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 121 122 /* 123 * HAMMER locks 124 */ 125 struct hammer_lock { 126 int refs; /* active references delay writes */ 127 int lockcount; /* lock count for exclusive/shared access */ 128 int wanted; 129 int exwanted; /* number of threads waiting for ex lock */ 130 struct thread *locktd; 131 }; 132 133 static __inline int 134 hammer_islocked(struct hammer_lock *lock) 135 { 136 return(lock->lockcount != 0); 137 } 138 139 static __inline int 140 hammer_isactive(struct hammer_lock *lock) 141 { 142 return(lock->refs != 0); 143 } 144 145 static __inline int 146 hammer_islastref(struct hammer_lock *lock) 147 { 148 return(lock->refs == 1); 149 } 150 151 /* 152 * Return if we specifically own the lock exclusively. 153 */ 154 static __inline int 155 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td) 156 { 157 if (lock->lockcount > 0 && lock->locktd == td) 158 return(1); 159 return(0); 160 } 161 162 /* 163 * Flush state, used by various structures 164 */ 165 typedef enum hammer_inode_state { 166 HAMMER_FST_IDLE, 167 HAMMER_FST_SETUP, 168 HAMMER_FST_FLUSH 169 } hammer_inode_state_t; 170 171 TAILQ_HEAD(hammer_record_list, hammer_record); 172 173 /* 174 * Pseudo-filesystem extended data tracking 175 */ 176 struct hammer_pfs_rb_tree; 177 struct hammer_pseudofs_inmem; 178 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem); 179 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node, 180 hammer_pfs_rb_compare, u_int32_t); 181 182 struct hammer_pseudofs_inmem { 183 RB_ENTRY(hammer_pseudofs_inmem) rb_node; 184 struct hammer_lock lock; 185 u_int32_t localization; 186 hammer_tid_t create_tid; 187 int flags; 188 udev_t fsid_udev; 189 struct hammer_pseudofs_data pfsd; 190 }; 191 192 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t; 193 194 #define HAMMER_PFSM_DELETED 0x0001 195 196 /* 197 * Cache object ids. A fixed number of objid cache structures are 198 * created to reserve object id's for newly created files in multiples 199 * of 100,000, localized to a particular directory, and recycled as 200 * needed. This allows parallel create operations in different 201 * directories to retain fairly localized object ids which in turn 202 * improves reblocking performance and layout. 203 */ 204 #define OBJID_CACHE_SIZE 1024 205 #define OBJID_CACHE_BULK 100000 206 207 typedef struct hammer_objid_cache { 208 TAILQ_ENTRY(hammer_objid_cache) entry; 209 struct hammer_inode *dip; 210 hammer_tid_t next_tid; 211 int count; 212 } *hammer_objid_cache_t; 213 214 /* 215 * Associate an inode with a B-Tree node to cache search start positions 216 */ 217 typedef struct hammer_node_cache { 218 TAILQ_ENTRY(hammer_node_cache) entry; 219 struct hammer_node *node; 220 struct hammer_inode *ip; 221 } *hammer_node_cache_t; 222 223 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache); 224 225 /* 226 * Structure used to organize flush groups. Flush groups must be 227 * organized into chunks in order to avoid blowing out the UNDO FIFO. 228 * Without this a 'sync' could end up flushing 50,000 inodes in a single 229 * transaction. 230 */ 231 struct hammer_flush_group { 232 TAILQ_ENTRY(hammer_flush_group) flush_entry; 233 TAILQ_HEAD(, hammer_inode) flush_list; 234 int unused01; /* inode load */ 235 int total_count; /* record load */ 236 int running; /* group is running */ 237 int closed; 238 int refs; 239 }; 240 241 typedef struct hammer_flush_group *hammer_flush_group_t; 242 243 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group); 244 245 /* 246 * Structure used to represent an inode in-memory. 247 * 248 * The record and data associated with an inode may be out of sync with 249 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag 250 * clear). 251 * 252 * An inode may also hold a cache of unsynchronized records, used for 253 * database and directories only. Unsynchronized regular file data is 254 * stored in the buffer cache. 255 * 256 * NOTE: A file which is created and destroyed within the initial 257 * synchronization period can wind up not doing any disk I/O at all. 258 * 259 * Finally, an inode may cache numerous disk-referencing B-Tree cursors. 260 */ 261 struct hammer_ino_rb_tree; 262 struct hammer_inode; 263 RB_HEAD(hammer_ino_rb_tree, hammer_inode); 264 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node, 265 hammer_ino_rb_compare, hammer_inode_info_t); 266 267 struct hammer_rec_rb_tree; 268 struct hammer_record; 269 RB_HEAD(hammer_rec_rb_tree, hammer_record); 270 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node, 271 hammer_rec_rb_compare, hammer_btree_leaf_elm_t); 272 273 TAILQ_HEAD(hammer_node_list, hammer_node); 274 275 struct hammer_inode { 276 RB_ENTRY(hammer_inode) rb_node; 277 hammer_inode_state_t flush_state; 278 hammer_flush_group_t flush_group; 279 TAILQ_ENTRY(hammer_inode) flush_entry; 280 struct hammer_record_list target_list; /* target of dependant recs */ 281 int64_t obj_id; /* (key) object identifier */ 282 hammer_tid_t obj_asof; /* (key) snapshot or 0 */ 283 u_int32_t obj_localization; /* (key) pseudo-fs */ 284 struct hammer_mount *hmp; 285 hammer_objid_cache_t objid_cache; 286 int flags; 287 int error; /* flush error */ 288 int cursor_ip_refs; /* sanity */ 289 int rsv_recs; 290 struct vnode *vp; 291 hammer_pseudofs_inmem_t pfsm; 292 struct lockf advlock; 293 struct hammer_lock lock; /* sync copy interlock */ 294 off_t trunc_off; 295 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */ 296 struct hammer_inode_data ino_data; /* in-memory cache */ 297 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */ 298 struct hammer_node_cache cache[2]; /* search initiate cache */ 299 300 /* 301 * When a demark is created to synchronize an inode to 302 * disk, certain fields are copied so the front-end VOPs 303 * can continue to run in parallel with the synchronization 304 * occuring in the background. 305 */ 306 int sync_flags; /* to-sync flags cache */ 307 off_t sync_trunc_off; /* to-sync truncation */ 308 off_t save_trunc_off; /* write optimization */ 309 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */ 310 struct hammer_inode_data sync_ino_data; /* to-sync cache */ 311 }; 312 313 typedef struct hammer_inode *hammer_inode_t; 314 315 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data) 316 317 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */ 318 /* (not including atime/mtime) */ 319 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */ 320 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */ 321 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */ 322 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */ 323 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */ 324 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */ 325 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */ 326 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */ 327 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */ 328 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */ 329 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */ 330 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */ 331 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */ 332 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */ 333 334 #define HAMMER_INODE_TRUNCATED 0x00010000 335 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/ 336 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */ 337 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */ 338 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */ 339 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */ 340 341 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \ 342 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \ 343 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \ 344 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING) 345 346 #define HAMMER_INODE_MODMASK_NOXDIRTY \ 347 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY) 348 349 #define HAMMER_FLUSH_GROUP_SIZE 64 350 351 #define HAMMER_FLUSH_SIGNAL 0x0001 352 #define HAMMER_FLUSH_RECURSION 0x0002 353 354 /* 355 * Used by the inode reclaim code to pipeline reclaims and avoid 356 * blowing out kernel memory or letting the flusher get too far 357 * behind. The reclaim wakes up when count reaches 0 or the 358 * timer expires. 359 */ 360 struct hammer_reclaim { 361 TAILQ_ENTRY(hammer_reclaim) entry; 362 int count; 363 }; 364 365 #define HAMMER_RECLAIM_FLUSH 2000 366 #define HAMMER_RECLAIM_WAIT 4000 367 368 /* 369 * Structure used to represent an unsynchronized record in-memory. These 370 * records typically represent directory entries. Only non-historical 371 * records are kept in-memory. 372 * 373 * Records are organized as a per-inode RB-Tree. If the inode is not 374 * on disk then neither are any records and the in-memory record tree 375 * represents the entire contents of the inode. If the inode is on disk 376 * then the on-disk B-Tree is scanned in parallel with the in-memory 377 * RB-Tree to synthesize the current state of the file. 378 * 379 * Records are also used to enforce the ordering of directory create/delete 380 * operations. A new inode will not be flushed to disk unless its related 381 * directory entry is also being flushed at the same time. A directory entry 382 * will not be removed unless its related inode is also being removed at the 383 * same time. 384 */ 385 typedef enum hammer_record_type { 386 HAMMER_MEM_RECORD_GENERAL, /* misc record */ 387 HAMMER_MEM_RECORD_INODE, /* inode record */ 388 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */ 389 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */ 390 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */ 391 } hammer_record_type_t; 392 393 struct hammer_record { 394 RB_ENTRY(hammer_record) rb_node; 395 TAILQ_ENTRY(hammer_record) target_entry; 396 hammer_inode_state_t flush_state; 397 hammer_flush_group_t flush_group; 398 hammer_record_type_t type; 399 struct hammer_lock lock; 400 struct hammer_reserve *resv; 401 struct hammer_inode *ip; 402 struct hammer_inode *target_ip; 403 struct hammer_btree_leaf_elm leaf; 404 union hammer_data_ondisk *data; 405 int flags; 406 hammer_off_t zone2_offset; /* direct-write only */ 407 }; 408 409 typedef struct hammer_record *hammer_record_t; 410 411 /* 412 * Record flags. Note that FE can only be set by the frontend if the 413 * record has not been interlocked by the backend w/ BE. 414 */ 415 #define HAMMER_RECF_ALLOCDATA 0x0001 416 #define HAMMER_RECF_ONRBTREE 0x0002 417 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */ 418 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */ 419 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */ 420 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */ 421 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */ 422 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */ 423 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/ 424 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/ 425 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */ 426 427 /* 428 * hammer_delete_at_cursor() flags 429 */ 430 #define HAMMER_DELETE_ADJUST 0x0001 431 #define HAMMER_DELETE_DESTROY 0x0002 432 433 /* 434 * In-memory structures representing on-disk structures. 435 */ 436 struct hammer_volume; 437 struct hammer_buffer; 438 struct hammer_node; 439 struct hammer_undo; 440 struct hammer_reserve; 441 442 RB_HEAD(hammer_vol_rb_tree, hammer_volume); 443 RB_HEAD(hammer_buf_rb_tree, hammer_buffer); 444 RB_HEAD(hammer_nod_rb_tree, hammer_node); 445 RB_HEAD(hammer_und_rb_tree, hammer_undo); 446 RB_HEAD(hammer_res_rb_tree, hammer_reserve); 447 448 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node, 449 hammer_vol_rb_compare, int32_t); 450 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node, 451 hammer_buf_rb_compare, hammer_off_t); 452 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node, 453 hammer_nod_rb_compare, hammer_off_t); 454 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node, 455 hammer_und_rb_compare, hammer_off_t); 456 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node, 457 hammer_res_rb_compare, hammer_off_t); 458 459 /* 460 * IO management - embedded at the head of various in-memory structures 461 * 462 * VOLUME - hammer_volume containing meta-data 463 * META_BUFFER - hammer_buffer containing meta-data 464 * DATA_BUFFER - hammer_buffer containing pure-data 465 * 466 * Dirty volume headers and dirty meta-data buffers are locked until the 467 * flusher can sequence them out. Dirty pure-data buffers can be written. 468 * Clean buffers can be passively released. 469 */ 470 typedef enum hammer_io_type { 471 HAMMER_STRUCTURE_VOLUME, 472 HAMMER_STRUCTURE_META_BUFFER, 473 HAMMER_STRUCTURE_UNDO_BUFFER, 474 HAMMER_STRUCTURE_DATA_BUFFER 475 } hammer_io_type_t; 476 477 union hammer_io_structure; 478 struct hammer_io; 479 480 struct worklist { 481 LIST_ENTRY(worklist) node; 482 }; 483 484 TAILQ_HEAD(hammer_io_list, hammer_io); 485 typedef struct hammer_io_list *hammer_io_list_t; 486 487 struct hammer_io { 488 struct worklist worklist; 489 struct hammer_lock lock; 490 enum hammer_io_type type; 491 struct hammer_mount *hmp; 492 struct hammer_volume *volume; 493 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */ 494 hammer_io_list_t mod_list; 495 struct buf *bp; 496 int64_t offset; /* zone-2 offset */ 497 int bytes; /* buffer cache buffer size */ 498 int loading; /* loading/unloading interlock */ 499 int modify_refs; 500 501 u_int modified : 1; /* bp's data was modified */ 502 u_int released : 1; /* bp released (w/ B_LOCKED set) */ 503 u_int running : 1; /* bp write IO in progress */ 504 u_int waiting : 1; /* someone is waiting on us */ 505 u_int validated : 1; /* ondisk has been validated */ 506 u_int waitdep : 1; /* flush waits for dependancies */ 507 u_int recovered : 1; /* has recovery ref */ 508 u_int waitmod : 1; /* waiting for modify_refs */ 509 u_int reclaim : 1; /* reclaim requested */ 510 u_int gencrc : 1; /* crc needs to be generated */ 511 u_int ioerror : 1; /* abort on io-error */ 512 }; 513 514 typedef struct hammer_io *hammer_io_t; 515 516 #define HAMMER_CLUSTER_SIZE (64 * 1024) 517 #if HAMMER_CLUSTER_SIZE > MAXBSIZE 518 #undef HAMMER_CLUSTER_SIZE 519 #define HAMMER_CLUSTER_SIZE MAXBSIZE 520 #endif 521 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE) 522 523 /* 524 * In-memory volume representing on-disk buffer 525 */ 526 struct hammer_volume { 527 struct hammer_io io; 528 RB_ENTRY(hammer_volume) rb_node; 529 struct hammer_volume_ondisk *ondisk; 530 int32_t vol_no; 531 int64_t nblocks; /* note: special calculation for statfs */ 532 int64_t buffer_base; /* base offset of buffer 0 */ 533 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */ 534 hammer_off_t maxraw_off; /* Maximum raw offset for device */ 535 char *vol_name; 536 struct vnode *devvp; 537 int vol_flags; 538 }; 539 540 typedef struct hammer_volume *hammer_volume_t; 541 542 /* 543 * In-memory buffer (other then volume, super-cluster, or cluster), 544 * representing an on-disk buffer. 545 */ 546 struct hammer_buffer { 547 struct hammer_io io; 548 RB_ENTRY(hammer_buffer) rb_node; 549 void *ondisk; 550 hammer_off_t zoneX_offset; 551 hammer_off_t zone2_offset; 552 struct hammer_reserve *resv; 553 struct hammer_node_list clist; 554 }; 555 556 typedef struct hammer_buffer *hammer_buffer_t; 557 558 /* 559 * In-memory B-Tree node, representing an on-disk B-Tree node. 560 * 561 * This is a hang-on structure which is backed by a hammer_buffer, 562 * indexed by a hammer_cluster, and used for fine-grained locking of 563 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer 564 * can contain multiple nodes representing wildly disassociated portions 565 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis. 566 * 567 * This structure uses a cluster-relative index to reduce the number 568 * of layers required to access it, and also because all on-disk B-Tree 569 * references are cluster-relative offsets. 570 */ 571 struct hammer_node { 572 struct hammer_lock lock; /* node-by-node lock */ 573 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */ 574 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */ 575 hammer_off_t node_offset; /* full offset spec */ 576 struct hammer_mount *hmp; 577 struct hammer_buffer *buffer; /* backing buffer */ 578 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */ 579 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */ 580 struct hammer_node_cache_list cache_list; /* passive caches */ 581 int flags; 582 int loading; /* load interlock */ 583 }; 584 585 #define HAMMER_NODE_DELETED 0x0001 586 #define HAMMER_NODE_FLUSH 0x0002 587 #define HAMMER_NODE_CRCGOOD 0x0004 588 #define HAMMER_NODE_NEEDSCRC 0x0008 589 #define HAMMER_NODE_NEEDSMIRROR 0x0010 590 591 typedef struct hammer_node *hammer_node_t; 592 593 /* 594 * List of locked nodes. 595 */ 596 struct hammer_node_locklist { 597 struct hammer_node_locklist *next; 598 hammer_node_t node; 599 }; 600 601 typedef struct hammer_node_locklist *hammer_node_locklist_t; 602 603 604 /* 605 * Common I/O management structure - embedded in in-memory structures 606 * which are backed by filesystem buffers. 607 */ 608 union hammer_io_structure { 609 struct hammer_io io; 610 struct hammer_volume volume; 611 struct hammer_buffer buffer; 612 }; 613 614 typedef union hammer_io_structure *hammer_io_structure_t; 615 616 /* 617 * The reserve structure prevents the blockmap from allocating 618 * out of a reserved bigblock. Such reservations are used by 619 * the direct-write mechanism. 620 * 621 * The structure is also used to hold off on reallocations of 622 * big blocks from the freemap until flush dependancies have 623 * been dealt with. 624 */ 625 struct hammer_reserve { 626 RB_ENTRY(hammer_reserve) rb_node; 627 TAILQ_ENTRY(hammer_reserve) delay_entry; 628 int flush_group; 629 int flags; 630 int refs; 631 int zone; 632 int append_off; 633 hammer_off_t zone_offset; 634 }; 635 636 typedef struct hammer_reserve *hammer_reserve_t; 637 638 #define HAMMER_RESF_ONDELAY 0x0001 639 #define HAMMER_RESF_LAYER2FREE 0x0002 640 641 #include "hammer_cursor.h" 642 643 /* 644 * The undo structure tracks recent undos to avoid laying down duplicate 645 * undos within a flush group, saving us a significant amount of overhead. 646 * 647 * This is strictly a heuristic. 648 */ 649 #define HAMMER_MAX_UNDOS 1024 650 #define HAMMER_MAX_FLUSHERS 4 651 652 struct hammer_undo { 653 RB_ENTRY(hammer_undo) rb_node; 654 TAILQ_ENTRY(hammer_undo) lru_entry; 655 hammer_off_t offset; 656 int bytes; 657 }; 658 659 typedef struct hammer_undo *hammer_undo_t; 660 661 struct hammer_flusher_info; 662 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info); 663 664 struct hammer_flusher { 665 int signal; /* flusher thread sequencer */ 666 int act; /* currently active flush group */ 667 int done; /* set to act when complete */ 668 int next; /* next flush group */ 669 int group_lock; /* lock sequencing of the next flush */ 670 int exiting; /* request master exit */ 671 thread_t td; /* master flusher thread */ 672 hammer_tid_t tid; /* last flushed transaction id */ 673 int finalize_want; /* serialize finalization */ 674 struct hammer_lock finalize_lock; /* serialize finalization */ 675 struct hammer_transaction trans; /* shared transaction */ 676 struct hammer_flusher_info_list run_list; 677 struct hammer_flusher_info_list ready_list; 678 }; 679 680 /* 681 * Internal hammer mount data structure 682 */ 683 struct hammer_mount { 684 struct mount *mp; 685 /*struct vnode *rootvp;*/ 686 struct hammer_ino_rb_tree rb_inos_root; 687 struct hammer_vol_rb_tree rb_vols_root; 688 struct hammer_nod_rb_tree rb_nods_root; 689 struct hammer_und_rb_tree rb_undo_root; 690 struct hammer_res_rb_tree rb_resv_root; 691 struct hammer_buf_rb_tree rb_bufs_root; 692 struct hammer_pfs_rb_tree rb_pfsm_root; 693 struct hammer_volume *rootvol; 694 struct hammer_base_elm root_btree_beg; 695 struct hammer_base_elm root_btree_end; 696 697 struct malloc_type *m_misc; 698 struct malloc_type *m_inodes; 699 700 int flags; /* HAMMER_MOUNT_xxx flags */ 701 int hflags; 702 int ronly; 703 int nvolumes; 704 int volume_iterator; 705 int master_id; /* -1 or 0-15 - clustering and mirroring */ 706 int version; /* hammer filesystem version to use */ 707 int rsv_inodes; /* reserved space due to dirty inodes */ 708 int64_t rsv_databytes; /* reserved space due to record data */ 709 int rsv_recs; /* reserved space due to dirty records */ 710 int rsv_fromdelay; /* bigblocks reserved due to flush delay */ 711 int undo_rec_limit; /* based on size of undo area */ 712 int last_newrecords; 713 int count_newrecords; 714 715 int inode_reclaims; /* inodes pending reclaim by flusher */ 716 int count_inodes; /* total number of inodes */ 717 int count_iqueued; /* inodes queued to flusher */ 718 719 struct hammer_flusher flusher; 720 721 u_int check_interrupt; 722 uuid_t fsid; 723 struct hammer_io_list volu_list; /* dirty undo buffers */ 724 struct hammer_io_list undo_list; /* dirty undo buffers */ 725 struct hammer_io_list data_list; /* dirty data buffers */ 726 struct hammer_io_list alt_data_list; /* dirty data buffers */ 727 struct hammer_io_list meta_list; /* dirty meta bufs */ 728 struct hammer_io_list lose_list; /* loose buffers */ 729 int locked_dirty_space; /* meta/volu count */ 730 int io_running_space; 731 int objid_cache_count; 732 int error; /* critical I/O error */ 733 struct krate krate; /* rate limited kprintf */ 734 hammer_tid_t asof; /* snapshot mount */ 735 hammer_tid_t next_tid; 736 hammer_tid_t flush_tid1; /* flusher tid sequencing */ 737 hammer_tid_t flush_tid2; /* flusher tid sequencing */ 738 int64_t copy_stat_freebigblocks; /* number of free bigblocks */ 739 740 struct netexport export; 741 struct hammer_lock sync_lock; 742 struct hammer_lock free_lock; 743 struct hammer_lock undo_lock; 744 struct hammer_lock blkmap_lock; 745 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES]; 746 struct hammer_undo undos[HAMMER_MAX_UNDOS]; 747 int undo_alloc; 748 TAILQ_HEAD(, hammer_undo) undo_lru_list; 749 TAILQ_HEAD(, hammer_reserve) delay_list; 750 struct hammer_flush_group_list flush_group_list; 751 hammer_flush_group_t next_flush_group; 752 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list; 753 TAILQ_HEAD(, hammer_reclaim) reclaim_list; 754 }; 755 756 typedef struct hammer_mount *hammer_mount_t; 757 758 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001 759 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002 760 761 struct hammer_sync_info { 762 int error; 763 int waitfor; 764 }; 765 766 #endif 767 768 /* 769 * checkspace slop (8MB chunks), higher numbers are more conservative. 770 */ 771 #define HAMMER_CHKSPC_REBLOCK 25 772 #define HAMMER_CHKSPC_MIRROR 20 773 #define HAMMER_CHKSPC_WRITE 20 774 #define HAMMER_CHKSPC_CREATE 20 775 #define HAMMER_CHKSPC_REMOVE 10 776 #define HAMMER_CHKSPC_EMERGENCY 0 777 778 #if defined(_KERNEL) 779 780 extern struct vop_ops hammer_vnode_vops; 781 extern struct vop_ops hammer_spec_vops; 782 extern struct vop_ops hammer_fifo_vops; 783 extern struct bio_ops hammer_bioops; 784 785 extern int hammer_debug_io; 786 extern int hammer_debug_general; 787 extern int hammer_debug_debug; 788 extern int hammer_debug_inode; 789 extern int hammer_debug_locks; 790 extern int hammer_debug_btree; 791 extern int hammer_debug_tid; 792 extern int hammer_debug_recover; 793 extern int hammer_debug_recover_faults; 794 extern int hammer_cluster_enable; 795 extern int hammer_count_fsyncs; 796 extern int hammer_count_inodes; 797 extern int hammer_count_iqueued; 798 extern int hammer_count_reclaiming; 799 extern int hammer_count_records; 800 extern int hammer_count_record_datas; 801 extern int hammer_count_volumes; 802 extern int hammer_count_buffers; 803 extern int hammer_count_nodes; 804 extern int64_t hammer_count_extra_space_used; 805 extern int64_t hammer_stats_btree_lookups; 806 extern int64_t hammer_stats_btree_searches; 807 extern int64_t hammer_stats_btree_inserts; 808 extern int64_t hammer_stats_btree_deletes; 809 extern int64_t hammer_stats_btree_elements; 810 extern int64_t hammer_stats_btree_splits; 811 extern int64_t hammer_stats_btree_iterations; 812 extern int64_t hammer_stats_record_iterations; 813 extern int64_t hammer_stats_file_read; 814 extern int64_t hammer_stats_file_write; 815 extern int64_t hammer_stats_file_iopsr; 816 extern int64_t hammer_stats_file_iopsw; 817 extern int64_t hammer_stats_disk_read; 818 extern int64_t hammer_stats_disk_write; 819 extern int64_t hammer_stats_inode_flushes; 820 extern int64_t hammer_stats_commits; 821 extern int hammer_count_dirtybufspace; 822 extern int hammer_count_refedbufs; 823 extern int hammer_count_reservations; 824 extern int hammer_count_io_running_read; 825 extern int hammer_count_io_running_write; 826 extern int hammer_count_io_locked; 827 extern int hammer_limit_dirtybufspace; 828 extern int hammer_limit_recs; 829 extern int hammer_bio_count; 830 extern int hammer_verify_zone; 831 extern int hammer_verify_data; 832 extern int hammer_write_mode; 833 extern int hammer_autoflush; 834 extern int64_t hammer_contention_count; 835 836 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip, 837 int error, const char *msg); 838 int hammer_vop_inactive(struct vop_inactive_args *); 839 int hammer_vop_reclaim(struct vop_reclaim_args *); 840 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp); 841 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans, 842 hammer_inode_t dip, int64_t obj_id, 843 hammer_tid_t asof, u_int32_t localization, 844 int flags, int *errorp); 845 void hammer_scan_inode_snapshots(hammer_mount_t hmp, 846 hammer_inode_info_t iinfo, 847 int (*callback)(hammer_inode_t ip, void *data), 848 void *data); 849 void hammer_put_inode(struct hammer_inode *ip); 850 void hammer_put_inode_ref(struct hammer_inode *ip); 851 void hammer_inode_waitreclaims(hammer_mount_t hmp); 852 void hammer_inode_waithard(hammer_mount_t hmp); 853 854 int hammer_unload_volume(hammer_volume_t volume, void *data __unused); 855 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused); 856 857 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused); 858 int hammer_install_volume(hammer_mount_t hmp, const char *volname, 859 struct vnode *devvp); 860 int hammer_mountcheck_volumes(hammer_mount_t hmp); 861 862 int hammer_mem_add(hammer_record_t record); 863 int hammer_ip_lookup(hammer_cursor_t cursor); 864 int hammer_ip_first(hammer_cursor_t cursor); 865 int hammer_ip_next(hammer_cursor_t cursor); 866 int hammer_ip_resolve_data(hammer_cursor_t cursor); 867 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip, 868 hammer_tid_t tid); 869 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags, 870 hammer_tid_t delete_tid, u_int32_t delete_ts, 871 int track, int64_t *stat_bytes); 872 int hammer_ip_check_directory_empty(hammer_transaction_t trans, 873 hammer_inode_t ip); 874 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor); 875 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor); 876 877 hammer_record_t 878 hammer_alloc_mem_record(hammer_inode_t ip, int data_len); 879 void hammer_flush_record_done(hammer_record_t record, int error); 880 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident); 881 void hammer_rel_mem_record(hammer_record_t record); 882 883 int hammer_cursor_up(hammer_cursor_t cursor); 884 int hammer_cursor_up_locked(hammer_cursor_t cursor); 885 int hammer_cursor_down(hammer_cursor_t cursor); 886 int hammer_cursor_upgrade(hammer_cursor_t cursor); 887 int hammer_cursor_upgrade_node(hammer_cursor_t cursor); 888 void hammer_cursor_downgrade(hammer_cursor_t cursor); 889 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, 890 int index); 891 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident); 892 int hammer_lock_ex_try(struct hammer_lock *lock); 893 void hammer_lock_sh(struct hammer_lock *lock); 894 int hammer_lock_sh_try(struct hammer_lock *lock); 895 int hammer_lock_upgrade(struct hammer_lock *lock); 896 void hammer_lock_downgrade(struct hammer_lock *lock); 897 int hammer_lock_status(struct hammer_lock *lock); 898 void hammer_unlock(struct hammer_lock *lock); 899 void hammer_ref(struct hammer_lock *lock); 900 void hammer_unref(struct hammer_lock *lock); 901 902 void hammer_sync_lock_ex(hammer_transaction_t trans); 903 void hammer_sync_lock_sh(hammer_transaction_t trans); 904 int hammer_sync_lock_sh_try(hammer_transaction_t trans); 905 void hammer_sync_unlock(hammer_transaction_t trans); 906 907 u_int32_t hammer_to_unix_xid(uuid_t *uuid); 908 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid); 909 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts); 910 u_int64_t hammer_timespec_to_time(struct timespec *ts); 911 int hammer_str_to_tid(const char *str, int *ispfsp, 912 hammer_tid_t *tidp, u_int32_t *localizationp); 913 int hammer_is_atatext(const char *name, int len); 914 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip); 915 void hammer_clear_objid(hammer_inode_t dip); 916 void hammer_destroy_objid_cache(hammer_mount_t hmp); 917 918 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset, 919 int bytes); 920 void hammer_clear_undo_history(hammer_mount_t hmp); 921 enum vtype hammer_get_vnode_type(u_int8_t obj_type); 922 int hammer_get_dtype(u_int8_t obj_type); 923 u_int8_t hammer_get_obj_type(enum vtype vtype); 924 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len, 925 u_int32_t *max_iterationsp); 926 int hammer_nohistory(hammer_inode_t ip); 927 928 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor, 929 hammer_node_cache_t cache, hammer_inode_t ip); 930 void hammer_normalize_cursor(hammer_cursor_t cursor); 931 void hammer_done_cursor(hammer_cursor_t cursor); 932 int hammer_recover_cursor(hammer_cursor_t cursor); 933 void hammer_unlock_cursor(hammer_cursor_t cursor); 934 int hammer_lock_cursor(hammer_cursor_t cursor); 935 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor); 936 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor); 937 938 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode); 939 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent, 940 int index); 941 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode, 942 int index); 943 void hammer_cursor_inserted_element(hammer_node_t node, int index); 944 void hammer_cursor_deleted_element(hammer_node_t node, int index); 945 946 int hammer_btree_lookup(hammer_cursor_t cursor); 947 int hammer_btree_first(hammer_cursor_t cursor); 948 int hammer_btree_last(hammer_cursor_t cursor); 949 int hammer_btree_extract(hammer_cursor_t cursor, int flags); 950 int hammer_btree_iterate(hammer_cursor_t cursor); 951 int hammer_btree_iterate_reverse(hammer_cursor_t cursor); 952 int hammer_btree_insert(hammer_cursor_t cursor, 953 hammer_btree_leaf_elm_t elm, int *doprop); 954 int hammer_btree_delete(hammer_cursor_t cursor); 955 void hammer_btree_do_propagation(hammer_cursor_t cursor, 956 hammer_pseudofs_inmem_t pfsm, 957 hammer_btree_leaf_elm_t leaf); 958 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2); 959 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key); 960 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid); 961 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid); 962 963 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node, 964 hammer_btree_elm_t elm); 965 int hammer_btree_lock_children(hammer_cursor_t cursor, 966 struct hammer_node_locklist **locklistp); 967 void hammer_btree_unlock_children(hammer_cursor_t cursor, 968 struct hammer_node_locklist **locklistp); 969 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node); 970 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans, 971 hammer_node_t node, int *parent_indexp, 972 int *errorp, int try_exclusive); 973 974 void hammer_print_btree_node(hammer_node_ondisk_t ondisk); 975 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i); 976 977 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off, 978 int *errorp, struct hammer_buffer **bufferp); 979 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off, 980 int *errorp, struct hammer_buffer **bufferp); 981 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes, 982 int *errorp, struct hammer_buffer **bufferp); 983 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes, 984 int *errorp, struct hammer_buffer **bufferp); 985 986 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp); 987 988 hammer_volume_t hammer_get_volume(hammer_mount_t hmp, 989 int32_t vol_no, int *errorp); 990 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset, 991 int bytes, int isnew, int *errorp); 992 void hammer_sync_buffers(hammer_mount_t hmp, 993 hammer_off_t base_offset, int bytes); 994 int hammer_del_buffers(hammer_mount_t hmp, 995 hammer_off_t base_offset, 996 hammer_off_t zone2_offset, int bytes, 997 int report_conflicts); 998 999 int hammer_ref_volume(hammer_volume_t volume); 1000 int hammer_ref_buffer(hammer_buffer_t buffer); 1001 void hammer_flush_buffer_nodes(hammer_buffer_t buffer); 1002 1003 void hammer_rel_volume(hammer_volume_t volume, int flush); 1004 void hammer_rel_buffer(hammer_buffer_t buffer, int flush); 1005 1006 int hammer_vfs_export(struct mount *mp, int op, 1007 const struct export_args *export); 1008 hammer_node_t hammer_get_node(hammer_transaction_t trans, 1009 hammer_off_t node_offset, int isnew, int *errorp); 1010 void hammer_ref_node(hammer_node_t node); 1011 hammer_node_t hammer_ref_node_safe(struct hammer_mount *hmp, 1012 hammer_node_cache_t cache, int *errorp); 1013 void hammer_rel_node(hammer_node_t node); 1014 void hammer_delete_node(hammer_transaction_t trans, 1015 hammer_node_t node); 1016 void hammer_cache_node(hammer_node_cache_t cache, 1017 hammer_node_t node); 1018 void hammer_uncache_node(hammer_node_cache_t cache); 1019 void hammer_flush_node(hammer_node_t node); 1020 1021 void hammer_dup_buffer(struct hammer_buffer **bufferp, 1022 struct hammer_buffer *buffer); 1023 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp); 1024 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len, 1025 u_int16_t rec_type, hammer_off_t *data_offsetp, 1026 struct hammer_buffer **data_bufferp, int *errorp); 1027 1028 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io, 1029 hammer_off_t zone1_offset, void *base, int len); 1030 1031 void hammer_put_volume(struct hammer_volume *volume, int flush); 1032 void hammer_put_buffer(struct hammer_buffer *buffer, int flush); 1033 1034 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans, 1035 hammer_off_t owner, int *errorp); 1036 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset, 1037 hammer_off_t owner, int *errorp); 1038 int hammer_checkspace(hammer_mount_t hmp, int slop); 1039 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone, 1040 int bytes, int *errorp); 1041 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone, 1042 int bytes, hammer_off_t *zone_offp, int *errorp); 1043 void hammer_blockmap_reserve_undo(hammer_mount_t hmp, hammer_reserve_t resv, 1044 hammer_off_t zone_offset, int bytes); 1045 void hammer_blockmap_reserve_complete(hammer_mount_t hmp, 1046 hammer_reserve_t resv); 1047 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv); 1048 void hammer_blockmap_free(hammer_transaction_t trans, 1049 hammer_off_t bmap_off, int bytes); 1050 int hammer_blockmap_finalize(hammer_transaction_t trans, 1051 hammer_reserve_t resv, 1052 hammer_off_t bmap_off, int bytes); 1053 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off, 1054 int *curp, int *errorp); 1055 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, 1056 int *errorp); 1057 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, 1058 int *errorp); 1059 int64_t hammer_undo_used(hammer_transaction_t trans); 1060 int64_t hammer_undo_space(hammer_transaction_t trans); 1061 int64_t hammer_undo_max(hammer_mount_t hmp); 1062 1063 void hammer_start_transaction(struct hammer_transaction *trans, 1064 struct hammer_mount *hmp); 1065 void hammer_simple_transaction(struct hammer_transaction *trans, 1066 struct hammer_mount *hmp); 1067 void hammer_start_transaction_fls(struct hammer_transaction *trans, 1068 struct hammer_mount *hmp); 1069 void hammer_done_transaction(struct hammer_transaction *trans); 1070 1071 void hammer_modify_inode(hammer_inode_t ip, int flags); 1072 void hammer_flush_inode(hammer_inode_t ip, int flags); 1073 void hammer_flush_inode_done(hammer_inode_t ip, int error); 1074 void hammer_wait_inode(hammer_inode_t ip); 1075 1076 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap, 1077 struct ucred *cred, struct hammer_inode *dip, 1078 hammer_pseudofs_inmem_t pfsm, 1079 struct hammer_inode **ipp); 1080 void hammer_rel_inode(hammer_inode_t ip, int flush); 1081 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused); 1082 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2); 1083 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused); 1084 1085 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip); 1086 void hammer_test_inode(hammer_inode_t dip); 1087 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp); 1088 1089 int hammer_ip_add_directory(struct hammer_transaction *trans, 1090 hammer_inode_t dip, const char *name, int bytes, 1091 hammer_inode_t nip); 1092 int hammer_ip_del_directory(struct hammer_transaction *trans, 1093 hammer_cursor_t cursor, hammer_inode_t dip, 1094 hammer_inode_t ip); 1095 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset, 1096 void *data, int bytes, int *errorp); 1097 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size); 1098 int hammer_ip_add_record(struct hammer_transaction *trans, 1099 hammer_record_t record); 1100 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip, 1101 int64_t ran_beg, int64_t ran_end, int truncating); 1102 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip, 1103 int *countp); 1104 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip, 1105 int64_t offset, void *data, int bytes); 1106 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec); 1107 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec); 1108 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans, 1109 u_int32_t localization, int *errorp); 1110 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred, 1111 hammer_pseudofs_inmem_t pfsm); 1112 int hammer_save_pseudofs(hammer_transaction_t trans, 1113 hammer_pseudofs_inmem_t pfsm); 1114 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization); 1115 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm); 1116 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag, 1117 struct ucred *cred); 1118 1119 void hammer_io_init(hammer_io_t io, hammer_volume_t volume, 1120 enum hammer_io_type type); 1121 int hammer_io_read(struct vnode *devvp, struct hammer_io *io, 1122 hammer_off_t limit); 1123 int hammer_io_new(struct vnode *devvp, struct hammer_io *io); 1124 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset); 1125 struct buf *hammer_io_release(struct hammer_io *io, int flush); 1126 void hammer_io_flush(struct hammer_io *io); 1127 void hammer_io_wait(struct hammer_io *io); 1128 void hammer_io_waitdep(struct hammer_io *io); 1129 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident); 1130 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio, 1131 hammer_btree_leaf_elm_t leaf); 1132 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record, 1133 struct bio *bio); 1134 void hammer_io_direct_wait(hammer_record_t record); 1135 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf); 1136 void hammer_io_write_interlock(hammer_io_t io); 1137 void hammer_io_done_interlock(hammer_io_t io); 1138 void hammer_io_clear_modify(struct hammer_io *io, int inval); 1139 void hammer_io_clear_modlist(struct hammer_io *io); 1140 void hammer_io_flush_sync(hammer_mount_t hmp); 1141 1142 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume, 1143 void *base, int len); 1144 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer, 1145 void *base, int len); 1146 void hammer_modify_volume_done(hammer_volume_t volume); 1147 void hammer_modify_buffer_done(hammer_buffer_t buffer); 1148 1149 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip, 1150 struct hammer_ioc_reblock *reblock); 1151 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip, 1152 struct hammer_ioc_prune *prune); 1153 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip, 1154 struct hammer_ioc_mirror_rw *mirror); 1155 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip, 1156 struct hammer_ioc_mirror_rw *mirror); 1157 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1158 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs); 1159 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1160 struct hammer_ioc_pseudofs_rw *pfs); 1161 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1162 struct hammer_ioc_pseudofs_rw *pfs); 1163 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1164 struct hammer_ioc_pseudofs_rw *pfs); 1165 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1166 struct hammer_ioc_pseudofs_rw *pfs); 1167 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip, 1168 struct hammer_ioc_pseudofs_rw *pfs); 1169 1170 int hammer_signal_check(hammer_mount_t hmp); 1171 1172 void hammer_flusher_create(hammer_mount_t hmp); 1173 void hammer_flusher_destroy(hammer_mount_t hmp); 1174 void hammer_flusher_sync(hammer_mount_t hmp); 1175 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg); 1176 int hammer_flusher_async_one(hammer_mount_t hmp); 1177 void hammer_flusher_wait(hammer_mount_t hmp, int seq); 1178 void hammer_flusher_wait_next(hammer_mount_t hmp); 1179 int hammer_flusher_meta_limit(hammer_mount_t hmp); 1180 int hammer_flusher_meta_halflimit(hammer_mount_t hmp); 1181 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter); 1182 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp); 1183 void hammer_flusher_finalize(hammer_transaction_t trans, int final); 1184 int hammer_flusher_haswork(hammer_mount_t hmp); 1185 1186 1187 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol); 1188 void hammer_recover_flush_buffers(hammer_mount_t hmp, 1189 hammer_volume_t root_volume, int final); 1190 1191 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap); 1192 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk); 1193 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf); 1194 1195 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap); 1196 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk); 1197 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk); 1198 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf); 1199 void hkprintf(const char *ctl, ...); 1200 udev_t hammer_fsid_to_udev(uuid_t *uuid); 1201 1202 1203 int hammer_blocksize(int64_t file_offset); 1204 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2); 1205 1206 #endif 1207 1208 static __inline void 1209 hammer_wait_mem_record(hammer_record_t record) 1210 { 1211 hammer_wait_mem_record_ident(record, "hmmwai"); 1212 } 1213 1214 static __inline void 1215 hammer_lock_ex(struct hammer_lock *lock) 1216 { 1217 hammer_lock_ex_ident(lock, "hmrlck"); 1218 } 1219 1220 /* 1221 * Indicate that a B-Tree node is being modified. 1222 */ 1223 static __inline void 1224 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node) 1225 { 1226 hammer_modify_buffer(trans, node->buffer, NULL, 0); 1227 } 1228 1229 static __inline void 1230 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node) 1231 { 1232 hammer_modify_buffer(trans, node->buffer, 1233 node->ondisk, sizeof(*node->ondisk)); 1234 } 1235 1236 static __inline void 1237 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node, 1238 void *base, int len) 1239 { 1240 hammer_crc_t *crcptr; 1241 1242 KKASSERT((char *)base >= (char *)node->ondisk && 1243 (char *)base + len <= 1244 (char *)node->ondisk + sizeof(*node->ondisk)); 1245 hammer_modify_buffer(trans, node->buffer, base, len); 1246 crcptr = &node->ondisk->crc; 1247 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t)); 1248 --node->buffer->io.modify_refs; /* only want one ref */ 1249 } 1250 1251 /* 1252 * Indicate that the specified modifications have been completed. 1253 * 1254 * Do not try to generate the crc here, it's very expensive to do and a 1255 * sequence of insertions or deletions can result in many calls to this 1256 * function on the same node. 1257 */ 1258 static __inline void 1259 hammer_modify_node_done(hammer_node_t node) 1260 { 1261 node->flags |= HAMMER_NODE_CRCGOOD; 1262 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) { 1263 node->flags |= HAMMER_NODE_NEEDSCRC; 1264 node->buffer->io.gencrc = 1; 1265 hammer_ref_node(node); 1266 } 1267 hammer_modify_buffer_done(node->buffer); 1268 } 1269 1270 #define hammer_modify_volume_field(trans, vol, field) \ 1271 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \ 1272 sizeof((vol)->ondisk->field)) 1273 1274 #define hammer_modify_node_field(trans, node, field) \ 1275 hammer_modify_node(trans, node, &(node)->ondisk->field, \ 1276 sizeof((node)->ondisk->field)) 1277 1278