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
35 #ifndef VFS_HAMMER_HAMMER_H_
36 #define VFS_HAMMER_HAMMER_H_
39 * This header file contains structures used internally by the HAMMERFS
40 * implementation. See hammer_disk.h for on-disk structures.
43 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/vnode.h>
55 #include <sys/dirent.h>
57 #include <sys/fcntl.h>
58 #include <sys/lockf.h>
60 #include <sys/event.h>
62 #include <sys/queue.h>
64 #include <sys/limits.h>
65 #include <sys/sysctl.h>
66 #include <vm/swap_pager.h>
67 #include <vm/vm_extern.h>
69 #include "hammer_disk.h"
70 #include "hammer_mount.h"
71 #include "hammer_ioctl.h"
72 #include "hammer_crc.h"
74 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
76 MALLOC_DECLARE(M_HAMMER);
81 #if !defined(KTR_HAMMER)
82 #define KTR_HAMMER KTR_ALL
84 /* KTR_INFO_MASTER_EXTERN(hammer); */
95 struct hammer_reserve;
99 * Key structure used for custom RB tree inode lookups. This prototypes
100 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
102 typedef struct hammer_inode_info {
103 int64_t obj_id; /* (key) object identifier */
104 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
105 uint32_t obj_localization; /* (key) pseudo-fs id for upper 16 bits */
107 hammer_btree_leaf_elm_t leaf;
109 } *hammer_inode_info_t;
111 typedef enum hammer_transaction_type {
115 } hammer_transaction_type_t;
118 * HAMMER Transaction tracking
120 typedef struct hammer_transaction {
121 hammer_transaction_type_t type;
122 struct hammer_mount *hmp;
128 struct hammer_volume *rootvol;
129 } *hammer_transaction_t;
131 #define HAMMER_TRANSF_NEWINODE 0x0001
132 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */
138 volatile u_int refs; /* active references */
139 volatile u_int lockval; /* lock count and control bits */
140 struct thread *lowner; /* owner if exclusively held */
141 struct thread *rowner; /* owner if exclusively held */
144 #define HAMMER_REFS_LOCKED 0x40000000 /* transition check */
145 #define HAMMER_REFS_WANTED 0x20000000 /* transition check */
146 #define HAMMER_REFS_CHECK 0x10000000 /* transition check */
148 #define HAMMER_REFS_FLAGS (HAMMER_REFS_LOCKED | \
149 HAMMER_REFS_WANTED | \
152 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000
153 #define HAMMER_LOCKF_WANTED 0x20000000
155 #define HAMMER_LIMIT_RECLAIMS 16384 /* maximum reclaims in-prog */
158 hammer_notlocked(struct hammer_lock *lock)
160 return(lock->lockval == 0);
164 hammer_islocked(struct hammer_lock *lock)
166 return(lock->lockval != 0);
170 * Returns the number of refs on the object.
173 hammer_isactive(struct hammer_lock *lock)
175 return(lock->refs & ~HAMMER_REFS_FLAGS);
179 hammer_oneref(struct hammer_lock *lock)
181 return((lock->refs & ~HAMMER_REFS_FLAGS) == 1);
185 hammer_norefs(struct hammer_lock *lock)
187 return((lock->refs & ~HAMMER_REFS_FLAGS) == 0);
191 hammer_norefsorlock(struct hammer_lock *lock)
193 return(lock->refs == 0);
197 hammer_refsorlock(struct hammer_lock *lock)
199 return(lock->refs != 0);
203 * Return if we specifically own the lock exclusively.
206 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
208 if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
209 lock->lowner == td) {
216 * Flush state, used by various structures
218 typedef enum hammer_inode_state {
222 } hammer_inode_state_t;
225 * Pseudo-filesystem extended data tracking
227 struct hammer_pseudofs_inmem;
228 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
229 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
230 hammer_pfs_rb_compare, uint32_t);
232 typedef struct hammer_pseudofs_inmem {
233 RB_ENTRY(hammer_pseudofs_inmem) rb_node;
234 struct hammer_lock lock;
235 uint32_t localization;
236 hammer_tid_t create_tid;
239 struct hammer_pseudofs_data pfsd;
240 } *hammer_pseudofs_inmem_t;
243 * Cache object ids. A fixed number of objid cache structures are
244 * created to reserve object id's for newly created files in multiples
245 * of 100,000, localized to a particular directory, and recycled as
246 * needed. This allows parallel create operations in different
247 * directories to retain fairly localized object ids which in turn
248 * improves reblocking performance and layout.
250 #define OBJID_CACHE_SIZE 2048
251 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */
252 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */
253 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1)
254 #define OBJID_CACHE_BULK_MASK64 ((uint64_t)(OBJID_CACHE_BULK - 1))
256 typedef struct hammer_objid_cache {
257 TAILQ_ENTRY(hammer_objid_cache) entry;
258 struct hammer_inode *dip;
259 hammer_tid_t base_tid;
263 } *hammer_objid_cache_t;
266 * Associate an inode with a B-Tree node to cache search start positions
268 typedef struct hammer_node_cache {
269 TAILQ_ENTRY(hammer_node_cache) entry;
270 struct hammer_node *node;
271 struct hammer_inode *ip;
272 } *hammer_node_cache_t;
274 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
277 * Structure used to organize flush groups. Flush groups must be
278 * organized into chunks in order to avoid blowing out the UNDO FIFO.
279 * Without this a 'sync' could end up flushing 50,000 inodes in a single
282 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
283 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
284 hammer_ino_rb_compare);
286 typedef struct hammer_flush_group {
287 TAILQ_ENTRY(hammer_flush_group) flush_entry;
288 struct hammer_fls_rb_tree flush_tree;
289 int seq; /* our seq no */
290 int total_count; /* record load */
291 int running; /* group is running */
294 } *hammer_flush_group_t;
296 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
299 * Structure used to represent an inode in-memory.
301 * The record and data associated with an inode may be out of sync with
302 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
305 * An inode may also hold a cache of unsynchronized records, used for
306 * database and directories only. Unsynchronized regular file data is
307 * stored in the buffer cache.
309 * NOTE: A file which is created and destroyed within the initial
310 * synchronization period can wind up not doing any disk I/O at all.
312 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
314 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
315 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
316 hammer_ino_rb_compare, hammer_inode_info_t);
318 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
319 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
320 hammer_redo_rb_compare, hammer_off_t);
322 struct hammer_record;
323 RB_HEAD(hammer_rec_rb_tree, hammer_record);
324 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
325 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
327 TAILQ_HEAD(hammer_record_list, hammer_record);
328 TAILQ_HEAD(hammer_node_list, hammer_node);
330 typedef struct hammer_inode {
331 RB_ENTRY(hammer_inode) rb_node;
332 hammer_inode_state_t flush_state;
333 hammer_flush_group_t flush_group;
334 RB_ENTRY(hammer_inode) rb_flsnode; /* when on flush list */
335 RB_ENTRY(hammer_inode) rb_redonode; /* when INODE_RDIRTY is set */
336 struct hammer_record_list target_list; /* target of dependant recs */
337 int64_t obj_id; /* (key) object identifier */
338 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
339 uint32_t obj_localization; /* (key) pseudo-fs id for upper 16 bits */
340 struct hammer_mount *hmp;
341 hammer_objid_cache_t objid_cache;
343 int error; /* flush error */
344 int cursor_ip_refs; /* sanity */
346 int cursor_exclreq_count;
350 hammer_pseudofs_inmem_t pfsm;
351 struct lockf advlock;
352 struct hammer_lock lock; /* sync copy interlock */
354 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
355 struct hammer_inode_data ino_data; /* in-memory cache */
356 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
360 * search initiate cache
361 * cache[0] - this inode
362 * cache[1] - related data, the content depends on situations
363 * cache[2] - for dip to cache ip to shortcut B-Tree search
364 * cache[3] - related data copied from dip to a new ip's cache[1]
366 struct hammer_node_cache cache[4];
369 * When a demark is created to synchronize an inode to
370 * disk, certain fields are copied so the front-end VOPs
371 * can continue to run in parallel with the synchronization
372 * occuring in the background.
374 int sync_flags; /* to-sync flags cache */
375 off_t sync_trunc_off; /* to-sync truncation */
376 off_t save_trunc_off; /* write optimization */
377 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
378 struct hammer_inode_data sync_ino_data; /* to-sync cache */
382 * Track the earliest offset in the UNDO/REDO FIFO containing
383 * REDO records. This is staged to the backend during flush
384 * sequences. While the inode is staged redo_fifo_next is used
385 * to track the earliest offset for rotation into redo_fifo_start
386 * on completion of the flush.
388 hammer_off_t redo_fifo_start;
389 hammer_off_t redo_fifo_next;
392 #define VTOI(vp) ((hammer_inode_t)(vp)->v_data)
395 * NOTE: DDIRTY does not include atime or mtime and does not include
396 * write-append size changes. SDIRTY handles write-append size
399 * REDO indicates that REDO logging is active, creating a definitive
400 * stream of REDO records in the UNDO/REDO log for writes and
401 * truncations, including boundary records when/if REDO is turned off.
402 * REDO is typically enabled by fsync() and turned off if excessive
403 * writes without an fsync() occurs.
405 * RDIRTY indicates that REDO records were laid down in the UNDO/REDO
406 * FIFO (even if REDO is turned off some might still be active) and
407 * still being tracked for this inode. See hammer_redo.c
409 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
410 /* (not including atime/mtime) */
411 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
412 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */
413 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
414 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
415 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
416 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
417 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
418 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
419 #define HAMMER_INODE_RECSW 0x0400 /* waiting on data record flush */
420 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
421 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
422 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */
423 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
424 #define HAMMER_INODE_FLUSHW 0x8000 /* someone waiting for flush */
426 #define HAMMER_INODE_TRUNCATED 0x00010000
427 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
428 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
429 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
430 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
431 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
432 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */
433 #define HAMMER_INODE_SDIRTY 0x01000000 /* in-memory ino_data.size is dirty*/
434 #define HAMMER_INODE_REDO 0x02000000 /* REDO logging active */
435 #define HAMMER_INODE_RDIRTY 0x04000000 /* REDO records active in fifo */
436 #define HAMMER_INODE_SLAVEFLUSH 0x08000000 /* being flushed by slave */
438 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY| \
439 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
440 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
441 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
443 #define HAMMER_INODE_MODMASK_NOXDIRTY \
444 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
446 #define HAMMER_INODE_MODMASK_NOREDO \
447 (HAMMER_INODE_DDIRTY| \
448 HAMMER_INODE_XDIRTY| \
449 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
451 #define HAMMER_FLUSH_SIGNAL 0x0001
452 #define HAMMER_FLUSH_RECURSION 0x0002
455 * Used by the inode reclaim code to pipeline reclaims and avoid
456 * blowing out kernel memory or letting the flusher get too far
457 * behind. The reclaim wakes up when count reaches 0 or the
460 struct hammer_reclaim {
461 TAILQ_ENTRY(hammer_reclaim) entry;
466 * Track who is creating the greatest burden on the
469 struct hammer_inostats {
470 pid_t pid; /* track user process */
471 int ltick; /* last tick */
472 int count; /* count (degenerates) */
475 #define HAMMER_INOSTATS_HSIZE 32
476 #define HAMMER_INOSTATS_HMASK (HAMMER_INOSTATS_HSIZE - 1)
479 * Structure used to represent an unsynchronized record in-memory. These
480 * records typically represent directory entries. Only non-historical
481 * records are kept in-memory.
483 * Records are organized as a per-inode RB-Tree. If the inode is not
484 * on disk then neither are any records and the in-memory record tree
485 * represents the entire contents of the inode. If the inode is on disk
486 * then the on-disk B-Tree is scanned in parallel with the in-memory
487 * RB-Tree to synthesize the current state of the file.
489 * Records are also used to enforce the ordering of directory create/delete
490 * operations. A new inode will not be flushed to disk unless its related
491 * directory entry is also being flushed at the same time. A directory entry
492 * will not be removed unless its related inode is also being removed at the
495 typedef enum hammer_record_type {
496 HAMMER_MEM_RECORD_GENERAL, /* misc record */
497 HAMMER_MEM_RECORD_INODE, /* inode record */
498 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
499 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
500 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
501 } hammer_record_type_t;
503 typedef struct hammer_record {
504 RB_ENTRY(hammer_record) rb_node;
505 TAILQ_ENTRY(hammer_record) target_entry;
506 hammer_inode_state_t flush_state;
507 hammer_flush_group_t flush_group;
508 hammer_record_type_t type;
509 struct hammer_lock lock;
510 struct hammer_reserve *resv;
512 hammer_inode_t target_ip;
513 struct hammer_btree_leaf_elm leaf;
514 hammer_data_ondisk_t data;
517 hammer_off_t zone2_offset; /* direct-write only */
521 * Record flags. Note that FE can only be set by the frontend if the
522 * record has not been interlocked by the backend w/ BE.
524 #define HAMMER_RECF_ALLOCDATA 0x0001
525 #define HAMMER_RECF_ONRBTREE 0x0002
526 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
527 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
528 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */
529 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
530 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
531 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
532 #define HAMMER_RECF_REDO 0x1000 /* REDO was laid down */
535 * These flags must be separate to deal with SMP races
537 #define HAMMER_RECG_DIRECT_IO 0x0001 /* related direct I/O running*/
538 #define HAMMER_RECG_DIRECT_WAIT 0x0002 /* related direct I/O running*/
539 #define HAMMER_RECG_DIRECT_INVAL 0x0004 /* buffer alias invalidation */
541 * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
543 #define HAMMER_CREATE_MODE_UMIRROR 0x0001
544 #define HAMMER_CREATE_MODE_SYS 0x0002
546 #define HAMMER_DELETE_ADJUST 0x0001
547 #define HAMMER_DELETE_DESTROY 0x0002
550 * In-memory structures representing on-disk structures.
552 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
553 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
554 RB_HEAD(hammer_nod_rb_tree, hammer_node);
555 RB_HEAD(hammer_und_rb_tree, hammer_undo);
556 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
557 RB_HEAD(hammer_mod_rb_tree, hammer_io);
559 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
560 hammer_vol_rb_compare, int32_t);
561 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
562 hammer_buf_rb_compare, hammer_off_t);
563 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
564 hammer_nod_rb_compare, hammer_off_t);
565 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
566 hammer_und_rb_compare, hammer_off_t);
567 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
568 hammer_res_rb_compare, hammer_off_t);
569 RB_PROTOTYPE2(hammer_mod_rb_tree, hammer_io, rb_node,
570 hammer_mod_rb_compare, hammer_off_t);
573 * IO management - embedded at the head of various in-memory structures
575 * VOLUME - hammer_volume containing meta-data
576 * META_BUFFER - hammer_buffer containing meta-data
577 * UNDO_BUFFER - hammer_buffer containing undo-data
578 * DATA_BUFFER - hammer_buffer containing pure-data
579 * DUMMY - hammer_buffer not containing valid data
581 * Dirty volume headers and dirty meta-data buffers are locked until the
582 * flusher can sequence them out. Dirty pure-data buffers can be written.
583 * Clean buffers can be passively released.
585 typedef enum hammer_io_type {
586 HAMMER_IOTYPE_VOLUME,
587 HAMMER_IOTYPE_META_BUFFER,
588 HAMMER_IOTYPE_UNDO_BUFFER,
589 HAMMER_IOTYPE_DATA_BUFFER,
593 typedef struct hammer_io {
594 struct hammer_lock lock;
595 hammer_io_type_t type;
596 struct hammer_mount *hmp;
597 struct hammer_volume *volume;
598 RB_ENTRY(hammer_io) rb_node; /* if modified */
599 TAILQ_ENTRY(hammer_io) iorun_entry; /* iorun_list */
600 struct hammer_mod_rb_tree *mod_root;
602 int64_t offset; /* volume offset */
603 int bytes; /* buffer cache buffer size */
607 * These can be modified at any time by the backend while holding
608 * io_token, due to bio_done and hammer_io_complete() callbacks.
610 u_int running : 1; /* bp write IO in progress */
611 u_int waiting : 1; /* someone is waiting on us */
612 u_int ioerror : 1; /* abort on io-error */
616 * These can only be modified by the frontend while holding
617 * fs_token, or by the backend while holding the io interlocked
618 * with no references (which will block the frontend when it
619 * tries to reference it).
621 * WARNING! SMP RACES will create havoc if the callbacks ever tried
622 * to modify any of these outside the above restrictions.
624 u_int modified : 1; /* bp's data was modified */
625 u_int released : 1; /* bp released (w/ B_LOCKED set) */
626 u_int waitdep : 1; /* flush waits for dependancies */
627 u_int recovered : 1; /* has recovery ref */
628 u_int waitmod : 1; /* waiting for modify_refs */
629 u_int reclaim : 1; /* reclaim requested */
630 u_int gencrc : 1; /* crc needs to be generated */
634 #define HAMMER_CLUSTER_SIZE (64 * 1024)
635 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
636 #undef HAMMER_CLUSTER_SIZE
637 #define HAMMER_CLUSTER_SIZE MAXBSIZE
641 * In-memory volume representing on-disk buffer
643 typedef struct hammer_volume {
644 struct hammer_io io; /* must be at offset 0 */
645 RB_ENTRY(hammer_volume) rb_node;
646 hammer_volume_ondisk_t ondisk;
648 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
654 #define HAMMER_ITOV(iop) ((hammer_volume_t)(iop))
657 * In-memory buffer representing an on-disk buffer.
659 typedef struct hammer_buffer {
660 struct hammer_io io; /* must be at offset 0 */
661 RB_ENTRY(hammer_buffer) rb_node;
663 hammer_off_t zoneX_offset;
664 hammer_off_t zone2_offset;
665 struct hammer_reserve *resv;
666 struct hammer_node_list node_list;
669 #define HAMMER_ITOB(iop) ((hammer_buffer_t)(iop))
672 * In-memory B-Tree node, representing an on-disk B-Tree node.
674 * This is a hang-on structure which is backed by a hammer_buffer,
675 * and used for fine-grained locking of B-Tree nodes in order to
676 * properly control lock ordering.
678 typedef struct hammer_node {
679 struct hammer_lock lock; /* node-by-node lock */
680 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
681 RB_ENTRY(hammer_node) rb_node; /* per-mount linkage */
682 hammer_off_t node_offset; /* full offset spec */
683 struct hammer_mount *hmp;
684 hammer_buffer_t buffer; /* backing buffer */
685 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
686 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */
687 struct hammer_node_cache_list cache_list; /* passive caches */
690 int cursor_exclreq_count;
694 #define HAMMER_NODE_DELETED 0x0001
695 #define HAMMER_NODE_FLUSH 0x0002
696 #define HAMMER_NODE_CRCGOOD 0x0004
697 #define HAMMER_NODE_NEEDSCRC 0x0008
698 #define HAMMER_NODE_NEEDSMIRROR 0x0010
699 #define HAMMER_NODE_CRCBAD 0x0020
700 #define HAMMER_NODE_NONLINEAR 0x0040 /* linear heuristic */
702 #define HAMMER_NODE_CRCANY (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
705 * List of locked nodes. This structure is used to lock potentially large
706 * numbers of nodes as an aid for complex B-Tree operations.
708 struct hammer_node_lock;
709 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
711 typedef struct hammer_node_lock {
712 TAILQ_ENTRY(hammer_node_lock) entry;
713 struct hammer_node_lock_list list;
714 struct hammer_node_lock *parent;
716 hammer_node_ondisk_t copy; /* copy of on-disk data */
717 int index; /* index of this node in parent */
718 int count; /* count children */
720 } *hammer_node_lock_t;
722 #define HAMMER_NODE_LOCK_UPDATED 0x0001
723 #define HAMMER_NODE_LOCK_LCACHE 0x0002
726 * The reserve structure prevents the blockmap from allocating
727 * out of a reserved big-block. Such reservations are used by
728 * the direct-write mechanism.
730 * The structure is also used to hold off on reallocations of
731 * big-blocks from the freemap until flush dependancies have
734 typedef struct hammer_reserve {
735 RB_ENTRY(hammer_reserve) rb_node;
736 TAILQ_ENTRY(hammer_reserve) delay_entry;
742 hammer_off_t zone_offset;
745 #define HAMMER_RESF_ONDELAY 0x0001
746 #define HAMMER_RESF_LAYER2FREE 0x0002
748 #include "hammer_cursor.h"
751 * The undo structure tracks recent undos to avoid laying down duplicate
752 * undos within a flush group, saving us a significant amount of overhead.
754 * This is strictly a heuristic.
756 #define HAMMER_MAX_UNDOS 1024
757 #define HAMMER_MAX_FLUSHERS 4
759 typedef struct hammer_undo {
760 RB_ENTRY(hammer_undo) rb_node;
761 TAILQ_ENTRY(hammer_undo) lru_entry;
766 struct hammer_flusher_info;
767 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
769 struct hammer_flusher {
770 int signal; /* flusher thread sequencer */
771 int done; /* last completed flush group */
772 int next; /* next unallocated flg seqno */
773 int group_lock; /* lock sequencing of the next flush */
774 int exiting; /* request master exit */
775 thread_t td; /* master flusher thread */
776 hammer_tid_t tid; /* last flushed transaction id */
777 int finalize_want; /* serialize finalization */
778 struct hammer_lock finalize_lock; /* serialize finalization */
779 struct hammer_transaction trans; /* shared transaction */
780 struct hammer_flusher_info_list run_list;
781 struct hammer_flusher_info_list ready_list;
784 #define HAMMER_FLUSH_UNDOS_RELAXED 0
785 #define HAMMER_FLUSH_UNDOS_FORCED 1
786 #define HAMMER_FLUSH_UNDOS_AUTO 2
788 * Internal hammer mount data structure
790 typedef struct hammer_mount {
792 struct hammer_ino_rb_tree rb_inos_root;
793 struct hammer_redo_rb_tree rb_redo_root;
794 struct hammer_vol_rb_tree rb_vols_root;
795 struct hammer_nod_rb_tree rb_nods_root;
796 struct hammer_und_rb_tree rb_undo_root;
797 struct hammer_res_rb_tree rb_resv_root;
798 struct hammer_buf_rb_tree rb_bufs_root;
799 struct hammer_pfs_rb_tree rb_pfsm_root;
800 hammer_volume_t rootvol;
801 struct hammer_base_elm root_btree_beg;
802 struct hammer_base_elm root_btree_end;
804 struct malloc_type *m_misc;
805 struct malloc_type *m_inodes;
807 int flags; /* HAMMER_MOUNT_xxx flags */
811 int master_id; /* default 0, no-mirror -1, otherwise 1-15 */
812 int version; /* hammer filesystem version to use */
813 int rsv_inodes; /* reserved space due to dirty inodes */
814 int64_t rsv_databytes; /* reserved space due to record data */
815 int rsv_recs; /* reserved space due to dirty records */
816 int rsv_fromdelay; /* big-blocks reserved due to flush delay */
817 int undo_rec_limit; /* based on size of undo area */
819 int volume_to_remove; /* volume that is currently being removed */
821 int count_inodes; /* total number of inodes */
822 int count_iqueued; /* inodes queued to flusher */
823 int count_reclaims; /* inodes pending reclaim by flusher */
825 struct hammer_flusher flusher;
827 u_int check_interrupt;
830 struct hammer_mod_rb_tree volu_root; /* dirty undo buffers */
831 struct hammer_mod_rb_tree undo_root; /* dirty undo buffers */
832 struct hammer_mod_rb_tree data_root; /* dirty data buffers */
833 struct hammer_mod_rb_tree meta_root; /* dirty meta bufs */
834 struct hammer_mod_rb_tree lose_root; /* loose buffers */
835 long locked_dirty_space; /* meta/volu count */
836 long io_running_space; /* io_token */
837 int objid_cache_count;
838 int error; /* critical I/O error */
839 struct krate krate; /* rate limited kprintf */
840 struct krate kdiag; /* rate limited kprintf */
841 hammer_tid_t asof; /* snapshot mount */
842 hammer_tid_t next_tid;
843 hammer_tid_t flush_tid1; /* flusher tid sequencing */
844 hammer_tid_t flush_tid2; /* flusher tid sequencing */
845 int64_t copy_stat_freebigblocks; /* number of free big-blocks */
846 uint32_t undo_seqno; /* UNDO/REDO FIFO seqno */
847 uint32_t recover_stage2_seqno; /* REDO recovery seqno */
848 hammer_off_t recover_stage2_offset; /* REDO recovery offset */
850 struct netexport export;
851 struct hammer_lock sync_lock;
852 struct hammer_lock undo_lock;
853 struct hammer_lock blkmap_lock;
854 struct hammer_lock snapshot_lock;
855 struct hammer_lock volume_lock;
856 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
857 struct hammer_undo undos[HAMMER_MAX_UNDOS];
859 TAILQ_HEAD(, hammer_undo) undo_lru_list;
860 TAILQ_HEAD(, hammer_reserve) delay_list;
861 struct hammer_flush_group_list flush_group_list;
862 hammer_flush_group_t fill_flush_group;
863 hammer_flush_group_t next_flush_group;
864 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
865 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
866 TAILQ_HEAD(, hammer_io) iorun_list;
868 struct lwkt_token fs_token; /* high level */
869 struct lwkt_token io_token; /* low level (IO callback) */
871 struct hammer_inostats inostats[HAMMER_INOSTATS_HSIZE];
872 uint64_t volume_map[4]; /* 256 bits bitfield */
875 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001
876 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002
877 #define HAMMER_MOUNT_REDO_SYNC 0x0004
878 #define HAMMER_MOUNT_REDO_RECOVERY_REQ 0x0008
879 #define HAMMER_MOUNT_REDO_RECOVERY_RUN 0x0010
881 #define HAMMER_VOLUME_NUMBER_FOREACH(hmp, n) \
882 for (n = 0; n < HAMMER_MAX_VOLUMES; n++) \
883 if (hammer_volume_number_test(hmp, n))
886 * Minium buffer cache bufs required to rebalance the B-Tree.
887 * This is because we must hold the children and the children's children
888 * locked. Even this might not be enough if things are horribly out
891 #define HAMMER_REBALANCE_MIN_BUFS \
892 (HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
894 #endif /* _KERNEL || _KERNEL_STRUCTURES */
898 * checkspace slop (8MB chunks), higher numbers are more conservative.
900 #define HAMMER_CHKSPC_REBLOCK 25
901 #define HAMMER_CHKSPC_MIRROR 20
902 #define HAMMER_CHKSPC_WRITE 20
903 #define HAMMER_CHKSPC_CREATE 20
904 #define HAMMER_CHKSPC_REMOVE 10
905 #define HAMMER_CHKSPC_EMERGENCY 0
907 extern struct vop_ops hammer_vnode_vops;
908 extern struct vop_ops hammer_spec_vops;
909 extern struct vop_ops hammer_fifo_vops;
911 extern int hammer_debug_io;
912 extern int hammer_debug_general;
913 extern int hammer_debug_inode;
914 extern int hammer_debug_locks;
915 extern int hammer_debug_btree;
916 extern int hammer_debug_tid;
917 extern int hammer_debug_recover;
918 extern int hammer_debug_critical;
919 extern int hammer_cluster_enable;
920 extern int hammer_tdmux_ticks;
921 extern int hammer_count_fsyncs;
922 extern int hammer_count_inodes;
923 extern int hammer_count_iqueued;
924 extern int hammer_count_reclaims;
925 extern int hammer_count_records;
926 extern int hammer_count_record_datas;
927 extern int hammer_count_volumes;
928 extern int hammer_count_buffers;
929 extern int hammer_count_nodes;
930 extern int64_t hammer_stats_btree_lookups;
931 extern int64_t hammer_stats_btree_searches;
932 extern int64_t hammer_stats_btree_inserts;
933 extern int64_t hammer_stats_btree_deletes;
934 extern int64_t hammer_stats_btree_elements;
935 extern int64_t hammer_stats_btree_splits;
936 extern int64_t hammer_stats_btree_iterations;
937 extern int64_t hammer_stats_btree_root_iterations;
938 extern int64_t hammer_stats_record_iterations;
939 extern int64_t hammer_stats_file_read;
940 extern int64_t hammer_stats_file_write;
941 extern int64_t hammer_stats_disk_read;
942 extern int64_t hammer_stats_disk_write;
943 extern int64_t hammer_stats_inode_flushes;
944 extern int64_t hammer_stats_commits;
945 extern int64_t hammer_stats_undo;
946 extern int64_t hammer_stats_redo;
947 extern long hammer_count_dirtybufspace;
948 extern int hammer_count_refedbufs;
949 extern int hammer_count_reservations;
950 extern long hammer_count_io_running_read;
951 extern long hammer_count_io_running_write;
952 extern int hammer_count_io_locked;
953 extern long hammer_limit_dirtybufspace;
954 extern int hammer_limit_recs;
955 extern int hammer_limit_inode_recs;
956 extern int hammer_limit_reclaims;
957 extern int hammer_live_dedup_cache_size;
958 extern int hammer_limit_redo;
959 extern int hammer_verify_zone;
960 extern int hammer_verify_data;
961 extern int hammer_double_buffer;
962 extern int hammer_btree_full_undo;
963 extern int hammer_yield_check;
964 extern int hammer_fsync_mode;
965 extern int hammer_autoflush;
966 extern int64_t hammer_contention_count;
968 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
969 int error, const char *msg);
970 int hammer_vop_inactive(struct vop_inactive_args *);
971 int hammer_vop_reclaim(struct vop_reclaim_args *);
972 int hammer_get_vnode(hammer_inode_t ip, struct vnode **vpp);
973 hammer_inode_t hammer_get_inode(hammer_transaction_t trans,
974 hammer_inode_t dip, int64_t obj_id,
975 hammer_tid_t asof, uint32_t localization,
976 int flags, int *errorp);
977 hammer_inode_t hammer_get_dummy_inode(hammer_transaction_t trans,
978 hammer_inode_t dip, int64_t obj_id,
979 hammer_tid_t asof, uint32_t localization,
980 int flags, int *errorp);
981 hammer_inode_t hammer_find_inode(hammer_transaction_t trans,
982 int64_t obj_id, hammer_tid_t asof,
983 uint32_t localization);
984 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
985 hammer_inode_info_t iinfo,
986 int (*callback)(hammer_inode_t ip, void *data),
988 void hammer_put_inode(hammer_inode_t ip);
989 void hammer_put_inode_ref(hammer_inode_t ip);
990 void hammer_inode_waitreclaims(hammer_transaction_t trans);
991 void hammer_inode_dirty(hammer_inode_t ip);
993 int hammer_unload_volume(hammer_volume_t volume, void *data);
994 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
996 int hammer_unload_buffer(hammer_buffer_t buffer, void *data);
997 int hammer_install_volume(hammer_mount_t hmp, const char *volname,
998 struct vnode *devvp, void *data);
999 int hammer_mountcheck_volumes(hammer_mount_t hmp);
1000 int hammer_get_installed_volumes(hammer_mount_t hmp);
1002 int hammer_mem_add(hammer_record_t record);
1003 int hammer_ip_lookup(hammer_cursor_t cursor);
1004 int hammer_ip_first(hammer_cursor_t cursor);
1005 int hammer_ip_next(hammer_cursor_t cursor);
1006 int hammer_ip_resolve_data(hammer_cursor_t cursor);
1007 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1009 int hammer_create_at_cursor(hammer_cursor_t cursor,
1010 hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1011 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1012 hammer_tid_t delete_tid, uint32_t delete_ts,
1013 int track, int64_t *stat_bytes);
1014 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
1016 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1017 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1020 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1021 void hammer_flush_record_done(hammer_record_t record, int error);
1022 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1023 void hammer_rel_mem_record(hammer_record_t record);
1025 int hammer_cursor_up(hammer_cursor_t cursor);
1026 int hammer_cursor_up_locked(hammer_cursor_t cursor);
1027 int hammer_cursor_down(hammer_cursor_t cursor);
1028 int hammer_cursor_upgrade(hammer_cursor_t cursor);
1029 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1030 void hammer_cursor_downgrade(hammer_cursor_t cursor);
1031 int hammer_cursor_upgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1032 void hammer_cursor_downgrade2(hammer_cursor_t c1, hammer_cursor_t c2);
1033 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1035 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1036 int hammer_lock_ex_try(struct hammer_lock *lock);
1037 void hammer_lock_sh(struct hammer_lock *lock);
1038 int hammer_lock_sh_try(struct hammer_lock *lock);
1039 int hammer_lock_upgrade(struct hammer_lock *lock, int shcount);
1040 void hammer_lock_downgrade(struct hammer_lock *lock, int shcount);
1041 int hammer_lock_status(struct hammer_lock *lock);
1042 void hammer_unlock(struct hammer_lock *lock);
1043 void hammer_ref(struct hammer_lock *lock);
1044 int hammer_ref_interlock(struct hammer_lock *lock);
1045 int hammer_ref_interlock_true(struct hammer_lock *lock);
1046 void hammer_ref_interlock_done(struct hammer_lock *lock);
1047 void hammer_rel(struct hammer_lock *lock);
1048 int hammer_rel_interlock(struct hammer_lock *lock, int locked);
1049 void hammer_rel_interlock_done(struct hammer_lock *lock, int orig_locked);
1050 int hammer_get_interlock(struct hammer_lock *lock);
1051 int hammer_try_interlock_norefs(struct hammer_lock *lock);
1052 void hammer_put_interlock(struct hammer_lock *lock, int error);
1054 void hammer_sync_lock_ex(hammer_transaction_t trans);
1055 void hammer_sync_lock_sh(hammer_transaction_t trans);
1056 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
1057 void hammer_sync_unlock(hammer_transaction_t trans);
1059 uint32_t hammer_to_unix_xid(hammer_uuid_t *uuid);
1060 void hammer_guid_to_uuid(hammer_uuid_t *uuid, uint32_t guid);
1061 void hammer_time_to_timespec(uint64_t xtime, struct timespec *ts);
1062 uint64_t hammer_timespec_to_time(struct timespec *ts);
1063 int hammer_str_to_tid(const char *str, int *ispfsp,
1064 hammer_tid_t *tidp, uint32_t *localizationp);
1065 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1067 void hammer_clear_objid(hammer_inode_t dip);
1068 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1070 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1072 void hammer_clear_undo_history(hammer_mount_t hmp);
1073 enum vtype hammer_get_vnode_type(uint8_t obj_type);
1074 int hammer_get_dtype(uint8_t obj_type);
1075 uint8_t hammer_get_obj_type(enum vtype vtype);
1076 int64_t hammer_direntry_namekey(hammer_inode_t dip, const void *name, int len,
1077 uint32_t *max_iterationsp);
1078 int hammer_nohistory(hammer_inode_t ip);
1080 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1081 hammer_node_cache_t cache, hammer_inode_t ip);
1082 void hammer_normalize_cursor(hammer_cursor_t cursor);
1083 void hammer_done_cursor(hammer_cursor_t cursor);
1084 int hammer_recover_cursor(hammer_cursor_t cursor);
1085 void hammer_unlock_cursor(hammer_cursor_t cursor);
1086 int hammer_lock_cursor(hammer_cursor_t cursor);
1087 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
1088 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1090 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1091 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1093 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1095 void hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1096 hammer_node_t onode, int oindex,
1097 hammer_node_t nnode, int nindex);
1098 void hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1099 hammer_node_t nparent, int nindex);
1100 void hammer_cursor_inserted_element(hammer_node_t node, int index);
1101 void hammer_cursor_deleted_element(hammer_node_t node, int index);
1102 void hammer_cursor_invalidate_cache(hammer_cursor_t cursor);
1104 int hammer_btree_lookup(hammer_cursor_t cursor);
1105 int hammer_btree_first(hammer_cursor_t cursor);
1106 int hammer_btree_last(hammer_cursor_t cursor);
1107 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
1108 int hammer_btree_iterate(hammer_cursor_t cursor);
1109 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1110 int hammer_btree_insert(hammer_cursor_t cursor,
1111 hammer_btree_leaf_elm_t elm, int *doprop);
1112 int hammer_btree_delete(hammer_cursor_t cursor, int *ndelete);
1113 void hammer_btree_do_propagation(hammer_cursor_t cursor,
1114 hammer_btree_leaf_elm_t leaf);
1115 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1116 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1117 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1118 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1120 int btree_set_parent_of_child(hammer_transaction_t trans,
1122 hammer_btree_elm_t elm);
1123 void hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1124 void hammer_btree_lcache_init(hammer_mount_t hmp, hammer_node_lock_t lcache,
1126 void hammer_btree_lcache_free(hammer_mount_t hmp, hammer_node_lock_t lcache);
1127 int hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1128 hammer_node_lock_t parent,
1129 hammer_node_lock_t lcache);
1130 void hammer_btree_lock_copy(hammer_cursor_t cursor,
1131 hammer_node_lock_t parent);
1132 int hammer_btree_sync_copy(hammer_cursor_t cursor,
1133 hammer_node_lock_t parent);
1134 void hammer_btree_unlock_children(hammer_mount_t hmp,
1135 hammer_node_lock_t parent,
1136 hammer_node_lock_t lcache);
1137 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1138 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1139 hammer_node_t node, int *parent_indexp,
1140 int *errorp, int try_exclusive);
1142 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1143 void hammer_print_btree_elm(hammer_btree_elm_t elm);
1145 void *hammer_bread(hammer_mount_t hmp, hammer_off_t off,
1146 int *errorp, hammer_buffer_t *bufferp);
1147 void *hammer_bnew(hammer_mount_t hmp, hammer_off_t off,
1148 int *errorp, hammer_buffer_t *bufferp);
1149 void *hammer_bread_ext(hammer_mount_t hmp, hammer_off_t off, int bytes,
1150 int *errorp, hammer_buffer_t *bufferp);
1151 void *hammer_bnew_ext(hammer_mount_t hmp, hammer_off_t off, int bytes,
1152 int *errorp, hammer_buffer_t *bufferp);
1154 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1156 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1157 int32_t vol_no, int *errorp);
1158 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1159 int bytes, int isnew, int *errorp);
1160 void hammer_sync_buffers(hammer_mount_t hmp,
1161 hammer_off_t base_offset, int bytes);
1162 int hammer_del_buffers(hammer_mount_t hmp,
1163 hammer_off_t base_offset,
1164 hammer_off_t zone2_offset, int bytes,
1165 int report_conflicts);
1167 int hammer_ref_volume(hammer_volume_t volume);
1168 int hammer_ref_buffer(hammer_buffer_t buffer);
1169 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1171 void hammer_rel_volume(hammer_volume_t volume, int locked);
1172 void hammer_rel_buffer(hammer_buffer_t buffer, int locked);
1174 int hammer_vfs_export(struct mount *mp, int op,
1175 const struct export_args *export);
1176 hammer_node_t hammer_get_node(hammer_transaction_t trans,
1177 hammer_off_t node_offset, int isnew, int *errorp);
1178 void hammer_ref_node(hammer_node_t node);
1179 hammer_node_t hammer_ref_node_safe(hammer_transaction_t trans,
1180 hammer_node_cache_t cache, int *errorp);
1181 void hammer_rel_node(hammer_node_t node);
1182 void hammer_delete_node(hammer_transaction_t trans,
1183 hammer_node_t node);
1184 void hammer_cache_node(hammer_node_cache_t cache,
1185 hammer_node_t node);
1186 void hammer_uncache_node(hammer_node_cache_t cache);
1187 void hammer_flush_node(hammer_node_t node, int locked);
1189 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1190 hammer_off_t hint, int *errorp);
1191 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1192 uint16_t rec_type, hammer_off_t *data_offsetp,
1193 hammer_buffer_t *data_bufferp,
1194 hammer_off_t hint, int *errorp);
1196 int hammer_generate_undo(hammer_transaction_t trans,
1197 hammer_off_t zone_offset, void *base, int len);
1198 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1199 hammer_off_t file_offset, uint32_t flags,
1200 void *base, int len);
1201 void hammer_generate_redo_sync(hammer_transaction_t trans);
1202 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1203 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1205 void hammer_format_undo(hammer_mount_t hmp, void *base, uint32_t seqno);
1206 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1208 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1209 hammer_off_t owner, int *errorp);
1210 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1211 hammer_off_t owner, int *errorp);
1212 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1213 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1214 int bytes, hammer_off_t hint, int *errorp);
1215 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1216 int bytes, hammer_off_t *zone_offp, int *errorp);
1217 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1218 hammer_reserve_t resv);
1219 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1220 void hammer_blockmap_free(hammer_transaction_t trans,
1221 hammer_off_t zone_offset, int bytes);
1222 int hammer_blockmap_dedup(hammer_transaction_t trans,
1223 hammer_off_t zone_offset, int bytes);
1224 int hammer_blockmap_finalize(hammer_transaction_t trans,
1225 hammer_reserve_t resv,
1226 hammer_off_t zone_offset, int bytes);
1227 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t zone_offset,
1228 int *curp, int *errorp);
1229 hammer_off_t hammer_blockmap_lookup_verify(hammer_mount_t hmp,
1230 hammer_off_t zone_offset, int *errorp);
1232 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t zone_offset,
1234 int64_t hammer_undo_used(hammer_transaction_t trans);
1235 int64_t hammer_undo_space(hammer_transaction_t trans);
1236 int64_t hammer_undo_max(hammer_mount_t hmp);
1237 int hammer_undo_reclaim(hammer_io_t io);
1239 void hammer_start_transaction(hammer_transaction_t trans,
1240 hammer_mount_t hmp);
1241 void hammer_simple_transaction(hammer_transaction_t trans,
1242 hammer_mount_t hmp);
1243 void hammer_start_transaction_fls(hammer_transaction_t trans,
1244 hammer_mount_t hmp);
1245 void hammer_done_transaction(hammer_transaction_t trans);
1246 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1248 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1249 void hammer_flush_inode(hammer_inode_t ip, int flags);
1250 void hammer_wait_inode(hammer_inode_t ip);
1252 int hammer_create_inode(hammer_transaction_t trans, struct vattr *vap,
1253 struct ucred *cred, hammer_inode_t dip,
1254 const char *name, int namelen,
1255 hammer_pseudofs_inmem_t pfsm,
1256 hammer_inode_t *ipp);
1257 void hammer_rel_inode(hammer_inode_t ip, int flush);
1258 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1259 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1260 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1261 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1263 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1264 void hammer_sync_inode_done(hammer_inode_t ip, int error);
1265 void hammer_test_inode(hammer_inode_t dip);
1266 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1267 int hammer_update_atime_quick(hammer_inode_t ip);
1269 int hammer_ip_add_direntry(hammer_transaction_t trans,
1270 hammer_inode_t dip, const char *name, int bytes,
1271 hammer_inode_t nip);
1272 int hammer_ip_del_direntry(hammer_transaction_t trans,
1273 hammer_cursor_t cursor, hammer_inode_t dip,
1275 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1276 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1277 void *data, int bytes, int *errorp);
1278 int hammer_ip_frontend_trunc(hammer_inode_t ip, off_t file_size);
1279 int hammer_ip_add_record(hammer_transaction_t trans,
1280 hammer_record_t record);
1281 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1282 int64_t ran_beg, int64_t ran_end, int truncating);
1283 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1285 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1286 int64_t offset, void *data, int bytes);
1287 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1288 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans,
1289 uint32_t localization, int *errorp);
1290 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1291 hammer_pseudofs_inmem_t pfsm, hammer_inode_t dip);
1292 int hammer_save_pseudofs(hammer_transaction_t trans,
1293 hammer_pseudofs_inmem_t pfsm);
1294 int hammer_unload_pseudofs(hammer_transaction_t trans, uint32_t localization);
1295 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1296 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1297 struct ucred *cred);
1299 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1300 hammer_io_type_t type);
1301 hammer_io_type_t hammer_zone_to_iotype(int zone);
1302 int hammer_io_read(struct vnode *devvp, hammer_io_t io, int limit);
1303 void hammer_io_advance(hammer_io_t io);
1304 int hammer_io_new(struct vnode *devvp, hammer_io_t io);
1305 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1306 struct buf *hammer_io_release(hammer_io_t io, int flush);
1307 void hammer_io_flush(hammer_io_t io, int reclaim);
1308 void hammer_io_wait(hammer_io_t io);
1309 void hammer_io_waitdep(hammer_io_t io);
1310 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1311 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1312 hammer_btree_leaf_elm_t leaf);
1313 int hammer_io_indirect_read(hammer_mount_t hmp, struct bio *bio,
1314 hammer_btree_leaf_elm_t leaf);
1315 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1316 hammer_record_t record);
1317 void hammer_io_direct_wait(hammer_record_t record);
1318 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1319 void hammer_io_write_interlock(hammer_io_t io);
1320 void hammer_io_done_interlock(hammer_io_t io);
1321 void hammer_io_clear_modify(hammer_io_t io, int inval);
1322 void hammer_io_clear_modlist(hammer_io_t io);
1323 void hammer_io_flush_sync(hammer_mount_t hmp);
1324 void hammer_io_clear_error(hammer_io_t io);
1325 void hammer_io_clear_error_noassert(hammer_io_t io);
1326 void hammer_io_notmeta(hammer_buffer_t buffer);
1327 void hammer_io_limit_backlog(hammer_mount_t hmp);
1329 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1330 void *base, int len);
1331 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1332 void *base, int len);
1333 void hammer_modify_volume_done(hammer_volume_t volume);
1334 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1336 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1337 struct hammer_ioc_reblock *reblock);
1338 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1339 struct hammer_ioc_rebalance *rebal);
1340 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1341 struct hammer_ioc_prune *prune);
1342 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1343 struct hammer_ioc_mirror_rw *mirror);
1344 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1345 struct hammer_ioc_mirror_rw *mirror);
1346 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1347 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1348 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1349 struct hammer_ioc_pseudofs_rw *pfs);
1350 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1351 struct hammer_ioc_pseudofs_rw *pfs);
1352 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1353 struct hammer_ioc_pseudofs_rw *pfs);
1354 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1355 struct hammer_ioc_pseudofs_rw *pfs);
1356 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1357 struct hammer_ioc_pseudofs_rw *pfs);
1358 int hammer_ioc_scan_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1359 struct hammer_ioc_pseudofs_rw *pfs);
1360 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1361 struct hammer_ioc_volume *ioc);
1362 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1363 struct hammer_ioc_volume *ioc);
1364 int hammer_ioc_volume_list(hammer_transaction_t trans, hammer_inode_t ip,
1365 struct hammer_ioc_volume_list *ioc);
1366 int hammer_ioc_dedup(hammer_transaction_t trans, hammer_inode_t ip,
1367 struct hammer_ioc_dedup *dedup);
1369 int hammer_signal_check(hammer_mount_t hmp);
1371 void hammer_flusher_create(hammer_mount_t hmp);
1372 void hammer_flusher_destroy(hammer_mount_t hmp);
1373 void hammer_flusher_sync(hammer_mount_t hmp);
1374 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1375 int hammer_flusher_async_one(hammer_mount_t hmp);
1376 int hammer_flusher_running(hammer_mount_t hmp);
1377 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1378 void hammer_flusher_wait_next(hammer_mount_t hmp);
1379 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1380 int hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1381 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1382 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1383 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1384 int hammer_flusher_haswork(hammer_mount_t hmp);
1385 int hammer_flush_dirty(hammer_mount_t hmp, int max_count);
1386 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1388 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1389 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1390 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1391 hammer_volume_t root_volume, int final);
1393 udev_t hammer_fsid_to_udev(hammer_uuid_t *uuid);
1396 int hammer_blocksize(int64_t file_offset);
1397 int hammer_blockoff(int64_t file_offset);
1398 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1401 * Shortcut for _hammer_checkspace(), used all over the code.
1404 hammer_checkspace(hammer_mount_t hmp, int slop)
1406 return(_hammer_checkspace(hmp, slop, NULL));
1409 static __inline void
1410 hammer_wait_mem_record(hammer_record_t record)
1412 hammer_wait_mem_record_ident(record, "hmmwai");
1415 static __inline void
1416 hammer_lock_ex(struct hammer_lock *lock)
1418 hammer_lock_ex_ident(lock, "hmrlck");
1421 static __inline void
1422 hammer_modify_volume_noundo(hammer_transaction_t trans, hammer_volume_t volume)
1424 hammer_modify_volume(trans, volume, NULL, 0);
1427 static __inline void
1428 hammer_modify_buffer_noundo(hammer_transaction_t trans, hammer_buffer_t buffer)
1430 hammer_modify_buffer(trans, buffer, NULL, 0);
1434 * Indicate that a B-Tree node is being modified.
1436 static __inline void
1437 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1439 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1440 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1443 static __inline void
1444 hammer_modify_node_all(hammer_transaction_t trans, hammer_node_t node)
1446 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1447 hammer_modify_buffer(trans, node->buffer,
1448 node->ondisk, sizeof(*node->ondisk));
1451 static __inline void
1452 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1453 void *base, int len)
1455 KKASSERT((char *)base >= (char *)node->ondisk &&
1456 (char *)base + len <=
1457 (char *)node->ondisk + sizeof(*node->ondisk));
1458 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1460 if (hammer_btree_full_undo) {
1461 hammer_modify_node_all(trans, node);
1463 hammer_modify_buffer(trans, node->buffer, base, len);
1464 hammer_modify_buffer(trans, node->buffer, &node->ondisk->crc,
1465 sizeof(hammer_crc_t));
1466 --node->buffer->io.modify_refs; /* only want one ref */
1471 * Indicate that the specified modifications have been completed.
1473 * Do not try to generate the crc here, it's very expensive to do and a
1474 * sequence of insertions or deletions can result in many calls to this
1475 * function on the same node.
1477 static __inline void
1478 hammer_modify_node_done(hammer_node_t node)
1480 node->flags |= HAMMER_NODE_CRCGOOD;
1481 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1482 node->flags |= HAMMER_NODE_NEEDSCRC;
1483 node->buffer->io.gencrc = 1;
1484 hammer_ref_node(node);
1486 hammer_modify_buffer_done(node->buffer);
1490 hammer_btree_extract_leaf(hammer_cursor_t cursor)
1492 return(hammer_btree_extract(cursor, 0));
1496 hammer_btree_extract_data(hammer_cursor_t cursor)
1498 return(hammer_btree_extract(cursor, HAMMER_CURSOR_GET_DATA));
1502 * Lookup a blockmap offset.
1504 static __inline hammer_off_t
1505 hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t zone_offset,
1508 #if defined INVARIANTS
1509 KKASSERT(hammer_is_zone_record(zone_offset));
1513 * We can actually skip blockmap verify by default,
1514 * as normal blockmaps are now direct-mapped onto the freemap
1515 * and so represent zone-2 addresses.
1517 if (hammer_verify_zone == 0) {
1519 return hammer_xlate_to_zone2(zone_offset);
1522 return hammer_blockmap_lookup_verify(hmp, zone_offset, errorp);
1525 #define hammer_modify_volume_field(trans, vol, field) \
1526 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1527 sizeof((vol)->ondisk->field))
1529 #define hammer_modify_node_field(trans, node, field) \
1530 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1531 sizeof((node)->ondisk->field))
1534 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1535 * created directories for HAMMER version 2 or greater and causes
1536 * directory entries to be placed the inode localization zone in
1537 * the B-Tree instead of the misc zone.
1539 * This greatly improves localization between directory entries and
1542 static __inline uint32_t
1543 hammer_dir_localization(hammer_inode_t dip)
1545 return(HAMMER_DIR_INODE_LOCALIZATION(&dip->ino_data));
1550 hammer_buf_peek_io(struct buf *bp)
1552 return((hammer_io_t)bp->b_priv);
1557 hammer_buf_attach_io(struct buf *bp, hammer_io_t io)
1559 /* struct buf and struct hammer_io are 1:1 */
1560 KKASSERT(hammer_buf_peek_io(bp) == NULL);
1565 __hammer_vol_index(int vol_no)
1567 return(vol_no >> 6);
1570 static __inline uint64_t
1571 __hammer_vol_low(int vol_no)
1573 return((uint64_t)1 << (vol_no & ((1 << 6) - 1)));
1576 static __inline void
1577 hammer_volume_number_add(hammer_mount_t hmp, hammer_volume_t vol)
1579 int i = __hammer_vol_index(vol->vol_no);
1580 hmp->volume_map[i] |= __hammer_vol_low(vol->vol_no);
1583 static __inline void
1584 hammer_volume_number_del(hammer_mount_t hmp, hammer_volume_t vol)
1586 int i = __hammer_vol_index(vol->vol_no);
1587 hmp->volume_map[i] &= ~__hammer_vol_low(vol->vol_no);
1591 hammer_volume_number_test(hammer_mount_t hmp, int n)
1593 int i = __hammer_vol_index(n);
1594 return((hmp->volume_map[i] & __hammer_vol_low(n)) != 0);
1597 #define hkprintf(format, args...) \
1598 kprintf("HAMMER: "format,## args)
1599 #define hvkprintf(vol, format, args...) \
1600 kprintf("HAMMER(%s) "format, vol->ondisk->vol_label,## args)
1601 #define hmkprintf(hmp, format, args...) \
1602 kprintf("HAMMER(%s) "format, hmp->mp->mnt_stat.f_mntfromname,## args)
1603 #define hdkprintf(format, args...) \
1604 kprintf("%s: "format, __func__,## args)
1606 #define hkrateprintf(rate , format, args...) \
1607 krateprintf(rate, "HAMMER: "format,## args)
1608 #define hvkrateprintf(rate, vol, format, args...) \
1609 krateprintf(rate, "HAMMER(%s) "format, vol->ondisk->vol_label,## args)
1610 #define hmkrateprintf(rate, hmp, format, args...) \
1611 krateprintf(rate, "HAMMER(%s) "format, hmp->mp->mnt_stat.f_mntfromname,## args)
1612 #define hdkrateprintf(rate, format, args...) \
1613 krateprintf(rate, "%s: "format, __func__,## args)
1615 #define hpanic(format, args...) \
1616 panic("%s: "format, __func__,## args)
1617 #endif /* _KERNEL */
1619 #endif /* !VFS_HAMMER_HAMMER_H_ */