2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
34 * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.94 2008/06/27 20:56:59 dillon Exp $
37 * This header file contains structures used internally by the HAMMERFS
38 * implementation. See hammer_disk.h for on-disk structures.
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/kernel.h>
45 #include <sys/systm.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/vnode.h>
53 #include <sys/globaldata.h>
54 #include <sys/lockf.h>
56 #include <sys/queue.h>
58 #include <sys/globaldata.h>
61 #include <sys/signal2.h>
62 #include "hammer_disk.h"
63 #include "hammer_mount.h"
64 #include "hammer_ioctl.h"
66 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
68 MALLOC_DECLARE(M_HAMMER);
73 #if !defined(KTR_HAMMER)
74 #define KTR_HAMMER KTR_ALL
76 KTR_INFO_MASTER_EXTERN(hammer);
84 * Key structure used for custom RB tree inode lookups. This prototypes
85 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
87 typedef struct hammer_inode_info {
88 int64_t obj_id; /* (key) object identifier */
89 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
90 u_int32_t obj_localization; /* (key) pseudo-fs */
92 struct hammer_btree_leaf_elm *leaf;
94 } *hammer_inode_info_t;
96 typedef enum hammer_transaction_type {
100 } hammer_transaction_type_t;
103 * HAMMER Transaction tracking
105 struct hammer_transaction {
106 hammer_transaction_type_t type;
107 struct hammer_mount *hmp;
112 struct hammer_volume *rootvol;
115 typedef struct hammer_transaction *hammer_transaction_t;
121 int refs; /* active references delay writes */
122 int lockcount; /* lock count for exclusive/shared access */
124 int exwanted; /* number of threads waiting for ex lock */
125 struct thread *locktd;
129 hammer_islocked(struct hammer_lock *lock)
131 return(lock->lockcount != 0);
135 hammer_isactive(struct hammer_lock *lock)
137 return(lock->refs != 0);
141 hammer_islastref(struct hammer_lock *lock)
143 return(lock->refs == 1);
147 * Return if we specifically own the lock exclusively.
150 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
152 if (lock->lockcount > 0 && lock->locktd == td)
158 * Flush state, used by various structures
160 typedef enum hammer_inode_state {
164 } hammer_inode_state_t;
166 TAILQ_HEAD(hammer_record_list, hammer_record);
169 * Cache object ids. A fixed number of objid cache structures are
170 * created to reserve object id's for newly created files in multiples
171 * of 100,000, localized to a particular directory, and recycled as
172 * needed. This allows parallel create operations in different
173 * directories to retain fairly localized object ids which in turn
174 * improves reblocking performance and layout.
176 #define OBJID_CACHE_SIZE 1024
177 #define OBJID_CACHE_BULK 100000
179 typedef struct hammer_objid_cache {
180 TAILQ_ENTRY(hammer_objid_cache) entry;
181 struct hammer_inode *dip;
182 hammer_tid_t next_tid;
184 } *hammer_objid_cache_t;
187 * Associate an inode with a B-Tree node to cache search start positions
189 typedef struct hammer_node_cache {
190 TAILQ_ENTRY(hammer_node_cache) entry;
191 struct hammer_node *node;
192 struct hammer_inode *ip;
193 } *hammer_node_cache_t;
195 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
198 * Structure used to represent an inode in-memory.
200 * The record and data associated with an inode may be out of sync with
201 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
204 * An inode may also hold a cache of unsynchronized records, used for
205 * database and directories only. Unsynchronized regular file data is
206 * stored in the buffer cache.
208 * NOTE: A file which is created and destroyed within the initial
209 * synchronization period can wind up not doing any disk I/O at all.
211 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
213 struct hammer_ino_rb_tree;
215 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
216 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
217 hammer_ino_rb_compare, hammer_inode_info_t);
219 struct hammer_rec_rb_tree;
220 struct hammer_record;
221 RB_HEAD(hammer_rec_rb_tree, hammer_record);
222 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
223 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
225 TAILQ_HEAD(hammer_node_list, hammer_node);
227 struct hammer_inode {
228 RB_ENTRY(hammer_inode) rb_node;
229 hammer_inode_state_t flush_state;
231 TAILQ_ENTRY(hammer_inode) flush_entry;
232 struct hammer_record_list target_list; /* target of dependant recs */
233 u_int64_t obj_id; /* (key) object identifier */
234 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
235 u_int32_t obj_localization; /* (key) pseudo-fs */
236 struct hammer_mount *hmp;
237 hammer_objid_cache_t objid_cache;
239 int error; /* flush error */
240 int cursor_ip_refs; /* sanity */
244 struct lockf advlock;
245 struct hammer_lock lock; /* sync copy interlock */
247 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
248 struct hammer_inode_data ino_data; /* in-memory cache */
249 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
250 struct hammer_node_cache cache[2]; /* search initiate cache */
253 * When a demark is created to synchronize an inode to
254 * disk, certain fields are copied so the front-end VOPs
255 * can continue to run in parallel with the synchronization
256 * occuring in the background.
258 int sync_flags; /* to-sync flags cache */
259 off_t sync_trunc_off; /* to-sync truncation */
260 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
261 struct hammer_inode_data sync_ino_data; /* to-sync cache */
264 typedef struct hammer_inode *hammer_inode_t;
266 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
268 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
269 /* (not including atime/mtime) */
270 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
271 #define HAMMER_INODE_UNUSED0004 0x0004
272 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
273 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
274 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
275 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
276 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
277 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
278 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */
279 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
280 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
281 #define HAMMER_INODE_REFLUSH 0x2000 /* pipelined flush during flush */
282 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
283 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
285 #define HAMMER_INODE_TRUNCATED 0x00010000
286 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
287 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
288 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
289 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
290 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
292 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \
293 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
294 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
295 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
297 #define HAMMER_INODE_MODMASK_NOXDIRTY \
298 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
300 #define HAMMER_FLUSH_GROUP_SIZE 64
302 #define HAMMER_FLUSH_SIGNAL 0x0001
303 #define HAMMER_FLUSH_RECURSION 0x0002
306 * Used by the inode reclaim code to pipeline reclaims and avoid
307 * blowing out kernel memory or letting the flusher get too far
310 struct hammer_reclaim {
311 TAILQ_ENTRY(hammer_reclaim) entry;
315 #define HAMMER_RECLAIM_FLUSH 2000
316 #define HAMMER_RECLAIM_WAIT 4000
319 * Structure used to represent an unsynchronized record in-memory. These
320 * records typically represent directory entries. Only non-historical
321 * records are kept in-memory.
323 * Records are organized as a per-inode RB-Tree. If the inode is not
324 * on disk then neither are any records and the in-memory record tree
325 * represents the entire contents of the inode. If the inode is on disk
326 * then the on-disk B-Tree is scanned in parallel with the in-memory
327 * RB-Tree to synthesize the current state of the file.
329 * Records are also used to enforce the ordering of directory create/delete
330 * operations. A new inode will not be flushed to disk unless its related
331 * directory entry is also being flushed at the same time. A directory entry
332 * will not be removed unless its related inode is also being removed at the
335 typedef enum hammer_record_type {
336 HAMMER_MEM_RECORD_GENERAL, /* misc record */
337 HAMMER_MEM_RECORD_INODE, /* inode record */
338 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
339 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
340 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
341 } hammer_record_type_t;
343 struct hammer_record {
344 RB_ENTRY(hammer_record) rb_node;
345 TAILQ_ENTRY(hammer_record) target_entry;
346 hammer_inode_state_t flush_state;
348 hammer_record_type_t type;
349 struct hammer_lock lock;
350 struct hammer_reserve *resv;
351 struct hammer_inode *ip;
352 struct hammer_inode *target_ip;
353 struct hammer_btree_leaf_elm leaf;
354 union hammer_data_ondisk *data;
358 typedef struct hammer_record *hammer_record_t;
361 * Record flags. Note that FE can only be set by the frontend if the
362 * record has not been interlocked by the backend w/ BE.
364 #define HAMMER_RECF_ALLOCDATA 0x0001
365 #define HAMMER_RECF_ONRBTREE 0x0002
366 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
367 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
368 #define HAMMER_RECF_UNUSED0010 0x0010
369 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
370 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
371 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
374 * In-memory structures representing on-disk structures.
376 struct hammer_volume;
377 struct hammer_buffer;
380 struct hammer_reserve;
382 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
383 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
384 RB_HEAD(hammer_nod_rb_tree, hammer_node);
385 RB_HEAD(hammer_und_rb_tree, hammer_undo);
386 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
388 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
389 hammer_vol_rb_compare, int32_t);
390 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
391 hammer_buf_rb_compare, hammer_off_t);
392 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
393 hammer_nod_rb_compare, hammer_off_t);
394 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
395 hammer_und_rb_compare, hammer_off_t);
396 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
397 hammer_res_rb_compare, hammer_off_t);
400 * IO management - embedded at the head of various in-memory structures
402 * VOLUME - hammer_volume containing meta-data
403 * META_BUFFER - hammer_buffer containing meta-data
404 * DATA_BUFFER - hammer_buffer containing pure-data
406 * Dirty volume headers and dirty meta-data buffers are locked until the
407 * flusher can sequence them out. Dirty pure-data buffers can be written.
408 * Clean buffers can be passively released.
410 typedef enum hammer_io_type {
411 HAMMER_STRUCTURE_VOLUME,
412 HAMMER_STRUCTURE_META_BUFFER,
413 HAMMER_STRUCTURE_UNDO_BUFFER,
414 HAMMER_STRUCTURE_DATA_BUFFER
417 union hammer_io_structure;
421 LIST_ENTRY(worklist) node;
424 TAILQ_HEAD(hammer_io_list, hammer_io);
425 typedef struct hammer_io_list *hammer_io_list_t;
428 struct worklist worklist;
429 struct hammer_lock lock;
430 enum hammer_io_type type;
431 struct hammer_mount *hmp;
432 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */
433 hammer_io_list_t mod_list;
435 int64_t offset; /* zone-2 offset */
436 int bytes; /* buffer cache buffer size */
437 int loading; /* loading/unloading interlock */
440 u_int modified : 1; /* bp's data was modified */
441 u_int released : 1; /* bp released (w/ B_LOCKED set) */
442 u_int running : 1; /* bp write IO in progress */
443 u_int waiting : 1; /* someone is waiting on us */
444 u_int validated : 1; /* ondisk has been validated */
445 u_int waitdep : 1; /* flush waits for dependancies */
446 u_int recovered : 1; /* has recovery ref */
447 u_int waitmod : 1; /* waiting for modify_refs */
448 u_int reclaim : 1; /* reclaim requested */
449 u_int gencrc : 1; /* crc needs to be generated */
452 typedef struct hammer_io *hammer_io_t;
454 #define HAMMER_CLUSTER_SIZE (64 * 1024)
455 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
456 #undef HAMMER_CLUSTER_SIZE
457 #define HAMMER_CLUSTER_SIZE MAXBSIZE
459 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
462 * In-memory volume representing on-disk buffer
464 struct hammer_volume {
466 RB_ENTRY(hammer_volume) rb_node;
467 struct hammer_volume_ondisk *ondisk;
469 int64_t nblocks; /* note: special calculation for statfs */
470 int64_t buffer_base; /* base offset of buffer 0 */
471 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
472 hammer_off_t maxraw_off; /* Maximum raw offset for device */
478 typedef struct hammer_volume *hammer_volume_t;
481 * In-memory buffer (other then volume, super-cluster, or cluster),
482 * representing an on-disk buffer.
484 struct hammer_buffer {
486 RB_ENTRY(hammer_buffer) rb_node;
488 struct hammer_volume *volume;
489 hammer_off_t zoneX_offset;
490 hammer_off_t zone2_offset;
491 struct hammer_reserve *resv;
492 struct hammer_node_list clist;
495 typedef struct hammer_buffer *hammer_buffer_t;
498 * In-memory B-Tree node, representing an on-disk B-Tree node.
500 * This is a hang-on structure which is backed by a hammer_buffer,
501 * indexed by a hammer_cluster, and used for fine-grained locking of
502 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
503 * can contain multiple nodes representing wildly disassociated portions
504 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
506 * This structure uses a cluster-relative index to reduce the number
507 * of layers required to access it, and also because all on-disk B-Tree
508 * references are cluster-relative offsets.
511 struct hammer_lock lock; /* node-by-node lock */
512 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
513 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */
514 hammer_off_t node_offset; /* full offset spec */
515 struct hammer_mount *hmp;
516 struct hammer_buffer *buffer; /* backing buffer */
517 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
518 struct hammer_node_cache_list cache_list; /* passive caches */
520 int loading; /* load interlock */
523 #define HAMMER_NODE_DELETED 0x0001
524 #define HAMMER_NODE_FLUSH 0x0002
525 #define HAMMER_NODE_CRCGOOD 0x0004
526 #define HAMMER_NODE_NEEDSCRC 0x0008
527 #define HAMMER_NODE_NEEDSMIRROR 0x0010
529 typedef struct hammer_node *hammer_node_t;
532 * List of locked nodes.
534 struct hammer_node_locklist {
535 struct hammer_node_locklist *next;
539 typedef struct hammer_node_locklist *hammer_node_locklist_t;
543 * Common I/O management structure - embedded in in-memory structures
544 * which are backed by filesystem buffers.
546 union hammer_io_structure {
548 struct hammer_volume volume;
549 struct hammer_buffer buffer;
552 typedef union hammer_io_structure *hammer_io_structure_t;
555 * The reserve structure prevents the blockmap from allocating
556 * out of a reserved bigblock. Such reservations are used by
557 * the direct-write mechanism.
559 * The structure is also used to hold off on reallocations of
560 * big blocks from the freemap until flush dependancies have
563 struct hammer_reserve {
564 RB_ENTRY(hammer_reserve) rb_node;
565 TAILQ_ENTRY(hammer_reserve) delay_entry;
570 hammer_off_t zone_offset;
573 typedef struct hammer_reserve *hammer_reserve_t;
575 #define HAMMER_RESF_ONDELAY 0x0001
577 #include "hammer_cursor.h"
580 * The undo structure tracks recent undos to avoid laying down duplicate
581 * undos within a flush group, saving us a significant amount of overhead.
583 * This is strictly a heuristic.
585 #define HAMMER_MAX_UNDOS 1024
586 #define HAMMER_MAX_FLUSHERS 4
589 RB_ENTRY(hammer_undo) rb_node;
590 TAILQ_ENTRY(hammer_undo) lru_entry;
595 typedef struct hammer_undo *hammer_undo_t;
597 struct hammer_flusher_info;
599 struct hammer_flusher {
600 int signal; /* flusher thread sequencer */
601 int act; /* currently active flush group */
602 int done; /* set to act when complete */
603 int next; /* next flush group */
604 int group_lock; /* lock sequencing of the next flush */
605 int exiting; /* request master exit */
606 int count; /* number of slave flushers */
607 int running; /* number of slave flushers running */
608 thread_t td; /* master flusher thread */
609 hammer_tid_t tid; /* last flushed transaction id */
610 int finalize_want; /* serialize finalization */
611 struct hammer_lock finalize_lock; /* serialize finalization */
612 struct hammer_transaction trans; /* shared transaction */
613 struct hammer_flusher_info *info[HAMMER_MAX_FLUSHERS];
617 * Internal hammer mount data structure
619 struct hammer_mount {
621 /*struct vnode *rootvp;*/
622 struct hammer_ino_rb_tree rb_inos_root;
623 struct hammer_vol_rb_tree rb_vols_root;
624 struct hammer_nod_rb_tree rb_nods_root;
625 struct hammer_und_rb_tree rb_undo_root;
626 struct hammer_res_rb_tree rb_resv_root;
627 struct hammer_buf_rb_tree rb_bufs_root;
628 struct hammer_volume *rootvol;
629 struct hammer_base_elm root_btree_beg;
630 struct hammer_base_elm root_btree_end;
636 int masterid; /* -1 or 0-15 - clustering and mirroring */
637 int rsv_inodes; /* reserved space due to dirty inodes */
638 int rsv_databufs; /* reserved space due to dirty buffers */
639 int rsv_databytes; /* reserved space due to record data */
640 int rsv_recs; /* reserved space due to dirty records */
642 int count_newrecords;
644 int inode_reclaims; /* inodes pending reclaim by flusher */
645 int count_inodes; /* total number of inodes */
646 int count_iqueued; /* inodes queued to flusher */
648 struct hammer_flusher flusher;
650 u_int check_interrupt;
653 struct hammer_io_list volu_list; /* dirty undo buffers */
654 struct hammer_io_list undo_list; /* dirty undo buffers */
655 struct hammer_io_list data_list; /* dirty data buffers */
656 struct hammer_io_list alt_data_list; /* dirty data buffers */
657 struct hammer_io_list meta_list; /* dirty meta bufs */
658 struct hammer_io_list lose_list; /* loose buffers */
659 int locked_dirty_count; /* meta/volu count */
660 int io_running_count;
661 int objid_cache_count;
662 hammer_tid_t asof; /* snapshot mount */
663 hammer_off_t next_tid;
664 int64_t copy_stat_freebigblocks; /* number of free bigblocks */
666 u_int32_t namekey_iterator;
667 struct netexport export;
668 struct hammer_lock sync_lock;
669 struct hammer_lock free_lock;
670 struct hammer_lock undo_lock;
671 struct hammer_lock blkmap_lock;
672 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
673 struct hammer_undo undos[HAMMER_MAX_UNDOS];
675 TAILQ_HEAD(, hammer_undo) undo_lru_list;
676 TAILQ_HEAD(, hammer_inode) flush_list;
677 TAILQ_HEAD(, hammer_reserve) delay_list;
678 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
679 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
682 typedef struct hammer_mount *hammer_mount_t;
684 #define HAMMER_MOUNT_UNUSED0001 0x0001
686 struct hammer_sync_info {
695 extern struct vop_ops hammer_vnode_vops;
696 extern struct vop_ops hammer_spec_vops;
697 extern struct vop_ops hammer_fifo_vops;
698 extern struct bio_ops hammer_bioops;
700 extern int hammer_debug_io;
701 extern int hammer_debug_general;
702 extern int hammer_debug_debug;
703 extern int hammer_debug_inode;
704 extern int hammer_debug_locks;
705 extern int hammer_debug_btree;
706 extern int hammer_debug_tid;
707 extern int hammer_debug_recover;
708 extern int hammer_debug_recover_faults;
709 extern int hammer_debug_cluster_enable;
710 extern int hammer_count_inodes;
711 extern int hammer_count_iqueued;
712 extern int hammer_count_reclaiming;
713 extern int hammer_count_records;
714 extern int hammer_count_record_datas;
715 extern int hammer_count_volumes;
716 extern int hammer_count_buffers;
717 extern int hammer_count_nodes;
718 extern int64_t hammer_stats_btree_lookups;
719 extern int64_t hammer_stats_btree_searches;
720 extern int64_t hammer_stats_btree_inserts;
721 extern int64_t hammer_stats_btree_deletes;
722 extern int64_t hammer_stats_btree_elements;
723 extern int64_t hammer_stats_btree_splits;
724 extern int64_t hammer_stats_btree_iterations;
725 extern int64_t hammer_stats_record_iterations;
726 extern int hammer_count_dirtybufs;
727 extern int hammer_count_refedbufs;
728 extern int hammer_count_reservations;
729 extern int hammer_count_io_running_read;
730 extern int hammer_count_io_running_write;
731 extern int hammer_count_io_locked;
732 extern int hammer_limit_dirtybufs;
733 extern int hammer_limit_iqueued;
734 extern int hammer_limit_recs;
735 extern int hammer_bio_count;
736 extern int hammer_verify_zone;
737 extern int hammer_write_mode;
738 extern int64_t hammer_contention_count;
740 int hammer_vop_inactive(struct vop_inactive_args *);
741 int hammer_vop_reclaim(struct vop_reclaim_args *);
742 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
743 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
744 hammer_inode_t dip, u_int64_t obj_id,
745 hammer_tid_t asof, u_int32_t localization,
746 int flags, int *errorp);
747 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
748 hammer_inode_info_t iinfo,
749 int (*callback)(hammer_inode_t ip, void *data),
751 void hammer_put_inode(struct hammer_inode *ip);
752 void hammer_put_inode_ref(struct hammer_inode *ip);
753 void hammer_inode_waitreclaims(hammer_mount_t hmp);
755 int hammer_unload_volume(hammer_volume_t volume, void *data __unused);
756 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
758 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
759 int hammer_install_volume(hammer_mount_t hmp, const char *volname);
761 int hammer_ip_lookup(hammer_cursor_t cursor);
762 int hammer_ip_first(hammer_cursor_t cursor);
763 int hammer_ip_next(hammer_cursor_t cursor);
764 int hammer_ip_resolve_data(hammer_cursor_t cursor);
765 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
767 int hammer_delete_at_cursor(hammer_cursor_t cursor, int64_t *stat_bytes);
768 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
770 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
771 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
775 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
776 void hammer_flush_record_done(hammer_record_t record, int error);
777 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
778 void hammer_rel_mem_record(hammer_record_t record);
780 int hammer_cursor_up(hammer_cursor_t cursor);
781 int hammer_cursor_up_locked(hammer_cursor_t cursor);
782 int hammer_cursor_down(hammer_cursor_t cursor);
783 int hammer_cursor_upgrade(hammer_cursor_t cursor);
784 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
785 void hammer_cursor_downgrade(hammer_cursor_t cursor);
786 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
788 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
789 int hammer_lock_ex_try(struct hammer_lock *lock);
790 void hammer_lock_sh(struct hammer_lock *lock);
791 void hammer_lock_sh_lowpri(struct hammer_lock *lock);
792 int hammer_lock_sh_try(struct hammer_lock *lock);
793 int hammer_lock_upgrade(struct hammer_lock *lock);
794 void hammer_lock_downgrade(struct hammer_lock *lock);
795 void hammer_unlock(struct hammer_lock *lock);
796 void hammer_ref(struct hammer_lock *lock);
797 void hammer_unref(struct hammer_lock *lock);
799 void hammer_sync_lock_ex(hammer_transaction_t trans);
800 void hammer_sync_lock_sh(hammer_transaction_t trans);
801 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
802 void hammer_sync_unlock(hammer_transaction_t trans);
804 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
805 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
806 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
807 u_int64_t hammer_timespec_to_time(struct timespec *ts);
808 hammer_tid_t hammer_str_to_tid(const char *str);
809 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip);
810 void hammer_clear_objid(hammer_inode_t dip);
811 void hammer_destroy_objid_cache(hammer_mount_t hmp);
813 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
815 void hammer_clear_undo_history(hammer_mount_t hmp);
816 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
817 int hammer_get_dtype(u_int8_t obj_type);
818 u_int8_t hammer_get_obj_type(enum vtype vtype);
819 int64_t hammer_directory_namekey(const void *name, int len);
820 int hammer_nohistory(hammer_inode_t ip);
822 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
823 hammer_node_cache_t cache, hammer_inode_t ip);
824 int hammer_reinit_cursor(hammer_cursor_t cursor);
825 void hammer_normalize_cursor(hammer_cursor_t cursor);
826 void hammer_done_cursor(hammer_cursor_t cursor);
827 void hammer_mem_done(hammer_cursor_t cursor);
829 int hammer_btree_lookup(hammer_cursor_t cursor);
830 int hammer_btree_first(hammer_cursor_t cursor);
831 int hammer_btree_last(hammer_cursor_t cursor);
832 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
833 int hammer_btree_iterate(hammer_cursor_t cursor);
834 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
835 int hammer_btree_insert(hammer_cursor_t cursor,
836 hammer_btree_leaf_elm_t elm);
837 int hammer_btree_delete(hammer_cursor_t cursor);
838 int hammer_btree_mirror_propagate(hammer_transaction_t trans,
839 hammer_node_t node, int index,
840 hammer_tid_t mirror_tid);
842 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
843 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
844 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
845 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
847 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
848 hammer_btree_elm_t elm);
849 int hammer_btree_lock_children(hammer_cursor_t cursor,
850 struct hammer_node_locklist **locklistp);
851 void hammer_btree_unlock_children(struct hammer_node_locklist **locklistp);
852 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
853 hammer_node_t hammer_btree_get_parent(hammer_node_t node, int *parent_indexp,
854 int *errorp, int try_exclusive);
856 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
857 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
859 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
860 int *errorp, struct hammer_buffer **bufferp);
861 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
862 int *errorp, struct hammer_buffer **bufferp);
863 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
864 int *errorp, struct hammer_buffer **bufferp);
865 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
866 int *errorp, struct hammer_buffer **bufferp);
868 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
870 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
871 int32_t vol_no, int *errorp);
872 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
873 int bytes, int isnew, int *errorp);
874 void hammer_del_buffers(hammer_mount_t hmp, hammer_off_t base_offset,
875 hammer_off_t zone2_offset, int bytes);
877 int hammer_ref_volume(hammer_volume_t volume);
878 int hammer_ref_buffer(hammer_buffer_t buffer);
879 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
881 void hammer_rel_volume(hammer_volume_t volume, int flush);
882 void hammer_rel_buffer(hammer_buffer_t buffer, int flush);
884 int hammer_vfs_export(struct mount *mp, int op,
885 const struct export_args *export);
886 hammer_node_t hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset,
887 int isnew, int *errorp);
888 void hammer_ref_node(hammer_node_t node);
889 hammer_node_t hammer_ref_node_safe(struct hammer_mount *hmp,
890 hammer_node_cache_t cache, int *errorp);
891 void hammer_rel_node(hammer_node_t node);
892 void hammer_delete_node(hammer_transaction_t trans,
894 void hammer_cache_node(hammer_node_cache_t cache,
896 void hammer_uncache_node(hammer_node_cache_t cache);
897 void hammer_flush_node(hammer_node_t node);
899 void hammer_dup_buffer(struct hammer_buffer **bufferp,
900 struct hammer_buffer *buffer);
901 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans, int *errorp);
902 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
903 u_int16_t rec_type, hammer_off_t *data_offsetp,
904 struct hammer_buffer **data_bufferp, int *errorp);
906 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
907 hammer_off_t zone1_offset, void *base, int len);
909 void hammer_put_volume(struct hammer_volume *volume, int flush);
910 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
912 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
913 hammer_off_t owner, int *errorp);
914 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
915 hammer_off_t owner, int *errorp);
916 int hammer_checkspace(hammer_mount_t hmp);
917 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
918 int bytes, int *errorp);
919 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
920 int bytes, hammer_off_t *zone_offp, int *errorp);
921 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
922 hammer_reserve_t resv);
923 void hammer_reserve_setdelay(hammer_mount_t hmp, hammer_reserve_t resv,
924 hammer_off_t zone2_offset);
925 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
926 void hammer_blockmap_free(hammer_transaction_t trans,
927 hammer_off_t bmap_off, int bytes);
928 void hammer_blockmap_finalize(hammer_transaction_t trans,
929 hammer_off_t bmap_off, int bytes);
930 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
931 int *curp, int *errorp);
932 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
934 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
936 int64_t hammer_undo_used(hammer_transaction_t trans);
937 int64_t hammer_undo_space(hammer_transaction_t trans);
938 int64_t hammer_undo_max(hammer_mount_t hmp);
940 void hammer_start_transaction(struct hammer_transaction *trans,
941 struct hammer_mount *hmp);
942 void hammer_simple_transaction(struct hammer_transaction *trans,
943 struct hammer_mount *hmp);
944 void hammer_start_transaction_fls(struct hammer_transaction *trans,
945 struct hammer_mount *hmp);
946 void hammer_done_transaction(struct hammer_transaction *trans);
948 void hammer_modify_inode(hammer_inode_t ip, int flags);
949 void hammer_flush_inode(hammer_inode_t ip, int flags);
950 void hammer_flush_inode_done(hammer_inode_t ip);
951 void hammer_wait_inode(hammer_inode_t ip);
953 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
954 struct ucred *cred, struct hammer_inode *dip,
955 int pseudofs, struct hammer_inode **ipp);
956 void hammer_rel_inode(hammer_inode_t ip, int flush);
957 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
958 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
960 int hammer_sync_inode(hammer_inode_t ip);
961 void hammer_test_inode(hammer_inode_t ip);
962 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
964 int hammer_ip_add_directory(struct hammer_transaction *trans,
965 hammer_inode_t dip, const char *name, int bytes,
967 int hammer_ip_del_directory(struct hammer_transaction *trans,
968 hammer_cursor_t cursor, hammer_inode_t dip,
970 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
971 void *data, int bytes, int *errorp);
972 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
973 int hammer_ip_add_record(struct hammer_transaction *trans,
974 hammer_record_t record);
975 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
976 int64_t ran_beg, int64_t ran_end, int truncating);
977 int hammer_ip_delete_range_all(hammer_cursor_t cursor, hammer_inode_t ip,
979 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
980 int64_t offset, void *data, int bytes);
981 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
982 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
984 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
987 void hammer_io_init(hammer_io_t io, hammer_mount_t hmp,
988 enum hammer_io_type type);
989 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
991 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
992 void hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
993 void hammer_io_release(struct hammer_io *io, int flush);
994 void hammer_io_flush(struct hammer_io *io);
995 void hammer_io_waitdep(struct hammer_io *io);
996 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
997 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio);
998 int hammer_io_direct_write(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf,
1000 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1001 void hammer_io_write_interlock(hammer_io_t io);
1002 void hammer_io_done_interlock(hammer_io_t io);
1003 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1004 void hammer_io_clear_modlist(struct hammer_io *io);
1005 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1006 void *base, int len);
1007 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1008 void *base, int len);
1009 void hammer_modify_volume_done(hammer_volume_t volume);
1010 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1012 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1013 struct hammer_ioc_reblock *reblock);
1014 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1015 struct hammer_ioc_prune *prune);
1016 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1017 struct hammer_ioc_mirror_rw *mirror);
1018 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1019 struct hammer_ioc_mirror_rw *mirror);
1021 int hammer_signal_check(hammer_mount_t hmp);
1023 void hammer_flusher_create(hammer_mount_t hmp);
1024 void hammer_flusher_destroy(hammer_mount_t hmp);
1025 void hammer_flusher_sync(hammer_mount_t hmp);
1026 void hammer_flusher_async(hammer_mount_t hmp);
1027 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1028 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1030 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol);
1031 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1032 hammer_volume_t root_volume, int final);
1034 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1035 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1036 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1038 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1039 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1040 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1041 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1042 void hkprintf(const char *ctl, ...);
1044 int hammer_blocksize(int64_t file_offset);
1045 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1049 static __inline void
1050 hammer_wait_mem_record(hammer_record_t record)
1052 hammer_wait_mem_record_ident(record, "hmmwai");
1055 static __inline void
1056 hammer_lock_ex(struct hammer_lock *lock)
1058 hammer_lock_ex_ident(lock, "hmrlck");
1062 * Indicate that a B-Tree node is being modified.
1064 static __inline void
1065 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1067 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1070 static __inline void
1071 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1073 hammer_modify_buffer(trans, node->buffer,
1074 node->ondisk, sizeof(*node->ondisk));
1077 static __inline void
1078 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1079 void *base, int len)
1081 hammer_crc_t *crcptr;
1083 KKASSERT((char *)base >= (char *)node->ondisk &&
1084 (char *)base + len <=
1085 (char *)node->ondisk + sizeof(*node->ondisk));
1086 hammer_modify_buffer(trans, node->buffer, base, len);
1087 crcptr = &node->ondisk->crc;
1088 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1089 --node->buffer->io.modify_refs; /* only want one ref */
1093 * Indicate that the specified modifications have been completed.
1095 * Do not try to generate the crc here, it's very expensive to do and a
1096 * sequence of insertions or deletions can result in many calls to this
1097 * function on the same node.
1099 static __inline void
1100 hammer_modify_node_done(hammer_node_t node)
1102 node->flags |= HAMMER_NODE_CRCGOOD;
1103 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1104 node->flags |= HAMMER_NODE_NEEDSCRC;
1105 node->buffer->io.gencrc = 1;
1106 hammer_ref_node(node);
1108 hammer_modify_buffer_done(node->buffer);
1111 #define hammer_modify_volume_field(trans, vol, field) \
1112 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1113 sizeof((vol)->ondisk->field))
1115 #define hammer_modify_node_field(trans, node, field) \
1116 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1117 sizeof((node)->ondisk->field))