| 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 1.55 2008/05/02 01:00:42 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/globaldata.h> |
| 53 | #include <sys/lockf.h> |
| 54 | #include <sys/buf.h> |
| 55 | #include <sys/queue.h> |
| 56 | #include <sys/globaldata.h> |
| 57 | |
| 58 | #include <sys/buf2.h> |
| 59 | #include <sys/signal2.h> |
| 60 | #include "hammer_disk.h" |
| 61 | #include "hammer_mount.h" |
| 62 | #include "hammer_ioctl.h" |
| 63 | |
| 64 | #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES) |
| 65 | |
| 66 | MALLOC_DECLARE(M_HAMMER); |
| 67 | |
| 68 | struct hammer_mount; |
| 69 | |
| 70 | /* |
| 71 | * Key structure used for custom RB tree inode lookups. This prototypes |
| 72 | * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info). |
| 73 | */ |
| 74 | typedef struct hammer_inode_info { |
| 75 | int64_t obj_id; /* (key) object identifier */ |
| 76 | hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */ |
| 77 | } *hammer_inode_info_t; |
| 78 | |
| 79 | typedef enum hammer_transaction_type { |
| 80 | HAMMER_TRANS_RO, |
| 81 | HAMMER_TRANS_STD, |
| 82 | HAMMER_TRANS_FLS |
| 83 | } hammer_transaction_type_t; |
| 84 | |
| 85 | /* |
| 86 | * HAMMER Transaction tracking |
| 87 | */ |
| 88 | struct hammer_transaction { |
| 89 | hammer_transaction_type_t type; |
| 90 | struct hammer_mount *hmp; |
| 91 | hammer_tid_t tid; |
| 92 | hammer_tid_t time; |
| 93 | struct hammer_volume *rootvol; |
| 94 | }; |
| 95 | |
| 96 | typedef struct hammer_transaction *hammer_transaction_t; |
| 97 | |
| 98 | /* |
| 99 | * HAMMER locks |
| 100 | */ |
| 101 | struct hammer_lock { |
| 102 | int refs; /* active references delay writes */ |
| 103 | int lockcount; /* lock count for exclusive/shared access */ |
| 104 | int wanted; |
| 105 | struct thread *locktd; |
| 106 | }; |
| 107 | |
| 108 | static __inline int |
| 109 | hammer_islocked(struct hammer_lock *lock) |
| 110 | { |
| 111 | return(lock->lockcount != 0); |
| 112 | } |
| 113 | |
| 114 | static __inline int |
| 115 | hammer_isactive(struct hammer_lock *lock) |
| 116 | { |
| 117 | return(lock->refs != 0); |
| 118 | } |
| 119 | |
| 120 | static __inline int |
| 121 | hammer_islastref(struct hammer_lock *lock) |
| 122 | { |
| 123 | return(lock->refs == 1); |
| 124 | } |
| 125 | |
| 126 | /* |
| 127 | * Return if we specifically own the lock exclusively. |
| 128 | */ |
| 129 | static __inline int |
| 130 | hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td) |
| 131 | { |
| 132 | if (lock->lockcount > 0 && lock->locktd == td) |
| 133 | return(1); |
| 134 | return(0); |
| 135 | } |
| 136 | |
| 137 | /* |
| 138 | * Flush state, used by various structures |
| 139 | */ |
| 140 | typedef enum hammer_inode_state { |
| 141 | HAMMER_FST_IDLE, |
| 142 | HAMMER_FST_SETUP, |
| 143 | HAMMER_FST_FLUSH |
| 144 | } hammer_inode_state_t; |
| 145 | |
| 146 | TAILQ_HEAD(hammer_record_list, hammer_record); |
| 147 | |
| 148 | /* |
| 149 | * Cache object ids. A fixed number of objid cache structures are |
| 150 | * created to reserve object id's for newly created files in multiples |
| 151 | * of 100,000, localized to a particular directory, and recycled as |
| 152 | * needed. This allows parallel create operations in different |
| 153 | * directories to retain fairly localized object ids which in turn |
| 154 | * improves reblocking performance and layout. |
| 155 | */ |
| 156 | #define OBJID_CACHE_SIZE 128 |
| 157 | #define OBJID_CACHE_BULK 100000 |
| 158 | |
| 159 | typedef struct hammer_objid_cache { |
| 160 | TAILQ_ENTRY(hammer_objid_cache) entry; |
| 161 | struct hammer_inode *dip; |
| 162 | hammer_tid_t next_tid; |
| 163 | int count; |
| 164 | } *hammer_objid_cache_t; |
| 165 | |
| 166 | /* |
| 167 | * Structure used to represent an inode in-memory. |
| 168 | * |
| 169 | * The record and data associated with an inode may be out of sync with |
| 170 | * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag |
| 171 | * clear). |
| 172 | * |
| 173 | * An inode may also hold a cache of unsynchronized records, used for |
| 174 | * database and directories only. Unsynchronized regular file data is |
| 175 | * stored in the buffer cache. |
| 176 | * |
| 177 | * NOTE: A file which is created and destroyed within the initial |
| 178 | * synchronization period can wind up not doing any disk I/O at all. |
| 179 | * |
| 180 | * Finally, an inode may cache numerous disk-referencing B-Tree cursors. |
| 181 | */ |
| 182 | struct hammer_ino_rb_tree; |
| 183 | struct hammer_inode; |
| 184 | RB_HEAD(hammer_ino_rb_tree, hammer_inode); |
| 185 | RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node, |
| 186 | hammer_ino_rb_compare, hammer_inode_info_t); |
| 187 | |
| 188 | struct hammer_rec_rb_tree; |
| 189 | struct hammer_record; |
| 190 | RB_HEAD(hammer_rec_rb_tree, hammer_record); |
| 191 | RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node, |
| 192 | hammer_rec_rb_compare, hammer_base_elm_t); |
| 193 | |
| 194 | TAILQ_HEAD(hammer_node_list, hammer_node); |
| 195 | |
| 196 | struct hammer_inode { |
| 197 | RB_ENTRY(hammer_inode) rb_node; |
| 198 | hammer_inode_state_t flush_state; |
| 199 | int flush_group; |
| 200 | TAILQ_ENTRY(hammer_inode) flush_entry; |
| 201 | struct hammer_record_list target_list; /* target of dependant recs */ |
| 202 | u_int64_t obj_id; /* (key) object identifier */ |
| 203 | hammer_tid_t obj_asof; /* (key) snapshot or 0 */ |
| 204 | struct hammer_mount *hmp; |
| 205 | hammer_objid_cache_t objid_cache; |
| 206 | int flags; |
| 207 | int error; /* flush error */ |
| 208 | int cursor_ip_refs; /* sanity */ |
| 209 | struct vnode *vp; |
| 210 | struct lockf advlock; |
| 211 | struct hammer_lock lock; /* sync copy interlock */ |
| 212 | TAILQ_HEAD(, bio) bio_list; /* BIOs to flush out */ |
| 213 | TAILQ_HEAD(, bio) bio_alt_list; /* BIOs to flush out */ |
| 214 | off_t trunc_off; |
| 215 | struct hammer_inode_record ino_rec; /* in-memory cache */ |
| 216 | struct hammer_inode_data ino_data; /* in-memory cache */ |
| 217 | struct hammer_rec_rb_tree rec_tree; /* in-memory cache */ |
| 218 | struct hammer_node *cache[2]; /* search initiate cache */ |
| 219 | |
| 220 | /* |
| 221 | * When a demark is created to synchronize an inode to |
| 222 | * disk, certain fields are copied so the front-end VOPs |
| 223 | * can continue to run in parallel with the synchronization |
| 224 | * occuring in the background. |
| 225 | */ |
| 226 | int sync_flags; /* to-sync flags cache */ |
| 227 | off_t sync_trunc_off; /* to-sync truncation */ |
| 228 | struct hammer_inode_record sync_ino_rec;/* to-sync cache */ |
| 229 | struct hammer_inode_data sync_ino_data; /* to-sync cache */ |
| 230 | }; |
| 231 | |
| 232 | typedef struct hammer_inode *hammer_inode_t; |
| 233 | |
| 234 | #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data) |
| 235 | |
| 236 | #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */ |
| 237 | #define HAMMER_INODE_RDIRTY 0x0002 /* in-memory ino_rec is dirty */ |
| 238 | #define HAMMER_INODE_ITIMES 0x0004 /* in-memory mtime/atime modified */ |
| 239 | #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */ |
| 240 | #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */ |
| 241 | #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */ |
| 242 | #define HAMMER_INODE_DELETED 0x0080 /* inode ready for deletion */ |
| 243 | #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */ |
| 244 | #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */ |
| 245 | #define HAMMER_INODE_GONE 0x0400 /* delete flushed out */ |
| 246 | #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */ |
| 247 | #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */ |
| 248 | #define HAMMER_INODE_REFLUSH 0x2000 /* pipelined flush during flush */ |
| 249 | #define HAMMER_INODE_WRITE_ALT 0x4000 /* strategy writes to alt bioq */ |
| 250 | #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */ |
| 251 | |
| 252 | #define HAMMER_INODE_TRUNCATED 0x00010000 |
| 253 | #define HAMMER_INODE_DELETING 0x00020000 /* Destroy the inode on-disk */ |
| 254 | |
| 255 | #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY|HAMMER_INODE_RDIRTY| \ |
| 256 | HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \ |
| 257 | HAMMER_INODE_ITIMES|HAMMER_INODE_TRUNCATED|\ |
| 258 | HAMMER_INODE_DELETING) |
| 259 | |
| 260 | #define HAMMER_INODE_MODMASK_NOXDIRTY \ |
| 261 | (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY) |
| 262 | |
| 263 | #define HAMMER_MAX_INODE_CURSORS 4 |
| 264 | |
| 265 | #define HAMMER_FLUSH_SIGNAL 0x0001 |
| 266 | #define HAMMER_FLUSH_FORCE 0x0002 |
| 267 | #define HAMMER_FLUSH_RECURSION 0x0004 |
| 268 | |
| 269 | /* |
| 270 | * Structure used to represent an unsynchronized record in-memory. These |
| 271 | * records typically represent directory entries. Only non-historical |
| 272 | * records are kept in-memory. |
| 273 | * |
| 274 | * Records are organized as a per-inode RB-Tree. If the inode is not |
| 275 | * on disk then neither are any records and the in-memory record tree |
| 276 | * represents the entire contents of the inode. If the inode is on disk |
| 277 | * then the on-disk B-Tree is scanned in parallel with the in-memory |
| 278 | * RB-Tree to synthesize the current state of the file. |
| 279 | * |
| 280 | * Records are also used to enforce the ordering of directory create/delete |
| 281 | * operations. A new inode will not be flushed to disk unless its related |
| 282 | * directory entry is also being flushed at the same time. A directory entry |
| 283 | * will not be removed unless its related inode is also being removed at the |
| 284 | * same time. |
| 285 | */ |
| 286 | typedef enum hammer_record_type { |
| 287 | HAMMER_MEM_RECORD_ADD, /* positive memory cache record */ |
| 288 | HAMMER_MEM_RECORD_DEL /* negative delete-on-disk record */ |
| 289 | } hammer_record_type_t; |
| 290 | |
| 291 | struct hammer_record { |
| 292 | RB_ENTRY(hammer_record) rb_node; |
| 293 | TAILQ_ENTRY(hammer_record) target_entry; |
| 294 | hammer_inode_state_t flush_state; |
| 295 | int flush_group; |
| 296 | hammer_record_type_t type; |
| 297 | struct hammer_lock lock; |
| 298 | struct hammer_inode *ip; |
| 299 | struct hammer_inode *target_ip; |
| 300 | union hammer_record_ondisk rec; |
| 301 | union hammer_data_ondisk *data; |
| 302 | int flags; |
| 303 | }; |
| 304 | |
| 305 | typedef struct hammer_record *hammer_record_t; |
| 306 | |
| 307 | /* |
| 308 | * Record flags. Note that FE can only be set by the frontend if the |
| 309 | * record has not been interlocked by the backend w/ BE. |
| 310 | */ |
| 311 | #define HAMMER_RECF_ALLOCDATA 0x0001 |
| 312 | #define HAMMER_RECF_ONRBTREE 0x0002 |
| 313 | #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */ |
| 314 | #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */ |
| 315 | #define HAMMER_RECF_INBAND 0x0010 |
| 316 | #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */ |
| 317 | #define HAMMER_RECF_WANTED 0x0040 |
| 318 | #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */ |
| 319 | |
| 320 | /* |
| 321 | * In-memory structures representing on-disk structures. |
| 322 | */ |
| 323 | struct hammer_volume; |
| 324 | struct hammer_buffer; |
| 325 | struct hammer_node; |
| 326 | RB_HEAD(hammer_vol_rb_tree, hammer_volume); |
| 327 | RB_HEAD(hammer_buf_rb_tree, hammer_buffer); |
| 328 | RB_HEAD(hammer_nod_rb_tree, hammer_node); |
| 329 | |
| 330 | RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node, |
| 331 | hammer_vol_rb_compare, int32_t); |
| 332 | RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node, |
| 333 | hammer_buf_rb_compare, hammer_off_t); |
| 334 | RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node, |
| 335 | hammer_nod_rb_compare, hammer_off_t); |
| 336 | |
| 337 | /* |
| 338 | * IO management - embedded at the head of various in-memory structures |
| 339 | * |
| 340 | * VOLUME - hammer_volume containing meta-data |
| 341 | * META_BUFFER - hammer_buffer containing meta-data |
| 342 | * DATA_BUFFER - hammer_buffer containing pure-data |
| 343 | * |
| 344 | * Dirty volume headers and dirty meta-data buffers are locked until the |
| 345 | * flusher can sequence them out. Dirty pure-data buffers can be written. |
| 346 | * Clean buffers can be passively released. |
| 347 | */ |
| 348 | typedef enum hammer_io_type { |
| 349 | HAMMER_STRUCTURE_VOLUME, |
| 350 | HAMMER_STRUCTURE_META_BUFFER, |
| 351 | HAMMER_STRUCTURE_UNDO_BUFFER, |
| 352 | HAMMER_STRUCTURE_DATA_BUFFER |
| 353 | } hammer_io_type_t; |
| 354 | |
| 355 | union hammer_io_structure; |
| 356 | struct hammer_io; |
| 357 | |
| 358 | struct worklist { |
| 359 | LIST_ENTRY(worklist) node; |
| 360 | }; |
| 361 | |
| 362 | TAILQ_HEAD(hammer_io_list, hammer_io); |
| 363 | typedef struct hammer_io_list *hammer_io_list_t; |
| 364 | |
| 365 | struct hammer_io { |
| 366 | struct worklist worklist; |
| 367 | struct hammer_lock lock; |
| 368 | enum hammer_io_type type; |
| 369 | struct hammer_mount *hmp; |
| 370 | TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */ |
| 371 | hammer_io_list_t mod_list; |
| 372 | struct buf *bp; |
| 373 | int64_t offset; |
| 374 | int loading; /* loading/unloading interlock */ |
| 375 | int modify_refs; |
| 376 | |
| 377 | u_int modified : 1; /* bp's data was modified */ |
| 378 | u_int released : 1; /* bp released (w/ B_LOCKED set) */ |
| 379 | u_int running : 1; /* bp write IO in progress */ |
| 380 | u_int waiting : 1; /* someone is waiting on us */ |
| 381 | u_int validated : 1; /* ondisk has been validated */ |
| 382 | u_int flush : 1; /* flush on last release */ |
| 383 | u_int waitdep : 1; /* flush waits for dependancies */ |
| 384 | }; |
| 385 | |
| 386 | typedef struct hammer_io *hammer_io_t; |
| 387 | |
| 388 | /* |
| 389 | * In-memory volume representing on-disk buffer |
| 390 | */ |
| 391 | struct hammer_volume { |
| 392 | struct hammer_io io; |
| 393 | RB_ENTRY(hammer_volume) rb_node; |
| 394 | struct hammer_buf_rb_tree rb_bufs_root; |
| 395 | struct hammer_volume_ondisk *ondisk; |
| 396 | int32_t vol_no; |
| 397 | int64_t nblocks; /* note: special calculation for statfs */ |
| 398 | int64_t buffer_base; /* base offset of buffer 0 */ |
| 399 | hammer_off_t maxbuf_off; /* Maximum buffer offset */ |
| 400 | char *vol_name; |
| 401 | struct vnode *devvp; |
| 402 | int vol_flags; |
| 403 | }; |
| 404 | |
| 405 | typedef struct hammer_volume *hammer_volume_t; |
| 406 | |
| 407 | /* |
| 408 | * In-memory buffer (other then volume, super-cluster, or cluster), |
| 409 | * representing an on-disk buffer. |
| 410 | */ |
| 411 | struct hammer_buffer { |
| 412 | struct hammer_io io; |
| 413 | RB_ENTRY(hammer_buffer) rb_node; |
| 414 | void *ondisk; |
| 415 | struct hammer_volume *volume; |
| 416 | hammer_off_t zone2_offset; |
| 417 | hammer_off_t zoneX_offset; |
| 418 | struct hammer_node_list clist; |
| 419 | }; |
| 420 | |
| 421 | typedef struct hammer_buffer *hammer_buffer_t; |
| 422 | |
| 423 | /* |
| 424 | * In-memory B-Tree node, representing an on-disk B-Tree node. |
| 425 | * |
| 426 | * This is a hang-on structure which is backed by a hammer_buffer, |
| 427 | * indexed by a hammer_cluster, and used for fine-grained locking of |
| 428 | * B-Tree nodes in order to properly control lock ordering. A hammer_buffer |
| 429 | * can contain multiple nodes representing wildly disassociated portions |
| 430 | * of the B-Tree so locking cannot be done on a buffer-by-buffer basis. |
| 431 | * |
| 432 | * This structure uses a cluster-relative index to reduce the number |
| 433 | * of layers required to access it, and also because all on-disk B-Tree |
| 434 | * references are cluster-relative offsets. |
| 435 | */ |
| 436 | struct hammer_node { |
| 437 | struct hammer_lock lock; /* node-by-node lock */ |
| 438 | TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */ |
| 439 | RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */ |
| 440 | hammer_off_t node_offset; /* full offset spec */ |
| 441 | struct hammer_mount *hmp; |
| 442 | struct hammer_buffer *buffer; /* backing buffer */ |
| 443 | hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */ |
| 444 | struct hammer_node **cache1; /* passive cache(s) */ |
| 445 | struct hammer_node **cache2; |
| 446 | int flags; |
| 447 | int loading; /* load interlock */ |
| 448 | }; |
| 449 | |
| 450 | #define HAMMER_NODE_DELETED 0x0001 |
| 451 | #define HAMMER_NODE_FLUSH 0x0002 |
| 452 | |
| 453 | typedef struct hammer_node *hammer_node_t; |
| 454 | |
| 455 | /* |
| 456 | * List of locked nodes. |
| 457 | */ |
| 458 | struct hammer_node_locklist { |
| 459 | struct hammer_node_locklist *next; |
| 460 | hammer_node_t node; |
| 461 | }; |
| 462 | |
| 463 | typedef struct hammer_node_locklist *hammer_node_locklist_t; |
| 464 | |
| 465 | |
| 466 | /* |
| 467 | * Common I/O management structure - embedded in in-memory structures |
| 468 | * which are backed by filesystem buffers. |
| 469 | */ |
| 470 | union hammer_io_structure { |
| 471 | struct hammer_io io; |
| 472 | struct hammer_volume volume; |
| 473 | struct hammer_buffer buffer; |
| 474 | }; |
| 475 | |
| 476 | typedef union hammer_io_structure *hammer_io_structure_t; |
| 477 | |
| 478 | /* |
| 479 | * Allocation holes are recorded for a short period of time in an attempt |
| 480 | * to use up the space. |
| 481 | */ |
| 482 | |
| 483 | #define HAMMER_MAX_HOLES 8 |
| 484 | |
| 485 | struct hammer_hole; |
| 486 | |
| 487 | struct hammer_holes { |
| 488 | TAILQ_HEAD(, hammer_hole) list; |
| 489 | int count; |
| 490 | }; |
| 491 | |
| 492 | typedef struct hammer_holes *hammer_holes_t; |
| 493 | |
| 494 | struct hammer_hole { |
| 495 | TAILQ_ENTRY(hammer_hole) entry; |
| 496 | hammer_off_t offset; |
| 497 | int bytes; |
| 498 | }; |
| 499 | |
| 500 | typedef struct hammer_hole *hammer_hole_t; |
| 501 | |
| 502 | #include "hammer_cursor.h" |
| 503 | |
| 504 | /* |
| 505 | * Internal hammer mount data structure |
| 506 | */ |
| 507 | struct hammer_mount { |
| 508 | struct mount *mp; |
| 509 | /*struct vnode *rootvp;*/ |
| 510 | struct hammer_ino_rb_tree rb_inos_root; |
| 511 | struct hammer_vol_rb_tree rb_vols_root; |
| 512 | struct hammer_nod_rb_tree rb_nods_root; |
| 513 | struct hammer_volume *rootvol; |
| 514 | struct hammer_base_elm root_btree_beg; |
| 515 | struct hammer_base_elm root_btree_end; |
| 516 | char *zbuf; /* HAMMER_BUFSIZE bytes worth of all-zeros */ |
| 517 | int hflags; |
| 518 | int ronly; |
| 519 | int nvolumes; |
| 520 | int volume_iterator; |
| 521 | int flusher_signal; /* flusher thread sequencer */ |
| 522 | int flusher_act; /* currently active flush group */ |
| 523 | int flusher_done; /* set to act when complete */ |
| 524 | int flusher_next; /* next flush group */ |
| 525 | int flusher_exiting; |
| 526 | int reclaim_count; |
| 527 | thread_t flusher_td; |
| 528 | u_int check_interrupt; |
| 529 | uuid_t fsid; |
| 530 | udev_t fsid_udev; |
| 531 | struct hammer_io_list volu_list; /* dirty undo buffers */ |
| 532 | struct hammer_io_list undo_list; /* dirty undo buffers */ |
| 533 | struct hammer_io_list data_list; /* dirty data buffers */ |
| 534 | struct hammer_io_list meta_list; /* dirty meta bufs */ |
| 535 | struct hammer_io_list lose_list; /* loose buffers */ |
| 536 | int locked_dirty_count; /* meta/volu count */ |
| 537 | int io_running_count; |
| 538 | int objid_cache_count; |
| 539 | hammer_tid_t asof; |
| 540 | hammer_off_t next_tid; |
| 541 | u_int32_t namekey_iterator; |
| 542 | hammer_off_t zone_limits[HAMMER_MAX_ZONES]; |
| 543 | struct netexport export; |
| 544 | struct hammer_lock sync_lock; |
| 545 | struct lock blockmap_lock; |
| 546 | struct hammer_blockmap blockmap[HAMMER_MAX_ZONES]; |
| 547 | struct hammer_holes holes[HAMMER_MAX_ZONES]; |
| 548 | TAILQ_HEAD(, hammer_inode) flush_list; |
| 549 | TAILQ_HEAD(, hammer_objid_cache) objid_cache_list; |
| 550 | }; |
| 551 | |
| 552 | typedef struct hammer_mount *hammer_mount_t; |
| 553 | |
| 554 | struct hammer_sync_info { |
| 555 | int error; |
| 556 | int waitfor; |
| 557 | }; |
| 558 | |
| 559 | #endif |
| 560 | |
| 561 | #if defined(_KERNEL) |
| 562 | |
| 563 | extern struct vop_ops hammer_vnode_vops; |
| 564 | extern struct vop_ops hammer_spec_vops; |
| 565 | extern struct vop_ops hammer_fifo_vops; |
| 566 | extern struct bio_ops hammer_bioops; |
| 567 | |
| 568 | extern int hammer_debug_general; |
| 569 | extern int hammer_debug_locks; |
| 570 | extern int hammer_debug_btree; |
| 571 | extern int hammer_debug_tid; |
| 572 | extern int hammer_debug_recover; |
| 573 | extern int hammer_debug_recover_faults; |
| 574 | extern int hammer_count_inodes; |
| 575 | extern int hammer_count_records; |
| 576 | extern int hammer_count_record_datas; |
| 577 | extern int hammer_count_volumes; |
| 578 | extern int hammer_count_buffers; |
| 579 | extern int hammer_count_nodes; |
| 580 | extern int hammer_count_dirtybufs; |
| 581 | extern int hammer_limit_dirtybufs; |
| 582 | extern int hammer_bio_count; |
| 583 | extern int64_t hammer_contention_count; |
| 584 | |
| 585 | int hammer_vop_inactive(struct vop_inactive_args *); |
| 586 | int hammer_vop_reclaim(struct vop_reclaim_args *); |
| 587 | int hammer_get_vnode(struct hammer_inode *ip, int lktype, |
| 588 | struct vnode **vpp); |
| 589 | struct hammer_inode *hammer_get_inode(hammer_transaction_t trans, |
| 590 | struct hammer_node **cache, |
| 591 | u_int64_t obj_id, hammer_tid_t asof, int flags, |
| 592 | int *errorp); |
| 593 | void hammer_put_inode(struct hammer_inode *ip); |
| 594 | void hammer_put_inode_ref(struct hammer_inode *ip); |
| 595 | |
| 596 | int hammer_unload_volume(hammer_volume_t volume, void *data __unused); |
| 597 | int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused); |
| 598 | int hammer_install_volume(hammer_mount_t hmp, const char *volname); |
| 599 | |
| 600 | int hammer_ip_lookup(hammer_cursor_t cursor, hammer_inode_t ip); |
| 601 | int hammer_ip_first(hammer_cursor_t cursor, hammer_inode_t ip); |
| 602 | int hammer_ip_next(hammer_cursor_t cursor); |
| 603 | int hammer_ip_resolve_record_and_data(hammer_cursor_t cursor); |
| 604 | int hammer_ip_resolve_data(hammer_cursor_t cursor); |
| 605 | int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_tid_t tid); |
| 606 | int hammer_delete_at_cursor(hammer_cursor_t cursor, int64_t *stat_bytes); |
| 607 | int hammer_ip_check_directory_empty(hammer_transaction_t trans, |
| 608 | hammer_cursor_t parent_cursor, hammer_inode_t ip); |
| 609 | int hammer_sync_hmp(hammer_mount_t hmp, int waitfor); |
| 610 | |
| 611 | hammer_record_t |
| 612 | hammer_alloc_mem_record(hammer_inode_t ip); |
| 613 | void hammer_flush_record_done(hammer_record_t record, int error); |
| 614 | void hammer_wait_mem_record(hammer_record_t record); |
| 615 | void hammer_rel_mem_record(hammer_record_t record); |
| 616 | |
| 617 | int hammer_cursor_up(hammer_cursor_t cursor); |
| 618 | int hammer_cursor_down(hammer_cursor_t cursor); |
| 619 | int hammer_cursor_upgrade(hammer_cursor_t cursor); |
| 620 | void hammer_cursor_downgrade(hammer_cursor_t cursor); |
| 621 | int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node, |
| 622 | int index); |
| 623 | void hammer_lock_ex(struct hammer_lock *lock); |
| 624 | int hammer_lock_ex_try(struct hammer_lock *lock); |
| 625 | void hammer_lock_sh(struct hammer_lock *lock); |
| 626 | int hammer_lock_upgrade(struct hammer_lock *lock); |
| 627 | void hammer_lock_downgrade(struct hammer_lock *lock); |
| 628 | void hammer_unlock(struct hammer_lock *lock); |
| 629 | void hammer_ref(struct hammer_lock *lock); |
| 630 | void hammer_unref(struct hammer_lock *lock); |
| 631 | |
| 632 | u_int32_t hammer_to_unix_xid(uuid_t *uuid); |
| 633 | void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid); |
| 634 | void hammer_to_timespec(hammer_tid_t tid, struct timespec *ts); |
| 635 | hammer_tid_t hammer_timespec_to_transid(struct timespec *ts); |
| 636 | hammer_tid_t hammer_now_tid(void); |
| 637 | hammer_tid_t hammer_str_to_tid(const char *str); |
| 638 | hammer_tid_t hammer_alloc_objid(hammer_transaction_t trans, hammer_inode_t dip); |
| 639 | void hammer_clear_objid(hammer_inode_t dip); |
| 640 | void hammer_destroy_objid_cache(hammer_mount_t hmp); |
| 641 | |
| 642 | enum vtype hammer_get_vnode_type(u_int8_t obj_type); |
| 643 | int hammer_get_dtype(u_int8_t obj_type); |
| 644 | u_int8_t hammer_get_obj_type(enum vtype vtype); |
| 645 | int64_t hammer_directory_namekey(void *name, int len); |
| 646 | |
| 647 | int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor, |
| 648 | struct hammer_node **cache); |
| 649 | |
| 650 | void hammer_done_cursor(hammer_cursor_t cursor); |
| 651 | void hammer_mem_done(hammer_cursor_t cursor); |
| 652 | |
| 653 | int hammer_btree_lookup(hammer_cursor_t cursor); |
| 654 | int hammer_btree_first(hammer_cursor_t cursor); |
| 655 | int hammer_btree_last(hammer_cursor_t cursor); |
| 656 | int hammer_btree_extract(hammer_cursor_t cursor, int flags); |
| 657 | int hammer_btree_iterate(hammer_cursor_t cursor); |
| 658 | int hammer_btree_iterate_reverse(hammer_cursor_t cursor); |
| 659 | int hammer_btree_insert(hammer_cursor_t cursor, hammer_btree_elm_t elm); |
| 660 | int hammer_btree_delete(hammer_cursor_t cursor); |
| 661 | int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2); |
| 662 | int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key); |
| 663 | int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid); |
| 664 | int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid); |
| 665 | |
| 666 | |
| 667 | int hammer_btree_lock_children(hammer_cursor_t cursor, |
| 668 | struct hammer_node_locklist **locklistp); |
| 669 | |
| 670 | void hammer_print_btree_node(hammer_node_ondisk_t ondisk); |
| 671 | void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i); |
| 672 | |
| 673 | void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off, |
| 674 | int *errorp, struct hammer_buffer **bufferp); |
| 675 | void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off, |
| 676 | int *errorp, struct hammer_buffer **bufferp); |
| 677 | |
| 678 | hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp); |
| 679 | int hammer_dowrite(hammer_transaction_t trans, hammer_inode_t ip, |
| 680 | struct bio *bio); |
| 681 | |
| 682 | hammer_volume_t hammer_get_volume(hammer_mount_t hmp, |
| 683 | int32_t vol_no, int *errorp); |
| 684 | hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, |
| 685 | hammer_off_t buf_offset, int isnew, int *errorp); |
| 686 | void hammer_uncache_buffer(struct hammer_mount *hmp, hammer_off_t off); |
| 687 | |
| 688 | int hammer_ref_volume(hammer_volume_t volume); |
| 689 | int hammer_ref_buffer(hammer_buffer_t buffer); |
| 690 | void hammer_flush_buffer_nodes(hammer_buffer_t buffer); |
| 691 | |
| 692 | void hammer_rel_volume(hammer_volume_t volume, int flush); |
| 693 | void hammer_rel_buffer(hammer_buffer_t buffer, int flush); |
| 694 | |
| 695 | int hammer_vfs_export(struct mount *mp, int op, |
| 696 | const struct export_args *export); |
| 697 | hammer_node_t hammer_get_node(hammer_mount_t hmp, |
| 698 | hammer_off_t node_offset, int *errorp); |
| 699 | void hammer_ref_node(hammer_node_t node); |
| 700 | hammer_node_t hammer_ref_node_safe(struct hammer_mount *hmp, |
| 701 | struct hammer_node **cache, int *errorp); |
| 702 | void hammer_rel_node(hammer_node_t node); |
| 703 | void hammer_delete_node(hammer_transaction_t trans, |
| 704 | hammer_node_t node); |
| 705 | void hammer_cache_node(hammer_node_t node, |
| 706 | struct hammer_node **cache); |
| 707 | void hammer_uncache_node(struct hammer_node **cache); |
| 708 | void hammer_flush_node(hammer_node_t node); |
| 709 | |
| 710 | void hammer_dup_buffer(struct hammer_buffer **bufferp, |
| 711 | struct hammer_buffer *buffer); |
| 712 | hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp); |
| 713 | void *hammer_alloc_record(hammer_transaction_t trans, |
| 714 | hammer_off_t *rec_offp, u_int16_t rec_type, |
| 715 | struct hammer_buffer **rec_bufferp, |
| 716 | int32_t data_len, void **datap, |
| 717 | struct hammer_buffer **data_bufferp, int *errorp); |
| 718 | void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len, |
| 719 | hammer_off_t *data_offsetp, |
| 720 | struct hammer_buffer **data_bufferp, int *errorp); |
| 721 | |
| 722 | int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io, |
| 723 | hammer_off_t zone1_offset, void *base, int len); |
| 724 | |
| 725 | void hammer_put_volume(struct hammer_volume *volume, int flush); |
| 726 | void hammer_put_buffer(struct hammer_buffer *buffer, int flush); |
| 727 | |
| 728 | hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans, |
| 729 | hammer_off_t owner, int *errorp); |
| 730 | void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset, |
| 731 | hammer_off_t owner, int *errorp); |
| 732 | hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone, |
| 733 | int bytes, int *errorp); |
| 734 | void hammer_blockmap_free(hammer_transaction_t trans, |
| 735 | hammer_off_t bmap_off, int bytes); |
| 736 | int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off, |
| 737 | int *curp, int *errorp); |
| 738 | hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, |
| 739 | int *errorp); |
| 740 | hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off, |
| 741 | int *errorp); |
| 742 | int64_t hammer_undo_space(hammer_mount_t hmp); |
| 743 | int64_t hammer_undo_max(hammer_mount_t hmp); |
| 744 | |
| 745 | |
| 746 | void hammer_start_transaction(struct hammer_transaction *trans, |
| 747 | struct hammer_mount *hmp); |
| 748 | void hammer_simple_transaction(struct hammer_transaction *trans, |
| 749 | struct hammer_mount *hmp); |
| 750 | void hammer_start_transaction_fls(struct hammer_transaction *trans, |
| 751 | struct hammer_mount *hmp); |
| 752 | void hammer_done_transaction(struct hammer_transaction *trans); |
| 753 | |
| 754 | void hammer_modify_inode(struct hammer_transaction *trans, |
| 755 | hammer_inode_t ip, int flags); |
| 756 | void hammer_flush_inode(hammer_inode_t ip, int flags); |
| 757 | void hammer_flush_inode_done(hammer_inode_t ip); |
| 758 | void hammer_wait_inode(hammer_inode_t ip); |
| 759 | |
| 760 | int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap, |
| 761 | struct ucred *cred, struct hammer_inode *dip, |
| 762 | struct hammer_inode **ipp); |
| 763 | void hammer_rel_inode(hammer_inode_t ip, int flush); |
| 764 | int hammer_sync_inode(hammer_inode_t ip); |
| 765 | void hammer_test_inode(hammer_inode_t ip); |
| 766 | |
| 767 | int hammer_ip_add_directory(struct hammer_transaction *trans, |
| 768 | hammer_inode_t dip, struct namecache *ncp, |
| 769 | hammer_inode_t nip); |
| 770 | int hammer_ip_del_directory(struct hammer_transaction *trans, |
| 771 | hammer_cursor_t cursor, hammer_inode_t dip, |
| 772 | hammer_inode_t ip); |
| 773 | int hammer_ip_add_record(struct hammer_transaction *trans, |
| 774 | hammer_record_t record); |
| 775 | int hammer_ip_delete_range(struct hammer_transaction *trans, |
| 776 | hammer_inode_t ip, int64_t ran_beg, int64_t ran_end); |
| 777 | int hammer_ip_delete_range_all(struct hammer_transaction *trans, |
| 778 | hammer_inode_t ip); |
| 779 | int hammer_ip_sync_data(struct hammer_transaction *trans, |
| 780 | hammer_inode_t ip, int64_t offset, |
| 781 | void *data, int bytes); |
| 782 | int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec); |
| 783 | |
| 784 | int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag, |
| 785 | struct ucred *cred); |
| 786 | |
| 787 | void hammer_io_init(hammer_io_t io, hammer_mount_t hmp, |
| 788 | enum hammer_io_type type); |
| 789 | void hammer_io_reinit(hammer_io_t io, enum hammer_io_type type); |
| 790 | int hammer_io_read(struct vnode *devvp, struct hammer_io *io); |
| 791 | int hammer_io_new(struct vnode *devvp, struct hammer_io *io); |
| 792 | void hammer_io_release(struct hammer_io *io); |
| 793 | void hammer_io_flush(struct hammer_io *io); |
| 794 | int hammer_io_checkflush(hammer_io_t io); |
| 795 | void hammer_io_clear_modify(struct hammer_io *io); |
| 796 | void hammer_io_waitdep(struct hammer_io *io); |
| 797 | |
| 798 | void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume, |
| 799 | void *base, int len); |
| 800 | void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer, |
| 801 | void *base, int len); |
| 802 | void hammer_modify_volume_done(hammer_volume_t volume); |
| 803 | void hammer_modify_buffer_done(hammer_buffer_t buffer); |
| 804 | |
| 805 | int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip, |
| 806 | struct hammer_ioc_reblock *reblock); |
| 807 | |
| 808 | void hammer_init_holes(hammer_mount_t hmp, hammer_holes_t holes); |
| 809 | void hammer_free_holes(hammer_mount_t hmp, hammer_holes_t holes); |
| 810 | int hammer_signal_check(hammer_mount_t hmp); |
| 811 | |
| 812 | void hammer_flusher_create(hammer_mount_t hmp); |
| 813 | void hammer_flusher_destroy(hammer_mount_t hmp); |
| 814 | void hammer_flusher_sync(hammer_mount_t hmp); |
| 815 | void hammer_flusher_async(hammer_mount_t hmp); |
| 816 | |
| 817 | int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol); |
| 818 | |
| 819 | #endif |
| 820 | |
| 821 | static __inline void |
| 822 | hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node) |
| 823 | { |
| 824 | hammer_modify_buffer(trans, node->buffer, NULL, 0); |
| 825 | } |
| 826 | |
| 827 | static __inline void |
| 828 | hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node) |
| 829 | { |
| 830 | hammer_modify_buffer(trans, node->buffer, |
| 831 | node->ondisk, sizeof(*node->ondisk)); |
| 832 | } |
| 833 | |
| 834 | static __inline void |
| 835 | hammer_modify_node(hammer_transaction_t trans, hammer_node_t node, |
| 836 | void *base, int len) |
| 837 | { |
| 838 | KKASSERT((char *)base >= (char *)node->ondisk && |
| 839 | (char *)base + len <= |
| 840 | (char *)node->ondisk + sizeof(*node->ondisk)); |
| 841 | hammer_modify_buffer(trans, node->buffer, base, len); |
| 842 | } |
| 843 | |
| 844 | static __inline void |
| 845 | hammer_modify_node_done(hammer_node_t node) |
| 846 | { |
| 847 | hammer_modify_buffer_done(node->buffer); |
| 848 | } |
| 849 | |
| 850 | static __inline void |
| 851 | hammer_modify_record(hammer_transaction_t trans, hammer_buffer_t buffer, |
| 852 | void *base, int len) |
| 853 | { |
| 854 | KKASSERT((char *)base >= (char *)buffer->ondisk && |
| 855 | (char *)base + len <= (char *)buffer->ondisk + HAMMER_BUFSIZE); |
| 856 | hammer_modify_buffer(trans, buffer, base, len); |
| 857 | } |
| 858 | |
| 859 | static __inline void |
| 860 | hammer_modify_record_done(hammer_buffer_t buffer) |
| 861 | { |
| 862 | hammer_modify_buffer_done(buffer); |
| 863 | } |
| 864 | |