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 1.130 2008/11/13 02:18:43 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>
54 #include <sys/globaldata.h>
55 #include <sys/lockf.h>
57 #include <sys/queue.h>
59 #include <sys/globaldata.h>
60 #include <sys/limits.h>
61 #include <vm/vm_extern.h>
64 #include <sys/signal2.h>
65 #include "hammer_disk.h"
66 #include "hammer_mount.h"
67 #include "hammer_ioctl.h"
69 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
71 MALLOC_DECLARE(M_HAMMER);
76 #if !defined(KTR_HAMMER)
77 #define KTR_HAMMER KTR_ALL
79 KTR_INFO_MASTER_EXTERN(hammer);
87 * Key structure used for custom RB tree inode lookups. This prototypes
88 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
90 typedef struct hammer_inode_info {
91 int64_t obj_id; /* (key) object identifier */
92 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
93 u_int32_t obj_localization; /* (key) pseudo-fs */
95 struct hammer_btree_leaf_elm *leaf;
97 } *hammer_inode_info_t;
99 typedef enum hammer_transaction_type {
103 } hammer_transaction_type_t;
106 * HAMMER Transaction tracking
108 struct hammer_transaction {
109 hammer_transaction_type_t type;
110 struct hammer_mount *hmp;
116 struct hammer_volume *rootvol;
119 typedef struct hammer_transaction *hammer_transaction_t;
121 #define HAMMER_TRANSF_NEWINODE 0x0001
122 #define HAMMER_TRANSF_DIDIO 0x0002
123 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */
129 volatile u_int refs; /* active references */
130 volatile u_int lockval; /* lock count and control bits */
131 struct thread *lowner; /* owner if exclusively held */
132 struct thread *rowner; /* owner if exclusively held */
135 #define HAMMER_REFS_LOCKED 0x40000000 /* transition check */
136 #define HAMMER_REFS_WANTED 0x20000000 /* transition check */
137 #define HAMMER_REFS_CHECK 0x10000000 /* transition check */
139 #define HAMMER_REFS_FLAGS (HAMMER_REFS_LOCKED | \
140 HAMMER_REFS_WANTED | \
143 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000
144 #define HAMMER_LOCKF_WANTED 0x20000000
147 hammer_notlocked(struct hammer_lock *lock)
149 return(lock->lockval == 0);
153 hammer_islocked(struct hammer_lock *lock)
155 return(lock->lockval != 0);
159 * Returns the number of refs on the object.
162 hammer_isactive(struct hammer_lock *lock)
164 return(lock->refs & ~HAMMER_REFS_FLAGS);
168 hammer_oneref(struct hammer_lock *lock)
170 return((lock->refs & ~HAMMER_REFS_FLAGS) == 1);
174 hammer_norefs(struct hammer_lock *lock)
176 return((lock->refs & ~HAMMER_REFS_FLAGS) == 0);
180 hammer_norefsorlock(struct hammer_lock *lock)
182 return(lock->refs == 0);
186 hammer_refsorlock(struct hammer_lock *lock)
188 return(lock->refs != 0);
192 * Return if we specifically own the lock exclusively.
195 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
197 if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
198 lock->lowner == td) {
205 * Flush state, used by various structures
207 typedef enum hammer_inode_state {
211 } hammer_inode_state_t;
213 TAILQ_HEAD(hammer_record_list, hammer_record);
216 * Pseudo-filesystem extended data tracking
218 struct hammer_pfs_rb_tree;
219 struct hammer_pseudofs_inmem;
220 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
221 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
222 hammer_pfs_rb_compare, u_int32_t);
224 struct hammer_pseudofs_inmem {
225 RB_ENTRY(hammer_pseudofs_inmem) rb_node;
226 struct hammer_lock lock;
227 u_int32_t localization;
228 hammer_tid_t create_tid;
231 struct hammer_pseudofs_data pfsd;
234 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
236 #define HAMMER_PFSM_DELETED 0x0001
239 * Cache object ids. A fixed number of objid cache structures are
240 * created to reserve object id's for newly created files in multiples
241 * of 100,000, localized to a particular directory, and recycled as
242 * needed. This allows parallel create operations in different
243 * directories to retain fairly localized object ids which in turn
244 * improves reblocking performance and layout.
246 #define OBJID_CACHE_SIZE 2048
247 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */
248 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */
249 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1)
250 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
252 typedef struct hammer_objid_cache {
253 TAILQ_ENTRY(hammer_objid_cache) entry;
254 struct hammer_inode *dip;
255 hammer_tid_t base_tid;
259 } *hammer_objid_cache_t;
262 * Associate an inode with a B-Tree node to cache search start positions
264 typedef struct hammer_node_cache {
265 TAILQ_ENTRY(hammer_node_cache) entry;
266 struct hammer_node *node;
267 struct hammer_inode *ip;
268 } *hammer_node_cache_t;
270 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
275 struct hammer_dedup_crc_rb_tree;
276 RB_HEAD(hammer_dedup_crc_rb_tree, hammer_dedup_cache);
277 RB_PROTOTYPE2(hammer_dedup_crc_rb_tree, hammer_dedup_cache, crc_entry,
278 hammer_dedup_crc_rb_compare, hammer_crc_t);
280 struct hammer_dedup_off_rb_tree;
281 RB_HEAD(hammer_dedup_off_rb_tree, hammer_dedup_cache);
282 RB_PROTOTYPE2(hammer_dedup_off_rb_tree, hammer_dedup_cache, off_entry,
283 hammer_dedup_off_rb_compare, hammer_off_t);
285 #define DEDUP_CACHE_SIZE 4096 /* XXX make it a dynamic tunable */
287 typedef struct hammer_dedup_cache {
288 RB_ENTRY(hammer_dedup_cache) crc_entry;
289 RB_ENTRY(hammer_dedup_cache) off_entry;
290 TAILQ_ENTRY(hammer_dedup_cache) lru_entry;
291 struct hammer_mount *hmp;
293 u_int32_t localization;
296 hammer_off_t data_offset;
298 } *hammer_dedup_cache_t;
301 * Structure used to organize flush groups. Flush groups must be
302 * organized into chunks in order to avoid blowing out the UNDO FIFO.
303 * Without this a 'sync' could end up flushing 50,000 inodes in a single
306 struct hammer_fls_rb_tree;
307 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
308 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
309 hammer_ino_rb_compare);
311 struct hammer_flush_group {
312 TAILQ_ENTRY(hammer_flush_group) flush_entry;
313 struct hammer_fls_rb_tree flush_tree;
314 int seq; /* our seq no */
315 int total_count; /* record load */
316 int running; /* group is running */
321 typedef struct hammer_flush_group *hammer_flush_group_t;
323 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
326 * Structure used to represent an inode in-memory.
328 * The record and data associated with an inode may be out of sync with
329 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
332 * An inode may also hold a cache of unsynchronized records, used for
333 * database and directories only. Unsynchronized regular file data is
334 * stored in the buffer cache.
336 * NOTE: A file which is created and destroyed within the initial
337 * synchronization period can wind up not doing any disk I/O at all.
339 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
341 struct hammer_ino_rb_tree;
343 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
344 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
345 hammer_ino_rb_compare, hammer_inode_info_t);
347 struct hammer_redo_rb_tree;
348 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
349 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
350 hammer_redo_rb_compare, hammer_off_t);
352 struct hammer_rec_rb_tree;
353 struct hammer_record;
354 RB_HEAD(hammer_rec_rb_tree, hammer_record);
355 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
356 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
358 TAILQ_HEAD(hammer_node_list, hammer_node);
360 struct hammer_inode {
361 RB_ENTRY(hammer_inode) rb_node;
362 hammer_inode_state_t flush_state;
363 hammer_flush_group_t flush_group;
364 RB_ENTRY(hammer_inode) rb_flsnode; /* when on flush list */
365 RB_ENTRY(hammer_inode) rb_redonode; /* when INODE_RDIRTY is set */
366 struct hammer_record_list target_list; /* target of dependant recs */
367 int64_t obj_id; /* (key) object identifier */
368 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
369 u_int32_t obj_localization; /* (key) pseudo-fs */
370 struct hammer_mount *hmp;
371 hammer_objid_cache_t objid_cache;
373 int error; /* flush error */
374 int cursor_ip_refs; /* sanity */
377 hammer_pseudofs_inmem_t pfsm;
378 struct lockf advlock;
379 struct hammer_lock lock; /* sync copy interlock */
381 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
382 struct hammer_inode_data ino_data; /* in-memory cache */
383 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
385 struct hammer_node_cache cache[4]; /* search initiate cache */
388 * When a demark is created to synchronize an inode to
389 * disk, certain fields are copied so the front-end VOPs
390 * can continue to run in parallel with the synchronization
391 * occuring in the background.
393 int sync_flags; /* to-sync flags cache */
394 off_t sync_trunc_off; /* to-sync truncation */
395 off_t save_trunc_off; /* write optimization */
396 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
397 struct hammer_inode_data sync_ino_data; /* to-sync cache */
401 * Track the earliest offset in the UNDO/REDO FIFO containing
402 * REDO records. This is staged to the backend during flush
403 * sequences. While the inode is staged redo_fifo_next is used
404 * to track the earliest offset for rotation into redo_fifo_start
405 * on completion of the flush.
407 hammer_off_t redo_fifo_start;
408 hammer_off_t redo_fifo_next;
411 typedef struct hammer_inode *hammer_inode_t;
413 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
416 * NOTE: DDIRTY does not include atime or mtime and does not include
417 * write-append size changes. SDIRTY handles write-append size
420 * REDO indicates that REDO logging is active, creating a definitive
421 * stream of REDO records in the UNDO/REDO log for writes and
422 * truncations, including boundary records when/if REDO is turned off.
423 * REDO is typically enabled by fsync() and turned off if excessive
424 * writes without an fsync() occurs.
426 * RDIRTY indicates that REDO records were laid down in the UNDO/REDO
427 * FIFO (even if REDO is turned off some might still be active) and
428 * still being tracked for this inode. See hammer_redo.c
430 /* (not including atime/mtime) */
431 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
432 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
433 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */
434 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
435 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
436 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
437 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
438 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
439 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
440 #define HAMMER_INODE_RECSW 0x0400 /* waiting on data record flush */
441 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
442 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
443 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */
444 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
445 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
447 #define HAMMER_INODE_TRUNCATED 0x00010000
448 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
449 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
450 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
451 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
452 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
453 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */
454 #define HAMMER_INODE_SDIRTY 0x01000000 /* in-memory ino_data.size is dirty*/
455 #define HAMMER_INODE_REDO 0x02000000 /* REDO logging active */
456 #define HAMMER_INODE_RDIRTY 0x04000000 /* REDO records active in fifo */
458 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY| \
459 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
460 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
461 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
463 #define HAMMER_INODE_MODMASK_NOXDIRTY \
464 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
466 #define HAMMER_INODE_MODMASK_NOREDO \
467 (HAMMER_INODE_DDIRTY| \
468 HAMMER_INODE_XDIRTY| \
469 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
471 #define HAMMER_FLUSH_SIGNAL 0x0001
472 #define HAMMER_FLUSH_RECURSION 0x0002
475 * Used by the inode reclaim code to pipeline reclaims and avoid
476 * blowing out kernel memory or letting the flusher get too far
477 * behind. The reclaim wakes up when count reaches 0 or the
480 struct hammer_reclaim {
481 TAILQ_ENTRY(hammer_reclaim) entry;
486 * Track who is creating the greatest burden on the
489 struct hammer_inostats {
490 pid_t pid; /* track user process */
491 int ltick; /* last tick */
492 int count; /* count (degenerates) */
495 #define HAMMER_INOSTATS_HSIZE 32
496 #define HAMMER_INOSTATS_HMASK (HAMMER_INOSTATS_HSIZE - 1)
499 * Structure used to represent an unsynchronized record in-memory. These
500 * records typically represent directory entries. Only non-historical
501 * records are kept in-memory.
503 * Records are organized as a per-inode RB-Tree. If the inode is not
504 * on disk then neither are any records and the in-memory record tree
505 * represents the entire contents of the inode. If the inode is on disk
506 * then the on-disk B-Tree is scanned in parallel with the in-memory
507 * RB-Tree to synthesize the current state of the file.
509 * Records are also used to enforce the ordering of directory create/delete
510 * operations. A new inode will not be flushed to disk unless its related
511 * directory entry is also being flushed at the same time. A directory entry
512 * will not be removed unless its related inode is also being removed at the
515 typedef enum hammer_record_type {
516 HAMMER_MEM_RECORD_GENERAL, /* misc record */
517 HAMMER_MEM_RECORD_INODE, /* inode record */
518 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
519 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
520 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
521 } hammer_record_type_t;
523 struct hammer_record {
524 RB_ENTRY(hammer_record) rb_node;
525 TAILQ_ENTRY(hammer_record) target_entry;
526 hammer_inode_state_t flush_state;
527 hammer_flush_group_t flush_group;
528 hammer_record_type_t type;
529 struct hammer_lock lock;
530 struct hammer_reserve *resv;
531 struct hammer_inode *ip;
532 struct hammer_inode *target_ip;
533 struct hammer_btree_leaf_elm leaf;
534 union hammer_data_ondisk *data;
537 hammer_off_t zone2_offset; /* direct-write only */
540 typedef struct hammer_record *hammer_record_t;
543 * Record flags. Note that FE can only be set by the frontend if the
544 * record has not been interlocked by the backend w/ BE.
546 #define HAMMER_RECF_ALLOCDATA 0x0001
547 #define HAMMER_RECF_ONRBTREE 0x0002
548 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
549 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
550 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */
551 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
552 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
553 #define HAMMER_RECF_DEDUPED 0x0080 /* will be live-dedup'ed */
554 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
555 #define HAMMER_RECF_REDO 0x1000 /* REDO was laid down */
558 * These flags must be separate to deal with SMP races
560 #define HAMMER_RECG_DIRECT_IO 0x0001 /* related direct I/O running*/
561 #define HAMMER_RECG_DIRECT_WAIT 0x0002 /* related direct I/O running*/
562 #define HAMMER_RECG_DIRECT_INVAL 0x0004 /* buffer alias invalidation */
564 * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
566 #define HAMMER_CREATE_MODE_UMIRROR 0x0001
567 #define HAMMER_CREATE_MODE_SYS 0x0002
569 #define HAMMER_DELETE_ADJUST 0x0001
570 #define HAMMER_DELETE_DESTROY 0x0002
573 * In-memory structures representing on-disk structures.
575 struct hammer_volume;
576 struct hammer_buffer;
579 struct hammer_reserve;
581 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
582 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
583 RB_HEAD(hammer_nod_rb_tree, hammer_node);
584 RB_HEAD(hammer_und_rb_tree, hammer_undo);
585 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
586 RB_HEAD(hammer_mod_rb_tree, hammer_io);
588 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
589 hammer_vol_rb_compare, int32_t);
590 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
591 hammer_buf_rb_compare, hammer_off_t);
592 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
593 hammer_nod_rb_compare, hammer_off_t);
594 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
595 hammer_und_rb_compare, hammer_off_t);
596 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
597 hammer_res_rb_compare, hammer_off_t);
598 RB_PROTOTYPE2(hammer_mod_rb_tree, hammer_io, rb_node,
599 hammer_mod_rb_compare, hammer_off_t);
602 * IO management - embedded at the head of various in-memory structures
604 * VOLUME - hammer_volume containing meta-data
605 * META_BUFFER - hammer_buffer containing meta-data
606 * DATA_BUFFER - hammer_buffer containing pure-data
608 * Dirty volume headers and dirty meta-data buffers are locked until the
609 * flusher can sequence them out. Dirty pure-data buffers can be written.
610 * Clean buffers can be passively released.
612 typedef enum hammer_io_type {
613 HAMMER_STRUCTURE_VOLUME,
614 HAMMER_STRUCTURE_META_BUFFER,
615 HAMMER_STRUCTURE_UNDO_BUFFER,
616 HAMMER_STRUCTURE_DATA_BUFFER,
617 HAMMER_STRUCTURE_DUMMY
620 union hammer_io_structure;
624 LIST_ENTRY(worklist) node;
627 TAILQ_HEAD(hammer_io_list, hammer_io);
628 typedef struct hammer_io_list *hammer_io_list_t;
631 struct worklist worklist;
632 struct hammer_lock lock;
633 enum hammer_io_type type;
634 struct hammer_mount *hmp;
635 struct hammer_volume *volume;
636 RB_ENTRY(hammer_io) rb_node; /* if modified */
637 TAILQ_ENTRY(hammer_io) iorun_entry; /* iorun_list */
638 struct hammer_mod_rb_tree *mod_root;
640 int64_t offset; /* zone-2 offset */
641 int bytes; /* buffer cache buffer size */
645 * These can be modified at any time by the backend while holding
646 * io_token, due to bio_done and hammer_io_complete() callbacks.
648 u_int running : 1; /* bp write IO in progress */
649 u_int waiting : 1; /* someone is waiting on us */
650 u_int ioerror : 1; /* abort on io-error */
654 * These can only be modified by the frontend while holding
655 * fs_token, or by the backend while holding the io interlocked
656 * with no references (which will block the frontend when it
657 * tries to reference it).
659 * WARNING! SMP RACES will create havoc if the callbacks ever tried
660 * to modify any of these outside the above restrictions.
662 u_int modified : 1; /* bp's data was modified */
663 u_int released : 1; /* bp released (w/ B_LOCKED set) */
664 u_int validated : 1; /* ondisk has been validated */
665 u_int waitdep : 1; /* flush waits for dependancies */
666 u_int recovered : 1; /* has recovery ref */
667 u_int waitmod : 1; /* waiting for modify_refs */
668 u_int reclaim : 1; /* reclaim requested */
669 u_int gencrc : 1; /* crc needs to be generated */
673 typedef struct hammer_io *hammer_io_t;
675 #define HAMMER_CLUSTER_SIZE (64 * 1024)
676 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
677 #undef HAMMER_CLUSTER_SIZE
678 #define HAMMER_CLUSTER_SIZE MAXBSIZE
680 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
683 * In-memory volume representing on-disk buffer
685 struct hammer_volume {
687 RB_ENTRY(hammer_volume) rb_node;
688 struct hammer_volume_ondisk *ondisk;
690 int64_t nblocks; /* note: special calculation for statfs */
691 int64_t buffer_base; /* base offset of buffer 0 */
692 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
693 hammer_off_t maxraw_off; /* Maximum raw offset for device */
699 typedef struct hammer_volume *hammer_volume_t;
702 * In-memory buffer (other then volume, super-cluster, or cluster),
703 * representing an on-disk buffer.
705 struct hammer_buffer {
707 RB_ENTRY(hammer_buffer) rb_node;
709 hammer_off_t zoneX_offset;
710 hammer_off_t zone2_offset;
711 struct hammer_reserve *resv;
712 struct hammer_node_list clist;
715 typedef struct hammer_buffer *hammer_buffer_t;
718 * In-memory B-Tree node, representing an on-disk B-Tree node.
720 * This is a hang-on structure which is backed by a hammer_buffer,
721 * indexed by a hammer_cluster, and used for fine-grained locking of
722 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
723 * can contain multiple nodes representing wildly disassociated portions
724 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
726 * This structure uses a cluster-relative index to reduce the number
727 * of layers required to access it, and also because all on-disk B-Tree
728 * references are cluster-relative offsets.
731 struct hammer_lock lock; /* node-by-node lock */
732 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
733 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */
734 hammer_off_t node_offset; /* full offset spec */
735 struct hammer_mount *hmp;
736 struct hammer_buffer *buffer; /* backing buffer */
737 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
738 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */
739 struct hammer_node_cache_list cache_list; /* passive caches */
743 #define HAMMER_NODE_DELETED 0x0001
744 #define HAMMER_NODE_FLUSH 0x0002
745 #define HAMMER_NODE_CRCGOOD 0x0004
746 #define HAMMER_NODE_NEEDSCRC 0x0008
747 #define HAMMER_NODE_NEEDSMIRROR 0x0010
748 #define HAMMER_NODE_CRCBAD 0x0020
749 #define HAMMER_NODE_NONLINEAR 0x0040 /* linear heuristic */
751 #define HAMMER_NODE_CRCANY (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
753 typedef struct hammer_node *hammer_node_t;
756 * List of locked nodes. This structure is used to lock potentially large
757 * numbers of nodes as an aid for complex B-Tree operations.
759 struct hammer_node_lock;
760 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
762 struct hammer_node_lock {
763 TAILQ_ENTRY(hammer_node_lock) entry;
764 struct hammer_node_lock_list list;
765 struct hammer_node_lock *parent;
767 hammer_node_ondisk_t copy; /* copy of on-disk data */
768 int index; /* index of this node in parent */
769 int count; /* count children */
773 typedef struct hammer_node_lock *hammer_node_lock_t;
775 #define HAMMER_NODE_LOCK_UPDATED 0x0001
776 #define HAMMER_NODE_LOCK_LCACHE 0x0002
779 * Common I/O management structure - embedded in in-memory structures
780 * which are backed by filesystem buffers.
782 union hammer_io_structure {
784 struct hammer_volume volume;
785 struct hammer_buffer buffer;
788 typedef union hammer_io_structure *hammer_io_structure_t;
791 * The reserve structure prevents the blockmap from allocating
792 * out of a reserved bigblock. Such reservations are used by
793 * the direct-write mechanism.
795 * The structure is also used to hold off on reallocations of
796 * big blocks from the freemap until flush dependancies have
799 struct hammer_reserve {
800 RB_ENTRY(hammer_reserve) rb_node;
801 TAILQ_ENTRY(hammer_reserve) delay_entry;
808 hammer_off_t zone_offset;
811 typedef struct hammer_reserve *hammer_reserve_t;
813 #define HAMMER_RESF_ONDELAY 0x0001
814 #define HAMMER_RESF_LAYER2FREE 0x0002
816 #include "hammer_cursor.h"
819 * The undo structure tracks recent undos to avoid laying down duplicate
820 * undos within a flush group, saving us a significant amount of overhead.
822 * This is strictly a heuristic.
824 #define HAMMER_MAX_UNDOS 1024
825 #define HAMMER_MAX_FLUSHERS 4
828 RB_ENTRY(hammer_undo) rb_node;
829 TAILQ_ENTRY(hammer_undo) lru_entry;
834 typedef struct hammer_undo *hammer_undo_t;
836 struct hammer_flusher_info;
837 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
839 struct hammer_flusher {
840 int signal; /* flusher thread sequencer */
841 int done; /* last completed flush group */
842 int next; /* next unallocated flg seqno */
843 int group_lock; /* lock sequencing of the next flush */
844 int exiting; /* request master exit */
845 thread_t td; /* master flusher thread */
846 hammer_tid_t tid; /* last flushed transaction id */
847 int finalize_want; /* serialize finalization */
848 struct hammer_lock finalize_lock; /* serialize finalization */
849 struct hammer_transaction trans; /* shared transaction */
850 struct hammer_flusher_info_list run_list;
851 struct hammer_flusher_info_list ready_list;
854 #define HAMMER_FLUSH_UNDOS_RELAXED 0
855 #define HAMMER_FLUSH_UNDOS_FORCED 1
856 #define HAMMER_FLUSH_UNDOS_AUTO 2
858 * Internal hammer mount data structure
860 struct hammer_mount {
862 /*struct vnode *rootvp;*/
863 struct hammer_ino_rb_tree rb_inos_root;
864 struct hammer_redo_rb_tree rb_redo_root;
865 struct hammer_vol_rb_tree rb_vols_root;
866 struct hammer_nod_rb_tree rb_nods_root;
867 struct hammer_und_rb_tree rb_undo_root;
868 struct hammer_res_rb_tree rb_resv_root;
869 struct hammer_buf_rb_tree rb_bufs_root;
870 struct hammer_pfs_rb_tree rb_pfsm_root;
872 struct hammer_dedup_crc_rb_tree rb_dedup_crc_root;
873 struct hammer_dedup_off_rb_tree rb_dedup_off_root;
875 struct hammer_volume *rootvol;
876 struct hammer_base_elm root_btree_beg;
877 struct hammer_base_elm root_btree_end;
879 struct malloc_type *m_misc;
880 struct malloc_type *m_inodes;
882 int flags; /* HAMMER_MOUNT_xxx flags */
887 int master_id; /* -1 or 0-15 - clustering and mirroring */
888 int version; /* hammer filesystem version to use */
889 int rsv_inodes; /* reserved space due to dirty inodes */
890 int64_t rsv_databytes; /* reserved space due to record data */
891 int rsv_recs; /* reserved space due to dirty records */
892 int rsv_fromdelay; /* bigblocks reserved due to flush delay */
893 int undo_rec_limit; /* based on size of undo area */
895 int count_newrecords;
897 int volume_to_remove; /* volume that is currently being removed */
899 int inode_reclaims; /* inodes pending reclaim by flusher */
900 int count_inodes; /* total number of inodes */
901 int count_iqueued; /* inodes queued to flusher */
903 struct hammer_flusher flusher;
905 u_int check_interrupt;
908 struct hammer_mod_rb_tree volu_root; /* dirty undo buffers */
909 struct hammer_mod_rb_tree undo_root; /* dirty undo buffers */
910 struct hammer_mod_rb_tree data_root; /* dirty data buffers */
911 struct hammer_mod_rb_tree meta_root; /* dirty meta bufs */
912 struct hammer_mod_rb_tree lose_root; /* loose buffers */
913 int locked_dirty_space; /* meta/volu count */
914 int io_running_space; /* io_token */
915 int io_running_wakeup; /* io_token */
916 int objid_cache_count;
917 int dedup_cache_count;
918 int error; /* critical I/O error */
919 struct krate krate; /* rate limited kprintf */
920 hammer_tid_t asof; /* snapshot mount */
921 hammer_tid_t next_tid;
922 hammer_tid_t flush_tid1; /* flusher tid sequencing */
923 hammer_tid_t flush_tid2; /* flusher tid sequencing */
924 int64_t copy_stat_freebigblocks; /* number of free bigblocks */
925 u_int32_t undo_seqno; /* UNDO/REDO FIFO seqno */
926 u_int32_t recover_stage2_seqno; /* REDO recovery seqno */
927 hammer_off_t recover_stage2_offset; /* REDO recovery offset */
929 struct netexport export;
930 struct hammer_lock sync_lock;
931 struct hammer_lock free_lock;
932 struct hammer_lock undo_lock;
933 struct hammer_lock blkmap_lock;
934 struct hammer_lock snapshot_lock;
935 struct hammer_lock volume_lock;
936 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
937 struct hammer_undo undos[HAMMER_MAX_UNDOS];
939 TAILQ_HEAD(, hammer_undo) undo_lru_list;
940 TAILQ_HEAD(, hammer_reserve) delay_list;
941 struct hammer_flush_group_list flush_group_list;
942 hammer_flush_group_t fill_flush_group;
943 hammer_flush_group_t next_flush_group;
944 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
945 TAILQ_HEAD(, hammer_dedup_cache) dedup_lru_list;
946 hammer_dedup_cache_t dedup_free_cache;
947 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
948 TAILQ_HEAD(, hammer_io) iorun_list;
950 struct lwkt_token fs_token; /* high level */
951 struct lwkt_token io_token; /* low level (IO callback) */
953 struct hammer_inostats inostats[HAMMER_INOSTATS_HSIZE];
956 typedef struct hammer_mount *hammer_mount_t;
958 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001
959 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002
960 #define HAMMER_MOUNT_REDO_SYNC 0x0004
961 #define HAMMER_MOUNT_REDO_RECOVERY_REQ 0x0008
962 #define HAMMER_MOUNT_REDO_RECOVERY_RUN 0x0010
964 struct hammer_sync_info {
970 * Minium buffer cache bufs required to rebalance the B-Tree.
971 * This is because we must hold the children and the children's children
972 * locked. Even this might not be enough if things are horribly out
975 #define HAMMER_REBALANCE_MIN_BUFS \
976 (HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
982 * checkspace slop (8MB chunks), higher numbers are more conservative.
984 #define HAMMER_CHKSPC_REBLOCK 25
985 #define HAMMER_CHKSPC_MIRROR 20
986 #define HAMMER_CHKSPC_WRITE 20
987 #define HAMMER_CHKSPC_CREATE 20
988 #define HAMMER_CHKSPC_REMOVE 10
989 #define HAMMER_CHKSPC_EMERGENCY 0
993 extern struct vop_ops hammer_vnode_vops;
994 extern struct vop_ops hammer_spec_vops;
995 extern struct vop_ops hammer_fifo_vops;
996 extern struct bio_ops hammer_bioops;
998 extern int hammer_debug_io;
999 extern int hammer_debug_general;
1000 extern int hammer_debug_debug;
1001 extern int hammer_debug_inode;
1002 extern int hammer_debug_locks;
1003 extern int hammer_debug_btree;
1004 extern int hammer_debug_tid;
1005 extern int hammer_debug_recover;
1006 extern int hammer_debug_recover_faults;
1007 extern int hammer_debug_critical;
1008 extern int hammer_cluster_enable;
1009 extern int hammer_live_dedup;
1010 extern int hammer_count_fsyncs;
1011 extern int hammer_count_inodes;
1012 extern int hammer_count_iqueued;
1013 extern int hammer_count_reclaiming;
1014 extern int hammer_count_records;
1015 extern int hammer_count_record_datas;
1016 extern int hammer_count_volumes;
1017 extern int hammer_count_buffers;
1018 extern int hammer_count_nodes;
1019 extern int64_t hammer_count_extra_space_used;
1020 extern int64_t hammer_stats_btree_lookups;
1021 extern int64_t hammer_stats_btree_searches;
1022 extern int64_t hammer_stats_btree_inserts;
1023 extern int64_t hammer_stats_btree_deletes;
1024 extern int64_t hammer_stats_btree_elements;
1025 extern int64_t hammer_stats_btree_splits;
1026 extern int64_t hammer_stats_btree_iterations;
1027 extern int64_t hammer_stats_btree_root_iterations;
1028 extern int64_t hammer_stats_record_iterations;
1029 extern int64_t hammer_stats_file_read;
1030 extern int64_t hammer_stats_file_write;
1031 extern int64_t hammer_stats_file_iopsr;
1032 extern int64_t hammer_stats_file_iopsw;
1033 extern int64_t hammer_stats_disk_read;
1034 extern int64_t hammer_stats_disk_write;
1035 extern int64_t hammer_stats_inode_flushes;
1036 extern int64_t hammer_stats_commits;
1037 extern int64_t hammer_stats_undo;
1038 extern int64_t hammer_stats_redo;
1039 extern int hammer_count_dirtybufspace;
1040 extern int hammer_count_refedbufs;
1041 extern int hammer_count_reservations;
1042 extern int hammer_count_io_running_read;
1043 extern int hammer_count_io_running_write;
1044 extern int hammer_count_io_locked;
1045 extern int hammer_limit_dirtybufspace;
1046 extern int hammer_limit_running_io;
1047 extern int hammer_limit_recs;
1048 extern int hammer_limit_inode_recs;
1049 extern int hammer_limit_reclaim;
1050 extern int hammer_live_dedup_cache_size;
1051 extern int hammer_limit_redo;
1052 extern int hammer_bio_count;
1053 extern int hammer_verify_zone;
1054 extern int hammer_verify_data;
1055 extern int hammer_write_mode;
1056 extern int hammer_double_buffer;
1057 extern int hammer_yield_check;
1058 extern int hammer_fsync_mode;
1059 extern int hammer_autoflush;
1060 extern int64_t hammer_contention_count;
1062 extern int64_t hammer_live_dedup_vnode_bcmps;
1063 extern int64_t hammer_live_dedup_device_bcmps;
1064 extern int64_t hammer_live_dedup_findblk_failures;
1065 extern int64_t hammer_live_dedup_bmap_saves;
1067 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
1068 int error, const char *msg);
1069 int hammer_vop_inactive(struct vop_inactive_args *);
1070 int hammer_vop_reclaim(struct vop_reclaim_args *);
1071 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
1072 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
1073 hammer_inode_t dip, int64_t obj_id,
1074 hammer_tid_t asof, u_int32_t localization,
1075 int flags, int *errorp);
1076 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
1077 hammer_inode_t dip, int64_t obj_id,
1078 hammer_tid_t asof, u_int32_t localization,
1079 int flags, int *errorp);
1080 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
1081 int64_t obj_id, hammer_tid_t asof,
1082 u_int32_t localization);
1083 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
1084 hammer_inode_info_t iinfo,
1085 int (*callback)(hammer_inode_t ip, void *data),
1087 void hammer_put_inode(struct hammer_inode *ip);
1088 void hammer_put_inode_ref(struct hammer_inode *ip);
1089 void hammer_inode_waitreclaims(hammer_transaction_t trans);
1091 int hammer_unload_volume(hammer_volume_t volume, void *data __unused);
1092 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
1094 int hammer_unload_buffer(hammer_buffer_t buffer, void *data);
1095 int hammer_install_volume(hammer_mount_t hmp, const char *volname,
1096 struct vnode *devvp);
1097 int hammer_mountcheck_volumes(hammer_mount_t hmp);
1099 int hammer_mem_add(hammer_record_t record);
1100 int hammer_ip_lookup(hammer_cursor_t cursor);
1101 int hammer_ip_first(hammer_cursor_t cursor);
1102 int hammer_ip_next(hammer_cursor_t cursor);
1103 int hammer_ip_resolve_data(hammer_cursor_t cursor);
1104 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1106 int hammer_create_at_cursor(hammer_cursor_t cursor,
1107 hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1108 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1109 hammer_tid_t delete_tid, u_int32_t delete_ts,
1110 int track, int64_t *stat_bytes);
1111 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
1113 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1114 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1117 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1118 void hammer_flush_record_done(hammer_record_t record, int error);
1119 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1120 void hammer_rel_mem_record(hammer_record_t record);
1122 int hammer_cursor_up(hammer_cursor_t cursor);
1123 int hammer_cursor_up_locked(hammer_cursor_t cursor);
1124 int hammer_cursor_down(hammer_cursor_t cursor);
1125 int hammer_cursor_upgrade(hammer_cursor_t cursor);
1126 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1127 void hammer_cursor_downgrade(hammer_cursor_t cursor);
1128 int hammer_cursor_upgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1129 void hammer_cursor_downgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1130 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1132 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1133 int hammer_lock_ex_try(struct hammer_lock *lock);
1134 void hammer_lock_sh(struct hammer_lock *lock);
1135 int hammer_lock_sh_try(struct hammer_lock *lock);
1136 int hammer_lock_upgrade(struct hammer_lock *lock, int shcount);
1137 void hammer_lock_downgrade(struct hammer_lock *lock, int shcount);
1138 int hammer_lock_status(struct hammer_lock *lock);
1139 void hammer_unlock(struct hammer_lock *lock);
1140 void hammer_ref(struct hammer_lock *lock);
1141 int hammer_ref_interlock(struct hammer_lock *lock);
1142 int hammer_ref_interlock_true(struct hammer_lock *lock);
1143 void hammer_ref_interlock_done(struct hammer_lock *lock);
1144 void hammer_rel(struct hammer_lock *lock);
1145 int hammer_rel_interlock(struct hammer_lock *lock, int locked);
1146 void hammer_rel_interlock_done(struct hammer_lock *lock, int orig_locked);
1147 int hammer_get_interlock(struct hammer_lock *lock);
1148 int hammer_try_interlock_norefs(struct hammer_lock *lock);
1149 void hammer_put_interlock(struct hammer_lock *lock, int error);
1151 void hammer_sync_lock_ex(hammer_transaction_t trans);
1152 void hammer_sync_lock_sh(hammer_transaction_t trans);
1153 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
1154 void hammer_sync_unlock(hammer_transaction_t trans);
1156 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
1157 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1158 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1159 u_int64_t hammer_timespec_to_time(struct timespec *ts);
1160 int hammer_str_to_tid(const char *str, int *ispfsp,
1161 hammer_tid_t *tidp, u_int32_t *localizationp);
1162 int hammer_is_atatext(const char *name, int len);
1163 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1165 void hammer_clear_objid(hammer_inode_t dip);
1166 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1168 int hammer_dedup_crc_rb_compare(hammer_dedup_cache_t dc1,
1169 hammer_dedup_cache_t dc2);
1170 int hammer_dedup_off_rb_compare(hammer_dedup_cache_t dc1,
1171 hammer_dedup_cache_t dc2);
1172 hammer_dedup_cache_t hammer_dedup_cache_add(hammer_inode_t ip,
1173 hammer_btree_leaf_elm_t leaf);
1174 hammer_dedup_cache_t hammer_dedup_cache_lookup(hammer_mount_t hmp,
1176 void hammer_dedup_cache_inval(hammer_mount_t hmp, hammer_off_t base_offset);
1177 void hammer_destroy_dedup_cache(hammer_mount_t hmp);
1178 void hammer_dump_dedup_cache(hammer_mount_t hmp);
1179 int hammer_dedup_validate(hammer_dedup_cache_t dcp, int zone, int bytes,
1182 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1184 void hammer_clear_undo_history(hammer_mount_t hmp);
1185 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
1186 int hammer_get_dtype(u_int8_t obj_type);
1187 u_int8_t hammer_get_obj_type(enum vtype vtype);
1188 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
1189 u_int32_t *max_iterationsp);
1190 int hammer_nohistory(hammer_inode_t ip);
1192 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1193 hammer_node_cache_t cache, hammer_inode_t ip);
1194 void hammer_normalize_cursor(hammer_cursor_t cursor);
1195 void hammer_done_cursor(hammer_cursor_t cursor);
1196 int hammer_recover_cursor(hammer_cursor_t cursor);
1197 void hammer_unlock_cursor(hammer_cursor_t cursor);
1198 int hammer_lock_cursor(hammer_cursor_t cursor);
1199 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
1200 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1202 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1203 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1205 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1207 void hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1208 hammer_node_t onode, int oindex,
1209 hammer_node_t nnode, int nindex);
1210 void hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1211 hammer_node_t nparent, int nindex);
1212 void hammer_cursor_inserted_element(hammer_node_t node, int index);
1213 void hammer_cursor_deleted_element(hammer_node_t node, int index);
1214 void hammer_cursor_invalidate_cache(hammer_cursor_t cursor);
1216 int hammer_btree_lookup(hammer_cursor_t cursor);
1217 int hammer_btree_first(hammer_cursor_t cursor);
1218 int hammer_btree_last(hammer_cursor_t cursor);
1219 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
1220 int hammer_btree_iterate(hammer_cursor_t cursor);
1221 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1222 int hammer_btree_insert(hammer_cursor_t cursor,
1223 hammer_btree_leaf_elm_t elm, int *doprop);
1224 int hammer_btree_delete(hammer_cursor_t cursor);
1225 void hammer_btree_do_propagation(hammer_cursor_t cursor,
1226 hammer_pseudofs_inmem_t pfsm,
1227 hammer_btree_leaf_elm_t leaf);
1228 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1229 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1230 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1231 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1233 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1234 hammer_btree_elm_t elm);
1235 void hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1236 void hammer_btree_lcache_init(hammer_mount_t hmp, hammer_node_lock_t lcache,
1238 void hammer_btree_lcache_free(hammer_mount_t hmp, hammer_node_lock_t lcache);
1239 int hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1240 hammer_node_lock_t parent,
1241 hammer_node_lock_t lcache);
1242 void hammer_btree_lock_copy(hammer_cursor_t cursor,
1243 hammer_node_lock_t parent);
1244 int hammer_btree_sync_copy(hammer_cursor_t cursor,
1245 hammer_node_lock_t parent);
1246 void hammer_btree_unlock_children(hammer_mount_t hmp,
1247 hammer_node_lock_t parent,
1248 hammer_node_lock_t lcache);
1249 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1250 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1251 hammer_node_t node, int *parent_indexp,
1252 int *errorp, int try_exclusive);
1254 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1255 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1257 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1258 int *errorp, struct hammer_buffer **bufferp);
1259 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1260 int *errorp, struct hammer_buffer **bufferp);
1261 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1262 int *errorp, struct hammer_buffer **bufferp);
1263 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1264 int *errorp, struct hammer_buffer **bufferp);
1266 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1268 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1269 int32_t vol_no, int *errorp);
1270 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1271 int bytes, int isnew, int *errorp);
1272 void hammer_sync_buffers(hammer_mount_t hmp,
1273 hammer_off_t base_offset, int bytes);
1274 int hammer_del_buffers(hammer_mount_t hmp,
1275 hammer_off_t base_offset,
1276 hammer_off_t zone2_offset, int bytes,
1277 int report_conflicts);
1279 int hammer_ref_volume(hammer_volume_t volume);
1280 int hammer_ref_buffer(hammer_buffer_t buffer);
1281 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1283 void hammer_rel_volume(hammer_volume_t volume, int locked);
1284 void hammer_rel_buffer(hammer_buffer_t buffer, int locked);
1286 int hammer_vfs_export(struct mount *mp, int op,
1287 const struct export_args *export);
1288 hammer_node_t hammer_get_node(hammer_transaction_t trans,
1289 hammer_off_t node_offset, int isnew, int *errorp);
1290 void hammer_ref_node(hammer_node_t node);
1291 hammer_node_t hammer_ref_node_safe(hammer_transaction_t trans,
1292 hammer_node_cache_t cache, int *errorp);
1293 void hammer_rel_node(hammer_node_t node);
1294 void hammer_delete_node(hammer_transaction_t trans,
1295 hammer_node_t node);
1296 void hammer_cache_node(hammer_node_cache_t cache,
1297 hammer_node_t node);
1298 void hammer_uncache_node(hammer_node_cache_t cache);
1299 void hammer_flush_node(hammer_node_t node, int locked);
1301 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1302 struct hammer_buffer *buffer);
1303 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1304 hammer_off_t hint, int *errorp);
1305 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1306 u_int16_t rec_type, hammer_off_t *data_offsetp,
1307 struct hammer_buffer **data_bufferp,
1308 hammer_off_t hint, int *errorp);
1310 int hammer_generate_undo(hammer_transaction_t trans,
1311 hammer_off_t zone1_offset, void *base, int len);
1312 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1313 hammer_off_t file_offset, u_int32_t flags,
1314 void *base, int len);
1315 void hammer_generate_redo_sync(hammer_transaction_t trans);
1316 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1317 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1319 void hammer_format_undo(void *base, u_int32_t seqno);
1320 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1322 void hammer_put_volume(struct hammer_volume *volume, int flush);
1323 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1325 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1326 hammer_off_t owner, int *errorp);
1327 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1328 hammer_off_t owner, int *errorp);
1329 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1330 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1331 int bytes, hammer_off_t hint, int *errorp);
1332 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1333 int bytes, hammer_off_t *zone_offp, int *errorp);
1334 hammer_reserve_t hammer_blockmap_reserve_dedup(hammer_mount_t hmp, int zone,
1335 int bytes, hammer_off_t zone_offset, int *errorp);
1336 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1337 hammer_reserve_t resv);
1338 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1339 void hammer_blockmap_free(hammer_transaction_t trans,
1340 hammer_off_t bmap_off, int bytes);
1341 int hammer_blockmap_dedup(hammer_transaction_t trans,
1342 hammer_off_t bmap_off, int bytes);
1343 int hammer_blockmap_finalize(hammer_transaction_t trans,
1344 hammer_reserve_t resv,
1345 hammer_off_t bmap_off, int bytes);
1346 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1347 int *curp, int *errorp);
1348 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1350 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1352 int64_t hammer_undo_used(hammer_transaction_t trans);
1353 int64_t hammer_undo_space(hammer_transaction_t trans);
1354 int64_t hammer_undo_max(hammer_mount_t hmp);
1355 int hammer_undo_reclaim(hammer_io_t io);
1357 void hammer_start_transaction(struct hammer_transaction *trans,
1358 struct hammer_mount *hmp);
1359 void hammer_simple_transaction(struct hammer_transaction *trans,
1360 struct hammer_mount *hmp);
1361 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1362 struct hammer_mount *hmp);
1363 void hammer_done_transaction(struct hammer_transaction *trans);
1364 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1366 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1367 void hammer_flush_inode(hammer_inode_t ip, int flags);
1368 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1369 void hammer_wait_inode(hammer_inode_t ip);
1371 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1372 struct ucred *cred, struct hammer_inode *dip,
1373 const char *name, int namelen,
1374 hammer_pseudofs_inmem_t pfsm,
1375 struct hammer_inode **ipp);
1376 void hammer_rel_inode(hammer_inode_t ip, int flush);
1377 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1378 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1379 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1380 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1382 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1383 void hammer_test_inode(hammer_inode_t dip);
1384 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1386 int hammer_ip_add_directory(struct hammer_transaction *trans,
1387 hammer_inode_t dip, const char *name, int bytes,
1388 hammer_inode_t nip);
1389 int hammer_ip_del_directory(struct hammer_transaction *trans,
1390 hammer_cursor_t cursor, hammer_inode_t dip,
1392 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1393 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1394 void *data, int bytes, int *errorp);
1395 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1396 int hammer_ip_add_record(struct hammer_transaction *trans,
1397 hammer_record_t record);
1398 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1399 int64_t ran_beg, int64_t ran_end, int truncating);
1400 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1402 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1403 int64_t offset, void *data, int bytes);
1404 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1405 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1406 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans,
1407 u_int32_t localization, int *errorp);
1408 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1409 hammer_pseudofs_inmem_t pfsm);
1410 int hammer_save_pseudofs(hammer_transaction_t trans,
1411 hammer_pseudofs_inmem_t pfsm);
1412 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1413 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1414 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1415 struct ucred *cred);
1417 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1418 enum hammer_io_type type);
1419 int hammer_io_read(struct vnode *devvp, struct hammer_io *io, int limit);
1420 void hammer_io_advance(struct hammer_io *io);
1421 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1422 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1423 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1424 void hammer_io_flush(struct hammer_io *io, int reclaim);
1425 void hammer_io_wait(struct hammer_io *io);
1426 void hammer_io_waitdep(struct hammer_io *io);
1427 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1428 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1429 hammer_btree_leaf_elm_t leaf);
1430 int hammer_io_indirect_read(hammer_mount_t hmp, struct bio *bio,
1431 hammer_btree_leaf_elm_t leaf);
1432 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1433 hammer_record_t record);
1434 void hammer_io_direct_wait(hammer_record_t record);
1435 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1436 void hammer_io_write_interlock(hammer_io_t io);
1437 void hammer_io_done_interlock(hammer_io_t io);
1438 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1439 void hammer_io_clear_modlist(struct hammer_io *io);
1440 void hammer_io_flush_sync(hammer_mount_t hmp);
1441 void hammer_io_clear_error(struct hammer_io *io);
1442 void hammer_io_clear_error_noassert(struct hammer_io *io);
1443 void hammer_io_notmeta(hammer_buffer_t buffer);
1444 void hammer_io_limit_backlog(hammer_mount_t hmp);
1446 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1447 void *base, int len);
1448 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1449 void *base, int len);
1450 void hammer_modify_volume_done(hammer_volume_t volume);
1451 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1453 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1454 struct hammer_ioc_reblock *reblock);
1455 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1456 struct hammer_ioc_rebalance *rebal);
1457 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1458 struct hammer_ioc_prune *prune);
1459 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1460 struct hammer_ioc_mirror_rw *mirror);
1461 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1462 struct hammer_ioc_mirror_rw *mirror);
1463 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1464 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1465 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1466 struct hammer_ioc_pseudofs_rw *pfs);
1467 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1468 struct hammer_ioc_pseudofs_rw *pfs);
1469 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1470 struct hammer_ioc_pseudofs_rw *pfs);
1471 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1472 struct hammer_ioc_pseudofs_rw *pfs);
1473 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1474 struct hammer_ioc_pseudofs_rw *pfs);
1475 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1476 struct hammer_ioc_volume *ioc);
1477 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1478 struct hammer_ioc_volume *ioc);
1479 int hammer_ioc_volume_list(hammer_transaction_t trans, hammer_inode_t ip,
1480 struct hammer_ioc_volume_list *ioc);
1481 int hammer_ioc_dedup(hammer_transaction_t trans, hammer_inode_t ip,
1482 struct hammer_ioc_dedup *dedup);
1484 int hammer_signal_check(hammer_mount_t hmp);
1486 void hammer_flusher_create(hammer_mount_t hmp);
1487 void hammer_flusher_destroy(hammer_mount_t hmp);
1488 void hammer_flusher_sync(hammer_mount_t hmp);
1489 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1490 int hammer_flusher_async_one(hammer_mount_t hmp);
1491 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1492 void hammer_flusher_wait_next(hammer_mount_t hmp);
1493 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1494 int hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1495 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1496 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1497 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1498 int hammer_flusher_haswork(hammer_mount_t hmp);
1499 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1501 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1502 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1503 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1504 hammer_volume_t root_volume, int final);
1506 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1507 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1508 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1510 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1511 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1512 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1513 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1514 void hkprintf(const char *ctl, ...) __printflike(1, 2);
1515 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1518 int hammer_blocksize(int64_t file_offset);
1519 int hammer_blockoff(int64_t file_offset);
1520 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1523 * Shortcut for _hammer_checkspace(), used all over the code.
1526 hammer_checkspace(hammer_mount_t hmp, int slop)
1528 return(_hammer_checkspace(hmp, slop, NULL));
1533 static __inline void
1534 hammer_wait_mem_record(hammer_record_t record)
1536 hammer_wait_mem_record_ident(record, "hmmwai");
1539 static __inline void
1540 hammer_lock_ex(struct hammer_lock *lock)
1542 hammer_lock_ex_ident(lock, "hmrlck");
1546 * Indicate that a B-Tree node is being modified.
1548 static __inline void
1549 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1551 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1552 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1555 static __inline void
1556 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1558 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1559 hammer_modify_buffer(trans, node->buffer,
1560 node->ondisk, sizeof(*node->ondisk));
1563 static __inline void
1564 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1565 void *base, int len)
1567 hammer_crc_t *crcptr;
1569 KKASSERT((char *)base >= (char *)node->ondisk &&
1570 (char *)base + len <=
1571 (char *)node->ondisk + sizeof(*node->ondisk));
1572 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1573 hammer_modify_buffer(trans, node->buffer, base, len);
1574 crcptr = &node->ondisk->crc;
1575 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1576 --node->buffer->io.modify_refs; /* only want one ref */
1580 * Indicate that the specified modifications have been completed.
1582 * Do not try to generate the crc here, it's very expensive to do and a
1583 * sequence of insertions or deletions can result in many calls to this
1584 * function on the same node.
1586 static __inline void
1587 hammer_modify_node_done(hammer_node_t node)
1589 node->flags |= HAMMER_NODE_CRCGOOD;
1590 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1591 node->flags |= HAMMER_NODE_NEEDSCRC;
1592 node->buffer->io.gencrc = 1;
1593 hammer_ref_node(node);
1595 hammer_modify_buffer_done(node->buffer);
1598 #define hammer_modify_volume_field(trans, vol, field) \
1599 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1600 sizeof((vol)->ondisk->field))
1602 #define hammer_modify_node_field(trans, node, field) \
1603 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1604 sizeof((node)->ondisk->field))
1607 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1608 * created directories for HAMMER version 2 or greater and causes
1609 * directory entries to be placed the inode localization zone in
1610 * the B-Tree instead of the misc zone.
1612 * This greatly improves localization between directory entries and
1615 static __inline u_int32_t
1616 hammer_dir_localization(hammer_inode_t dip)
1618 if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1619 return(HAMMER_LOCALIZE_INODE);
1621 return(HAMMER_LOCALIZE_MISC);