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 <sys/mplock2.h>
66 #include "hammer_disk.h"
67 #include "hammer_mount.h"
68 #include "hammer_ioctl.h"
70 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
72 MALLOC_DECLARE(M_HAMMER
);
77 #if !defined(KTR_HAMMER)
78 #define KTR_HAMMER KTR_ALL
80 KTR_INFO_MASTER_EXTERN(hammer
);
88 * Key structure used for custom RB tree inode lookups. This prototypes
89 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
91 typedef struct hammer_inode_info
{
92 int64_t obj_id
; /* (key) object identifier */
93 hammer_tid_t obj_asof
; /* (key) snapshot transid or 0 */
94 u_int32_t obj_localization
; /* (key) pseudo-fs */
96 struct hammer_btree_leaf_elm
*leaf
;
98 } *hammer_inode_info_t
;
100 typedef enum hammer_transaction_type
{
104 } hammer_transaction_type_t
;
107 * HAMMER Transaction tracking
109 struct hammer_transaction
{
110 hammer_transaction_type_t type
;
111 struct hammer_mount
*hmp
;
117 struct hammer_volume
*rootvol
;
120 typedef struct hammer_transaction
*hammer_transaction_t
;
122 #define HAMMER_TRANSF_NEWINODE 0x0001
123 #define HAMMER_TRANSF_DIDIO 0x0002
124 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */
130 volatile u_int refs
; /* active references */
131 volatile u_int lockval
; /* lock count and control bits */
132 struct thread
*lowner
; /* owner if exclusively held */
133 struct thread
*rowner
; /* owner if exclusively held */
136 #define HAMMER_REFS_LOCKED 0x40000000 /* transition check */
137 #define HAMMER_REFS_WANTED 0x20000000 /* transition check */
138 #define HAMMER_REFS_CHECK 0x10000000 /* transition check */
140 #define HAMMER_REFS_FLAGS (HAMMER_REFS_LOCKED | \
141 HAMMER_REFS_WANTED | \
144 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000
145 #define HAMMER_LOCKF_WANTED 0x20000000
148 hammer_notlocked(struct hammer_lock
*lock
)
150 return(lock
->lockval
== 0);
154 hammer_islocked(struct hammer_lock
*lock
)
156 return(lock
->lockval
!= 0);
160 * Returns the number of refs on the object.
163 hammer_isactive(struct hammer_lock
*lock
)
165 return(lock
->refs
& ~HAMMER_REFS_FLAGS
);
169 hammer_oneref(struct hammer_lock
*lock
)
171 return((lock
->refs
& ~HAMMER_REFS_FLAGS
) == 1);
175 hammer_norefs(struct hammer_lock
*lock
)
177 return((lock
->refs
& ~HAMMER_REFS_FLAGS
) == 0);
181 hammer_norefsorlock(struct hammer_lock
*lock
)
183 return(lock
->refs
== 0);
187 hammer_refsorlock(struct hammer_lock
*lock
)
189 return(lock
->refs
!= 0);
193 * Return if we specifically own the lock exclusively.
196 hammer_lock_excl_owned(struct hammer_lock
*lock
, thread_t td
)
198 if ((lock
->lockval
& HAMMER_LOCKF_EXCLUSIVE
) &&
199 lock
->lowner
== td
) {
206 * Flush state, used by various structures
208 typedef enum hammer_inode_state
{
212 } hammer_inode_state_t
;
214 TAILQ_HEAD(hammer_record_list
, hammer_record
);
217 * Pseudo-filesystem extended data tracking
219 struct hammer_pfs_rb_tree
;
220 struct hammer_pseudofs_inmem
;
221 RB_HEAD(hammer_pfs_rb_tree
, hammer_pseudofs_inmem
);
222 RB_PROTOTYPE2(hammer_pfs_rb_tree
, hammer_pseudofs_inmem
, rb_node
,
223 hammer_pfs_rb_compare
, u_int32_t
);
225 struct hammer_pseudofs_inmem
{
226 RB_ENTRY(hammer_pseudofs_inmem
) rb_node
;
227 struct hammer_lock lock
;
228 u_int32_t localization
;
229 hammer_tid_t create_tid
;
232 struct hammer_pseudofs_data pfsd
;
235 typedef struct hammer_pseudofs_inmem
*hammer_pseudofs_inmem_t
;
237 #define HAMMER_PFSM_DELETED 0x0001
240 * Cache object ids. A fixed number of objid cache structures are
241 * created to reserve object id's for newly created files in multiples
242 * of 100,000, localized to a particular directory, and recycled as
243 * needed. This allows parallel create operations in different
244 * directories to retain fairly localized object ids which in turn
245 * improves reblocking performance and layout.
247 #define OBJID_CACHE_SIZE 1024
248 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */
249 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */
250 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1)
251 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
253 typedef struct hammer_objid_cache
{
254 TAILQ_ENTRY(hammer_objid_cache
) entry
;
255 struct hammer_inode
*dip
;
256 hammer_tid_t base_tid
;
260 } *hammer_objid_cache_t
;
263 * Associate an inode with a B-Tree node to cache search start positions
265 typedef struct hammer_node_cache
{
266 TAILQ_ENTRY(hammer_node_cache
) entry
;
267 struct hammer_node
*node
;
268 struct hammer_inode
*ip
;
269 } *hammer_node_cache_t
;
271 TAILQ_HEAD(hammer_node_cache_list
, hammer_node_cache
);
274 * Structure used to organize flush groups. Flush groups must be
275 * organized into chunks in order to avoid blowing out the UNDO FIFO.
276 * Without this a 'sync' could end up flushing 50,000 inodes in a single
279 struct hammer_fls_rb_tree
;
280 RB_HEAD(hammer_fls_rb_tree
, hammer_inode
);
281 RB_PROTOTYPE(hammer_fls_rb_tree
, hammer_inode
, rb_flsnode
,
282 hammer_ino_rb_compare
);
284 struct hammer_flush_group
{
285 TAILQ_ENTRY(hammer_flush_group
) flush_entry
;
286 struct hammer_fls_rb_tree flush_tree
;
288 int total_count
; /* record load */
289 int running
; /* group is running */
294 typedef struct hammer_flush_group
*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 struct hammer_ino_rb_tree
;
316 RB_HEAD(hammer_ino_rb_tree
, hammer_inode
);
317 RB_PROTOTYPEX(hammer_ino_rb_tree
, INFO
, hammer_inode
, rb_node
,
318 hammer_ino_rb_compare
, hammer_inode_info_t
);
320 struct hammer_redo_rb_tree
;
321 RB_HEAD(hammer_redo_rb_tree
, hammer_inode
);
322 RB_PROTOTYPE2(hammer_redo_rb_tree
, hammer_inode
, rb_redonode
,
323 hammer_redo_rb_compare
, hammer_off_t
);
325 struct hammer_rec_rb_tree
;
326 struct hammer_record
;
327 RB_HEAD(hammer_rec_rb_tree
, hammer_record
);
328 RB_PROTOTYPEX(hammer_rec_rb_tree
, INFO
, hammer_record
, rb_node
,
329 hammer_rec_rb_compare
, hammer_btree_leaf_elm_t
);
331 TAILQ_HEAD(hammer_node_list
, hammer_node
);
333 struct hammer_inode
{
334 RB_ENTRY(hammer_inode
) rb_node
;
335 hammer_inode_state_t flush_state
;
336 hammer_flush_group_t flush_group
;
337 RB_ENTRY(hammer_inode
) rb_flsnode
; /* when on flush list */
338 RB_ENTRY(hammer_inode
) rb_redonode
; /* when INODE_RDIRTY is set */
339 struct hammer_record_list target_list
; /* target of dependant recs */
340 int64_t obj_id
; /* (key) object identifier */
341 hammer_tid_t obj_asof
; /* (key) snapshot or 0 */
342 u_int32_t obj_localization
; /* (key) pseudo-fs */
343 struct hammer_mount
*hmp
;
344 hammer_objid_cache_t objid_cache
;
346 int error
; /* flush error */
347 int cursor_ip_refs
; /* sanity */
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 */
358 struct hammer_node_cache cache
[4]; /* search initiate cache */
361 * When a demark is created to synchronize an inode to
362 * disk, certain fields are copied so the front-end VOPs
363 * can continue to run in parallel with the synchronization
364 * occuring in the background.
366 int sync_flags
; /* to-sync flags cache */
367 off_t sync_trunc_off
; /* to-sync truncation */
368 off_t save_trunc_off
; /* write optimization */
369 struct hammer_btree_leaf_elm sync_ino_leaf
; /* to-sync cache */
370 struct hammer_inode_data sync_ino_data
; /* to-sync cache */
374 * Track the earliest offset in the UNDO/REDO FIFO containing
375 * REDO records. This is staged to the backend during flush
376 * sequences. While the inode is staged redo_fifo_next is used
377 * to track the earliest offset for rotation into redo_fifo_start
378 * on completion of the flush.
380 hammer_off_t redo_fifo_start
;
381 hammer_off_t redo_fifo_next
;
384 typedef struct hammer_inode
*hammer_inode_t
;
386 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
389 * NOTE: DDIRTY does not include atime or mtime and does not include
390 * write-append size changes. SDIRTY handles write-append size
393 * REDO indicates that REDO logging is active, creating a definitive
394 * stream of REDO records in the UNDO/REDO log for writes and
395 * truncations, including boundary records when/if REDO is turned off.
396 * REDO is typically enabled by fsync() and turned off if excessive
397 * writes without an fsync() occurs.
399 * RDIRTY indicates that REDO records were laid down in the UNDO/REDO
400 * FIFO (even if REDO is turned off some might still be active) and
401 * still being tracked for this inode. See hammer_redo.c
403 /* (not including atime/mtime) */
404 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
405 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
406 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */
407 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
408 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
409 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
410 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
411 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
412 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
413 #define HAMMER_INODE_UNUSED0400 0x0400
414 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
415 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
416 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */
417 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
418 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
420 #define HAMMER_INODE_TRUNCATED 0x00010000
421 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
422 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
423 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
424 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
425 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
426 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */
427 #define HAMMER_INODE_SDIRTY 0x01000000 /* in-memory ino_data.size is dirty*/
428 #define HAMMER_INODE_REDO 0x02000000 /* REDO logging active */
429 #define HAMMER_INODE_RDIRTY 0x04000000 /* REDO records active in fifo */
431 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY| \
432 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
433 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
434 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
436 #define HAMMER_INODE_MODMASK_NOXDIRTY \
437 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
439 #define HAMMER_INODE_MODMASK_NOREDO \
440 (HAMMER_INODE_DDIRTY| \
441 HAMMER_INODE_XDIRTY| \
442 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
444 #define HAMMER_FLUSH_SIGNAL 0x0001
445 #define HAMMER_FLUSH_RECURSION 0x0002
448 * Used by the inode reclaim code to pipeline reclaims and avoid
449 * blowing out kernel memory or letting the flusher get too far
450 * behind. The reclaim wakes up when count reaches 0 or the
453 struct hammer_reclaim
{
454 TAILQ_ENTRY(hammer_reclaim
) entry
;
458 #define HAMMER_RECLAIM_WAIT 4000 /* default vfs.hammer.limit_reclaim */
461 * Track who is creating the greatest burden on the
464 struct hammer_inostats
{
465 pid_t pid
; /* track user process */
466 int ltick
; /* last tick */
467 int count
; /* count (degenerates) */
470 #define HAMMER_INOSTATS_HSIZE 32
471 #define HAMMER_INOSTATS_HMASK (HAMMER_INOSTATS_HSIZE - 1)
474 * Structure used to represent an unsynchronized record in-memory. These
475 * records typically represent directory entries. Only non-historical
476 * records are kept in-memory.
478 * Records are organized as a per-inode RB-Tree. If the inode is not
479 * on disk then neither are any records and the in-memory record tree
480 * represents the entire contents of the inode. If the inode is on disk
481 * then the on-disk B-Tree is scanned in parallel with the in-memory
482 * RB-Tree to synthesize the current state of the file.
484 * Records are also used to enforce the ordering of directory create/delete
485 * operations. A new inode will not be flushed to disk unless its related
486 * directory entry is also being flushed at the same time. A directory entry
487 * will not be removed unless its related inode is also being removed at the
490 typedef enum hammer_record_type
{
491 HAMMER_MEM_RECORD_GENERAL
, /* misc record */
492 HAMMER_MEM_RECORD_INODE
, /* inode record */
493 HAMMER_MEM_RECORD_ADD
, /* positive memory cache record */
494 HAMMER_MEM_RECORD_DEL
, /* negative delete-on-disk record */
495 HAMMER_MEM_RECORD_DATA
/* bulk-data record w/on-disk ref */
496 } hammer_record_type_t
;
498 struct hammer_record
{
499 RB_ENTRY(hammer_record
) rb_node
;
500 TAILQ_ENTRY(hammer_record
) target_entry
;
501 hammer_inode_state_t flush_state
;
502 hammer_flush_group_t flush_group
;
503 hammer_record_type_t type
;
504 struct hammer_lock lock
;
505 struct hammer_reserve
*resv
;
506 struct hammer_inode
*ip
;
507 struct hammer_inode
*target_ip
;
508 struct hammer_btree_leaf_elm leaf
;
509 union hammer_data_ondisk
*data
;
511 hammer_off_t zone2_offset
; /* direct-write only */
514 typedef struct hammer_record
*hammer_record_t
;
517 * Record flags. Note that FE can only be set by the frontend if the
518 * record has not been interlocked by the backend w/ BE.
520 #define HAMMER_RECF_ALLOCDATA 0x0001
521 #define HAMMER_RECF_ONRBTREE 0x0002
522 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
523 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
524 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */
525 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
526 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
527 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
528 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/
529 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/
530 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */
531 #define HAMMER_RECF_REDO 0x1000 /* REDO was laid down */
534 * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
536 #define HAMMER_CREATE_MODE_UMIRROR 0x0001
537 #define HAMMER_CREATE_MODE_SYS 0x0002
539 #define HAMMER_DELETE_ADJUST 0x0001
540 #define HAMMER_DELETE_DESTROY 0x0002
543 * In-memory structures representing on-disk structures.
545 struct hammer_volume
;
546 struct hammer_buffer
;
549 struct hammer_reserve
;
551 RB_HEAD(hammer_vol_rb_tree
, hammer_volume
);
552 RB_HEAD(hammer_buf_rb_tree
, hammer_buffer
);
553 RB_HEAD(hammer_nod_rb_tree
, hammer_node
);
554 RB_HEAD(hammer_und_rb_tree
, hammer_undo
);
555 RB_HEAD(hammer_res_rb_tree
, hammer_reserve
);
557 RB_PROTOTYPE2(hammer_vol_rb_tree
, hammer_volume
, rb_node
,
558 hammer_vol_rb_compare
, int32_t);
559 RB_PROTOTYPE2(hammer_buf_rb_tree
, hammer_buffer
, rb_node
,
560 hammer_buf_rb_compare
, hammer_off_t
);
561 RB_PROTOTYPE2(hammer_nod_rb_tree
, hammer_node
, rb_node
,
562 hammer_nod_rb_compare
, hammer_off_t
);
563 RB_PROTOTYPE2(hammer_und_rb_tree
, hammer_undo
, rb_node
,
564 hammer_und_rb_compare
, hammer_off_t
);
565 RB_PROTOTYPE2(hammer_res_rb_tree
, hammer_reserve
, rb_node
,
566 hammer_res_rb_compare
, hammer_off_t
);
569 * IO management - embedded at the head of various in-memory structures
571 * VOLUME - hammer_volume containing meta-data
572 * META_BUFFER - hammer_buffer containing meta-data
573 * DATA_BUFFER - hammer_buffer containing pure-data
575 * Dirty volume headers and dirty meta-data buffers are locked until the
576 * flusher can sequence them out. Dirty pure-data buffers can be written.
577 * Clean buffers can be passively released.
579 typedef enum hammer_io_type
{
580 HAMMER_STRUCTURE_VOLUME
,
581 HAMMER_STRUCTURE_META_BUFFER
,
582 HAMMER_STRUCTURE_UNDO_BUFFER
,
583 HAMMER_STRUCTURE_DATA_BUFFER
,
584 HAMMER_STRUCTURE_DUMMY
587 union hammer_io_structure
;
591 LIST_ENTRY(worklist
) node
;
594 TAILQ_HEAD(hammer_io_list
, hammer_io
);
595 typedef struct hammer_io_list
*hammer_io_list_t
;
598 struct worklist worklist
;
599 struct hammer_lock lock
;
600 enum hammer_io_type type
;
601 struct hammer_mount
*hmp
;
602 struct hammer_volume
*volume
;
603 TAILQ_ENTRY(hammer_io
) mod_entry
; /* list entry if modified */
604 TAILQ_ENTRY(hammer_io
) iorun_entry
; /* iorun_list */
605 hammer_io_list_t mod_list
;
607 int64_t offset
; /* zone-2 offset */
608 int bytes
; /* buffer cache buffer size */
611 u_int modified
: 1; /* bp's data was modified */
612 u_int released
: 1; /* bp released (w/ B_LOCKED set) */
613 u_int running
: 1; /* bp write IO in progress */
614 u_int waiting
: 1; /* someone is waiting on us */
615 u_int validated
: 1; /* ondisk has been validated */
616 u_int waitdep
: 1; /* flush waits for dependancies */
617 u_int recovered
: 1; /* has recovery ref */
618 u_int waitmod
: 1; /* waiting for modify_refs */
619 u_int reclaim
: 1; /* reclaim requested */
620 u_int gencrc
: 1; /* crc needs to be generated */
621 u_int ioerror
: 1; /* abort on io-error */
624 typedef struct hammer_io
*hammer_io_t
;
626 #define HAMMER_CLUSTER_SIZE (64 * 1024)
627 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
628 #undef HAMMER_CLUSTER_SIZE
629 #define HAMMER_CLUSTER_SIZE MAXBSIZE
631 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
634 * In-memory volume representing on-disk buffer
636 struct hammer_volume
{
638 RB_ENTRY(hammer_volume
) rb_node
;
639 struct hammer_volume_ondisk
*ondisk
;
641 int64_t nblocks
; /* note: special calculation for statfs */
642 int64_t buffer_base
; /* base offset of buffer 0 */
643 hammer_off_t maxbuf_off
; /* Maximum buffer offset (zone-2) */
644 hammer_off_t maxraw_off
; /* Maximum raw offset for device */
650 typedef struct hammer_volume
*hammer_volume_t
;
653 * In-memory buffer (other then volume, super-cluster, or cluster),
654 * representing an on-disk buffer.
656 struct hammer_buffer
{
658 RB_ENTRY(hammer_buffer
) rb_node
;
660 hammer_off_t zoneX_offset
;
661 hammer_off_t zone2_offset
;
662 struct hammer_reserve
*resv
;
663 struct hammer_node_list clist
;
666 typedef struct hammer_buffer
*hammer_buffer_t
;
669 * In-memory B-Tree node, representing an on-disk B-Tree node.
671 * This is a hang-on structure which is backed by a hammer_buffer,
672 * indexed by a hammer_cluster, and used for fine-grained locking of
673 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
674 * can contain multiple nodes representing wildly disassociated portions
675 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
677 * This structure uses a cluster-relative index to reduce the number
678 * of layers required to access it, and also because all on-disk B-Tree
679 * references are cluster-relative offsets.
682 struct hammer_lock lock
; /* node-by-node lock */
683 TAILQ_ENTRY(hammer_node
) entry
; /* per-buffer linkage */
684 RB_ENTRY(hammer_node
) rb_node
; /* per-cluster linkage */
685 hammer_off_t node_offset
; /* full offset spec */
686 struct hammer_mount
*hmp
;
687 struct hammer_buffer
*buffer
; /* backing buffer */
688 hammer_node_ondisk_t ondisk
; /* ptr to on-disk structure */
689 TAILQ_HEAD(, hammer_cursor
) cursor_list
; /* deadlock recovery */
690 struct hammer_node_cache_list cache_list
; /* passive caches */
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)
704 typedef struct hammer_node
*hammer_node_t
;
707 * List of locked nodes. This structure is used to lock potentially large
708 * numbers of nodes as an aid for complex B-Tree operations.
710 struct hammer_node_lock
;
711 TAILQ_HEAD(hammer_node_lock_list
, hammer_node_lock
);
713 struct hammer_node_lock
{
714 TAILQ_ENTRY(hammer_node_lock
) entry
;
715 struct hammer_node_lock_list list
;
716 struct hammer_node_lock
*parent
;
718 hammer_node_ondisk_t copy
; /* copy of on-disk data */
719 int index
; /* index of this node in parent */
720 int count
; /* count children */
724 typedef struct hammer_node_lock
*hammer_node_lock_t
;
726 #define HAMMER_NODE_LOCK_UPDATED 0x0001
729 * Common I/O management structure - embedded in in-memory structures
730 * which are backed by filesystem buffers.
732 union hammer_io_structure
{
734 struct hammer_volume volume
;
735 struct hammer_buffer buffer
;
738 typedef union hammer_io_structure
*hammer_io_structure_t
;
741 * The reserve structure prevents the blockmap from allocating
742 * out of a reserved bigblock. Such reservations are used by
743 * the direct-write mechanism.
745 * The structure is also used to hold off on reallocations of
746 * big blocks from the freemap until flush dependancies have
749 struct hammer_reserve
{
750 RB_ENTRY(hammer_reserve
) rb_node
;
751 TAILQ_ENTRY(hammer_reserve
) delay_entry
;
757 hammer_off_t zone_offset
;
760 typedef struct hammer_reserve
*hammer_reserve_t
;
762 #define HAMMER_RESF_ONDELAY 0x0001
763 #define HAMMER_RESF_LAYER2FREE 0x0002
765 #include "hammer_cursor.h"
768 * The undo structure tracks recent undos to avoid laying down duplicate
769 * undos within a flush group, saving us a significant amount of overhead.
771 * This is strictly a heuristic.
773 #define HAMMER_MAX_UNDOS 1024
774 #define HAMMER_MAX_FLUSHERS 4
777 RB_ENTRY(hammer_undo
) rb_node
;
778 TAILQ_ENTRY(hammer_undo
) lru_entry
;
783 typedef struct hammer_undo
*hammer_undo_t
;
785 struct hammer_flusher_info
;
786 TAILQ_HEAD(hammer_flusher_info_list
, hammer_flusher_info
);
788 struct hammer_flusher
{
789 int signal
; /* flusher thread sequencer */
790 int act
; /* currently active flush group */
791 int done
; /* set to act when complete */
792 int next
; /* next flush group */
793 int group_lock
; /* lock sequencing of the next flush */
794 int exiting
; /* request master exit */
795 thread_t td
; /* master flusher thread */
796 hammer_tid_t tid
; /* last flushed transaction id */
797 int finalize_want
; /* serialize finalization */
798 struct hammer_lock finalize_lock
; /* serialize finalization */
799 struct hammer_transaction trans
; /* shared transaction */
800 struct hammer_flusher_info_list run_list
;
801 struct hammer_flusher_info_list ready_list
;
804 #define HAMMER_FLUSH_UNDOS_RELAXED 0
805 #define HAMMER_FLUSH_UNDOS_FORCED 1
806 #define HAMMER_FLUSH_UNDOS_AUTO 2
808 * Internal hammer mount data structure
810 struct hammer_mount
{
812 /*struct vnode *rootvp;*/
813 struct hammer_ino_rb_tree rb_inos_root
;
814 struct hammer_redo_rb_tree rb_redo_root
;
815 struct hammer_vol_rb_tree rb_vols_root
;
816 struct hammer_nod_rb_tree rb_nods_root
;
817 struct hammer_und_rb_tree rb_undo_root
;
818 struct hammer_res_rb_tree rb_resv_root
;
819 struct hammer_buf_rb_tree rb_bufs_root
;
820 struct hammer_pfs_rb_tree rb_pfsm_root
;
821 struct hammer_volume
*rootvol
;
822 struct hammer_base_elm root_btree_beg
;
823 struct hammer_base_elm root_btree_end
;
825 struct malloc_type
*m_misc
;
826 struct malloc_type
*m_inodes
;
828 int flags
; /* HAMMER_MOUNT_xxx flags */
833 int master_id
; /* -1 or 0-15 - clustering and mirroring */
834 int version
; /* hammer filesystem version to use */
835 int rsv_inodes
; /* reserved space due to dirty inodes */
836 int64_t rsv_databytes
; /* reserved space due to record data */
837 int rsv_recs
; /* reserved space due to dirty records */
838 int rsv_fromdelay
; /* bigblocks reserved due to flush delay */
839 int undo_rec_limit
; /* based on size of undo area */
841 int count_newrecords
;
843 int volume_to_remove
; /* volume that is currently being removed */
845 int inode_reclaims
; /* inodes pending reclaim by flusher */
846 int count_inodes
; /* total number of inodes */
847 int count_iqueued
; /* inodes queued to flusher */
849 struct hammer_flusher flusher
;
851 u_int check_interrupt
;
854 struct hammer_io_list volu_list
; /* dirty undo buffers */
855 struct hammer_io_list undo_list
; /* dirty undo buffers */
856 struct hammer_io_list data_list
; /* dirty data buffers */
857 struct hammer_io_list alt_data_list
; /* dirty data buffers */
858 struct hammer_io_list meta_list
; /* dirty meta bufs */
859 struct hammer_io_list lose_list
; /* loose buffers */
860 int locked_dirty_space
; /* meta/volu count */
861 int io_running_space
;
862 int objid_cache_count
;
863 int error
; /* critical I/O error */
864 struct krate krate
; /* rate limited kprintf */
865 hammer_tid_t asof
; /* snapshot mount */
866 hammer_tid_t next_tid
;
867 hammer_tid_t flush_tid1
; /* flusher tid sequencing */
868 hammer_tid_t flush_tid2
; /* flusher tid sequencing */
869 int64_t copy_stat_freebigblocks
; /* number of free bigblocks */
870 u_int32_t undo_seqno
; /* UNDO/REDO FIFO seqno */
872 struct netexport export
;
873 struct hammer_lock sync_lock
;
874 struct hammer_lock free_lock
;
875 struct hammer_lock undo_lock
;
876 struct hammer_lock blkmap_lock
;
877 struct hammer_lock snapshot_lock
;
878 struct hammer_lock volume_lock
;
879 struct hammer_blockmap blockmap
[HAMMER_MAX_ZONES
];
880 struct hammer_undo undos
[HAMMER_MAX_UNDOS
];
882 TAILQ_HEAD(, hammer_undo
) undo_lru_list
;
883 TAILQ_HEAD(, hammer_reserve
) delay_list
;
884 struct hammer_flush_group_list flush_group_list
;
885 hammer_flush_group_t next_flush_group
;
886 TAILQ_HEAD(, hammer_objid_cache
) objid_cache_list
;
887 TAILQ_HEAD(, hammer_reclaim
) reclaim_list
;
888 TAILQ_HEAD(, hammer_io
) iorun_list
;
890 struct hammer_inostats inostats
[HAMMER_INOSTATS_HSIZE
];
893 typedef struct hammer_mount
*hammer_mount_t
;
895 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001
896 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002
897 #define HAMMER_MOUNT_REDO_SYNC 0x0004
899 struct hammer_sync_info
{
905 * Minium buffer cache bufs required to rebalance the B-Tree.
906 * This is because we must hold the children and the children's children
907 * locked. Even this might not be enough if things are horribly out
910 #define HAMMER_REBALANCE_MIN_BUFS \
911 (HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
917 * checkspace slop (8MB chunks), higher numbers are more conservative.
919 #define HAMMER_CHKSPC_REBLOCK 25
920 #define HAMMER_CHKSPC_MIRROR 20
921 #define HAMMER_CHKSPC_WRITE 20
922 #define HAMMER_CHKSPC_CREATE 20
923 #define HAMMER_CHKSPC_REMOVE 10
924 #define HAMMER_CHKSPC_EMERGENCY 0
928 extern struct vop_ops hammer_vnode_vops
;
929 extern struct vop_ops hammer_spec_vops
;
930 extern struct vop_ops hammer_fifo_vops
;
931 extern struct bio_ops hammer_bioops
;
933 extern int hammer_debug_io
;
934 extern int hammer_debug_general
;
935 extern int hammer_debug_debug
;
936 extern int hammer_debug_inode
;
937 extern int hammer_debug_locks
;
938 extern int hammer_debug_btree
;
939 extern int hammer_debug_tid
;
940 extern int hammer_debug_recover
;
941 extern int hammer_debug_recover_faults
;
942 extern int hammer_debug_critical
;
943 extern int hammer_cluster_enable
;
944 extern int hammer_count_fsyncs
;
945 extern int hammer_count_inodes
;
946 extern int hammer_count_iqueued
;
947 extern int hammer_count_reclaiming
;
948 extern int hammer_count_records
;
949 extern int hammer_count_record_datas
;
950 extern int hammer_count_volumes
;
951 extern int hammer_count_buffers
;
952 extern int hammer_count_nodes
;
953 extern int64_t hammer_count_extra_space_used
;
954 extern int64_t hammer_stats_btree_lookups
;
955 extern int64_t hammer_stats_btree_searches
;
956 extern int64_t hammer_stats_btree_inserts
;
957 extern int64_t hammer_stats_btree_deletes
;
958 extern int64_t hammer_stats_btree_elements
;
959 extern int64_t hammer_stats_btree_splits
;
960 extern int64_t hammer_stats_btree_iterations
;
961 extern int64_t hammer_stats_btree_root_iterations
;
962 extern int64_t hammer_stats_record_iterations
;
963 extern int64_t hammer_stats_file_read
;
964 extern int64_t hammer_stats_file_write
;
965 extern int64_t hammer_stats_file_iopsr
;
966 extern int64_t hammer_stats_file_iopsw
;
967 extern int64_t hammer_stats_disk_read
;
968 extern int64_t hammer_stats_disk_write
;
969 extern int64_t hammer_stats_inode_flushes
;
970 extern int64_t hammer_stats_commits
;
971 extern int64_t hammer_stats_undo
;
972 extern int64_t hammer_stats_redo
;
973 extern int hammer_count_dirtybufspace
;
974 extern int hammer_count_refedbufs
;
975 extern int hammer_count_reservations
;
976 extern int hammer_count_io_running_read
;
977 extern int hammer_count_io_running_write
;
978 extern int hammer_count_io_locked
;
979 extern int hammer_limit_dirtybufspace
;
980 extern int hammer_limit_recs
;
981 extern int hammer_limit_inode_recs
;
982 extern int hammer_limit_reclaim
;
983 extern int hammer_limit_redo
;
984 extern int hammer_bio_count
;
985 extern int hammer_verify_zone
;
986 extern int hammer_verify_data
;
987 extern int hammer_write_mode
;
988 extern int hammer_yield_check
;
989 extern int hammer_fsync_mode
;
990 extern int hammer_autoflush
;
991 extern int64_t hammer_contention_count
;
993 void hammer_critical_error(hammer_mount_t hmp
, hammer_inode_t ip
,
994 int error
, const char *msg
);
995 int hammer_vop_inactive(struct vop_inactive_args
*);
996 int hammer_vop_reclaim(struct vop_reclaim_args
*);
997 int hammer_get_vnode(struct hammer_inode
*ip
, struct vnode
**vpp
);
998 struct hammer_inode
*hammer_get_inode(hammer_transaction_t trans
,
999 hammer_inode_t dip
, int64_t obj_id
,
1000 hammer_tid_t asof
, u_int32_t localization
,
1001 int flags
, int *errorp
);
1002 struct hammer_inode
*hammer_get_dummy_inode(hammer_transaction_t trans
,
1003 hammer_inode_t dip
, int64_t obj_id
,
1004 hammer_tid_t asof
, u_int32_t localization
,
1005 int flags
, int *errorp
);
1006 struct hammer_inode
*hammer_find_inode(hammer_transaction_t trans
,
1007 int64_t obj_id
, hammer_tid_t asof
,
1008 u_int32_t localization
);
1009 void hammer_scan_inode_snapshots(hammer_mount_t hmp
,
1010 hammer_inode_info_t iinfo
,
1011 int (*callback
)(hammer_inode_t ip
, void *data
),
1013 void hammer_put_inode(struct hammer_inode
*ip
);
1014 void hammer_put_inode_ref(struct hammer_inode
*ip
);
1015 void hammer_inode_waitreclaims(hammer_transaction_t trans
);
1017 int hammer_unload_volume(hammer_volume_t volume
, void *data __unused
);
1018 int hammer_adjust_volume_mode(hammer_volume_t volume
, void *data __unused
);
1020 int hammer_unload_buffer(hammer_buffer_t buffer
, void *data
);
1021 int hammer_install_volume(hammer_mount_t hmp
, const char *volname
,
1022 struct vnode
*devvp
);
1023 int hammer_mountcheck_volumes(hammer_mount_t hmp
);
1025 int hammer_mem_add(hammer_record_t record
);
1026 int hammer_ip_lookup(hammer_cursor_t cursor
);
1027 int hammer_ip_first(hammer_cursor_t cursor
);
1028 int hammer_ip_next(hammer_cursor_t cursor
);
1029 int hammer_ip_resolve_data(hammer_cursor_t cursor
);
1030 int hammer_ip_delete_record(hammer_cursor_t cursor
, hammer_inode_t ip
,
1032 int hammer_create_at_cursor(hammer_cursor_t cursor
,
1033 hammer_btree_leaf_elm_t leaf
, void *udata
, int mode
);
1034 int hammer_delete_at_cursor(hammer_cursor_t cursor
, int delete_flags
,
1035 hammer_tid_t delete_tid
, u_int32_t delete_ts
,
1036 int track
, int64_t *stat_bytes
);
1037 int hammer_ip_check_directory_empty(hammer_transaction_t trans
,
1039 int hammer_sync_hmp(hammer_mount_t hmp
, int waitfor
);
1040 int hammer_queue_inodes_flusher(hammer_mount_t hmp
, int waitfor
);
1043 hammer_alloc_mem_record(hammer_inode_t ip
, int data_len
);
1044 void hammer_flush_record_done(hammer_record_t record
, int error
);
1045 void hammer_wait_mem_record_ident(hammer_record_t record
, const char *ident
);
1046 void hammer_rel_mem_record(hammer_record_t record
);
1048 int hammer_cursor_up(hammer_cursor_t cursor
);
1049 int hammer_cursor_up_locked(hammer_cursor_t cursor
);
1050 int hammer_cursor_down(hammer_cursor_t cursor
);
1051 int hammer_cursor_upgrade(hammer_cursor_t cursor
);
1052 int hammer_cursor_upgrade_node(hammer_cursor_t cursor
);
1053 void hammer_cursor_downgrade(hammer_cursor_t cursor
);
1054 int hammer_cursor_seek(hammer_cursor_t cursor
, hammer_node_t node
,
1056 void hammer_lock_ex_ident(struct hammer_lock
*lock
, const char *ident
);
1057 int hammer_lock_ex_try(struct hammer_lock
*lock
);
1058 void hammer_lock_sh(struct hammer_lock
*lock
);
1059 int hammer_lock_sh_try(struct hammer_lock
*lock
);
1060 int hammer_lock_upgrade(struct hammer_lock
*lock
);
1061 void hammer_lock_downgrade(struct hammer_lock
*lock
);
1062 int hammer_lock_status(struct hammer_lock
*lock
);
1063 void hammer_unlock(struct hammer_lock
*lock
);
1064 void hammer_ref(struct hammer_lock
*lock
);
1065 int hammer_ref_interlock(struct hammer_lock
*lock
);
1066 int hammer_ref_interlock_true(struct hammer_lock
*lock
);
1067 void hammer_ref_interlock_done(struct hammer_lock
*lock
);
1068 void hammer_rel(struct hammer_lock
*lock
);
1069 int hammer_rel_interlock(struct hammer_lock
*lock
, int locked
);
1070 void hammer_rel_interlock_done(struct hammer_lock
*lock
, int orig_locked
);
1071 int hammer_get_interlock(struct hammer_lock
*lock
);
1072 int hammer_try_interlock_norefs(struct hammer_lock
*lock
);
1073 void hammer_put_interlock(struct hammer_lock
*lock
, int error
);
1075 void hammer_sync_lock_ex(hammer_transaction_t trans
);
1076 void hammer_sync_lock_sh(hammer_transaction_t trans
);
1077 int hammer_sync_lock_sh_try(hammer_transaction_t trans
);
1078 void hammer_sync_unlock(hammer_transaction_t trans
);
1080 u_int32_t
hammer_to_unix_xid(uuid_t
*uuid
);
1081 void hammer_guid_to_uuid(uuid_t
*uuid
, u_int32_t guid
);
1082 void hammer_time_to_timespec(u_int64_t xtime
, struct timespec
*ts
);
1083 u_int64_t
hammer_timespec_to_time(struct timespec
*ts
);
1084 int hammer_str_to_tid(const char *str
, int *ispfsp
,
1085 hammer_tid_t
*tidp
, u_int32_t
*localizationp
);
1086 int hammer_is_atatext(const char *name
, int len
);
1087 hammer_tid_t
hammer_alloc_objid(hammer_mount_t hmp
, hammer_inode_t dip
,
1089 void hammer_clear_objid(hammer_inode_t dip
);
1090 void hammer_destroy_objid_cache(hammer_mount_t hmp
);
1092 int hammer_enter_undo_history(hammer_mount_t hmp
, hammer_off_t offset
,
1094 void hammer_clear_undo_history(hammer_mount_t hmp
);
1095 enum vtype
hammer_get_vnode_type(u_int8_t obj_type
);
1096 int hammer_get_dtype(u_int8_t obj_type
);
1097 u_int8_t
hammer_get_obj_type(enum vtype vtype
);
1098 int64_t hammer_directory_namekey(hammer_inode_t dip
, const void *name
, int len
,
1099 u_int32_t
*max_iterationsp
);
1100 int hammer_nohistory(hammer_inode_t ip
);
1102 int hammer_init_cursor(hammer_transaction_t trans
, hammer_cursor_t cursor
,
1103 hammer_node_cache_t cache
, hammer_inode_t ip
);
1104 void hammer_normalize_cursor(hammer_cursor_t cursor
);
1105 void hammer_done_cursor(hammer_cursor_t cursor
);
1106 int hammer_recover_cursor(hammer_cursor_t cursor
);
1107 void hammer_unlock_cursor(hammer_cursor_t cursor
);
1108 int hammer_lock_cursor(hammer_cursor_t cursor
);
1109 hammer_cursor_t
hammer_push_cursor(hammer_cursor_t ocursor
);
1110 void hammer_pop_cursor(hammer_cursor_t ocursor
, hammer_cursor_t ncursor
);
1112 void hammer_cursor_replaced_node(hammer_node_t onode
, hammer_node_t nnode
);
1113 void hammer_cursor_removed_node(hammer_node_t onode
, hammer_node_t parent
,
1115 void hammer_cursor_split_node(hammer_node_t onode
, hammer_node_t nnode
,
1117 void hammer_cursor_moved_element(hammer_node_t oparent
, int pindex
,
1118 hammer_node_t onode
, int oindex
,
1119 hammer_node_t nnode
, int nindex
);
1120 void hammer_cursor_parent_changed(hammer_node_t node
, hammer_node_t oparent
,
1121 hammer_node_t nparent
, int nindex
);
1122 void hammer_cursor_inserted_element(hammer_node_t node
, int index
);
1123 void hammer_cursor_deleted_element(hammer_node_t node
, int index
);
1125 int hammer_btree_lookup(hammer_cursor_t cursor
);
1126 int hammer_btree_first(hammer_cursor_t cursor
);
1127 int hammer_btree_last(hammer_cursor_t cursor
);
1128 int hammer_btree_extract(hammer_cursor_t cursor
, int flags
);
1129 int hammer_btree_iterate(hammer_cursor_t cursor
);
1130 int hammer_btree_iterate_reverse(hammer_cursor_t cursor
);
1131 int hammer_btree_insert(hammer_cursor_t cursor
,
1132 hammer_btree_leaf_elm_t elm
, int *doprop
);
1133 int hammer_btree_delete(hammer_cursor_t cursor
);
1134 void hammer_btree_do_propagation(hammer_cursor_t cursor
,
1135 hammer_pseudofs_inmem_t pfsm
,
1136 hammer_btree_leaf_elm_t leaf
);
1137 int hammer_btree_cmp(hammer_base_elm_t key1
, hammer_base_elm_t key2
);
1138 int hammer_btree_chkts(hammer_tid_t ts
, hammer_base_elm_t key
);
1139 int hammer_btree_correct_rhb(hammer_cursor_t cursor
, hammer_tid_t tid
);
1140 int hammer_btree_correct_lhb(hammer_cursor_t cursor
, hammer_tid_t tid
);
1142 int btree_set_parent(hammer_transaction_t trans
, hammer_node_t node
,
1143 hammer_btree_elm_t elm
);
1144 void hammer_node_lock_init(hammer_node_lock_t parent
, hammer_node_t node
);
1145 int hammer_btree_lock_children(hammer_cursor_t cursor
, int depth
,
1146 hammer_node_lock_t parent
);
1147 void hammer_btree_lock_copy(hammer_cursor_t cursor
,
1148 hammer_node_lock_t parent
);
1149 int hammer_btree_sync_copy(hammer_cursor_t cursor
,
1150 hammer_node_lock_t parent
);
1151 void hammer_btree_unlock_children(hammer_cursor_t cursor
,
1152 hammer_node_lock_t parent
);
1153 int hammer_btree_search_node(hammer_base_elm_t elm
, hammer_node_ondisk_t node
);
1154 hammer_node_t
hammer_btree_get_parent(hammer_transaction_t trans
,
1155 hammer_node_t node
, int *parent_indexp
,
1156 int *errorp
, int try_exclusive
);
1158 void hammer_print_btree_node(hammer_node_ondisk_t ondisk
);
1159 void hammer_print_btree_elm(hammer_btree_elm_t elm
, u_int8_t type
, int i
);
1161 void *hammer_bread(struct hammer_mount
*hmp
, hammer_off_t off
,
1162 int *errorp
, struct hammer_buffer
**bufferp
);
1163 void *hammer_bnew(struct hammer_mount
*hmp
, hammer_off_t off
,
1164 int *errorp
, struct hammer_buffer
**bufferp
);
1165 void *hammer_bread_ext(struct hammer_mount
*hmp
, hammer_off_t off
, int bytes
,
1166 int *errorp
, struct hammer_buffer
**bufferp
);
1167 void *hammer_bnew_ext(struct hammer_mount
*hmp
, hammer_off_t off
, int bytes
,
1168 int *errorp
, struct hammer_buffer
**bufferp
);
1170 hammer_volume_t
hammer_get_root_volume(hammer_mount_t hmp
, int *errorp
);
1172 hammer_volume_t
hammer_get_volume(hammer_mount_t hmp
,
1173 int32_t vol_no
, int *errorp
);
1174 hammer_buffer_t
hammer_get_buffer(hammer_mount_t hmp
, hammer_off_t buf_offset
,
1175 int bytes
, int isnew
, int *errorp
);
1176 void hammer_sync_buffers(hammer_mount_t hmp
,
1177 hammer_off_t base_offset
, int bytes
);
1178 int hammer_del_buffers(hammer_mount_t hmp
,
1179 hammer_off_t base_offset
,
1180 hammer_off_t zone2_offset
, int bytes
,
1181 int report_conflicts
);
1183 int hammer_ref_volume(hammer_volume_t volume
);
1184 int hammer_ref_buffer(hammer_buffer_t buffer
);
1185 void hammer_flush_buffer_nodes(hammer_buffer_t buffer
);
1187 void hammer_rel_volume(hammer_volume_t volume
, int locked
);
1188 void hammer_rel_buffer(hammer_buffer_t buffer
, int locked
);
1190 int hammer_vfs_export(struct mount
*mp
, int op
,
1191 const struct export_args
*export
);
1192 hammer_node_t
hammer_get_node(hammer_transaction_t trans
,
1193 hammer_off_t node_offset
, int isnew
, int *errorp
);
1194 void hammer_ref_node(hammer_node_t node
);
1195 hammer_node_t
hammer_ref_node_safe(hammer_transaction_t trans
,
1196 hammer_node_cache_t cache
, int *errorp
);
1197 void hammer_rel_node(hammer_node_t node
);
1198 void hammer_delete_node(hammer_transaction_t trans
,
1199 hammer_node_t node
);
1200 void hammer_cache_node(hammer_node_cache_t cache
,
1201 hammer_node_t node
);
1202 void hammer_uncache_node(hammer_node_cache_t cache
);
1203 void hammer_flush_node(hammer_node_t node
, int locked
);
1205 void hammer_dup_buffer(struct hammer_buffer
**bufferp
,
1206 struct hammer_buffer
*buffer
);
1207 hammer_node_t
hammer_alloc_btree(hammer_transaction_t trans
,
1208 hammer_off_t hint
, int *errorp
);
1209 void *hammer_alloc_data(hammer_transaction_t trans
, int32_t data_len
,
1210 u_int16_t rec_type
, hammer_off_t
*data_offsetp
,
1211 struct hammer_buffer
**data_bufferp
,
1212 hammer_off_t hint
, int *errorp
);
1214 int hammer_generate_undo(hammer_transaction_t trans
,
1215 hammer_off_t zone1_offset
, void *base
, int len
);
1216 int hammer_generate_redo(hammer_transaction_t trans
, hammer_inode_t ip
,
1217 hammer_off_t file_offset
, u_int32_t flags
,
1218 void *base
, int len
);
1219 void hammer_generate_redo_sync(hammer_transaction_t trans
);
1220 void hammer_redo_fifo_start_flush(hammer_inode_t ip
);
1221 void hammer_redo_fifo_end_flush(hammer_inode_t ip
);
1223 void hammer_format_undo(void *base
, u_int32_t seqno
);
1224 int hammer_upgrade_undo_4(hammer_transaction_t trans
);
1226 void hammer_put_volume(struct hammer_volume
*volume
, int flush
);
1227 void hammer_put_buffer(struct hammer_buffer
*buffer
, int flush
);
1229 hammer_off_t
hammer_freemap_alloc(hammer_transaction_t trans
,
1230 hammer_off_t owner
, int *errorp
);
1231 void hammer_freemap_free(hammer_transaction_t trans
, hammer_off_t phys_offset
,
1232 hammer_off_t owner
, int *errorp
);
1233 int _hammer_checkspace(hammer_mount_t hmp
, int slop
, int64_t *resp
);
1234 hammer_off_t
hammer_blockmap_alloc(hammer_transaction_t trans
, int zone
,
1235 int bytes
, hammer_off_t hint
, int *errorp
);
1236 hammer_reserve_t
hammer_blockmap_reserve(hammer_mount_t hmp
, int zone
,
1237 int bytes
, hammer_off_t
*zone_offp
, int *errorp
);
1238 void hammer_blockmap_reserve_complete(hammer_mount_t hmp
,
1239 hammer_reserve_t resv
);
1240 void hammer_reserve_clrdelay(hammer_mount_t hmp
, hammer_reserve_t resv
);
1241 void hammer_blockmap_free(hammer_transaction_t trans
,
1242 hammer_off_t bmap_off
, int bytes
);
1243 int hammer_blockmap_finalize(hammer_transaction_t trans
,
1244 hammer_reserve_t resv
,
1245 hammer_off_t bmap_off
, int bytes
);
1246 int hammer_blockmap_getfree(hammer_mount_t hmp
, hammer_off_t bmap_off
,
1247 int *curp
, int *errorp
);
1248 hammer_off_t
hammer_blockmap_lookup(hammer_mount_t hmp
, hammer_off_t bmap_off
,
1250 hammer_off_t
hammer_undo_lookup(hammer_mount_t hmp
, hammer_off_t bmap_off
,
1252 int64_t hammer_undo_used(hammer_transaction_t trans
);
1253 int64_t hammer_undo_space(hammer_transaction_t trans
);
1254 int64_t hammer_undo_max(hammer_mount_t hmp
);
1255 int hammer_undo_reclaim(hammer_io_t io
);
1257 void hammer_start_transaction(struct hammer_transaction
*trans
,
1258 struct hammer_mount
*hmp
);
1259 void hammer_simple_transaction(struct hammer_transaction
*trans
,
1260 struct hammer_mount
*hmp
);
1261 void hammer_start_transaction_fls(struct hammer_transaction
*trans
,
1262 struct hammer_mount
*hmp
);
1263 void hammer_done_transaction(struct hammer_transaction
*trans
);
1264 hammer_tid_t
hammer_alloc_tid(hammer_mount_t hmp
, int count
);
1266 void hammer_modify_inode(hammer_transaction_t trans
, hammer_inode_t ip
, int flags
);
1267 void hammer_flush_inode(hammer_inode_t ip
, int flags
);
1268 void hammer_flush_inode_done(hammer_inode_t ip
, int error
);
1269 void hammer_wait_inode(hammer_inode_t ip
);
1271 int hammer_create_inode(struct hammer_transaction
*trans
, struct vattr
*vap
,
1272 struct ucred
*cred
, struct hammer_inode
*dip
,
1273 const char *name
, int namelen
,
1274 hammer_pseudofs_inmem_t pfsm
,
1275 struct hammer_inode
**ipp
);
1276 void hammer_rel_inode(hammer_inode_t ip
, int flush
);
1277 int hammer_reload_inode(hammer_inode_t ip
, void *arg __unused
);
1278 int hammer_ino_rb_compare(hammer_inode_t ip1
, hammer_inode_t ip2
);
1279 int hammer_redo_rb_compare(hammer_inode_t ip1
, hammer_inode_t ip2
);
1280 int hammer_destroy_inode_callback(hammer_inode_t ip
, void *data __unused
);
1282 int hammer_sync_inode(hammer_transaction_t trans
, hammer_inode_t ip
);
1283 void hammer_test_inode(hammer_inode_t dip
);
1284 void hammer_inode_unloadable_check(hammer_inode_t ip
, int getvp
);
1286 int hammer_ip_add_directory(struct hammer_transaction
*trans
,
1287 hammer_inode_t dip
, const char *name
, int bytes
,
1288 hammer_inode_t nip
);
1289 int hammer_ip_del_directory(struct hammer_transaction
*trans
,
1290 hammer_cursor_t cursor
, hammer_inode_t dip
,
1292 void hammer_ip_replace_bulk(hammer_mount_t hmp
, hammer_record_t record
);
1293 hammer_record_t
hammer_ip_add_bulk(hammer_inode_t ip
, off_t file_offset
,
1294 void *data
, int bytes
, int *errorp
);
1295 int hammer_ip_frontend_trunc(struct hammer_inode
*ip
, off_t file_size
);
1296 int hammer_ip_add_record(struct hammer_transaction
*trans
,
1297 hammer_record_t record
);
1298 int hammer_ip_delete_range(hammer_cursor_t cursor
, hammer_inode_t ip
,
1299 int64_t ran_beg
, int64_t ran_end
, int truncating
);
1300 int hammer_ip_delete_clean(hammer_cursor_t cursor
, hammer_inode_t ip
,
1302 int hammer_ip_sync_data(hammer_cursor_t cursor
, hammer_inode_t ip
,
1303 int64_t offset
, void *data
, int bytes
);
1304 int hammer_ip_sync_record(hammer_transaction_t trans
, hammer_record_t rec
);
1305 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor
, hammer_record_t rec
);
1306 hammer_pseudofs_inmem_t
hammer_load_pseudofs(hammer_transaction_t trans
,
1307 u_int32_t localization
, int *errorp
);
1308 int hammer_mkroot_pseudofs(hammer_transaction_t trans
, struct ucred
*cred
,
1309 hammer_pseudofs_inmem_t pfsm
);
1310 int hammer_save_pseudofs(hammer_transaction_t trans
,
1311 hammer_pseudofs_inmem_t pfsm
);
1312 int hammer_unload_pseudofs(hammer_transaction_t trans
, u_int32_t localization
);
1313 void hammer_rel_pseudofs(hammer_mount_t hmp
, hammer_pseudofs_inmem_t pfsm
);
1314 int hammer_ioctl(hammer_inode_t ip
, u_long com
, caddr_t data
, int fflag
,
1315 struct ucred
*cred
);
1317 void hammer_io_init(hammer_io_t io
, hammer_volume_t volume
,
1318 enum hammer_io_type type
);
1319 int hammer_io_read(struct vnode
*devvp
, struct hammer_io
*io
,
1320 hammer_off_t limit
);
1321 void hammer_io_advance(struct hammer_io
*io
);
1322 int hammer_io_new(struct vnode
*devvp
, struct hammer_io
*io
);
1323 int hammer_io_inval(hammer_volume_t volume
, hammer_off_t zone2_offset
);
1324 struct buf
*hammer_io_release(struct hammer_io
*io
, int flush
);
1325 void hammer_io_flush(struct hammer_io
*io
, int reclaim
);
1326 void hammer_io_wait(struct hammer_io
*io
);
1327 void hammer_io_waitdep(struct hammer_io
*io
);
1328 void hammer_io_wait_all(hammer_mount_t hmp
, const char *ident
, int doflush
);
1329 int hammer_io_direct_read(hammer_mount_t hmp
, struct bio
*bio
,
1330 hammer_btree_leaf_elm_t leaf
);
1331 int hammer_io_direct_write(hammer_mount_t hmp
, struct bio
*bio
,
1332 hammer_record_t record
);
1333 void hammer_io_direct_wait(hammer_record_t record
);
1334 void hammer_io_direct_uncache(hammer_mount_t hmp
, hammer_btree_leaf_elm_t leaf
);
1335 void hammer_io_write_interlock(hammer_io_t io
);
1336 void hammer_io_done_interlock(hammer_io_t io
);
1337 void hammer_io_clear_modify(struct hammer_io
*io
, int inval
);
1338 void hammer_io_clear_modlist(struct hammer_io
*io
);
1339 void hammer_io_flush_sync(hammer_mount_t hmp
);
1340 void hammer_io_clear_error(struct hammer_io
*io
);
1341 void hammer_io_notmeta(hammer_buffer_t buffer
);
1343 void hammer_modify_volume(hammer_transaction_t trans
, hammer_volume_t volume
,
1344 void *base
, int len
);
1345 void hammer_modify_buffer(hammer_transaction_t trans
, hammer_buffer_t buffer
,
1346 void *base
, int len
);
1347 void hammer_modify_volume_done(hammer_volume_t volume
);
1348 void hammer_modify_buffer_done(hammer_buffer_t buffer
);
1350 int hammer_ioc_reblock(hammer_transaction_t trans
, hammer_inode_t ip
,
1351 struct hammer_ioc_reblock
*reblock
);
1352 int hammer_ioc_rebalance(hammer_transaction_t trans
, hammer_inode_t ip
,
1353 struct hammer_ioc_rebalance
*rebal
);
1354 int hammer_ioc_prune(hammer_transaction_t trans
, hammer_inode_t ip
,
1355 struct hammer_ioc_prune
*prune
);
1356 int hammer_ioc_mirror_read(hammer_transaction_t trans
, hammer_inode_t ip
,
1357 struct hammer_ioc_mirror_rw
*mirror
);
1358 int hammer_ioc_mirror_write(hammer_transaction_t trans
, hammer_inode_t ip
,
1359 struct hammer_ioc_mirror_rw
*mirror
);
1360 int hammer_ioc_set_pseudofs(hammer_transaction_t trans
, hammer_inode_t ip
,
1361 struct ucred
*cred
, struct hammer_ioc_pseudofs_rw
*pfs
);
1362 int hammer_ioc_get_pseudofs(hammer_transaction_t trans
, hammer_inode_t ip
,
1363 struct hammer_ioc_pseudofs_rw
*pfs
);
1364 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans
, hammer_inode_t ip
,
1365 struct hammer_ioc_pseudofs_rw
*pfs
);
1366 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans
, hammer_inode_t ip
,
1367 struct hammer_ioc_pseudofs_rw
*pfs
);
1368 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans
, hammer_inode_t ip
,
1369 struct hammer_ioc_pseudofs_rw
*pfs
);
1370 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans
, hammer_inode_t ip
,
1371 struct hammer_ioc_pseudofs_rw
*pfs
);
1372 int hammer_ioc_volume_add(hammer_transaction_t trans
, hammer_inode_t ip
,
1373 struct hammer_ioc_volume
*ioc
);
1374 int hammer_ioc_volume_del(hammer_transaction_t trans
, hammer_inode_t ip
,
1375 struct hammer_ioc_volume
*ioc
);
1377 int hammer_signal_check(hammer_mount_t hmp
);
1379 void hammer_flusher_create(hammer_mount_t hmp
);
1380 void hammer_flusher_destroy(hammer_mount_t hmp
);
1381 void hammer_flusher_sync(hammer_mount_t hmp
);
1382 int hammer_flusher_async(hammer_mount_t hmp
, hammer_flush_group_t flg
);
1383 int hammer_flusher_async_one(hammer_mount_t hmp
);
1384 void hammer_flusher_wait(hammer_mount_t hmp
, int seq
);
1385 void hammer_flusher_wait_next(hammer_mount_t hmp
);
1386 int hammer_flusher_meta_limit(hammer_mount_t hmp
);
1387 int hammer_flusher_meta_halflimit(hammer_mount_t hmp
);
1388 int hammer_flusher_undo_exhausted(hammer_transaction_t trans
, int quarter
);
1389 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp
);
1390 void hammer_flusher_finalize(hammer_transaction_t trans
, int final
);
1391 int hammer_flusher_haswork(hammer_mount_t hmp
);
1392 void hammer_flusher_flush_undos(hammer_mount_t hmp
, int already_flushed
);
1394 int hammer_recover_stage1(hammer_mount_t hmp
, hammer_volume_t rootvol
);
1395 int hammer_recover_stage2(hammer_mount_t hmp
, hammer_volume_t rootvol
);
1396 void hammer_recover_flush_buffers(hammer_mount_t hmp
,
1397 hammer_volume_t root_volume
, int final
);
1399 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap
);
1400 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk
);
1401 void hammer_crc_set_leaf(void *data
, hammer_btree_leaf_elm_t leaf
);
1403 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap
);
1404 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk
);
1405 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk
);
1406 int hammer_crc_test_leaf(void *data
, hammer_btree_leaf_elm_t leaf
);
1407 void hkprintf(const char *ctl
, ...);
1408 udev_t
hammer_fsid_to_udev(uuid_t
*uuid
);
1411 int hammer_blocksize(int64_t file_offset
);
1412 int hammer_blockoff(int64_t file_offset
);
1413 int64_t hammer_blockdemarc(int64_t file_offset1
, int64_t file_offset2
);
1416 * Shortcut for _hammer_checkspace(), used all over the code.
1419 hammer_checkspace(hammer_mount_t hmp
, int slop
)
1421 return(_hammer_checkspace(hmp
, slop
, NULL
));
1426 static __inline
void
1427 hammer_wait_mem_record(hammer_record_t record
)
1429 hammer_wait_mem_record_ident(record
, "hmmwai");
1432 static __inline
void
1433 hammer_lock_ex(struct hammer_lock
*lock
)
1435 hammer_lock_ex_ident(lock
, "hmrlck");
1439 * Indicate that a B-Tree node is being modified.
1441 static __inline
void
1442 hammer_modify_node_noundo(hammer_transaction_t trans
, hammer_node_t node
)
1444 KKASSERT((node
->flags
& HAMMER_NODE_CRCBAD
) == 0);
1445 hammer_modify_buffer(trans
, node
->buffer
, NULL
, 0);
1448 static __inline
void
1449 hammer_modify_node_all(hammer_transaction_t trans
, struct hammer_node
*node
)
1451 KKASSERT((node
->flags
& HAMMER_NODE_CRCBAD
) == 0);
1452 hammer_modify_buffer(trans
, node
->buffer
,
1453 node
->ondisk
, sizeof(*node
->ondisk
));
1456 static __inline
void
1457 hammer_modify_node(hammer_transaction_t trans
, hammer_node_t node
,
1458 void *base
, int len
)
1460 hammer_crc_t
*crcptr
;
1462 KKASSERT((char *)base
>= (char *)node
->ondisk
&&
1463 (char *)base
+ len
<=
1464 (char *)node
->ondisk
+ sizeof(*node
->ondisk
));
1465 KKASSERT((node
->flags
& HAMMER_NODE_CRCBAD
) == 0);
1466 hammer_modify_buffer(trans
, node
->buffer
, base
, len
);
1467 crcptr
= &node
->ondisk
->crc
;
1468 hammer_modify_buffer(trans
, node
->buffer
, crcptr
, sizeof(hammer_crc_t
));
1469 --node
->buffer
->io
.modify_refs
; /* only want one ref */
1473 * Indicate that the specified modifications have been completed.
1475 * Do not try to generate the crc here, it's very expensive to do and a
1476 * sequence of insertions or deletions can result in many calls to this
1477 * function on the same node.
1479 static __inline
void
1480 hammer_modify_node_done(hammer_node_t node
)
1482 node
->flags
|= HAMMER_NODE_CRCGOOD
;
1483 if ((node
->flags
& HAMMER_NODE_NEEDSCRC
) == 0) {
1484 node
->flags
|= HAMMER_NODE_NEEDSCRC
;
1485 node
->buffer
->io
.gencrc
= 1;
1486 hammer_ref_node(node
);
1488 hammer_modify_buffer_done(node
->buffer
);
1491 #define hammer_modify_volume_field(trans, vol, field) \
1492 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1493 sizeof((vol)->ondisk->field))
1495 #define hammer_modify_node_field(trans, node, field) \
1496 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1497 sizeof((node)->ondisk->field))
1500 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1501 * created directories for HAMMER version 2 or greater and causes
1502 * directory entries to be placed the inode localization zone in
1503 * the B-Tree instead of the misc zone.
1505 * This greatly improves localization between directory entries and
1508 static __inline u_int32_t
1509 hammer_dir_localization(hammer_inode_t dip
)
1511 if (dip
->ino_data
.cap_flags
& HAMMER_INODE_CAP_DIR_LOCAL_INO
)
1512 return(HAMMER_LOCALIZE_INODE
);
1514 return(HAMMER_LOCALIZE_MISC
);