1297 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0); 1298 hammer_modify_buffer(trans, node->buffer, NULL, 0); 1299 } 1300 1301 static __inline void 1302 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node) 1303 { 1304 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0); 1305 hammer_modify_buffer(trans, node->buffer, 1306 node->ondisk, sizeof(*node->ondisk)); 1307 } 1308 1309 static __inline void 1310 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node, 1311 void *base, int len) 1312 { 1313 hammer_crc_t *crcptr; 1314 1315 KKASSERT((char *)base >= (char *)node->ondisk && 1316 (char *)base + len <= 1317 (char *)node->ondisk + sizeof(*node->ondisk)); 1318 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0); 1319 hammer_modify_buffer(trans, node->buffer, base, len); 1320 crcptr = &node->ondisk->crc; 1321 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t)); 1322 --node->buffer->io.modify_refs; /* only want one ref */ 1323 } 1324 1325 /* 1326 * Indicate that the specified modifications have been completed. 1327 * 1328 * Do not try to generate the crc here, it's very expensive to do and a 1329 * sequence of insertions or deletions can result in many calls to this 1330 * function on the same node. 1331 */ 1332 static __inline void 1333 hammer_modify_node_done(hammer_node_t node) 1334 { 1335 node->flags |= HAMMER_NODE_CRCGOOD; 1336 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) { 1337 node->flags |= HAMMER_NODE_NEEDSCRC; 1338 node->buffer->io.gencrc = 1; 1339 hammer_ref_node(node); 1340 } 1341 hammer_modify_buffer_done(node->buffer); 1342 } 1343 1344 #define hammer_modify_volume_field(trans, vol, field) \ 1345 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \ 1346 sizeof((vol)->ondisk->field)) 1347 1348 #define hammer_modify_node_field(trans, node, field) \ 1349 hammer_modify_node(trans, node, &(node)->ondisk->field, \ 1350 sizeof((node)->ondisk->field)) 1351 1352 /* 1353 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly 1354 * created directories for HAMMER version 2 or greater and causes 1355 * directory entries to be placed the inode localization zone in 1356 * the B-Tree instead of the misc zone. 1357 * 1358 * This greatly improves localization between directory entries and 1359 * inodes 1360 */ 1361 static __inline u_int32_t 1362 hammer_dir_localization(hammer_inode_t dip) 1363 { 1364 if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) 1365 return(HAMMER_LOCALIZE_INODE); 1366 else 1367 return(HAMMER_LOCALIZE_MISC); 1368 } 1369