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>
63 #include <sys/signal2.h>
64 #include <sys/mplock2.h>
65 #include "hammer_disk.h"
66 #include "hammer_mount.h"
67 #include "hammer_ioctl.h"
69 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
71 MALLOC_DECLARE(M_HAMMER);
76 #if !defined(KTR_HAMMER)
77 #define KTR_HAMMER KTR_ALL
79 KTR_INFO_MASTER_EXTERN(hammer);
87 * Key structure used for custom RB tree inode lookups. This prototypes
88 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
90 typedef struct hammer_inode_info {
91 int64_t obj_id; /* (key) object identifier */
92 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
93 u_int32_t obj_localization; /* (key) pseudo-fs */
95 struct hammer_btree_leaf_elm *leaf;
97 } *hammer_inode_info_t;
99 typedef enum hammer_transaction_type {
103 } hammer_transaction_type_t;
106 * HAMMER Transaction tracking
108 struct hammer_transaction {
109 hammer_transaction_type_t type;
110 struct hammer_mount *hmp;
116 struct hammer_volume *rootvol;
119 typedef struct hammer_transaction *hammer_transaction_t;
121 #define HAMMER_TRANSF_NEWINODE 0x0001
122 #define HAMMER_TRANSF_DIDIO 0x0002
123 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */
129 int refs; /* active references delay writes */
130 volatile u_int lockval; /* lock count and control bits */
131 struct thread *owner; /* owner if exclusively held */
134 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000
135 #define HAMMER_LOCKF_WANTED 0x80000000
138 hammer_notlocked(struct hammer_lock *lock)
140 return(lock->lockval == 0);
144 hammer_islocked(struct hammer_lock *lock)
146 return(lock->lockval != 0);
150 hammer_isactive(struct hammer_lock *lock)
152 return(lock->refs != 0);
156 hammer_islastref(struct hammer_lock *lock)
158 return(lock->refs == 1);
162 * Return if we specifically own the lock exclusively.
165 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
167 if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
175 * Flush state, used by various structures
177 typedef enum hammer_inode_state {
181 } hammer_inode_state_t;
183 TAILQ_HEAD(hammer_record_list, hammer_record);
186 * Pseudo-filesystem extended data tracking
188 struct hammer_pfs_rb_tree;
189 struct hammer_pseudofs_inmem;
190 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
191 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
192 hammer_pfs_rb_compare, u_int32_t);
194 struct hammer_pseudofs_inmem {
195 RB_ENTRY(hammer_pseudofs_inmem) rb_node;
196 struct hammer_lock lock;
197 u_int32_t localization;
198 hammer_tid_t create_tid;
201 struct hammer_pseudofs_data pfsd;
204 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
206 #define HAMMER_PFSM_DELETED 0x0001
209 * Cache object ids. A fixed number of objid cache structures are
210 * created to reserve object id's for newly created files in multiples
211 * of 100,000, localized to a particular directory, and recycled as
212 * needed. This allows parallel create operations in different
213 * directories to retain fairly localized object ids which in turn
214 * improves reblocking performance and layout.
216 #define OBJID_CACHE_SIZE 1024
217 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */
218 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */
219 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1)
220 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
222 typedef struct hammer_objid_cache {
223 TAILQ_ENTRY(hammer_objid_cache) entry;
224 struct hammer_inode *dip;
225 hammer_tid_t base_tid;
229 } *hammer_objid_cache_t;
232 * Associate an inode with a B-Tree node to cache search start positions
234 typedef struct hammer_node_cache {
235 TAILQ_ENTRY(hammer_node_cache) entry;
236 struct hammer_node *node;
237 struct hammer_inode *ip;
238 } *hammer_node_cache_t;
240 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
243 * Structure used to organize flush groups. Flush groups must be
244 * organized into chunks in order to avoid blowing out the UNDO FIFO.
245 * Without this a 'sync' could end up flushing 50,000 inodes in a single
248 struct hammer_fls_rb_tree;
249 RB_HEAD(hammer_fls_rb_tree, hammer_inode);
250 RB_PROTOTYPE(hammer_fls_rb_tree, hammer_inode, rb_flsnode,
251 hammer_ino_rb_compare);
253 struct hammer_flush_group {
254 TAILQ_ENTRY(hammer_flush_group) flush_entry;
255 struct hammer_fls_rb_tree flush_tree;
257 int total_count; /* record load */
258 int running; /* group is running */
263 typedef struct hammer_flush_group *hammer_flush_group_t;
265 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
268 * Structure used to represent an inode in-memory.
270 * The record and data associated with an inode may be out of sync with
271 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
274 * An inode may also hold a cache of unsynchronized records, used for
275 * database and directories only. Unsynchronized regular file data is
276 * stored in the buffer cache.
278 * NOTE: A file which is created and destroyed within the initial
279 * synchronization period can wind up not doing any disk I/O at all.
281 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
283 struct hammer_ino_rb_tree;
285 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
286 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
287 hammer_ino_rb_compare, hammer_inode_info_t);
289 struct hammer_redo_rb_tree;
290 RB_HEAD(hammer_redo_rb_tree, hammer_inode);
291 RB_PROTOTYPE2(hammer_redo_rb_tree, hammer_inode, rb_redonode,
292 hammer_redo_rb_compare, hammer_off_t);
294 struct hammer_rec_rb_tree;
295 struct hammer_record;
296 RB_HEAD(hammer_rec_rb_tree, hammer_record);
297 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
298 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
300 TAILQ_HEAD(hammer_node_list, hammer_node);
302 struct hammer_inode {
303 RB_ENTRY(hammer_inode) rb_node;
304 hammer_inode_state_t flush_state;
305 hammer_flush_group_t flush_group;
306 RB_ENTRY(hammer_inode) rb_flsnode; /* when on flush list */
307 RB_ENTRY(hammer_inode) rb_redonode; /* when INODE_RDIRTY is set */
308 struct hammer_record_list target_list; /* target of dependant recs */
309 int64_t obj_id; /* (key) object identifier */
310 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
311 u_int32_t obj_localization; /* (key) pseudo-fs */
312 struct hammer_mount *hmp;
313 hammer_objid_cache_t objid_cache;
315 int error; /* flush error */
316 int cursor_ip_refs; /* sanity */
319 hammer_pseudofs_inmem_t pfsm;
320 struct lockf advlock;
321 struct hammer_lock lock; /* sync copy interlock */
323 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
324 struct hammer_inode_data ino_data; /* in-memory cache */
325 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
327 struct hammer_node_cache cache[4]; /* search initiate cache */
330 * When a demark is created to synchronize an inode to
331 * disk, certain fields are copied so the front-end VOPs
332 * can continue to run in parallel with the synchronization
333 * occuring in the background.
335 int sync_flags; /* to-sync flags cache */
336 off_t sync_trunc_off; /* to-sync truncation */
337 off_t save_trunc_off; /* write optimization */
338 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
339 struct hammer_inode_data sync_ino_data; /* to-sync cache */
343 * Track the earliest offset in the UNDO/REDO FIFO containing
344 * REDO records. This is staged to the backend during flush
345 * sequences. While the inode is staged redo_fifo_next is used
346 * to track the earliest offset for rotation into redo_fifo_start
347 * on completion of the flush.
349 hammer_off_t redo_fifo_start;
350 hammer_off_t redo_fifo_next;
353 typedef struct hammer_inode *hammer_inode_t;
355 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
358 * NOTE: DDIRTY does not include atime or mtime and does not include
359 * write-append size changes. SDIRTY handles write-append size
362 * REDO indicates that REDO logging is active, creating a definitive
363 * stream of REDO records in the UNDO/REDO log for writes and
364 * truncations, including boundary records when/if REDO is turned off.
365 * REDO is typically enabled by fsync() and turned off if excessive
366 * writes without an fsync() occurs.
368 * RDIRTY indicates that REDO records were laid down in the UNDO/REDO
369 * FIFO (even if REDO is turned off some might still be active) and
370 * still being tracked for this inode. See hammer_redo.c
372 /* (not including atime/mtime) */
373 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
374 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
375 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */
376 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
377 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
378 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
379 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
380 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
381 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
382 #define HAMMER_INODE_UNUSED0400 0x0400
383 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
384 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
385 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */
386 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
387 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
389 #define HAMMER_INODE_TRUNCATED 0x00010000
390 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
391 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
392 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
393 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
394 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
395 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */
396 #define HAMMER_INODE_SDIRTY 0x01000000 /* in-memory ino_data.size is dirty*/
397 #define HAMMER_INODE_REDO 0x02000000 /* REDO logging active */
398 #define HAMMER_INODE_RDIRTY 0x04000000 /* REDO records active in fifo */
400 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY|HAMMER_INODE_SDIRTY| \
401 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
402 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
403 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
405 #define HAMMER_INODE_MODMASK_NOXDIRTY \
406 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
408 #define HAMMER_INODE_MODMASK_NOREDO \
409 (HAMMER_INODE_DDIRTY| \
410 HAMMER_INODE_XDIRTY| \
411 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
413 #define HAMMER_FLUSH_SIGNAL 0x0001
414 #define HAMMER_FLUSH_RECURSION 0x0002
417 * Used by the inode reclaim code to pipeline reclaims and avoid
418 * blowing out kernel memory or letting the flusher get too far
419 * behind. The reclaim wakes up when count reaches 0 or the
422 struct hammer_reclaim {
423 TAILQ_ENTRY(hammer_reclaim) entry;
427 #define HAMMER_RECLAIM_WAIT 4000 /* default vfs.hammer.limit_reclaim */
430 * Track who is creating the greatest burden on the
433 struct hammer_inostats {
434 pid_t pid; /* track user process */
435 int ltick; /* last tick */
436 int count; /* count (degenerates) */
439 #define HAMMER_INOSTATS_HSIZE 32
440 #define HAMMER_INOSTATS_HMASK (HAMMER_INOSTATS_HSIZE - 1)
443 * Structure used to represent an unsynchronized record in-memory. These
444 * records typically represent directory entries. Only non-historical
445 * records are kept in-memory.
447 * Records are organized as a per-inode RB-Tree. If the inode is not
448 * on disk then neither are any records and the in-memory record tree
449 * represents the entire contents of the inode. If the inode is on disk
450 * then the on-disk B-Tree is scanned in parallel with the in-memory
451 * RB-Tree to synthesize the current state of the file.
453 * Records are also used to enforce the ordering of directory create/delete
454 * operations. A new inode will not be flushed to disk unless its related
455 * directory entry is also being flushed at the same time. A directory entry
456 * will not be removed unless its related inode is also being removed at the
459 typedef enum hammer_record_type {
460 HAMMER_MEM_RECORD_GENERAL, /* misc record */
461 HAMMER_MEM_RECORD_INODE, /* inode record */
462 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
463 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
464 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
465 } hammer_record_type_t;
467 struct hammer_record {
468 RB_ENTRY(hammer_record) rb_node;
469 TAILQ_ENTRY(hammer_record) target_entry;
470 hammer_inode_state_t flush_state;
471 hammer_flush_group_t flush_group;
472 hammer_record_type_t type;
473 struct hammer_lock lock;
474 struct hammer_reserve *resv;
475 struct hammer_inode *ip;
476 struct hammer_inode *target_ip;
477 struct hammer_btree_leaf_elm leaf;
478 union hammer_data_ondisk *data;
480 hammer_off_t zone2_offset; /* direct-write only */
483 typedef struct hammer_record *hammer_record_t;
486 * Record flags. Note that FE can only be set by the frontend if the
487 * record has not been interlocked by the backend w/ BE.
489 #define HAMMER_RECF_ALLOCDATA 0x0001
490 #define HAMMER_RECF_ONRBTREE 0x0002
491 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
492 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
493 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */
494 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
495 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
496 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
497 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/
498 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/
499 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */
500 #define HAMMER_RECF_REDO 0x1000 /* REDO was laid down */
503 * hammer_create_at_cursor() and hammer_delete_at_cursor() flags.
505 #define HAMMER_CREATE_MODE_UMIRROR 0x0001
506 #define HAMMER_CREATE_MODE_SYS 0x0002
508 #define HAMMER_DELETE_ADJUST 0x0001
509 #define HAMMER_DELETE_DESTROY 0x0002
512 * In-memory structures representing on-disk structures.
514 struct hammer_volume;
515 struct hammer_buffer;
518 struct hammer_reserve;
520 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
521 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
522 RB_HEAD(hammer_nod_rb_tree, hammer_node);
523 RB_HEAD(hammer_und_rb_tree, hammer_undo);
524 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
526 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
527 hammer_vol_rb_compare, int32_t);
528 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
529 hammer_buf_rb_compare, hammer_off_t);
530 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
531 hammer_nod_rb_compare, hammer_off_t);
532 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
533 hammer_und_rb_compare, hammer_off_t);
534 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
535 hammer_res_rb_compare, hammer_off_t);
538 * IO management - embedded at the head of various in-memory structures
540 * VOLUME - hammer_volume containing meta-data
541 * META_BUFFER - hammer_buffer containing meta-data
542 * DATA_BUFFER - hammer_buffer containing pure-data
544 * Dirty volume headers and dirty meta-data buffers are locked until the
545 * flusher can sequence them out. Dirty pure-data buffers can be written.
546 * Clean buffers can be passively released.
548 typedef enum hammer_io_type {
549 HAMMER_STRUCTURE_VOLUME,
550 HAMMER_STRUCTURE_META_BUFFER,
551 HAMMER_STRUCTURE_UNDO_BUFFER,
552 HAMMER_STRUCTURE_DATA_BUFFER,
553 HAMMER_STRUCTURE_DUMMY
556 union hammer_io_structure;
560 LIST_ENTRY(worklist) node;
563 TAILQ_HEAD(hammer_io_list, hammer_io);
564 typedef struct hammer_io_list *hammer_io_list_t;
567 struct worklist worklist;
568 struct hammer_lock lock;
569 enum hammer_io_type type;
570 struct hammer_mount *hmp;
571 struct hammer_volume *volume;
572 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */
573 TAILQ_ENTRY(hammer_io) iorun_entry; /* iorun_list */
574 hammer_io_list_t mod_list;
576 int64_t offset; /* zone-2 offset */
577 int bytes; /* buffer cache buffer size */
578 int loading; /* loading/unloading interlock */
581 u_int modified : 1; /* bp's data was modified */
582 u_int released : 1; /* bp released (w/ B_LOCKED set) */
583 u_int running : 1; /* bp write IO in progress */
584 u_int waiting : 1; /* someone is waiting on us */
585 u_int validated : 1; /* ondisk has been validated */
586 u_int waitdep : 1; /* flush waits for dependancies */
587 u_int recovered : 1; /* has recovery ref */
588 u_int waitmod : 1; /* waiting for modify_refs */
589 u_int reclaim : 1; /* reclaim requested */
590 u_int gencrc : 1; /* crc needs to be generated */
591 u_int ioerror : 1; /* abort on io-error */
594 typedef struct hammer_io *hammer_io_t;
596 #define HAMMER_CLUSTER_SIZE (64 * 1024)
597 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
598 #undef HAMMER_CLUSTER_SIZE
599 #define HAMMER_CLUSTER_SIZE MAXBSIZE
601 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
604 * In-memory volume representing on-disk buffer
606 struct hammer_volume {
608 RB_ENTRY(hammer_volume) rb_node;
609 struct hammer_volume_ondisk *ondisk;
611 int64_t nblocks; /* note: special calculation for statfs */
612 int64_t buffer_base; /* base offset of buffer 0 */
613 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
614 hammer_off_t maxraw_off; /* Maximum raw offset for device */
620 typedef struct hammer_volume *hammer_volume_t;
623 * In-memory buffer (other then volume, super-cluster, or cluster),
624 * representing an on-disk buffer.
626 struct hammer_buffer {
628 RB_ENTRY(hammer_buffer) rb_node;
630 hammer_off_t zoneX_offset;
631 hammer_off_t zone2_offset;
632 struct hammer_reserve *resv;
633 struct hammer_node_list clist;
636 typedef struct hammer_buffer *hammer_buffer_t;
639 * In-memory B-Tree node, representing an on-disk B-Tree node.
641 * This is a hang-on structure which is backed by a hammer_buffer,
642 * indexed by a hammer_cluster, and used for fine-grained locking of
643 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
644 * can contain multiple nodes representing wildly disassociated portions
645 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
647 * This structure uses a cluster-relative index to reduce the number
648 * of layers required to access it, and also because all on-disk B-Tree
649 * references are cluster-relative offsets.
652 struct hammer_lock lock; /* node-by-node lock */
653 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
654 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */
655 hammer_off_t node_offset; /* full offset spec */
656 struct hammer_mount *hmp;
657 struct hammer_buffer *buffer; /* backing buffer */
658 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
659 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */
660 struct hammer_node_cache_list cache_list; /* passive caches */
662 int loading; /* load interlock */
665 #define HAMMER_NODE_DELETED 0x0001
666 #define HAMMER_NODE_FLUSH 0x0002
667 #define HAMMER_NODE_CRCGOOD 0x0004
668 #define HAMMER_NODE_NEEDSCRC 0x0008
669 #define HAMMER_NODE_NEEDSMIRROR 0x0010
670 #define HAMMER_NODE_CRCBAD 0x0020
671 #define HAMMER_NODE_NONLINEAR 0x0040 /* linear heuristic */
673 #define HAMMER_NODE_CRCANY (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
675 typedef struct hammer_node *hammer_node_t;
678 * List of locked nodes. This structure is used to lock potentially large
679 * numbers of nodes as an aid for complex B-Tree operations.
681 struct hammer_node_lock;
682 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
684 struct hammer_node_lock {
685 TAILQ_ENTRY(hammer_node_lock) entry;
686 struct hammer_node_lock_list list;
687 struct hammer_node_lock *parent;
689 hammer_node_ondisk_t copy; /* copy of on-disk data */
690 int index; /* index of this node in parent */
691 int count; /* count children */
695 typedef struct hammer_node_lock *hammer_node_lock_t;
697 #define HAMMER_NODE_LOCK_UPDATED 0x0001
700 * Common I/O management structure - embedded in in-memory structures
701 * which are backed by filesystem buffers.
703 union hammer_io_structure {
705 struct hammer_volume volume;
706 struct hammer_buffer buffer;
709 typedef union hammer_io_structure *hammer_io_structure_t;
712 * The reserve structure prevents the blockmap from allocating
713 * out of a reserved bigblock. Such reservations are used by
714 * the direct-write mechanism.
716 * The structure is also used to hold off on reallocations of
717 * big blocks from the freemap until flush dependancies have
720 struct hammer_reserve {
721 RB_ENTRY(hammer_reserve) rb_node;
722 TAILQ_ENTRY(hammer_reserve) delay_entry;
728 hammer_off_t zone_offset;
731 typedef struct hammer_reserve *hammer_reserve_t;
733 #define HAMMER_RESF_ONDELAY 0x0001
734 #define HAMMER_RESF_LAYER2FREE 0x0002
736 #include "hammer_cursor.h"
739 * The undo structure tracks recent undos to avoid laying down duplicate
740 * undos within a flush group, saving us a significant amount of overhead.
742 * This is strictly a heuristic.
744 #define HAMMER_MAX_UNDOS 1024
745 #define HAMMER_MAX_FLUSHERS 4
748 RB_ENTRY(hammer_undo) rb_node;
749 TAILQ_ENTRY(hammer_undo) lru_entry;
754 typedef struct hammer_undo *hammer_undo_t;
756 struct hammer_flusher_info;
757 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
759 struct hammer_flusher {
760 int signal; /* flusher thread sequencer */
761 int act; /* currently active flush group */
762 int done; /* set to act when complete */
763 int next; /* next flush group */
764 int group_lock; /* lock sequencing of the next flush */
765 int exiting; /* request master exit */
766 thread_t td; /* master flusher thread */
767 hammer_tid_t tid; /* last flushed transaction id */
768 int finalize_want; /* serialize finalization */
769 struct hammer_lock finalize_lock; /* serialize finalization */
770 struct hammer_transaction trans; /* shared transaction */
771 struct hammer_flusher_info_list run_list;
772 struct hammer_flusher_info_list ready_list;
775 #define HAMMER_FLUSH_UNDOS_RELAXED 0
776 #define HAMMER_FLUSH_UNDOS_FORCED 1
777 #define HAMMER_FLUSH_UNDOS_AUTO 2
779 * Internal hammer mount data structure
781 struct hammer_mount {
783 /*struct vnode *rootvp;*/
784 struct hammer_ino_rb_tree rb_inos_root;
785 struct hammer_redo_rb_tree rb_redo_root;
786 struct hammer_vol_rb_tree rb_vols_root;
787 struct hammer_nod_rb_tree rb_nods_root;
788 struct hammer_und_rb_tree rb_undo_root;
789 struct hammer_res_rb_tree rb_resv_root;
790 struct hammer_buf_rb_tree rb_bufs_root;
791 struct hammer_pfs_rb_tree rb_pfsm_root;
792 struct hammer_volume *rootvol;
793 struct hammer_base_elm root_btree_beg;
794 struct hammer_base_elm root_btree_end;
796 struct malloc_type *m_misc;
797 struct malloc_type *m_inodes;
799 int flags; /* HAMMER_MOUNT_xxx flags */
804 int master_id; /* -1 or 0-15 - clustering and mirroring */
805 int version; /* hammer filesystem version to use */
806 int rsv_inodes; /* reserved space due to dirty inodes */
807 int64_t rsv_databytes; /* reserved space due to record data */
808 int rsv_recs; /* reserved space due to dirty records */
809 int rsv_fromdelay; /* bigblocks reserved due to flush delay */
810 int undo_rec_limit; /* based on size of undo area */
812 int count_newrecords;
814 int volume_to_remove; /* volume that is currently being removed */
816 int inode_reclaims; /* inodes pending reclaim by flusher */
817 int count_inodes; /* total number of inodes */
818 int count_iqueued; /* inodes queued to flusher */
820 struct hammer_flusher flusher;
822 u_int check_interrupt;
825 struct hammer_io_list volu_list; /* dirty undo buffers */
826 struct hammer_io_list undo_list; /* dirty undo buffers */
827 struct hammer_io_list data_list; /* dirty data buffers */
828 struct hammer_io_list alt_data_list; /* dirty data buffers */
829 struct hammer_io_list meta_list; /* dirty meta bufs */
830 struct hammer_io_list lose_list; /* loose buffers */
831 int locked_dirty_space; /* meta/volu count */
832 int io_running_space;
833 int objid_cache_count;
834 int error; /* critical I/O error */
835 struct krate krate; /* rate limited kprintf */
836 hammer_tid_t asof; /* snapshot mount */
837 hammer_tid_t next_tid;
838 hammer_tid_t flush_tid1; /* flusher tid sequencing */
839 hammer_tid_t flush_tid2; /* flusher tid sequencing */
840 int64_t copy_stat_freebigblocks; /* number of free bigblocks */
841 u_int32_t undo_seqno; /* UNDO/REDO FIFO seqno */
843 struct netexport export;
844 struct hammer_lock sync_lock;
845 struct hammer_lock free_lock;
846 struct hammer_lock undo_lock;
847 struct hammer_lock blkmap_lock;
848 struct hammer_lock snapshot_lock;
849 struct hammer_lock volume_lock;
850 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
851 struct hammer_undo undos[HAMMER_MAX_UNDOS];
853 TAILQ_HEAD(, hammer_undo) undo_lru_list;
854 TAILQ_HEAD(, hammer_reserve) delay_list;
855 struct hammer_flush_group_list flush_group_list;
856 hammer_flush_group_t next_flush_group;
857 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
858 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
859 TAILQ_HEAD(, hammer_io) iorun_list;
861 struct hammer_inostats inostats[HAMMER_INOSTATS_HSIZE];
864 typedef struct hammer_mount *hammer_mount_t;
866 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001
867 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002
868 #define HAMMER_MOUNT_REDO_SYNC 0x0004
870 struct hammer_sync_info {
876 * Minium buffer cache bufs required to rebalance the B-Tree.
877 * This is because we must hold the children and the children's children
878 * locked. Even this might not be enough if things are horribly out
881 #define HAMMER_REBALANCE_MIN_BUFS \
882 (HAMMER_BTREE_LEAF_ELMS * HAMMER_BTREE_LEAF_ELMS)
888 * checkspace slop (8MB chunks), higher numbers are more conservative.
890 #define HAMMER_CHKSPC_REBLOCK 25
891 #define HAMMER_CHKSPC_MIRROR 20
892 #define HAMMER_CHKSPC_WRITE 20
893 #define HAMMER_CHKSPC_CREATE 20
894 #define HAMMER_CHKSPC_REMOVE 10
895 #define HAMMER_CHKSPC_EMERGENCY 0
899 extern struct vop_ops hammer_vnode_vops;
900 extern struct vop_ops hammer_spec_vops;
901 extern struct vop_ops hammer_fifo_vops;
902 extern struct bio_ops hammer_bioops;
904 extern int hammer_debug_io;
905 extern int hammer_debug_general;
906 extern int hammer_debug_debug;
907 extern int hammer_debug_inode;
908 extern int hammer_debug_locks;
909 extern int hammer_debug_btree;
910 extern int hammer_debug_tid;
911 extern int hammer_debug_recover;
912 extern int hammer_debug_recover_faults;
913 extern int hammer_debug_critical;
914 extern int hammer_cluster_enable;
915 extern int hammer_count_fsyncs;
916 extern int hammer_count_inodes;
917 extern int hammer_count_iqueued;
918 extern int hammer_count_reclaiming;
919 extern int hammer_count_records;
920 extern int hammer_count_record_datas;
921 extern int hammer_count_volumes;
922 extern int hammer_count_buffers;
923 extern int hammer_count_nodes;
924 extern int64_t hammer_count_extra_space_used;
925 extern int64_t hammer_stats_btree_lookups;
926 extern int64_t hammer_stats_btree_searches;
927 extern int64_t hammer_stats_btree_inserts;
928 extern int64_t hammer_stats_btree_deletes;
929 extern int64_t hammer_stats_btree_elements;
930 extern int64_t hammer_stats_btree_splits;
931 extern int64_t hammer_stats_btree_iterations;
932 extern int64_t hammer_stats_btree_root_iterations;
933 extern int64_t hammer_stats_record_iterations;
934 extern int64_t hammer_stats_file_read;
935 extern int64_t hammer_stats_file_write;
936 extern int64_t hammer_stats_file_iopsr;
937 extern int64_t hammer_stats_file_iopsw;
938 extern int64_t hammer_stats_disk_read;
939 extern int64_t hammer_stats_disk_write;
940 extern int64_t hammer_stats_inode_flushes;
941 extern int64_t hammer_stats_commits;
942 extern int64_t hammer_stats_undo;
943 extern int64_t hammer_stats_redo;
944 extern int hammer_count_dirtybufspace;
945 extern int hammer_count_refedbufs;
946 extern int hammer_count_reservations;
947 extern int hammer_count_io_running_read;
948 extern int hammer_count_io_running_write;
949 extern int hammer_count_io_locked;
950 extern int hammer_limit_dirtybufspace;
951 extern int hammer_limit_recs;
952 extern int hammer_limit_inode_recs;
953 extern int hammer_limit_reclaim;
954 extern int hammer_limit_redo;
955 extern int hammer_bio_count;
956 extern int hammer_verify_zone;
957 extern int hammer_verify_data;
958 extern int hammer_write_mode;
959 extern int hammer_yield_check;
960 extern int hammer_fsync_mode;
961 extern int hammer_autoflush;
962 extern int64_t hammer_contention_count;
964 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
965 int error, const char *msg);
966 int hammer_vop_inactive(struct vop_inactive_args *);
967 int hammer_vop_reclaim(struct vop_reclaim_args *);
968 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
969 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
970 hammer_inode_t dip, int64_t obj_id,
971 hammer_tid_t asof, u_int32_t localization,
972 int flags, int *errorp);
973 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
974 hammer_inode_t dip, int64_t obj_id,
975 hammer_tid_t asof, u_int32_t localization,
976 int flags, int *errorp);
977 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
978 int64_t obj_id, hammer_tid_t asof,
979 u_int32_t localization);
980 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
981 hammer_inode_info_t iinfo,
982 int (*callback)(hammer_inode_t ip, void *data),
984 void hammer_put_inode(struct hammer_inode *ip);
985 void hammer_put_inode_ref(struct hammer_inode *ip);
986 void hammer_inode_waitreclaims(hammer_transaction_t trans);
988 int hammer_unload_volume(hammer_volume_t volume, void *data __unused);
989 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
991 int hammer_unload_buffer(hammer_buffer_t buffer, void *data);
992 int hammer_install_volume(hammer_mount_t hmp, const char *volname,
993 struct vnode *devvp);
994 int hammer_mountcheck_volumes(hammer_mount_t hmp);
996 int hammer_mem_add(hammer_record_t record);
997 int hammer_ip_lookup(hammer_cursor_t cursor);
998 int hammer_ip_first(hammer_cursor_t cursor);
999 int hammer_ip_next(hammer_cursor_t cursor);
1000 int hammer_ip_resolve_data(hammer_cursor_t cursor);
1001 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
1003 int hammer_create_at_cursor(hammer_cursor_t cursor,
1004 hammer_btree_leaf_elm_t leaf, void *udata, int mode);
1005 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
1006 hammer_tid_t delete_tid, u_int32_t delete_ts,
1007 int track, int64_t *stat_bytes);
1008 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
1010 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
1011 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
1014 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
1015 void hammer_flush_record_done(hammer_record_t record, int error);
1016 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
1017 void hammer_rel_mem_record(hammer_record_t record);
1019 int hammer_cursor_up(hammer_cursor_t cursor);
1020 int hammer_cursor_up_locked(hammer_cursor_t cursor);
1021 int hammer_cursor_down(hammer_cursor_t cursor);
1022 int hammer_cursor_upgrade(hammer_cursor_t cursor);
1023 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
1024 void hammer_cursor_downgrade(hammer_cursor_t cursor);
1025 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
1027 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
1028 int hammer_lock_ex_try(struct hammer_lock *lock);
1029 void hammer_lock_sh(struct hammer_lock *lock);
1030 int hammer_lock_sh_try(struct hammer_lock *lock);
1031 int hammer_lock_upgrade(struct hammer_lock *lock);
1032 void hammer_lock_downgrade(struct hammer_lock *lock);
1033 int hammer_lock_status(struct hammer_lock *lock);
1034 void hammer_unlock(struct hammer_lock *lock);
1035 void hammer_ref(struct hammer_lock *lock);
1036 void hammer_unref(struct hammer_lock *lock);
1038 void hammer_sync_lock_ex(hammer_transaction_t trans);
1039 void hammer_sync_lock_sh(hammer_transaction_t trans);
1040 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
1041 void hammer_sync_unlock(hammer_transaction_t trans);
1043 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
1044 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1045 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1046 u_int64_t hammer_timespec_to_time(struct timespec *ts);
1047 int hammer_str_to_tid(const char *str, int *ispfsp,
1048 hammer_tid_t *tidp, u_int32_t *localizationp);
1049 int hammer_is_atatext(const char *name, int len);
1050 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
1052 void hammer_clear_objid(hammer_inode_t dip);
1053 void hammer_destroy_objid_cache(hammer_mount_t hmp);
1055 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
1057 void hammer_clear_undo_history(hammer_mount_t hmp);
1058 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
1059 int hammer_get_dtype(u_int8_t obj_type);
1060 u_int8_t hammer_get_obj_type(enum vtype vtype);
1061 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
1062 u_int32_t *max_iterationsp);
1063 int hammer_nohistory(hammer_inode_t ip);
1065 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
1066 hammer_node_cache_t cache, hammer_inode_t ip);
1067 void hammer_normalize_cursor(hammer_cursor_t cursor);
1068 void hammer_done_cursor(hammer_cursor_t cursor);
1069 int hammer_recover_cursor(hammer_cursor_t cursor);
1070 void hammer_unlock_cursor(hammer_cursor_t cursor);
1071 int hammer_lock_cursor(hammer_cursor_t cursor);
1072 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
1073 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
1075 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
1076 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
1078 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
1080 void hammer_cursor_moved_element(hammer_node_t oparent, int pindex,
1081 hammer_node_t onode, int oindex,
1082 hammer_node_t nnode, int nindex);
1083 void hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
1084 hammer_node_t nparent, int nindex);
1085 void hammer_cursor_inserted_element(hammer_node_t node, int index);
1086 void hammer_cursor_deleted_element(hammer_node_t node, int index);
1088 int hammer_btree_lookup(hammer_cursor_t cursor);
1089 int hammer_btree_first(hammer_cursor_t cursor);
1090 int hammer_btree_last(hammer_cursor_t cursor);
1091 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
1092 int hammer_btree_iterate(hammer_cursor_t cursor);
1093 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
1094 int hammer_btree_insert(hammer_cursor_t cursor,
1095 hammer_btree_leaf_elm_t elm, int *doprop);
1096 int hammer_btree_delete(hammer_cursor_t cursor);
1097 void hammer_btree_do_propagation(hammer_cursor_t cursor,
1098 hammer_pseudofs_inmem_t pfsm,
1099 hammer_btree_leaf_elm_t leaf);
1100 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1101 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1102 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1103 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1105 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1106 hammer_btree_elm_t elm);
1107 void hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1108 int hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1109 hammer_node_lock_t parent);
1110 void hammer_btree_lock_copy(hammer_cursor_t cursor,
1111 hammer_node_lock_t parent);
1112 int hammer_btree_sync_copy(hammer_cursor_t cursor,
1113 hammer_node_lock_t parent);
1114 void hammer_btree_unlock_children(hammer_cursor_t cursor,
1115 hammer_node_lock_t parent);
1116 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1117 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1118 hammer_node_t node, int *parent_indexp,
1119 int *errorp, int try_exclusive);
1121 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1122 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1124 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1125 int *errorp, struct hammer_buffer **bufferp);
1126 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1127 int *errorp, struct hammer_buffer **bufferp);
1128 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1129 int *errorp, struct hammer_buffer **bufferp);
1130 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1131 int *errorp, struct hammer_buffer **bufferp);
1133 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1135 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1136 int32_t vol_no, int *errorp);
1137 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1138 int bytes, int isnew, int *errorp);
1139 void hammer_sync_buffers(hammer_mount_t hmp,
1140 hammer_off_t base_offset, int bytes);
1141 int hammer_del_buffers(hammer_mount_t hmp,
1142 hammer_off_t base_offset,
1143 hammer_off_t zone2_offset, int bytes,
1144 int report_conflicts);
1146 int hammer_ref_volume(hammer_volume_t volume);
1147 int hammer_ref_buffer(hammer_buffer_t buffer);
1148 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1150 void hammer_rel_volume(hammer_volume_t volume, int flush);
1151 void hammer_rel_buffer(hammer_buffer_t buffer, int flush);
1153 int hammer_vfs_export(struct mount *mp, int op,
1154 const struct export_args *export);
1155 hammer_node_t hammer_get_node(hammer_transaction_t trans,
1156 hammer_off_t node_offset, int isnew, int *errorp);
1157 void hammer_ref_node(hammer_node_t node);
1158 hammer_node_t hammer_ref_node_safe(hammer_transaction_t trans,
1159 hammer_node_cache_t cache, int *errorp);
1160 void hammer_rel_node(hammer_node_t node);
1161 void hammer_delete_node(hammer_transaction_t trans,
1162 hammer_node_t node);
1163 void hammer_cache_node(hammer_node_cache_t cache,
1164 hammer_node_t node);
1165 void hammer_uncache_node(hammer_node_cache_t cache);
1166 void hammer_flush_node(hammer_node_t node);
1168 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1169 struct hammer_buffer *buffer);
1170 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1171 hammer_off_t hint, int *errorp);
1172 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1173 u_int16_t rec_type, hammer_off_t *data_offsetp,
1174 struct hammer_buffer **data_bufferp,
1175 hammer_off_t hint, int *errorp);
1177 int hammer_generate_undo(hammer_transaction_t trans,
1178 hammer_off_t zone1_offset, void *base, int len);
1179 int hammer_generate_redo(hammer_transaction_t trans, hammer_inode_t ip,
1180 hammer_off_t file_offset, u_int32_t flags,
1181 void *base, int len);
1182 void hammer_generate_redo_sync(hammer_transaction_t trans);
1183 void hammer_redo_fifo_start_flush(hammer_inode_t ip);
1184 void hammer_redo_fifo_end_flush(hammer_inode_t ip);
1186 void hammer_format_undo(void *base, u_int32_t seqno);
1187 int hammer_upgrade_undo_4(hammer_transaction_t trans);
1189 void hammer_put_volume(struct hammer_volume *volume, int flush);
1190 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1192 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1193 hammer_off_t owner, int *errorp);
1194 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1195 hammer_off_t owner, int *errorp);
1196 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1197 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1198 int bytes, hammer_off_t hint, int *errorp);
1199 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1200 int bytes, hammer_off_t *zone_offp, int *errorp);
1201 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1202 hammer_reserve_t resv);
1203 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1204 void hammer_blockmap_free(hammer_transaction_t trans,
1205 hammer_off_t bmap_off, int bytes);
1206 int hammer_blockmap_finalize(hammer_transaction_t trans,
1207 hammer_reserve_t resv,
1208 hammer_off_t bmap_off, int bytes);
1209 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1210 int *curp, int *errorp);
1211 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1213 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1215 int64_t hammer_undo_used(hammer_transaction_t trans);
1216 int64_t hammer_undo_space(hammer_transaction_t trans);
1217 int64_t hammer_undo_max(hammer_mount_t hmp);
1218 int hammer_undo_reclaim(hammer_io_t io);
1220 void hammer_start_transaction(struct hammer_transaction *trans,
1221 struct hammer_mount *hmp);
1222 void hammer_simple_transaction(struct hammer_transaction *trans,
1223 struct hammer_mount *hmp);
1224 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1225 struct hammer_mount *hmp);
1226 void hammer_done_transaction(struct hammer_transaction *trans);
1227 hammer_tid_t hammer_alloc_tid(hammer_mount_t hmp, int count);
1229 void hammer_modify_inode(hammer_transaction_t trans, hammer_inode_t ip, int flags);
1230 void hammer_flush_inode(hammer_inode_t ip, int flags);
1231 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1232 void hammer_wait_inode(hammer_inode_t ip);
1234 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1235 struct ucred *cred, struct hammer_inode *dip,
1236 const char *name, int namelen,
1237 hammer_pseudofs_inmem_t pfsm,
1238 struct hammer_inode **ipp);
1239 void hammer_rel_inode(hammer_inode_t ip, int flush);
1240 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1241 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1242 int hammer_redo_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1243 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1245 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1246 void hammer_test_inode(hammer_inode_t dip);
1247 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1249 int hammer_ip_add_directory(struct hammer_transaction *trans,
1250 hammer_inode_t dip, const char *name, int bytes,
1251 hammer_inode_t nip);
1252 int hammer_ip_del_directory(struct hammer_transaction *trans,
1253 hammer_cursor_t cursor, hammer_inode_t dip,
1255 void hammer_ip_replace_bulk(hammer_mount_t hmp, hammer_record_t record);
1256 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1257 void *data, int bytes, int *errorp);
1258 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1259 int hammer_ip_add_record(struct hammer_transaction *trans,
1260 hammer_record_t record);
1261 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1262 int64_t ran_beg, int64_t ran_end, int truncating);
1263 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1265 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1266 int64_t offset, void *data, int bytes);
1267 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1268 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1269 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans,
1270 u_int32_t localization, int *errorp);
1271 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1272 hammer_pseudofs_inmem_t pfsm);
1273 int hammer_save_pseudofs(hammer_transaction_t trans,
1274 hammer_pseudofs_inmem_t pfsm);
1275 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1276 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1277 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1278 struct ucred *cred);
1280 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1281 enum hammer_io_type type);
1282 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1283 hammer_off_t limit);
1284 void hammer_io_advance(struct hammer_io *io);
1285 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1286 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1287 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1288 void hammer_io_flush(struct hammer_io *io, int reclaim);
1289 void hammer_io_wait(struct hammer_io *io);
1290 void hammer_io_waitdep(struct hammer_io *io);
1291 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident, int doflush);
1292 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1293 hammer_btree_leaf_elm_t leaf);
1294 int hammer_io_direct_write(hammer_mount_t hmp, struct bio *bio,
1295 hammer_record_t record);
1296 void hammer_io_direct_wait(hammer_record_t record);
1297 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1298 void hammer_io_write_interlock(hammer_io_t io);
1299 void hammer_io_done_interlock(hammer_io_t io);
1300 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1301 void hammer_io_clear_modlist(struct hammer_io *io);
1302 void hammer_io_flush_sync(hammer_mount_t hmp);
1303 void hammer_io_clear_error(struct hammer_io *io);
1304 void hammer_io_notmeta(hammer_buffer_t buffer);
1306 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1307 void *base, int len);
1308 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1309 void *base, int len);
1310 void hammer_modify_volume_done(hammer_volume_t volume);
1311 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1313 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1314 struct hammer_ioc_reblock *reblock);
1315 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1316 struct hammer_ioc_rebalance *rebal);
1317 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1318 struct hammer_ioc_prune *prune);
1319 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1320 struct hammer_ioc_mirror_rw *mirror);
1321 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1322 struct hammer_ioc_mirror_rw *mirror);
1323 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1324 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1325 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1326 struct hammer_ioc_pseudofs_rw *pfs);
1327 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1328 struct hammer_ioc_pseudofs_rw *pfs);
1329 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1330 struct hammer_ioc_pseudofs_rw *pfs);
1331 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1332 struct hammer_ioc_pseudofs_rw *pfs);
1333 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1334 struct hammer_ioc_pseudofs_rw *pfs);
1335 int hammer_ioc_volume_add(hammer_transaction_t trans, hammer_inode_t ip,
1336 struct hammer_ioc_volume *ioc);
1337 int hammer_ioc_volume_del(hammer_transaction_t trans, hammer_inode_t ip,
1338 struct hammer_ioc_volume *ioc);
1340 int hammer_signal_check(hammer_mount_t hmp);
1342 void hammer_flusher_create(hammer_mount_t hmp);
1343 void hammer_flusher_destroy(hammer_mount_t hmp);
1344 void hammer_flusher_sync(hammer_mount_t hmp);
1345 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1346 int hammer_flusher_async_one(hammer_mount_t hmp);
1347 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1348 void hammer_flusher_wait_next(hammer_mount_t hmp);
1349 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1350 int hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1351 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1352 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1353 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1354 int hammer_flusher_haswork(hammer_mount_t hmp);
1355 void hammer_flusher_flush_undos(hammer_mount_t hmp, int already_flushed);
1357 int hammer_recover_stage1(hammer_mount_t hmp, hammer_volume_t rootvol);
1358 int hammer_recover_stage2(hammer_mount_t hmp, hammer_volume_t rootvol);
1359 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1360 hammer_volume_t root_volume, int final);
1362 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1363 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1364 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1366 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1367 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1368 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1369 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1370 void hkprintf(const char *ctl, ...);
1371 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1374 int hammer_blocksize(int64_t file_offset);
1375 int hammer_blockoff(int64_t file_offset);
1376 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1379 * Shortcut for _hammer_checkspace(), used all over the code.
1382 hammer_checkspace(hammer_mount_t hmp, int slop)
1384 return(_hammer_checkspace(hmp, slop, NULL));
1389 static __inline void
1390 hammer_wait_mem_record(hammer_record_t record)
1392 hammer_wait_mem_record_ident(record, "hmmwai");
1395 static __inline void
1396 hammer_lock_ex(struct hammer_lock *lock)
1398 hammer_lock_ex_ident(lock, "hmrlck");
1402 * Indicate that a B-Tree node is being modified.
1404 static __inline void
1405 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1407 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1408 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1411 static __inline void
1412 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1414 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1415 hammer_modify_buffer(trans, node->buffer,
1416 node->ondisk, sizeof(*node->ondisk));
1419 static __inline void
1420 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1421 void *base, int len)
1423 hammer_crc_t *crcptr;
1425 KKASSERT((char *)base >= (char *)node->ondisk &&
1426 (char *)base + len <=
1427 (char *)node->ondisk + sizeof(*node->ondisk));
1428 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1429 hammer_modify_buffer(trans, node->buffer, base, len);
1430 crcptr = &node->ondisk->crc;
1431 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1432 --node->buffer->io.modify_refs; /* only want one ref */
1436 * Indicate that the specified modifications have been completed.
1438 * Do not try to generate the crc here, it's very expensive to do and a
1439 * sequence of insertions or deletions can result in many calls to this
1440 * function on the same node.
1442 static __inline void
1443 hammer_modify_node_done(hammer_node_t node)
1445 node->flags |= HAMMER_NODE_CRCGOOD;
1446 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1447 node->flags |= HAMMER_NODE_NEEDSCRC;
1448 node->buffer->io.gencrc = 1;
1449 hammer_ref_node(node);
1451 hammer_modify_buffer_done(node->buffer);
1454 #define hammer_modify_volume_field(trans, vol, field) \
1455 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1456 sizeof((vol)->ondisk->field))
1458 #define hammer_modify_node_field(trans, node, field) \
1459 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1460 sizeof((node)->ondisk->field))
1463 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1464 * created directories for HAMMER version 2 or greater and causes
1465 * directory entries to be placed the inode localization zone in
1466 * the B-Tree instead of the misc zone.
1468 * This greatly improves localization between directory entries and
1471 static __inline u_int32_t
1472 hammer_dir_localization(hammer_inode_t dip)
1474 if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1475 return(HAMMER_LOCALIZE_INODE);
1477 return(HAMMER_LOCALIZE_MISC);