2 * Copyright (c) 2011-2015 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific, prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * HAMMER2 IN-MEMORY CACHE OF MEDIA STRUCTURES
39 * This header file contains structures used internally by the HAMMER2
40 * implementation. See hammer2_disk.h for on-disk structures.
42 * There is an in-memory representation of all on-media data structure.
43 * Almost everything is represented by a hammer2_chain structure in-memory.
44 * Other higher-level structures typically map to chains.
46 * A great deal of data is accessed simply via its buffer cache buffer,
47 * which is mapped for the duration of the chain's lock. Hammer2 must
48 * implement its own buffer cache layer on top of the system layer to
49 * allow for different threads to lock different sub-block-sized buffers.
51 * When modifications are made to a chain a new filesystem block must be
52 * allocated. Multiple modifications do not typically allocate new blocks
53 * until the current block has been flushed. Flushes do not block the
54 * front-end unless the front-end operation crosses the current inode being
57 * The in-memory representation may remain cached (for example in order to
58 * placemark clustering locks) even after the related data has been
62 #ifndef _VFS_HAMMER2_HAMMER2_H_
63 #define _VFS_HAMMER2_HAMMER2_H_
65 #include <sys/param.h>
66 #include <sys/types.h>
67 #include <sys/kernel.h>
69 #include <sys/systm.h>
71 #include <sys/malloc.h>
72 #include <sys/mount.h>
73 #include <sys/vnode.h>
75 #include <sys/mountctl.h>
78 #include <sys/thread.h>
79 #include <sys/globaldata.h>
80 #include <sys/lockf.h>
82 #include <sys/queue.h>
83 #include <sys/limits.h>
85 #include <sys/mutex.h>
86 #include <sys/kern_syscall.h>
88 #include <sys/signal2.h>
90 #include <sys/mutex2.h>
91 #include <sys/thread2.h>
93 #include "hammer2_disk.h"
94 #include "hammer2_mount.h"
95 #include "hammer2_ioctl.h"
100 struct hammer2_cluster;
101 struct hammer2_inode;
105 struct hammer2_state;
107 struct hammer2_syncthr;
108 struct hammer2_vop_info;
111 * Mutex and lock shims. Hammer2 requires support for asynchronous and
112 * abortable locks, and both exclusive and shared spinlocks. Normal
113 * synchronous non-abortable locks can be substituted for spinlocks.
115 typedef mtx_t hammer2_mtx_t;
116 typedef mtx_link_t hammer2_mtx_link_t;
117 typedef mtx_state_t hammer2_mtx_state_t;
119 typedef struct spinlock hammer2_spin_t;
121 #define hammer2_mtx_ex mtx_lock_ex_quick
122 #define hammer2_mtx_sh mtx_lock_sh_quick
123 #define hammer2_mtx_unlock mtx_unlock
124 #define hammer2_mtx_owned mtx_owned
125 #define hammer2_mtx_init mtx_init
126 #define hammer2_mtx_temp_release mtx_lock_temp_release
127 #define hammer2_mtx_temp_restore mtx_lock_temp_restore
128 #define hammer2_mtx_refs mtx_lockrefs
130 #define hammer2_spin_init spin_init
131 #define hammer2_spin_sh spin_lock_shared
132 #define hammer2_spin_ex spin_lock
133 #define hammer2_spin_unsh spin_unlock_shared
134 #define hammer2_spin_unex spin_unlock
137 * General lock support
141 hammer2_mtx_upgrade(hammer2_mtx_t *mtx)
145 if (mtx_islocked_ex(mtx)) {
149 mtx_lock_ex_quick(mtx);
156 * Downgrade an inode lock from exclusive to shared only if the inode
157 * lock was previously shared. If the inode lock was previously exclusive,
162 hammer2_mtx_downgrade(hammer2_mtx_t *mtx, int wasexclusive)
164 if (wasexclusive == 0)
169 * The xid tracks internal transactional updates.
171 * XXX fix-me, really needs to be 64-bits
173 typedef uint32_t hammer2_xid_t;
175 #define HAMMER2_XID_MIN 0x00000000U
176 #define HAMMER2_XID_MAX 0x7FFFFFFFU
179 * The chain structure tracks a portion of the media topology from the
180 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
181 * data blocks, and freemap nodes and leafs.
183 * The chain structure utilizes a simple singly-homed topology and the
184 * chain's in-memory topology will move around as the chains do, due mainly
185 * to renames and indirect block creation.
187 * Block Table Updates
189 * Block table updates for insertions and updates are delayed until the
190 * flush. This allows us to avoid having to modify the parent chain
191 * all the way to the root.
193 * Block table deletions are performed immediately (modifying the parent
194 * in the process) because the flush code uses the chain structure to
195 * track delayed updates and the chain will be (likely) gone or moved to
196 * another location in the topology after a deletion.
198 * A prior iteration of the code tried to keep the relationship intact
199 * on deletes by doing a delete-duplicate operation on the chain, but
200 * it added way too much complexity to the codebase.
202 * Flush Synchronization
204 * The flush code must flush modified chains bottom-up. Because chain
205 * structures can shift around and are NOT topologically stable,
206 * modified chains are independently indexed for the flush. As the flush
207 * runs it modifies (or further modifies) and updates the parents,
208 * propagating the flush all the way to the volume root.
210 * Modifying front-end operations can occur during a flush but will block
211 * in two cases: (1) when the front-end tries to operate on the inode
212 * currently in the midst of being flushed and (2) if the front-end
213 * crosses an inode currently being flushed (such as during a rename).
214 * So, for example, if you rename directory "x" to "a/b/c/d/e/f/g/x" and
215 * the flusher is currently working on "a/b/c", the rename will block
216 * temporarily in order to ensure that "x" exists in one place or the
219 * Meta-data statistics are updated by the flusher. The front-end will
220 * make estimates but meta-data must be fully synchronized only during a
221 * flush in order to ensure that it remains correct across a crash.
223 * Multiple flush synchronizations can theoretically be in-flight at the
224 * same time but the implementation is not coded to handle the case and
225 * currently serializes them.
229 * Snapshots currently require the subdirectory tree being snapshotted
230 * to be flushed. The snapshot then creates a new super-root inode which
231 * copies the flushed blockdata of the directory or file that was
236 * - Note that the radix tree runs in powers of 2 only so sub-trees
237 * cannot straddle edges.
239 RB_HEAD(hammer2_chain_tree, hammer2_chain);
240 TAILQ_HEAD(h2_flush_list, hammer2_chain);
241 TAILQ_HEAD(h2_core_list, hammer2_chain);
242 TAILQ_HEAD(h2_iocb_list, hammer2_iocb);
244 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
245 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
248 * Core topology for chain (embedded in chain). Protected by a spinlock.
250 struct hammer2_chain_core {
252 struct hammer2_chain_tree rbtree; /* sub-chains */
253 int live_zero; /* blockref array opt */
254 u_int live_count; /* live (not deleted) chains in tree */
255 u_int chain_count; /* live + deleted chains under core */
256 int generation; /* generation number (inserts only) */
259 typedef struct hammer2_chain_core hammer2_chain_core_t;
261 RB_HEAD(hammer2_io_tree, hammer2_io);
264 * IOCB - IO callback (into chain, cluster, or manual request)
266 struct hammer2_iocb {
267 TAILQ_ENTRY(hammer2_iocb) entry;
268 void (*callback)(struct hammer2_iocb *iocb);
269 struct hammer2_io *dio;
270 struct hammer2_cluster *cluster;
271 struct hammer2_chain *chain;
279 typedef struct hammer2_iocb hammer2_iocb_t;
281 #define HAMMER2_IOCB_INTERLOCK 0x00000001
282 #define HAMMER2_IOCB_ONQ 0x00000002
283 #define HAMMER2_IOCB_DONE 0x00000004
284 #define HAMMER2_IOCB_INPROG 0x00000008
285 #define HAMMER2_IOCB_UNUSED10 0x00000010
286 #define HAMMER2_IOCB_QUICK 0x00010000
287 #define HAMMER2_IOCB_ZERO 0x00020000
288 #define HAMMER2_IOCB_READ 0x00040000
289 #define HAMMER2_IOCB_WAKEUP 0x00080000
292 * DIO - Management structure wrapping system buffer cache.
294 * Used for multiple purposes including concurrent management
295 * if small requests by chains into larger DIOs.
298 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
299 struct h2_iocb_list iocbq;
300 struct spinlock spin;
301 struct hammer2_dev *hmp;
306 int act; /* activity */
309 typedef struct hammer2_io hammer2_io_t;
311 #define HAMMER2_DIO_INPROG 0x80000000 /* bio in progress */
312 #define HAMMER2_DIO_GOOD 0x40000000 /* dio->bp is stable */
313 #define HAMMER2_DIO_WAITING 0x20000000 /* (old) */
314 #define HAMMER2_DIO_DIRTY 0x10000000 /* flush on last drop */
316 #define HAMMER2_DIO_MASK 0x0FFFFFFF
319 * Primary chain structure keeps track of the topology in-memory.
321 struct hammer2_chain {
323 hammer2_chain_core_t core;
324 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
325 hammer2_blockref_t bref;
326 struct hammer2_chain *parent;
327 struct hammer2_state *state; /* if active cache msg */
328 struct hammer2_dev *hmp;
329 struct hammer2_pfs *pmp; /* A PFS or super-root (spmp) */
331 hammer2_xid_t flush_xid; /* flush sequencing */
332 hammer2_io_t *dio; /* physical data buffer */
333 u_int bytes; /* physical data size */
337 int error; /* on-lock data error state */
339 hammer2_media_data_t *data; /* data pointer shortcut */
340 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
343 typedef struct hammer2_chain hammer2_chain_t;
345 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
346 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
349 * Special notes on flags:
351 * INITIAL - This flag allows a chain to be created and for storage to
352 * be allocated without having to immediately instantiate the
353 * related buffer. The data is assumed to be all-zeros. It
354 * is primarily used for indirect blocks.
356 * MODIFIED - The chain's media data has been modified.
358 * UPDATE - Chain might not be modified but parent blocktable needs update
360 * FICTITIOUS - Faked chain as a placeholder for an error condition. This
361 * chain is unsuitable for I/O.
363 * BMAPPED - Indicates that the chain is present in the parent blockmap.
365 * BMAPUPD - Indicates that the chain is present but needs to be updated
366 * in the parent blockmap.
368 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
369 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
370 #define HAMMER2_CHAIN_DESTROY 0x00000004
371 #define HAMMER2_CHAIN_UNUSED0008 0x00000008
372 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
373 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
374 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
375 #define HAMMER2_CHAIN_DEFERRED 0x00000080 /* flush depth defer */
376 #define HAMMER2_CHAIN_IOFLUSH 0x00000100 /* bawrite on put */
377 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
378 #define HAMMER2_CHAIN_FICTITIOUS 0x00000400 /* unsuitable for I/O */
379 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
380 #define HAMMER2_CHAIN_DELAYED 0x00001000 /* delayed flush */
381 #define HAMMER2_CHAIN_COUNTEDBREFS 0x00002000 /* block table stats */
382 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
383 #define HAMMER2_CHAIN_UNUSED00008000 0x00008000
384 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
385 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
386 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
387 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
388 #define HAMMER2_CHAIN_IOINPROG 0x00100000 /* I/O interlock */
389 #define HAMMER2_CHAIN_IOSIGNAL 0x00200000 /* I/O interlock */
390 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
392 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
393 HAMMER2_CHAIN_UPDATE | \
394 HAMMER2_CHAIN_ONFLUSH)
397 * Hammer2 error codes, used by chain->error and cluster->error. The error
398 * code is typically set on-lock unless no I/O was requested, and set on
399 * I/O otherwise. If set for a cluster it generally means that the cluster
400 * code could not find a valid copy to present.
402 * IO - An I/O error occurred
403 * CHECK - I/O succeeded but did not match the check code
404 * INCOMPLETE - A cluster is not complete enough to use, or
405 * a chain cannot be loaded because its parent has an error.
407 * NOTE: API allows callers to check zero/non-zero to determine if an error
410 * NOTE: Chain's data field is usually NULL on an IO error but not necessarily
411 * NULL on other errors. Check chain->error, not chain->data.
413 #define HAMMER2_ERROR_NONE 0
414 #define HAMMER2_ERROR_IO 1 /* device I/O error */
415 #define HAMMER2_ERROR_CHECK 2 /* check code mismatch */
416 #define HAMMER2_ERROR_INCOMPLETE 3 /* incomplete cluster */
417 #define HAMMER2_ERROR_DEPTH 4 /* temporary depth limit */
420 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
422 * NOTE: MATCHIND allows an indirect block / freemap node to be returned
423 * when the passed key range matches the radix. Remember that key_end
424 * is inclusive (e.g. {0x000,0xFFF}, not {0x000,0x1000}).
426 * NOTE: NODIRECT prevents a lookup of offset 0 in an inode from returning
427 * the inode itself if the inode is in DIRECTDATA mode (i.e. file is
430 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
431 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
432 #define HAMMER2_LOOKUP_NODIRECT 0x00000004 /* no offset=0 DD */
433 #define HAMMER2_LOOKUP_SHARED 0x00000100
434 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
435 #define HAMMER2_LOOKUP_ALLNODES 0x00000400 /* allow NULL focus */
436 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
439 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
441 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
442 * blocks in the INITIAL-create state.
444 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
445 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
446 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
447 #define HAMMER2_MODIFY_NOREALLOC 0x00000010
450 * Flags passed to hammer2_chain_lock()
452 * NOTE: RDONLY is set to optimize cluster operations when *no* modifications
453 * will be made to either the cluster being locked or any underlying
454 * cluster. It allows the cluster to lock and access data for a subset
455 * of available nodes instead of all available nodes.
457 #define HAMMER2_RESOLVE_NEVER 1
458 #define HAMMER2_RESOLVE_MAYBE 2
459 #define HAMMER2_RESOLVE_ALWAYS 3
460 #define HAMMER2_RESOLVE_MASK 0x0F
462 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
463 #define HAMMER2_RESOLVE_UNUSED20 0x20
464 #define HAMMER2_RESOLVE_RDONLY 0x40 /* higher level op flag */
467 * Flags passed to hammer2_chain_delete()
469 #define HAMMER2_DELETE_PERMANENT 0x0001
470 #define HAMMER2_DELETE_NOSTATS 0x0002
472 #define HAMMER2_INSERT_NOSTATS 0x0002
473 #define HAMMER2_INSERT_PFSROOT 0x0004
476 * Flags passed to hammer2_chain_delete_duplicate()
478 #define HAMMER2_DELDUP_RECORE 0x0001
481 * Cluster different types of storage together for allocations
483 #define HAMMER2_FREECACHE_INODE 0
484 #define HAMMER2_FREECACHE_INDIR 1
485 #define HAMMER2_FREECACHE_DATA 2
486 #define HAMMER2_FREECACHE_UNUSED3 3
487 #define HAMMER2_FREECACHE_TYPES 4
490 * hammer2_freemap_alloc() block preference
492 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
495 * BMAP read-ahead maximum parameters
497 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
498 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
501 * hammer2_freemap_adjust()
503 #define HAMMER2_FREEMAP_DORECOVER 1
504 #define HAMMER2_FREEMAP_DOMAYFREE 2
505 #define HAMMER2_FREEMAP_DOREALFREE 3
508 * HAMMER2 cluster - A set of chains representing the same entity.
510 * hammer2_cluster typically represents a temporary set of representitive
511 * chains. The one exception is that a hammer2_cluster is embedded in
512 * hammer2_inode. This embedded cluster is ONLY used to track the
513 * representitive chains and cannot be directly locked.
515 * A cluster is usually temporary (and thus per-thread) for locking purposes,
516 * allowing us to embed the asynchronous storage required for cluster
517 * operations in the cluster itself and adjust the state and status without
518 * having to worry too much about SMP issues.
520 * The exception is the cluster embedded in the hammer2_inode structure.
521 * This is used to cache the cluster state on an inode-by-inode basis.
522 * Individual hammer2_chain structures not incorporated into clusters might
523 * also stick around to cache miscellanious elements.
525 * Because the cluster is a 'working copy' and is usually subject to cluster
526 * quorum rules, it is quite possible for us to end up with an insufficient
527 * number of live chains to execute an operation. If an insufficient number
528 * of chains remain in a working copy, the operation may have to be
529 * downgraded, retried, stall until the requisit number of chains are
530 * available, or possibly even error out depending on the mount type.
532 * A cluster's focus is set when it is locked. The focus can only be set
533 * to a chain still part of the synchronized set.
535 #define HAMMER2_MAXCLUSTER 8
537 struct hammer2_cluster_item {
539 hammer2_mtx_link_t async_link;
541 hammer2_chain_t *chain;
543 struct hammer2_cluster *cluster; /* link back to cluster */
549 typedef struct hammer2_cluster_item hammer2_cluster_item_t;
552 * INVALID - Invalid for focus, i.e. not part of synchronized set.
553 * Once set, this bit is sticky across operations.
555 * FEMOD - Indicates that front-end modifying operations can
556 * mess with this entry and MODSYNC will copy also
559 #define HAMMER2_CITEM_INVALID 0x00000001
560 #define HAMMER2_CITEM_FEMOD 0x00000002
562 struct hammer2_cluster {
563 int refs; /* track for deallocation */
565 struct hammer2_pfs *pmp;
568 int error; /* error code valid on lock */
571 hammer2_chain_t *focus; /* current focus (or mod) */
572 hammer2_cluster_item_t array[HAMMER2_MAXCLUSTER];
575 typedef struct hammer2_cluster hammer2_cluster_t;
578 * WRHARD - Hard mounts can write fully synchronized
579 * RDHARD - Hard mounts can read fully synchronized
580 * UNHARD - Unsynchronized masters present
581 * NOHARD - No masters visible
582 * WRSOFT - Soft mounts can write to at least the SOFT_MASTER
583 * RDSOFT - Soft mounts can read from at least a SOFT_SLAVE
584 * UNSOFT - Unsynchronized slaves present
585 * NOSOFT - No slaves visible
586 * RDSLAVE - slaves are accessible (possibly unsynchronized or remote).
587 * MSYNCED - All masters are fully synchronized
588 * SSYNCED - All known local slaves are fully synchronized to masters
590 * All available masters are always incorporated. All PFSs belonging to a
591 * cluster (master, slave, copy, whatever) always try to synchronize the
592 * total number of known masters in the PFSs root inode.
594 * A cluster might have access to many slaves, copies, or caches, but we
595 * have a limited number of cluster slots. Any such elements which are
596 * directly mounted from block device(s) will always be incorporated. Note
597 * that SSYNCED only applies to such elements which are directly mounted,
598 * not to any remote slaves, copies, or caches that could be available. These
599 * bits are used to monitor and drive our synchronization threads.
601 * When asking the question 'is any data accessible at all', then a simple
602 * test against (RDHARD|RDSOFT|RDSLAVE) gives you the answer. If any of
603 * these bits are set the object can be read with certain caveats:
604 * RDHARD - no caveats. RDSOFT - authoritative but might not be synchronized.
605 * and RDSLAVE - not authoritative, has some data but it could be old or
608 * When both soft and hard mounts are available, data will be read and written
609 * via the soft mount only. But all might be in the cluster because
610 * background synchronization threads still need to do their work.
612 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode struct */
613 #define HAMMER2_CLUSTER_UNUSED2 0x00000002
614 #define HAMMER2_CLUSTER_LOCKED 0x00000004 /* cluster lks not recursive */
615 #define HAMMER2_CLUSTER_WRHARD 0x00000100 /* hard-mount can write */
616 #define HAMMER2_CLUSTER_RDHARD 0x00000200 /* hard-mount can read */
617 #define HAMMER2_CLUSTER_UNHARD 0x00000400 /* unsynchronized masters */
618 #define HAMMER2_CLUSTER_NOHARD 0x00000800 /* no masters visible */
619 #define HAMMER2_CLUSTER_WRSOFT 0x00001000 /* soft-mount can write */
620 #define HAMMER2_CLUSTER_RDSOFT 0x00002000 /* soft-mount can read */
621 #define HAMMER2_CLUSTER_UNSOFT 0x00004000 /* unsynchronized slaves */
622 #define HAMMER2_CLUSTER_NOSOFT 0x00008000 /* no slaves visible */
623 #define HAMMER2_CLUSTER_MSYNCED 0x00010000 /* all masters synchronized */
624 #define HAMMER2_CLUSTER_SSYNCED 0x00020000 /* known slaves synchronized */
626 #define HAMMER2_CLUSTER_ANYDATA ( HAMMER2_CLUSTER_RDHARD | \
627 HAMMER2_CLUSTER_RDSOFT | \
628 HAMMER2_CLUSTER_RDSLAVE)
630 #define HAMMER2_CLUSTER_RDOK ( HAMMER2_CLUSTER_RDHARD | \
631 HAMMER2_CLUSTER_RDSOFT)
633 #define HAMMER2_CLUSTER_WROK ( HAMMER2_CLUSTER_WRHARD | \
634 HAMMER2_CLUSTER_WRSOFT)
636 #define HAMMER2_CLUSTER_ZFLAGS ( HAMMER2_CLUSTER_WRHARD | \
637 HAMMER2_CLUSTER_RDHARD | \
638 HAMMER2_CLUSTER_WRSOFT | \
639 HAMMER2_CLUSTER_RDSOFT | \
640 HAMMER2_CLUSTER_MSYNCED | \
641 HAMMER2_CLUSTER_SSYNCED)
644 * Helper functions (cluster must be locked for flags to be valid).
648 hammer2_cluster_rdok(hammer2_cluster_t *cluster)
650 return (cluster->flags & HAMMER2_CLUSTER_RDOK);
655 hammer2_cluster_wrok(hammer2_cluster_t *cluster)
657 return (cluster->flags & HAMMER2_CLUSTER_WROK);
660 RB_HEAD(hammer2_inode_tree, hammer2_inode);
665 * NOTE: The inode-embedded cluster is never used directly for I/O (since
666 * it may be shared). Instead it will be replicated-in and synchronized
667 * back out if changed.
669 struct hammer2_inode {
670 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
671 hammer2_mtx_t lock; /* inode lock */
672 struct hammer2_pfs *pmp; /* PFS mount */
673 struct hammer2_inode *pip; /* parent inode */
675 struct spinlock cluster_spin; /* update cluster */
676 hammer2_cluster_t cluster;
677 struct lockf advlock;
680 u_int refs; /* +vpref, +flushref */
681 uint8_t comp_heuristic;
682 hammer2_off_t size; /* cache file size */
683 uint64_t mtime; /* cache mtime */
686 typedef struct hammer2_inode hammer2_inode_t;
688 #define HAMMER2_INODE_MODIFIED 0x0001
689 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
690 #define HAMMER2_INODE_RENAME_INPROG 0x0004
691 #define HAMMER2_INODE_ONRBTREE 0x0008
692 #define HAMMER2_INODE_RESIZED 0x0010
693 #define HAMMER2_INODE_MTIME 0x0020
694 #define HAMMER2_INODE_ISUNLINKED 0x0040
695 #define HAMMER2_INODE_DATAGOOD 0x0080 /* inode meta-data */
697 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
698 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
702 * inode-unlink side-structure
704 struct hammer2_inode_unlink {
705 TAILQ_ENTRY(hammer2_inode_unlink) entry;
708 TAILQ_HEAD(h2_unlk_list, hammer2_inode_unlink);
710 typedef struct hammer2_inode_unlink hammer2_inode_unlink_t;
713 * A hammer2 transaction and flush sequencing structure.
715 * This global structure is tied into hammer2_dev and is used
716 * to sequence modifying operations and flushes. These operations
717 * run on whole cluster PFSs, not individual nodes (at this level),
718 * so we do not record mirror_tid here.
720 struct hammer2_trans {
721 TAILQ_ENTRY(hammer2_trans) entry;
722 struct hammer2_pfs *pmp;
723 hammer2_xid_t sync_xid; /* transaction sequencer */
724 hammer2_tid_t inode_tid; /* inode number assignment */
725 hammer2_tid_t modify_tid; /* modify transaction id */
726 thread_t td; /* pointer */
729 uint8_t inodes_created;
733 typedef struct hammer2_trans hammer2_trans_t;
735 #define HAMMER2_TRANS_ISFLUSH 0x0001 /* formal flush */
736 #define HAMMER2_TRANS_CONCURRENT 0x0002 /* concurrent w/flush */
737 #define HAMMER2_TRANS_BUFCACHE 0x0004 /* from bioq strategy write */
738 #define HAMMER2_TRANS_NEWINODE 0x0008 /* caller allocating inode */
739 #define HAMMER2_TRANS_KEEPMODIFY 0x0010 /* do not change bref.modify */
740 #define HAMMER2_TRANS_PREFLUSH 0x0020 /* preflush state */
742 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
743 #define HAMMER2_FREEMAP_HEUR_TYPES 8
744 #define HAMMER2_FREEMAP_HEUR (HAMMER2_FREEMAP_HEUR_NRADIX * \
745 HAMMER2_FREEMAP_HEUR_TYPES)
748 * Transaction Rendezvous
750 TAILQ_HEAD(hammer2_trans_queue, hammer2_trans);
752 struct hammer2_trans_manage {
753 hammer2_xid_t flush_xid; /* last flush transaction */
754 hammer2_xid_t alloc_xid;
755 struct lock translk; /* lockmgr lock */
756 struct hammer2_trans_queue transq; /* modifying transactions */
757 int flushcnt; /* track flush trans */
760 typedef struct hammer2_trans_manage hammer2_trans_manage_t;
763 * hammer2_vop_info - container for VOP operation.
765 * This structure is used to distribute a VOP operation across multiple
766 * nodes. It provides a rendezvous for concurrent node execution and
767 * can be detached from the frontend operation to allow the frontend to
770 struct hammer2_vop_info {
771 hammer2_inode_t *dip;
775 void (*xio_func)(struct hammer2_syncthr *thr,
776 struct hammer2_vop_info *info);
779 typedef struct hammer2_vop_info hammer2_vop_info_t;
782 * Cluster node synchronization and operation thread element.
784 * Multiple syncthr's can hang off of a hammer2_pfs structure, typically one
785 * for each block device that is part of the PFS. Synchronization threads
786 * for PFSs accessed over the network are handled by their respective hosts.
788 * Synchronization threads are responsible for keeping a local node
789 * synchronized to the greater cluster.
791 * A syncthr can also hang off each hammer2_dev's super-root PFS (spmp).
792 * This thread is responsible for automatic bulkfree and dedup scans.
794 struct hammer2_syncthr {
795 struct hammer2_pfs *pmp;
799 int clindex; /* sync_thrs[] array index */
800 hammer2_trans_t trans;
804 typedef struct hammer2_syncthr hammer2_syncthr_t;
806 #define HAMMER2_SYNCTHR_UNMOUNTING 0x0001 /* unmount request */
807 #define HAMMER2_SYNCTHR_DEV 0x0002 /* related to dev, not pfs */
808 #define HAMMER2_SYNCTHR_UNUSED04 0x0004
809 #define HAMMER2_SYNCTHR_REMASTER 0x0008 /* remaster request */
810 #define HAMMER2_SYNCTHR_STOP 0x0010 /* exit request */
811 #define HAMMER2_SYNCTHR_FREEZE 0x0020 /* force idle */
812 #define HAMMER2_SYNCTHR_FROZEN 0x0040 /* restart */
816 * Global (per partition) management structure, represents a hard block
817 * device. Typically referenced by hammer2_chain structures when applicable.
818 * Typically not used for network-managed elements.
820 * Note that a single hammer2_dev can be indirectly tied to multiple system
821 * mount points. There is no direct relationship. System mounts are
822 * per-cluster-id, not per-block-device, and a single hard mount might contain
823 * many PFSs and those PFSs might combine together in various ways to form
824 * the set of available clusters.
827 struct vnode *devvp; /* device vnode */
828 int ronly; /* read-only mount */
829 int mount_count; /* number of actively mounted PFSs */
830 TAILQ_ENTRY(hammer2_dev) mntentry; /* hammer2_mntlist */
832 struct malloc_type *mchain;
835 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
836 struct spinlock io_spin; /* iotree access */
837 struct hammer2_io_tree iotree;
839 hammer2_chain_t vchain; /* anchor chain (topology) */
840 hammer2_chain_t fchain; /* anchor chain (freemap) */
841 struct spinlock list_spin;
842 struct h2_flush_list flushq; /* flush seeds */
843 struct hammer2_pfs *spmp; /* super-root pmp for transactions */
844 struct lock vollk; /* lockmgr lock */
845 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR];
846 int volhdrno; /* last volhdrno written */
847 char devrepname[64]; /* for kprintf */
848 hammer2_volume_data_t voldata;
849 hammer2_volume_data_t volsync; /* synchronized voldata */
852 typedef struct hammer2_dev hammer2_dev_t;
855 * Helper functions (cluster must be locked for flags to be valid).
859 hammer2_chain_rdok(hammer2_chain_t *chain)
861 return (chain->error == 0);
866 hammer2_chain_wrok(hammer2_chain_t *chain)
868 return (chain->error == 0 && chain->hmp->ronly == 0);
872 * Per-cluster management structure. This structure will be tied to a
873 * system mount point if the system is mounting the PFS, but is also used
874 * to manage clusters encountered during the super-root scan or received
875 * via LNK_SPANs that might not be mounted.
877 * This structure is also used to represent the super-root that hangs off
878 * of a hard mount point. The super-root is not really a cluster element.
879 * In this case the spmp_hmp field will be non-NULL. It's just easier to do
880 * this than to special case super-root manipulation in the hammer2_chain*
881 * code as being only hammer2_dev-related.
883 * pfs_mode and pfs_nmasters are rollup fields which critically describes
884 * how elements of the cluster act on the cluster. pfs_mode is only applicable
885 * when a PFS is mounted by the system. pfs_nmasters is our best guess as to
886 * how many masters have been configured for a cluster and is always
887 * applicable. pfs_types[] is an array with 1:1 correspondance to the
888 * iroot cluster and describes the PFS types of the nodes making up the
891 * WARNING! Portions of this structure have deferred initialization. In
892 * particular, if not mounted there will be no ihidden or wthread.
893 * umounted network PFSs will also be missing iroot and numerous
894 * other fields will not be initialized prior to mount.
896 * Synchronization threads are chain-specific and only applicable
897 * to local hard PFS entries. A hammer2_pfs structure may contain
898 * more than one when multiple hard PFSs are present on the local
899 * machine which require synchronization monitoring. Most PFSs
900 * (such as snapshots) are 1xMASTER PFSs which do not need a
901 * synchronization thread.
903 * WARNING! The chains making up pfs->iroot's cluster are accounted for in
904 * hammer2_dev->mount_count when the pfs is associated with a mount
909 TAILQ_ENTRY(hammer2_pfs) mntentry; /* hammer2_pfslist */
911 hammer2_dev_t *spmp_hmp; /* only if super-root pmp */
912 hammer2_inode_t *iroot; /* PFS root inode */
913 hammer2_inode_t *ihidden; /* PFS hidden directory */
914 uint8_t pfs_types[HAMMER2_MAXCLUSTER];
915 char *pfs_names[HAMMER2_MAXCLUSTER];
916 hammer2_trans_manage_t tmanage; /* transaction management */
917 struct lock lock; /* PFS lock for certain ops */
918 struct netexport export; /* nfs export */
919 int ronly; /* read-only mount */
920 struct malloc_type *minode;
921 struct malloc_type *mmsg;
922 struct spinlock inum_spin; /* inumber lookup */
923 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
924 hammer2_tid_t modify_tid; /* modify transaction id */
925 hammer2_tid_t inode_tid; /* inode allocator */
926 uint8_t pfs_nmasters; /* total masters */
927 uint8_t pfs_mode; /* operating mode PFSMODE */
932 uint32_t inmem_dirty_chains;
933 int count_lwinprog; /* logical write in prog */
934 struct spinlock list_spin;
935 struct h2_unlk_list unlinkq; /* last-close unlink */
936 hammer2_syncthr_t sync_thrs[HAMMER2_MAXCLUSTER];
937 thread_t wthread_td; /* write thread td */
938 struct bio_queue_head wthread_bioq; /* logical buffer bioq */
939 hammer2_mtx_t wthread_mtx; /* interlock */
940 int wthread_destroy;/* termination sequencing */
941 uint32_t flags; /* cached cluster flags */
944 typedef struct hammer2_pfs hammer2_pfs_t;
946 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
947 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
949 #define HAMMER2_LWINPROG_WAITING 0x80000000
950 #define HAMMER2_LWINPROG_MASK 0x7FFFFFFF
955 #define HAMMER2_BULK_ABORT 0x00000001
962 MALLOC_DECLARE(M_HAMMER2);
964 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
965 #define ITOV(ip) ((ip)->vp)
968 * Currently locked chains retain the locked buffer cache buffer for
969 * indirect blocks, and indirect blocks can be one of two sizes. The
970 * device buffer has to match the case to avoid deadlocking recursive
971 * chains that might otherwise try to access different offsets within
972 * the same device buffer.
976 hammer2_devblkradix(int radix)
979 if (radix <= HAMMER2_LBUFRADIX) {
980 return (HAMMER2_LBUFRADIX);
982 return (HAMMER2_PBUFRADIX);
985 return (HAMMER2_PBUFRADIX);
989 * XXX almost time to remove this. DIO uses PBUFSIZE exclusively now.
993 hammer2_devblksize(size_t bytes)
996 if (bytes <= HAMMER2_LBUFSIZE) {
997 return(HAMMER2_LBUFSIZE);
999 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
1000 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
1001 return (HAMMER2_PBUFSIZE);
1004 return (HAMMER2_PBUFSIZE);
1010 MPTOPMP(struct mount *mp)
1012 return ((hammer2_pfs_t *)mp->mnt_data);
1015 #define LOCKSTART int __nlocks = curthread->td_locks
1016 #define LOCKENTER (++curthread->td_locks)
1017 #define LOCKEXIT (--curthread->td_locks)
1018 #define LOCKSTOP KKASSERT(curthread->td_locks == __nlocks)
1020 extern struct vop_ops hammer2_vnode_vops;
1021 extern struct vop_ops hammer2_spec_vops;
1022 extern struct vop_ops hammer2_fifo_vops;
1024 extern int hammer2_debug;
1025 extern int hammer2_cluster_enable;
1026 extern int hammer2_hardlink_enable;
1027 extern int hammer2_flush_pipe;
1028 extern int hammer2_synchronous_flush;
1029 extern int hammer2_dio_count;
1030 extern long hammer2_limit_dirty_chains;
1031 extern long hammer2_iod_file_read;
1032 extern long hammer2_iod_meta_read;
1033 extern long hammer2_iod_indr_read;
1034 extern long hammer2_iod_fmap_read;
1035 extern long hammer2_iod_volu_read;
1036 extern long hammer2_iod_file_write;
1037 extern long hammer2_iod_meta_write;
1038 extern long hammer2_iod_indr_write;
1039 extern long hammer2_iod_fmap_write;
1040 extern long hammer2_iod_volu_write;
1041 extern long hammer2_ioa_file_read;
1042 extern long hammer2_ioa_meta_read;
1043 extern long hammer2_ioa_indr_read;
1044 extern long hammer2_ioa_fmap_read;
1045 extern long hammer2_ioa_volu_read;
1046 extern long hammer2_ioa_file_write;
1047 extern long hammer2_ioa_meta_write;
1048 extern long hammer2_ioa_indr_write;
1049 extern long hammer2_ioa_fmap_write;
1050 extern long hammer2_ioa_volu_write;
1052 extern struct objcache *cache_buffer_read;
1053 extern struct objcache *cache_buffer_write;
1054 extern struct objcache *cache_vop_info;
1057 extern int write_thread_wakeup;
1062 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
1063 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
1065 int hammer2_signal_check(time_t *timep);
1066 const char *hammer2_error_str(int error);
1068 hammer2_cluster_t *hammer2_inode_lock(hammer2_inode_t *ip, int how);
1069 void hammer2_inode_unlock(hammer2_inode_t *ip, hammer2_cluster_t *cluster);
1070 hammer2_mtx_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
1071 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip,
1072 hammer2_mtx_state_t ostate);
1073 int hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
1074 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int);
1076 void hammer2_dev_exlock(hammer2_dev_t *hmp);
1077 void hammer2_dev_shlock(hammer2_dev_t *hmp);
1078 void hammer2_dev_unlock(hammer2_dev_t *hmp);
1080 int hammer2_get_dtype(const hammer2_inode_data_t *ipdata);
1081 int hammer2_get_vtype(const hammer2_inode_data_t *ipdata);
1082 u_int8_t hammer2_get_obj_type(enum vtype vtype);
1083 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1084 u_int64_t hammer2_timespec_to_time(const struct timespec *ts);
1085 u_int32_t hammer2_to_unix_xid(const uuid_t *uuid);
1086 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1087 hammer2_xid_t hammer2_trans_newxid(hammer2_pfs_t *pmp);
1088 void hammer2_trans_manage_init(hammer2_trans_manage_t *tman);
1090 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
1091 int hammer2_getradix(size_t bytes);
1093 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
1094 hammer2_key_t *lbasep, hammer2_key_t *leofp);
1095 int hammer2_calc_physical(hammer2_inode_t *ip,
1096 const hammer2_inode_data_t *ipdata,
1097 hammer2_key_t lbase);
1098 void hammer2_update_time(uint64_t *timep);
1099 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
1104 struct vnode *hammer2_igetv(hammer2_inode_t *ip, hammer2_cluster_t *cparent,
1106 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfs_t *pmp,
1107 hammer2_tid_t inum);
1108 hammer2_inode_t *hammer2_inode_get(hammer2_pfs_t *pmp,
1109 hammer2_inode_t *dip, hammer2_cluster_t *cluster);
1110 void hammer2_inode_free(hammer2_inode_t *ip);
1111 void hammer2_inode_ref(hammer2_inode_t *ip);
1112 void hammer2_inode_drop(hammer2_inode_t *ip);
1113 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
1114 hammer2_cluster_t *cluster);
1115 void hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1118 void hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfs_t *pmp);
1120 hammer2_inode_t *hammer2_inode_create(hammer2_trans_t *trans,
1121 hammer2_inode_t *dip,
1122 struct vattr *vap, struct ucred *cred,
1123 const uint8_t *name, size_t name_len,
1124 hammer2_cluster_t **clusterp,
1125 int flags, int *errorp);
1126 int hammer2_inode_connect(hammer2_trans_t *trans,
1127 hammer2_cluster_t **clusterp, int hlink,
1128 hammer2_inode_t *dip, hammer2_cluster_t *dcluster,
1129 const uint8_t *name, size_t name_len,
1131 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
1132 hammer2_inode_t *tdip);
1133 void hammer2_inode_fsync(hammer2_trans_t *trans, hammer2_inode_t *ip,
1134 hammer2_cluster_t *cparent);
1135 int hammer2_unlink_file(hammer2_trans_t *trans, hammer2_inode_t *dip,
1136 const uint8_t *name, size_t name_len, int isdir,
1137 int *hlinkp, struct nchandle *nch, int nlinks);
1138 int hammer2_hardlink_consolidate(hammer2_trans_t *trans,
1139 hammer2_inode_t *ip, hammer2_cluster_t **clusterp,
1140 hammer2_inode_t *cdip, hammer2_cluster_t *cdcluster,
1142 int hammer2_hardlink_deconsolidate(hammer2_trans_t *trans, hammer2_inode_t *dip,
1143 hammer2_chain_t **chainp, hammer2_chain_t **ochainp);
1144 int hammer2_hardlink_find(hammer2_inode_t *dip, hammer2_cluster_t **cparentp,
1145 hammer2_cluster_t **clusterp);
1146 int hammer2_parent_find(hammer2_cluster_t **cparentp,
1147 hammer2_cluster_t *cluster);
1148 void hammer2_inode_install_hidden(hammer2_pfs_t *pmp);
1153 void hammer2_voldata_lock(hammer2_dev_t *hmp);
1154 void hammer2_voldata_unlock(hammer2_dev_t *hmp);
1155 void hammer2_voldata_modify(hammer2_dev_t *hmp);
1156 hammer2_chain_t *hammer2_chain_alloc(hammer2_dev_t *hmp,
1158 hammer2_trans_t *trans,
1159 hammer2_blockref_t *bref);
1160 void hammer2_chain_core_init(hammer2_chain_t *chain);
1161 void hammer2_chain_ref(hammer2_chain_t *chain);
1162 void hammer2_chain_drop(hammer2_chain_t *chain);
1163 void hammer2_chain_lock(hammer2_chain_t *chain, int how);
1164 void hammer2_chain_load_data(hammer2_chain_t *chain);
1165 const hammer2_media_data_t *hammer2_chain_rdata(hammer2_chain_t *chain);
1166 hammer2_media_data_t *hammer2_chain_wdata(hammer2_chain_t *chain);
1171 void hammer2_cluster_load_async(hammer2_cluster_t *cluster,
1172 void (*callback)(hammer2_iocb_t *iocb),
1174 void hammer2_chain_moved(hammer2_chain_t *chain);
1175 void hammer2_chain_modify(hammer2_trans_t *trans,
1176 hammer2_chain_t *chain, int flags);
1177 void hammer2_chain_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
1178 hammer2_chain_t *parent,
1179 hammer2_chain_t *chain,
1180 int nradix, int flags);
1181 void hammer2_chain_unlock(hammer2_chain_t *chain);
1182 void hammer2_chain_wait(hammer2_chain_t *chain);
1183 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
1184 hammer2_blockref_t *bref);
1185 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
1186 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
1187 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
1188 hammer2_key_t *key_nextp,
1189 hammer2_key_t key_beg, hammer2_key_t key_end,
1190 int *cache_indexp, int flags);
1191 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
1192 hammer2_chain_t *chain,
1193 hammer2_key_t *key_nextp,
1194 hammer2_key_t key_beg, hammer2_key_t key_end,
1195 int *cache_indexp, int flags);
1196 hammer2_chain_t *hammer2_chain_scan(hammer2_chain_t *parent,
1197 hammer2_chain_t *chain,
1198 int *cache_indexp, int flags);
1200 int hammer2_chain_create(hammer2_trans_t *trans, hammer2_chain_t **parentp,
1201 hammer2_chain_t **chainp,
1203 hammer2_key_t key, int keybits,
1204 int type, size_t bytes, int flags);
1205 void hammer2_chain_rename(hammer2_trans_t *trans, hammer2_blockref_t *bref,
1206 hammer2_chain_t **parentp,
1207 hammer2_chain_t *chain, int flags);
1208 int hammer2_chain_snapshot(hammer2_trans_t *trans, hammer2_chain_t **chainp,
1209 hammer2_ioc_pfs_t *pfs);
1210 void hammer2_chain_delete(hammer2_trans_t *trans, hammer2_chain_t *parent,
1211 hammer2_chain_t *chain, int flags);
1212 void hammer2_chain_delete_duplicate(hammer2_trans_t *trans,
1213 hammer2_chain_t **chainp, int flags);
1214 void hammer2_flush(hammer2_trans_t *trans, hammer2_chain_t *chain, int istop);
1215 void hammer2_delayed_flush(hammer2_trans_t *trans, hammer2_chain_t *chain);
1216 void hammer2_chain_commit(hammer2_trans_t *trans, hammer2_chain_t *chain);
1217 void hammer2_chain_setflush(hammer2_trans_t *trans, hammer2_chain_t *chain);
1218 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
1219 hammer2_blockref_t *base, int count);
1221 void hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata);
1222 int hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata);
1225 void hammer2_pfs_memory_wait(hammer2_pfs_t *pmp);
1226 void hammer2_pfs_memory_inc(hammer2_pfs_t *pmp);
1227 void hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp);
1229 void hammer2_base_delete(hammer2_trans_t *trans, hammer2_chain_t *chain,
1230 hammer2_blockref_t *base, int count,
1231 int *cache_indexp, hammer2_chain_t *child);
1232 void hammer2_base_insert(hammer2_trans_t *trans, hammer2_chain_t *chain,
1233 hammer2_blockref_t *base, int count,
1234 int *cache_indexp, hammer2_chain_t *child);
1239 void hammer2_trans_init(hammer2_trans_t *trans, hammer2_pfs_t *pmp,
1241 void hammer2_trans_done(hammer2_trans_t *trans);
1242 void hammer2_trans_assert_strategy(hammer2_pfs_t *pmp);
1247 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
1248 int fflag, struct ucred *cred);
1253 void hammer2_io_putblk(hammer2_io_t **diop);
1254 void hammer2_io_cleanup(hammer2_dev_t *hmp, struct hammer2_io_tree *tree);
1255 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
1256 void hammer2_io_getblk(hammer2_dev_t *hmp, off_t lbase, int lsize,
1257 hammer2_iocb_t *iocb);
1258 void hammer2_io_complete(hammer2_iocb_t *iocb);
1259 void hammer2_io_callback(struct bio *bio);
1260 void hammer2_iocb_wait(hammer2_iocb_t *iocb);
1261 int hammer2_io_new(hammer2_dev_t *hmp, off_t lbase, int lsize,
1262 hammer2_io_t **diop);
1263 int hammer2_io_newnz(hammer2_dev_t *hmp, off_t lbase, int lsize,
1264 hammer2_io_t **diop);
1265 int hammer2_io_newq(hammer2_dev_t *hmp, off_t lbase, int lsize,
1266 hammer2_io_t **diop);
1267 int hammer2_io_bread(hammer2_dev_t *hmp, off_t lbase, int lsize,
1268 hammer2_io_t **diop);
1269 void hammer2_io_bawrite(hammer2_io_t **diop);
1270 void hammer2_io_bdwrite(hammer2_io_t **diop);
1271 int hammer2_io_bwrite(hammer2_io_t **diop);
1272 int hammer2_io_isdirty(hammer2_io_t *dio);
1273 void hammer2_io_setdirty(hammer2_io_t *dio);
1274 void hammer2_io_setinval(hammer2_io_t *dio, u_int bytes);
1275 void hammer2_io_brelse(hammer2_io_t **diop);
1276 void hammer2_io_bqrelse(hammer2_io_t **diop);
1281 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
1282 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
1287 void hammer2_clusterctl_wakeup(kdmsg_iocom_t *iocom);
1288 void hammer2_volconf_update(hammer2_dev_t *hmp, int index);
1289 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
1290 void hammer2_bioq_sync(hammer2_pfs_t *pmp);
1291 int hammer2_vfs_sync(struct mount *mp, int waitflags);
1292 hammer2_pfs_t *hammer2_pfsalloc(hammer2_cluster_t *cluster,
1293 const hammer2_inode_data_t *ripdata,
1294 hammer2_tid_t modify_tid);
1296 void hammer2_lwinprog_ref(hammer2_pfs_t *pmp);
1297 void hammer2_lwinprog_drop(hammer2_pfs_t *pmp);
1298 void hammer2_lwinprog_wait(hammer2_pfs_t *pmp);
1303 int hammer2_freemap_alloc(hammer2_trans_t *trans, hammer2_chain_t *chain,
1305 void hammer2_freemap_adjust(hammer2_trans_t *trans, hammer2_dev_t *hmp,
1306 hammer2_blockref_t *bref, int how);
1311 int hammer2_cluster_need_resize(hammer2_cluster_t *cluster, int bytes);
1312 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
1313 const hammer2_media_data_t *hammer2_cluster_rdata(hammer2_cluster_t *cluster);
1314 hammer2_media_data_t *hammer2_cluster_wdata(hammer2_cluster_t *cluster);
1315 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
1316 int hammer2_cluster_modified(hammer2_cluster_t *cluster);
1317 int hammer2_cluster_duplicated(hammer2_cluster_t *cluster);
1318 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
1319 void hammer2_cluster_setflush(hammer2_trans_t *trans,
1320 hammer2_cluster_t *cluster);
1321 void hammer2_cluster_setmethod_check(hammer2_trans_t *trans,
1322 hammer2_cluster_t *cluster, int check_algo);
1323 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfs_t *pmp,
1324 hammer2_trans_t *trans,
1325 hammer2_blockref_t *bref);
1326 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
1327 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1328 void hammer2_cluster_wait(hammer2_cluster_t *cluster);
1329 void hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1330 void hammer2_cluster_lock_except(hammer2_cluster_t *cluster, int idx, int how);
1331 void hammer2_cluster_resolve(hammer2_cluster_t *cluster);
1332 void hammer2_cluster_forcegood(hammer2_cluster_t *cluster);
1333 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster);
1334 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1335 void hammer2_cluster_unlock_except(hammer2_cluster_t *cluster, int idx);
1336 void hammer2_cluster_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
1337 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1338 int nradix, int flags);
1339 hammer2_inode_data_t *hammer2_cluster_modify_ip(hammer2_trans_t *trans,
1340 hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1342 void hammer2_cluster_modify(hammer2_trans_t *trans, hammer2_cluster_t *cluster,
1344 void hammer2_cluster_modsync(hammer2_cluster_t *cluster);
1345 hammer2_cluster_t *hammer2_cluster_lookup_init(hammer2_cluster_t *cparent,
1347 void hammer2_cluster_lookup_done(hammer2_cluster_t *cparent);
1348 hammer2_cluster_t *hammer2_cluster_lookup(hammer2_cluster_t *cparent,
1349 hammer2_key_t *key_nextp,
1350 hammer2_key_t key_beg, hammer2_key_t key_end,
1352 hammer2_cluster_t *hammer2_cluster_next(hammer2_cluster_t *cparent,
1353 hammer2_cluster_t *cluster,
1354 hammer2_key_t *key_nextp,
1355 hammer2_key_t key_beg, hammer2_key_t key_end,
1357 void hammer2_cluster_next_single_chain(hammer2_cluster_t *cparent,
1358 hammer2_cluster_t *cluster,
1359 hammer2_key_t *key_nextp,
1360 hammer2_key_t key_beg,
1361 hammer2_key_t key_end,
1363 hammer2_cluster_t *hammer2_cluster_scan(hammer2_cluster_t *cparent,
1364 hammer2_cluster_t *cluster, int flags);
1365 int hammer2_cluster_create(hammer2_trans_t *trans, hammer2_cluster_t *cparent,
1366 hammer2_cluster_t **clusterp,
1367 hammer2_key_t key, int keybits,
1368 int type, size_t bytes, int flags);
1369 void hammer2_cluster_rename(hammer2_trans_t *trans, hammer2_blockref_t *bref,
1370 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1372 void hammer2_cluster_delete(hammer2_trans_t *trans, hammer2_cluster_t *pcluster,
1373 hammer2_cluster_t *cluster, int flags);
1374 int hammer2_cluster_snapshot(hammer2_trans_t *trans,
1375 hammer2_cluster_t *ocluster, hammer2_ioc_pfs_t *pfs);
1376 hammer2_cluster_t *hammer2_cluster_parent(hammer2_cluster_t *cluster);
1378 int hammer2_bulk_scan(hammer2_trans_t *trans, hammer2_chain_t *parent,
1379 int (*func)(hammer2_chain_t *chain, void *info),
1381 int hammer2_bulkfree_pass(hammer2_dev_t *hmp,
1382 struct hammer2_ioc_bulkfree *bfi);
1387 void hammer2_iocom_init(hammer2_dev_t *hmp);
1388 void hammer2_iocom_uninit(hammer2_dev_t *hmp);
1389 void hammer2_cluster_reconnect(hammer2_dev_t *hmp, struct file *fp);
1394 void hammer2_syncthr_create(hammer2_syncthr_t *thr, hammer2_pfs_t *pmp,
1395 int clindex, void (*func)(void *arg));
1396 void hammer2_syncthr_delete(hammer2_syncthr_t *thr);
1397 void hammer2_syncthr_remaster(hammer2_syncthr_t *thr);
1398 void hammer2_syncthr_freeze(hammer2_syncthr_t *thr);
1399 void hammer2_syncthr_unfreeze(hammer2_syncthr_t *thr);
1400 void hammer2_syncthr_primary(void *arg);
1403 * hammer2_strategy.c
1405 int hammer2_vop_strategy(struct vop_strategy_args *ap);
1406 int hammer2_vop_bmap(struct vop_bmap_args *ap);
1408 #endif /* !_KERNEL */
1409 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */