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_thread;
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
136 TAILQ_HEAD(hammer2_xop_list, hammer2_xop_head);
138 typedef struct hammer2_xop_list hammer2_xop_list_t;
142 * General lock support
146 hammer2_mtx_upgrade(hammer2_mtx_t *mtx)
150 if (mtx_islocked_ex(mtx)) {
154 mtx_lock_ex_quick(mtx);
161 * Downgrade an inode lock from exclusive to shared only if the inode
162 * lock was previously shared. If the inode lock was previously exclusive,
167 hammer2_mtx_downgrade(hammer2_mtx_t *mtx, int wasexclusive)
169 if (wasexclusive == 0)
174 * The xid tracks internal transactional updates.
176 * XXX fix-me, really needs to be 64-bits
178 typedef uint32_t hammer2_xid_t;
180 #define HAMMER2_XID_MIN 0x00000000U
181 #define HAMMER2_XID_MAX 0x7FFFFFFFU
184 * The chain structure tracks a portion of the media topology from the
185 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
186 * data blocks, and freemap nodes and leafs.
188 * The chain structure utilizes a simple singly-homed topology and the
189 * chain's in-memory topology will move around as the chains do, due mainly
190 * to renames and indirect block creation.
192 * Block Table Updates
194 * Block table updates for insertions and updates are delayed until the
195 * flush. This allows us to avoid having to modify the parent chain
196 * all the way to the root.
198 * Block table deletions are performed immediately (modifying the parent
199 * in the process) because the flush code uses the chain structure to
200 * track delayed updates and the chain will be (likely) gone or moved to
201 * another location in the topology after a deletion.
203 * A prior iteration of the code tried to keep the relationship intact
204 * on deletes by doing a delete-duplicate operation on the chain, but
205 * it added way too much complexity to the codebase.
207 * Flush Synchronization
209 * The flush code must flush modified chains bottom-up. Because chain
210 * structures can shift around and are NOT topologically stable,
211 * modified chains are independently indexed for the flush. As the flush
212 * runs it modifies (or further modifies) and updates the parents,
213 * propagating the flush all the way to the volume root.
215 * Modifying front-end operations can occur during a flush but will block
216 * in two cases: (1) when the front-end tries to operate on the inode
217 * currently in the midst of being flushed and (2) if the front-end
218 * crosses an inode currently being flushed (such as during a rename).
219 * So, for example, if you rename directory "x" to "a/b/c/d/e/f/g/x" and
220 * the flusher is currently working on "a/b/c", the rename will block
221 * temporarily in order to ensure that "x" exists in one place or the
224 * Meta-data statistics are updated by the flusher. The front-end will
225 * make estimates but meta-data must be fully synchronized only during a
226 * flush in order to ensure that it remains correct across a crash.
228 * Multiple flush synchronizations can theoretically be in-flight at the
229 * same time but the implementation is not coded to handle the case and
230 * currently serializes them.
234 * Snapshots currently require the subdirectory tree being snapshotted
235 * to be flushed. The snapshot then creates a new super-root inode which
236 * copies the flushed blockdata of the directory or file that was
241 * - Note that the radix tree runs in powers of 2 only so sub-trees
242 * cannot straddle edges.
244 RB_HEAD(hammer2_chain_tree, hammer2_chain);
245 TAILQ_HEAD(h2_flush_list, hammer2_chain);
246 TAILQ_HEAD(h2_core_list, hammer2_chain);
247 TAILQ_HEAD(h2_iocb_list, hammer2_iocb);
249 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
250 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
253 * Core topology for chain (embedded in chain). Protected by a spinlock.
255 struct hammer2_chain_core {
257 struct hammer2_chain_tree rbtree; /* sub-chains */
258 int live_zero; /* blockref array opt */
259 u_int live_count; /* live (not deleted) chains in tree */
260 u_int chain_count; /* live + deleted chains under core */
261 int generation; /* generation number (inserts only) */
264 typedef struct hammer2_chain_core hammer2_chain_core_t;
266 RB_HEAD(hammer2_io_tree, hammer2_io);
269 * IOCB - IO callback (into chain, cluster, or manual request)
271 struct hammer2_iocb {
272 TAILQ_ENTRY(hammer2_iocb) entry;
273 void (*callback)(struct hammer2_iocb *iocb);
274 struct hammer2_io *dio;
275 struct hammer2_cluster *cluster;
276 struct hammer2_chain *chain;
284 typedef struct hammer2_iocb hammer2_iocb_t;
286 #define HAMMER2_IOCB_INTERLOCK 0x00000001
287 #define HAMMER2_IOCB_ONQ 0x00000002
288 #define HAMMER2_IOCB_DONE 0x00000004
289 #define HAMMER2_IOCB_INPROG 0x00000008
290 #define HAMMER2_IOCB_UNUSED10 0x00000010
291 #define HAMMER2_IOCB_QUICK 0x00010000
292 #define HAMMER2_IOCB_ZERO 0x00020000
293 #define HAMMER2_IOCB_READ 0x00040000
294 #define HAMMER2_IOCB_WAKEUP 0x00080000
297 * DIO - Management structure wrapping system buffer cache.
299 * Used for multiple purposes including concurrent management
300 * if small requests by chains into larger DIOs.
303 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
304 struct h2_iocb_list iocbq;
305 struct spinlock spin;
306 struct hammer2_dev *hmp;
311 int act; /* activity */
314 typedef struct hammer2_io hammer2_io_t;
316 #define HAMMER2_DIO_INPROG 0x80000000 /* bio in progress */
317 #define HAMMER2_DIO_GOOD 0x40000000 /* dio->bp is stable */
318 #define HAMMER2_DIO_WAITING 0x20000000 /* (old) */
319 #define HAMMER2_DIO_DIRTY 0x10000000 /* flush on last drop */
321 #define HAMMER2_DIO_MASK 0x0FFFFFFF
324 * Primary chain structure keeps track of the topology in-memory.
326 struct hammer2_chain {
328 hammer2_chain_core_t core;
329 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
330 hammer2_blockref_t bref;
331 struct hammer2_chain *parent;
332 struct hammer2_state *state; /* if active cache msg */
333 struct hammer2_dev *hmp;
334 struct hammer2_pfs *pmp; /* A PFS or super-root (spmp) */
336 hammer2_xid_t flush_xid; /* flush sequencing */
337 hammer2_io_t *dio; /* physical data buffer */
338 u_int bytes; /* physical data size */
342 int error; /* on-lock data error state */
344 hammer2_media_data_t *data; /* data pointer shortcut */
345 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
348 typedef struct hammer2_chain hammer2_chain_t;
350 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
351 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
354 * Special notes on flags:
356 * INITIAL - This flag allows a chain to be created and for storage to
357 * be allocated without having to immediately instantiate the
358 * related buffer. The data is assumed to be all-zeros. It
359 * is primarily used for indirect blocks.
361 * MODIFIED - The chain's media data has been modified.
363 * UPDATE - Chain might not be modified but parent blocktable needs update
365 * FICTITIOUS - Faked chain as a placeholder for an error condition. This
366 * chain is unsuitable for I/O.
368 * BMAPPED - Indicates that the chain is present in the parent blockmap.
370 * BMAPUPD - Indicates that the chain is present but needs to be updated
371 * in the parent blockmap.
373 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
374 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
375 #define HAMMER2_CHAIN_DESTROY 0x00000004
376 #define HAMMER2_CHAIN_UNUSED0008 0x00000008
377 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
378 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
379 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
380 #define HAMMER2_CHAIN_DEFERRED 0x00000080 /* flush depth defer */
381 #define HAMMER2_CHAIN_IOFLUSH 0x00000100 /* bawrite on put */
382 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
383 #define HAMMER2_CHAIN_FICTITIOUS 0x00000400 /* unsuitable for I/O */
384 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
385 #define HAMMER2_CHAIN_DELAYED 0x00001000 /* delayed flush */
386 #define HAMMER2_CHAIN_COUNTEDBREFS 0x00002000 /* block table stats */
387 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
388 #define HAMMER2_CHAIN_UNUSED00008000 0x00008000
389 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
390 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
391 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
392 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
393 #define HAMMER2_CHAIN_IOINPROG 0x00100000 /* I/O interlock */
394 #define HAMMER2_CHAIN_IOSIGNAL 0x00200000 /* I/O interlock */
395 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
397 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
398 HAMMER2_CHAIN_UPDATE | \
399 HAMMER2_CHAIN_ONFLUSH)
402 * Hammer2 error codes, used by chain->error and cluster->error. The error
403 * code is typically set on-lock unless no I/O was requested, and set on
404 * I/O otherwise. If set for a cluster it generally means that the cluster
405 * code could not find a valid copy to present.
407 * IO - An I/O error occurred
408 * CHECK - I/O succeeded but did not match the check code
409 * INCOMPLETE - A cluster is not complete enough to use, or
410 * a chain cannot be loaded because its parent has an error.
412 * NOTE: API allows callers to check zero/non-zero to determine if an error
415 * NOTE: Chain's data field is usually NULL on an IO error but not necessarily
416 * NULL on other errors. Check chain->error, not chain->data.
418 #define HAMMER2_ERROR_NONE 0
419 #define HAMMER2_ERROR_IO 1 /* device I/O error */
420 #define HAMMER2_ERROR_CHECK 2 /* check code mismatch */
421 #define HAMMER2_ERROR_INCOMPLETE 3 /* incomplete cluster */
422 #define HAMMER2_ERROR_DEPTH 4 /* temporary depth limit */
425 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
428 * NOLOCK - Input and output chains are referenced only and not
429 * locked. Output chain might be temporarily locked
432 * NODATA - Asks that the chain->data not be resolved in order
435 * NODIRECT - Prevents a lookup of offset 0 in an inode from returning
436 * the inode itself if the inode is in DIRECTDATA mode
437 * (i.e. file is <= 512 bytes). Used by the synchronization
438 * code to prevent confusion.
440 * SHARED - The input chain is expected to be locked shared,
441 * and the output chain is locked shared.
443 * MATCHIND - Allows an indirect block / freemap node to be returned
444 * when the passed key range matches the radix. Remember
445 * that key_end is inclusive (e.g. {0x000,0xFFF},
446 * not {0x000,0x1000}).
448 * (Cannot be used for remote or cluster ops).
450 * ALLNODES - Allows NULL focus.
452 * ALWAYS - Always resolve the data. If ALWAYS and NODATA are both
453 * missing, bulk file data is not resolved but inodes and
454 * other meta-data will.
456 * NOUNLOCK - Used by hammer2_chain_next() to leave the lock on
457 * the input chain intact. The chain is still dropped.
458 * This allows the caller to add a reference to the chain
459 * and retain it in a locked state (used by the
460 * XOP/feed/collect code).
462 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
463 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
464 #define HAMMER2_LOOKUP_NODIRECT 0x00000004 /* no offset=0 DD */
465 #define HAMMER2_LOOKUP_SHARED 0x00000100
466 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
467 #define HAMMER2_LOOKUP_ALLNODES 0x00000400 /* allow NULL focus */
468 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
469 #define HAMMER2_LOOKUP_NOUNLOCK 0x00001000 /* leave lock intact */
472 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
474 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
475 * blocks in the INITIAL-create state.
477 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
478 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
479 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
480 #define HAMMER2_MODIFY_NOREALLOC 0x00000010
483 * Flags passed to hammer2_chain_lock()
485 * NOTE: RDONLY is set to optimize cluster operations when *no* modifications
486 * will be made to either the cluster being locked or any underlying
487 * cluster. It allows the cluster to lock and access data for a subset
488 * of available nodes instead of all available nodes.
490 #define HAMMER2_RESOLVE_NEVER 1
491 #define HAMMER2_RESOLVE_MAYBE 2
492 #define HAMMER2_RESOLVE_ALWAYS 3
493 #define HAMMER2_RESOLVE_MASK 0x0F
495 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
496 #define HAMMER2_RESOLVE_UNUSED20 0x20
497 #define HAMMER2_RESOLVE_RDONLY 0x40 /* higher level op flag */
500 * Flags passed to hammer2_chain_delete()
502 #define HAMMER2_DELETE_PERMANENT 0x0001
503 #define HAMMER2_DELETE_NOSTATS 0x0002
505 #define HAMMER2_INSERT_NOSTATS 0x0002
506 #define HAMMER2_INSERT_PFSROOT 0x0004
509 * Flags passed to hammer2_chain_delete_duplicate()
511 #define HAMMER2_DELDUP_RECORE 0x0001
514 * Cluster different types of storage together for allocations
516 #define HAMMER2_FREECACHE_INODE 0
517 #define HAMMER2_FREECACHE_INDIR 1
518 #define HAMMER2_FREECACHE_DATA 2
519 #define HAMMER2_FREECACHE_UNUSED3 3
520 #define HAMMER2_FREECACHE_TYPES 4
523 * hammer2_freemap_alloc() block preference
525 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
528 * BMAP read-ahead maximum parameters
530 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
531 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
534 * hammer2_freemap_adjust()
536 #define HAMMER2_FREEMAP_DORECOVER 1
537 #define HAMMER2_FREEMAP_DOMAYFREE 2
538 #define HAMMER2_FREEMAP_DOREALFREE 3
541 * HAMMER2 cluster - A set of chains representing the same entity.
543 * hammer2_cluster typically represents a temporary set of representitive
544 * chains. The one exception is that a hammer2_cluster is embedded in
545 * hammer2_inode. This embedded cluster is ONLY used to track the
546 * representitive chains and cannot be directly locked.
548 * A cluster is usually temporary (and thus per-thread) for locking purposes,
549 * allowing us to embed the asynchronous storage required for cluster
550 * operations in the cluster itself and adjust the state and status without
551 * having to worry too much about SMP issues.
553 * The exception is the cluster embedded in the hammer2_inode structure.
554 * This is used to cache the cluster state on an inode-by-inode basis.
555 * Individual hammer2_chain structures not incorporated into clusters might
556 * also stick around to cache miscellanious elements.
558 * Because the cluster is a 'working copy' and is usually subject to cluster
559 * quorum rules, it is quite possible for us to end up with an insufficient
560 * number of live chains to execute an operation. If an insufficient number
561 * of chains remain in a working copy, the operation may have to be
562 * downgraded, retried, stall until the requisit number of chains are
563 * available, or possibly even error out depending on the mount type.
565 * A cluster's focus is set when it is locked. The focus can only be set
566 * to a chain still part of the synchronized set.
568 #define HAMMER2_MAXCLUSTER 8
569 #define HAMMER2_XOPFIFO 16
570 #define HAMMER2_XOPFIFO_MASK (HAMMER2_XOPFIFO - 1)
571 #define HAMMER2_XOPGROUPS 16
572 #define HAMMER2_XOPGROUPS_MASK (HAMMER2_XOPGROUPS - 1)
573 #define HAMMER2_XOPMASK_VOP 0x80000000U
575 struct hammer2_cluster_item {
577 hammer2_mtx_link_t async_link;
579 hammer2_chain_t *chain;
581 struct hammer2_cluster *cluster; /* link back to cluster */
587 typedef struct hammer2_cluster_item hammer2_cluster_item_t;
590 * INVALID - Invalid for focus, i.e. not part of synchronized set.
591 * Once set, this bit is sticky across operations.
593 * FEMOD - Indicates that front-end modifying operations can
594 * mess with this entry and MODSYNC will copy also
597 #define HAMMER2_CITEM_INVALID 0x00000001
598 #define HAMMER2_CITEM_FEMOD 0x00000002
599 #define HAMMER2_CITEM_NULL 0x00000004
601 struct hammer2_cluster {
602 int refs; /* track for deallocation */
604 struct hammer2_pfs *pmp;
607 int error; /* error code valid on lock */
610 hammer2_chain_t *focus; /* current focus (or mod) */
611 hammer2_cluster_item_t array[HAMMER2_MAXCLUSTER];
614 typedef struct hammer2_cluster hammer2_cluster_t;
617 * WRHARD - Hard mounts can write fully synchronized
618 * RDHARD - Hard mounts can read fully synchronized
619 * UNHARD - Unsynchronized masters present
620 * NOHARD - No masters visible
621 * WRSOFT - Soft mounts can write to at least the SOFT_MASTER
622 * RDSOFT - Soft mounts can read from at least a SOFT_SLAVE
623 * UNSOFT - Unsynchronized slaves present
624 * NOSOFT - No slaves visible
625 * RDSLAVE - slaves are accessible (possibly unsynchronized or remote).
626 * MSYNCED - All masters are fully synchronized
627 * SSYNCED - All known local slaves are fully synchronized to masters
629 * All available masters are always incorporated. All PFSs belonging to a
630 * cluster (master, slave, copy, whatever) always try to synchronize the
631 * total number of known masters in the PFSs root inode.
633 * A cluster might have access to many slaves, copies, or caches, but we
634 * have a limited number of cluster slots. Any such elements which are
635 * directly mounted from block device(s) will always be incorporated. Note
636 * that SSYNCED only applies to such elements which are directly mounted,
637 * not to any remote slaves, copies, or caches that could be available. These
638 * bits are used to monitor and drive our synchronization threads.
640 * When asking the question 'is any data accessible at all', then a simple
641 * test against (RDHARD|RDSOFT|RDSLAVE) gives you the answer. If any of
642 * these bits are set the object can be read with certain caveats:
643 * RDHARD - no caveats. RDSOFT - authoritative but might not be synchronized.
644 * and RDSLAVE - not authoritative, has some data but it could be old or
647 * When both soft and hard mounts are available, data will be read and written
648 * via the soft mount only. But all might be in the cluster because
649 * background synchronization threads still need to do their work.
651 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode struct */
652 #define HAMMER2_CLUSTER_UNUSED2 0x00000002
653 #define HAMMER2_CLUSTER_LOCKED 0x00000004 /* cluster lks not recursive */
654 #define HAMMER2_CLUSTER_WRHARD 0x00000100 /* hard-mount can write */
655 #define HAMMER2_CLUSTER_RDHARD 0x00000200 /* hard-mount can read */
656 #define HAMMER2_CLUSTER_UNHARD 0x00000400 /* unsynchronized masters */
657 #define HAMMER2_CLUSTER_NOHARD 0x00000800 /* no masters visible */
658 #define HAMMER2_CLUSTER_WRSOFT 0x00001000 /* soft-mount can write */
659 #define HAMMER2_CLUSTER_RDSOFT 0x00002000 /* soft-mount can read */
660 #define HAMMER2_CLUSTER_UNSOFT 0x00004000 /* unsynchronized slaves */
661 #define HAMMER2_CLUSTER_NOSOFT 0x00008000 /* no slaves visible */
662 #define HAMMER2_CLUSTER_MSYNCED 0x00010000 /* all masters synchronized */
663 #define HAMMER2_CLUSTER_SSYNCED 0x00020000 /* known slaves synchronized */
665 #define HAMMER2_CLUSTER_ANYDATA ( HAMMER2_CLUSTER_RDHARD | \
666 HAMMER2_CLUSTER_RDSOFT | \
667 HAMMER2_CLUSTER_RDSLAVE)
669 #define HAMMER2_CLUSTER_RDOK ( HAMMER2_CLUSTER_RDHARD | \
670 HAMMER2_CLUSTER_RDSOFT)
672 #define HAMMER2_CLUSTER_WROK ( HAMMER2_CLUSTER_WRHARD | \
673 HAMMER2_CLUSTER_WRSOFT)
675 #define HAMMER2_CLUSTER_ZFLAGS ( HAMMER2_CLUSTER_WRHARD | \
676 HAMMER2_CLUSTER_RDHARD | \
677 HAMMER2_CLUSTER_WRSOFT | \
678 HAMMER2_CLUSTER_RDSOFT | \
679 HAMMER2_CLUSTER_MSYNCED | \
680 HAMMER2_CLUSTER_SSYNCED)
683 * Helper functions (cluster must be locked for flags to be valid).
687 hammer2_cluster_rdok(hammer2_cluster_t *cluster)
689 return (cluster->flags & HAMMER2_CLUSTER_RDOK);
694 hammer2_cluster_wrok(hammer2_cluster_t *cluster)
696 return (cluster->flags & HAMMER2_CLUSTER_WROK);
699 RB_HEAD(hammer2_inode_tree, hammer2_inode);
704 * NOTE: The inode-embedded cluster is never used directly for I/O (since
705 * it may be shared). Instead it will be replicated-in and synchronized
706 * back out if changed.
708 struct hammer2_inode {
709 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
710 hammer2_mtx_t lock; /* inode lock */
711 struct hammer2_pfs *pmp; /* PFS mount */
712 struct hammer2_inode *pip; /* parent inode */
714 struct spinlock cluster_spin; /* update cluster */
715 hammer2_cluster_t cluster;
716 struct lockf advlock;
718 u_int refs; /* +vpref, +flushref */
719 uint8_t comp_heuristic;
720 hammer2_inode_meta_t meta; /* copy of meta-data */
721 hammer2_blockref_t bref; /* copy of bref statistics */
725 typedef struct hammer2_inode hammer2_inode_t;
728 * MODIFIED - Inode is in a modified state, ip->meta may have changes.
729 * RESIZED - Inode truncated (any) or inode extended beyond
732 #define HAMMER2_INODE_MODIFIED 0x0001
733 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
734 #define HAMMER2_INODE_RENAME_INPROG 0x0004
735 #define HAMMER2_INODE_ONRBTREE 0x0008
736 #define HAMMER2_INODE_RESIZED 0x0010 /* requires inode_fsync */
737 #define HAMMER2_INODE_UNUSED0020 0x0020
738 #define HAMMER2_INODE_ISUNLINKED 0x0040
739 #define HAMMER2_INODE_METAGOOD 0x0080 /* inode meta-data good */
741 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
742 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
746 * inode-unlink side-structure
748 struct hammer2_inode_unlink {
749 TAILQ_ENTRY(hammer2_inode_unlink) entry;
752 TAILQ_HEAD(h2_unlk_list, hammer2_inode_unlink);
754 typedef struct hammer2_inode_unlink hammer2_inode_unlink_t;
757 * A hammer2 transaction and flush sequencing structure.
759 * This global structure is tied into hammer2_dev and is used
760 * to sequence modifying operations and flushes. These operations
761 * run on whole cluster PFSs, not individual nodes (at this level),
762 * so we do not record mirror_tid here.
764 struct hammer2_trans {
765 TAILQ_ENTRY(hammer2_trans) entry;
766 struct hammer2_pfs *pmp;
767 hammer2_xid_t sync_xid; /* transaction sequencer */
768 hammer2_tid_t inode_tid; /* inode number assignment */
769 hammer2_tid_t modify_tid; /* modify transaction id */
770 thread_t td; /* pointer */
773 uint8_t inodes_created;
777 typedef struct hammer2_trans hammer2_trans_t;
779 #define HAMMER2_TRANS_ISFLUSH 0x0001 /* formal flush */
780 #define HAMMER2_TRANS_CONCURRENT 0x0002 /* concurrent w/flush */
781 #define HAMMER2_TRANS_BUFCACHE 0x0004 /* from bioq strategy write */
782 #define HAMMER2_TRANS_NEWINODE 0x0008 /* caller allocating inode */
783 #define HAMMER2_TRANS_KEEPMODIFY 0x0010 /* do not change bref.modify */
784 #define HAMMER2_TRANS_PREFLUSH 0x0020 /* preflush state */
786 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
787 #define HAMMER2_FREEMAP_HEUR_TYPES 8
788 #define HAMMER2_FREEMAP_HEUR (HAMMER2_FREEMAP_HEUR_NRADIX * \
789 HAMMER2_FREEMAP_HEUR_TYPES)
792 * Transaction Rendezvous
794 TAILQ_HEAD(hammer2_trans_queue, hammer2_trans);
796 struct hammer2_trans_manage {
797 hammer2_xid_t flush_xid; /* last flush transaction */
798 hammer2_xid_t alloc_xid;
799 struct lock translk; /* lockmgr lock */
800 struct hammer2_trans_queue transq; /* modifying transactions */
801 int flushcnt; /* track flush trans */
804 typedef struct hammer2_trans_manage hammer2_trans_manage_t;
807 * Hammer2 support thread element.
809 * Potentially many support threads can hang off of hammer2, primarily
810 * off the hammer2_pfs structure. Typically:
812 * td x Nodes A synchronization thread for each node.
813 * td x Nodes x workers Worker threads for frontend operations.
814 * td x 1 Bioq thread for logical buffer writes.
816 * In addition, the synchronization thread(s) associated with the
817 * super-root PFS (spmp) for a node is responsible for automatic bulkfree
820 struct hammer2_thread {
821 struct hammer2_pfs *pmp;
825 int clindex; /* cluster element index */
827 hammer2_trans_t trans;
828 struct lock lk; /* thread control lock */
829 hammer2_xop_list_t xopq;
832 typedef struct hammer2_thread hammer2_thread_t;
834 #define HAMMER2_THREAD_UNMOUNTING 0x0001 /* unmount request */
835 #define HAMMER2_THREAD_DEV 0x0002 /* related to dev, not pfs */
836 #define HAMMER2_THREAD_UNUSED04 0x0004
837 #define HAMMER2_THREAD_REMASTER 0x0008 /* remaster request */
838 #define HAMMER2_THREAD_STOP 0x0010 /* exit request */
839 #define HAMMER2_THREAD_FREEZE 0x0020 /* force idle */
840 #define HAMMER2_THREAD_FROZEN 0x0040 /* restart */
844 * hammer2_xop - container for VOP/XOP operation (allocated, not on stack).
846 * This structure is used to distribute a VOP operation across multiple
847 * nodes. It provides a rendezvous for concurrent node execution and
848 * can be detached from the frontend operation to allow the frontend to
851 typedef void (*hammer2_xop_func_t)(union hammer2_xop *xop, int clidx);
853 typedef struct hammer2_xop_fifo {
854 TAILQ_ENTRY(hammer2_xop_head) entry;
855 hammer2_chain_t *array[HAMMER2_XOPFIFO];
856 int errors[HAMMER2_XOPFIFO];
860 } hammer2_xop_fifo_t;
862 struct hammer2_xop_head {
863 hammer2_xop_func_t func;
864 struct hammer2_inode *ip;
865 struct hammer2_xop_group *xgrp;
866 uint32_t check_counter;
873 hammer2_xop_fifo_t collect[HAMMER2_MAXCLUSTER];
874 hammer2_cluster_t cluster; /* help collections */
877 typedef struct hammer2_xop_head hammer2_xop_head_t;
879 struct hammer2_xop_readdir {
880 hammer2_xop_head_t head;
883 typedef struct hammer2_xop_readdir hammer2_xop_readdir_t;
886 hammer2_xop_head_t head;
887 hammer2_xop_readdir_t xop_readdir;
890 typedef union hammer2_xop hammer2_xop_t;
893 * hammer2_xop_group - Manage XOP support threads.
895 struct hammer2_xop_group {
896 hammer2_thread_t thrs[HAMMER2_MAXCLUSTER];
900 typedef struct hammer2_xop_group hammer2_xop_group_t;
903 * Global (per partition) management structure, represents a hard block
904 * device. Typically referenced by hammer2_chain structures when applicable.
905 * Typically not used for network-managed elements.
907 * Note that a single hammer2_dev can be indirectly tied to multiple system
908 * mount points. There is no direct relationship. System mounts are
909 * per-cluster-id, not per-block-device, and a single hard mount might contain
910 * many PFSs and those PFSs might combine together in various ways to form
911 * the set of available clusters.
914 struct vnode *devvp; /* device vnode */
915 int ronly; /* read-only mount */
916 int mount_count; /* number of actively mounted PFSs */
917 TAILQ_ENTRY(hammer2_dev) mntentry; /* hammer2_mntlist */
919 struct malloc_type *mchain;
922 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
923 struct spinlock io_spin; /* iotree access */
924 struct hammer2_io_tree iotree;
926 hammer2_chain_t vchain; /* anchor chain (topology) */
927 hammer2_chain_t fchain; /* anchor chain (freemap) */
928 struct spinlock list_spin;
929 struct h2_flush_list flushq; /* flush seeds */
930 struct hammer2_pfs *spmp; /* super-root pmp for transactions */
931 struct lock vollk; /* lockmgr lock */
932 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR];
933 int volhdrno; /* last volhdrno written */
934 char devrepname[64]; /* for kprintf */
935 hammer2_volume_data_t voldata;
936 hammer2_volume_data_t volsync; /* synchronized voldata */
939 typedef struct hammer2_dev hammer2_dev_t;
942 * Helper functions (cluster must be locked for flags to be valid).
946 hammer2_chain_rdok(hammer2_chain_t *chain)
948 return (chain->error == 0);
953 hammer2_chain_wrok(hammer2_chain_t *chain)
955 return (chain->error == 0 && chain->hmp->ronly == 0);
959 * Per-cluster management structure. This structure will be tied to a
960 * system mount point if the system is mounting the PFS, but is also used
961 * to manage clusters encountered during the super-root scan or received
962 * via LNK_SPANs that might not be mounted.
964 * This structure is also used to represent the super-root that hangs off
965 * of a hard mount point. The super-root is not really a cluster element.
966 * In this case the spmp_hmp field will be non-NULL. It's just easier to do
967 * this than to special case super-root manipulation in the hammer2_chain*
968 * code as being only hammer2_dev-related.
970 * pfs_mode and pfs_nmasters are rollup fields which critically describes
971 * how elements of the cluster act on the cluster. pfs_mode is only applicable
972 * when a PFS is mounted by the system. pfs_nmasters is our best guess as to
973 * how many masters have been configured for a cluster and is always
974 * applicable. pfs_types[] is an array with 1:1 correspondance to the
975 * iroot cluster and describes the PFS types of the nodes making up the
978 * WARNING! Portions of this structure have deferred initialization. In
979 * particular, if not mounted there will be no ihidden or wthread.
980 * umounted network PFSs will also be missing iroot and numerous
981 * other fields will not be initialized prior to mount.
983 * Synchronization threads are chain-specific and only applicable
984 * to local hard PFS entries. A hammer2_pfs structure may contain
985 * more than one when multiple hard PFSs are present on the local
986 * machine which require synchronization monitoring. Most PFSs
987 * (such as snapshots) are 1xMASTER PFSs which do not need a
988 * synchronization thread.
990 * WARNING! The chains making up pfs->iroot's cluster are accounted for in
991 * hammer2_dev->mount_count when the pfs is associated with a mount
996 TAILQ_ENTRY(hammer2_pfs) mntentry; /* hammer2_pfslist */
998 hammer2_dev_t *spmp_hmp; /* only if super-root pmp */
999 hammer2_inode_t *iroot; /* PFS root inode */
1000 hammer2_inode_t *ihidden; /* PFS hidden directory */
1001 uint8_t pfs_types[HAMMER2_MAXCLUSTER];
1002 char *pfs_names[HAMMER2_MAXCLUSTER];
1003 hammer2_trans_manage_t tmanage; /* transaction management */
1004 struct lock lock; /* PFS lock for certain ops */
1005 struct netexport export; /* nfs export */
1006 int ronly; /* read-only mount */
1007 struct malloc_type *minode;
1008 struct malloc_type *mmsg;
1009 struct spinlock inum_spin; /* inumber lookup */
1010 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
1011 hammer2_tid_t modify_tid; /* modify transaction id */
1012 hammer2_tid_t inode_tid; /* inode allocator */
1013 uint8_t pfs_nmasters; /* total masters */
1014 uint8_t pfs_mode; /* operating mode PFSMODE */
1019 uint32_t inmem_dirty_chains;
1020 int count_lwinprog; /* logical write in prog */
1021 struct spinlock list_spin;
1022 struct h2_unlk_list unlinkq; /* last-close unlink */
1023 hammer2_thread_t sync_thrs[HAMMER2_MAXCLUSTER];
1024 thread_t wthread_td; /* write thread td */
1025 struct bio_queue_head wthread_bioq; /* logical buffer bioq */
1026 hammer2_mtx_t wthread_mtx; /* interlock */
1027 int wthread_destroy;/* termination sequencing */
1028 uint32_t flags; /* cached cluster flags */
1029 hammer2_xop_group_t xop_groups[HAMMER2_XOPGROUPS];
1032 typedef struct hammer2_pfs hammer2_pfs_t;
1034 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
1035 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
1037 #define HAMMER2_LWINPROG_WAITING 0x80000000
1038 #define HAMMER2_LWINPROG_MASK 0x7FFFFFFF
1041 * hammer2_cluster_check
1043 #define HAMMER2_CHECK_NULL 0x00000001
1048 #define HAMMER2_BULK_ABORT 0x00000001
1053 #if defined(_KERNEL)
1055 MALLOC_DECLARE(M_HAMMER2);
1057 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
1058 #define ITOV(ip) ((ip)->vp)
1061 * Currently locked chains retain the locked buffer cache buffer for
1062 * indirect blocks, and indirect blocks can be one of two sizes. The
1063 * device buffer has to match the case to avoid deadlocking recursive
1064 * chains that might otherwise try to access different offsets within
1065 * the same device buffer.
1069 hammer2_devblkradix(int radix)
1072 if (radix <= HAMMER2_LBUFRADIX) {
1073 return (HAMMER2_LBUFRADIX);
1075 return (HAMMER2_PBUFRADIX);
1078 return (HAMMER2_PBUFRADIX);
1082 * XXX almost time to remove this. DIO uses PBUFSIZE exclusively now.
1086 hammer2_devblksize(size_t bytes)
1089 if (bytes <= HAMMER2_LBUFSIZE) {
1090 return(HAMMER2_LBUFSIZE);
1092 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
1093 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
1094 return (HAMMER2_PBUFSIZE);
1097 return (HAMMER2_PBUFSIZE);
1103 MPTOPMP(struct mount *mp)
1105 return ((hammer2_pfs_t *)mp->mnt_data);
1108 #define LOCKSTART int __nlocks = curthread->td_locks
1109 #define LOCKENTER (++curthread->td_locks)
1110 #define LOCKEXIT (--curthread->td_locks)
1111 #define LOCKSTOP KKASSERT(curthread->td_locks == __nlocks)
1113 extern struct vop_ops hammer2_vnode_vops;
1114 extern struct vop_ops hammer2_spec_vops;
1115 extern struct vop_ops hammer2_fifo_vops;
1117 extern int hammer2_debug;
1118 extern int hammer2_cluster_enable;
1119 extern int hammer2_hardlink_enable;
1120 extern int hammer2_flush_pipe;
1121 extern int hammer2_synchronous_flush;
1122 extern int hammer2_dio_count;
1123 extern long hammer2_limit_dirty_chains;
1124 extern long hammer2_iod_file_read;
1125 extern long hammer2_iod_meta_read;
1126 extern long hammer2_iod_indr_read;
1127 extern long hammer2_iod_fmap_read;
1128 extern long hammer2_iod_volu_read;
1129 extern long hammer2_iod_file_write;
1130 extern long hammer2_iod_meta_write;
1131 extern long hammer2_iod_indr_write;
1132 extern long hammer2_iod_fmap_write;
1133 extern long hammer2_iod_volu_write;
1134 extern long hammer2_ioa_file_read;
1135 extern long hammer2_ioa_meta_read;
1136 extern long hammer2_ioa_indr_read;
1137 extern long hammer2_ioa_fmap_read;
1138 extern long hammer2_ioa_volu_read;
1139 extern long hammer2_ioa_file_write;
1140 extern long hammer2_ioa_meta_write;
1141 extern long hammer2_ioa_indr_write;
1142 extern long hammer2_ioa_fmap_write;
1143 extern long hammer2_ioa_volu_write;
1145 extern struct objcache *cache_buffer_read;
1146 extern struct objcache *cache_buffer_write;
1147 extern struct objcache *cache_xops;
1150 extern int write_thread_wakeup;
1155 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
1156 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
1158 int hammer2_signal_check(time_t *timep);
1159 const char *hammer2_error_str(int error);
1161 void hammer2_inode_lock(hammer2_inode_t *ip, int how);
1162 void hammer2_inode_unlock(hammer2_inode_t *ip, hammer2_cluster_t *cluster);
1163 hammer2_cluster_t *hammer2_inode_cluster(hammer2_inode_t *ip, int how);
1164 hammer2_chain_t *hammer2_inode_chain(hammer2_inode_t *ip, int clindex, int how);
1165 hammer2_mtx_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
1166 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip,
1167 hammer2_mtx_state_t ostate);
1168 int hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
1169 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int);
1171 void hammer2_dev_exlock(hammer2_dev_t *hmp);
1172 void hammer2_dev_shlock(hammer2_dev_t *hmp);
1173 void hammer2_dev_unlock(hammer2_dev_t *hmp);
1175 int hammer2_get_dtype(const hammer2_inode_data_t *ipdata);
1176 int hammer2_get_vtype(uint8_t type);
1177 u_int8_t hammer2_get_obj_type(enum vtype vtype);
1178 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1179 u_int64_t hammer2_timespec_to_time(const struct timespec *ts);
1180 u_int32_t hammer2_to_unix_xid(const uuid_t *uuid);
1181 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1182 hammer2_xid_t hammer2_trans_newxid(hammer2_pfs_t *pmp);
1183 void hammer2_trans_manage_init(hammer2_trans_manage_t *tman);
1185 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
1186 int hammer2_getradix(size_t bytes);
1188 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
1189 hammer2_key_t *lbasep, hammer2_key_t *leofp);
1190 int hammer2_calc_physical(hammer2_inode_t *ip, hammer2_key_t lbase);
1191 void hammer2_update_time(uint64_t *timep);
1192 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
1197 struct vnode *hammer2_igetv(hammer2_inode_t *ip, hammer2_cluster_t *cparent,
1199 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfs_t *pmp,
1200 hammer2_tid_t inum);
1201 hammer2_inode_t *hammer2_inode_get(hammer2_pfs_t *pmp,
1202 hammer2_inode_t *dip, hammer2_cluster_t *cluster);
1203 void hammer2_inode_free(hammer2_inode_t *ip);
1204 void hammer2_inode_ref(hammer2_inode_t *ip);
1205 void hammer2_inode_drop(hammer2_inode_t *ip);
1206 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
1207 hammer2_cluster_t *cluster);
1208 void hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1210 void hammer2_inode_modify(hammer2_trans_t *trans, hammer2_inode_t *ip);
1211 void hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfs_t *pmp);
1213 hammer2_inode_t *hammer2_inode_create(hammer2_trans_t *trans,
1214 hammer2_inode_t *dip,
1215 struct vattr *vap, struct ucred *cred,
1216 const uint8_t *name, size_t name_len,
1217 hammer2_cluster_t **clusterp,
1218 int flags, int *errorp);
1219 int hammer2_inode_connect(hammer2_trans_t *trans,
1220 hammer2_inode_t *ip, hammer2_cluster_t **clusterp,
1222 hammer2_inode_t *dip, hammer2_cluster_t *dcluster,
1223 const uint8_t *name, size_t name_len,
1225 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
1226 hammer2_inode_t *tdip);
1227 void hammer2_inode_fsync(hammer2_trans_t *trans, hammer2_inode_t *ip,
1228 hammer2_cluster_t *cparent);
1229 int hammer2_unlink_file(hammer2_trans_t *trans,
1230 hammer2_inode_t *dip, hammer2_inode_t *ip,
1231 const uint8_t *name, size_t name_len, int isdir,
1232 int *hlinkp, struct nchandle *nch, int nlinks);
1233 int hammer2_hardlink_consolidate(hammer2_trans_t *trans,
1234 hammer2_inode_t *ip, hammer2_cluster_t **clusterp,
1235 hammer2_inode_t *cdip, hammer2_cluster_t *cdcluster,
1237 int hammer2_hardlink_deconsolidate(hammer2_trans_t *trans, hammer2_inode_t *dip,
1238 hammer2_chain_t **chainp, hammer2_chain_t **ochainp);
1239 int hammer2_hardlink_find(hammer2_inode_t *dip, hammer2_cluster_t **cparentp,
1240 hammer2_cluster_t **clusterp);
1241 int hammer2_parent_find(hammer2_cluster_t **cparentp,
1242 hammer2_cluster_t *cluster);
1243 void hammer2_inode_install_hidden(hammer2_pfs_t *pmp);
1248 void hammer2_voldata_lock(hammer2_dev_t *hmp);
1249 void hammer2_voldata_unlock(hammer2_dev_t *hmp);
1250 void hammer2_voldata_modify(hammer2_dev_t *hmp);
1251 hammer2_chain_t *hammer2_chain_alloc(hammer2_dev_t *hmp,
1253 hammer2_trans_t *trans,
1254 hammer2_blockref_t *bref);
1255 void hammer2_chain_core_init(hammer2_chain_t *chain);
1256 void hammer2_chain_ref(hammer2_chain_t *chain);
1257 void hammer2_chain_drop(hammer2_chain_t *chain);
1258 void hammer2_chain_lock(hammer2_chain_t *chain, int how);
1259 void hammer2_chain_load_data(hammer2_chain_t *chain);
1260 const hammer2_media_data_t *hammer2_chain_rdata(hammer2_chain_t *chain);
1261 hammer2_media_data_t *hammer2_chain_wdata(hammer2_chain_t *chain);
1266 void hammer2_cluster_load_async(hammer2_cluster_t *cluster,
1267 void (*callback)(hammer2_iocb_t *iocb),
1269 void hammer2_chain_moved(hammer2_chain_t *chain);
1270 void hammer2_chain_modify(hammer2_trans_t *trans,
1271 hammer2_chain_t *chain, int flags);
1272 void hammer2_chain_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
1273 hammer2_chain_t *parent,
1274 hammer2_chain_t *chain,
1275 int nradix, int flags);
1276 void hammer2_chain_unlock(hammer2_chain_t *chain);
1277 void hammer2_chain_wait(hammer2_chain_t *chain);
1278 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
1279 hammer2_blockref_t *bref);
1280 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
1281 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
1282 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
1283 hammer2_key_t *key_nextp,
1284 hammer2_key_t key_beg, hammer2_key_t key_end,
1285 int *cache_indexp, int flags);
1286 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
1287 hammer2_chain_t *chain,
1288 hammer2_key_t *key_nextp,
1289 hammer2_key_t key_beg, hammer2_key_t key_end,
1290 int *cache_indexp, int flags);
1291 hammer2_chain_t *hammer2_chain_scan(hammer2_chain_t *parent,
1292 hammer2_chain_t *chain,
1293 int *cache_indexp, int flags);
1295 int hammer2_chain_create(hammer2_trans_t *trans, hammer2_chain_t **parentp,
1296 hammer2_chain_t **chainp,
1298 hammer2_key_t key, int keybits,
1299 int type, size_t bytes, int flags);
1300 void hammer2_chain_rename(hammer2_trans_t *trans, hammer2_blockref_t *bref,
1301 hammer2_chain_t **parentp,
1302 hammer2_chain_t *chain, int flags);
1303 int hammer2_chain_snapshot(hammer2_trans_t *trans, hammer2_chain_t **chainp,
1304 hammer2_ioc_pfs_t *pmp);
1305 void hammer2_chain_delete(hammer2_trans_t *trans, hammer2_chain_t *parent,
1306 hammer2_chain_t *chain, int flags);
1307 void hammer2_chain_delete_duplicate(hammer2_trans_t *trans,
1308 hammer2_chain_t **chainp, int flags);
1309 void hammer2_flush(hammer2_trans_t *trans, hammer2_chain_t *chain, int istop);
1310 void hammer2_delayed_flush(hammer2_trans_t *trans, hammer2_chain_t *chain);
1311 void hammer2_chain_commit(hammer2_trans_t *trans, hammer2_chain_t *chain);
1312 void hammer2_chain_setflush(hammer2_trans_t *trans, hammer2_chain_t *chain);
1313 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
1314 hammer2_blockref_t *base, int count);
1316 void hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata);
1317 int hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata);
1320 void hammer2_pfs_memory_wait(hammer2_pfs_t *pmp);
1321 void hammer2_pfs_memory_inc(hammer2_pfs_t *pmp);
1322 void hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp);
1324 void hammer2_base_delete(hammer2_trans_t *trans, hammer2_chain_t *chain,
1325 hammer2_blockref_t *base, int count,
1326 int *cache_indexp, hammer2_chain_t *child);
1327 void hammer2_base_insert(hammer2_trans_t *trans, hammer2_chain_t *chain,
1328 hammer2_blockref_t *base, int count,
1329 int *cache_indexp, hammer2_chain_t *child);
1334 void hammer2_trans_init(hammer2_trans_t *trans, hammer2_pfs_t *pmp,
1336 void hammer2_trans_done(hammer2_trans_t *trans);
1337 void hammer2_trans_assert_strategy(hammer2_pfs_t *pmp);
1342 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
1343 int fflag, struct ucred *cred);
1348 void hammer2_io_putblk(hammer2_io_t **diop);
1349 void hammer2_io_cleanup(hammer2_dev_t *hmp, struct hammer2_io_tree *tree);
1350 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
1351 void hammer2_io_getblk(hammer2_dev_t *hmp, off_t lbase, int lsize,
1352 hammer2_iocb_t *iocb);
1353 void hammer2_io_complete(hammer2_iocb_t *iocb);
1354 void hammer2_io_callback(struct bio *bio);
1355 void hammer2_iocb_wait(hammer2_iocb_t *iocb);
1356 int hammer2_io_new(hammer2_dev_t *hmp, off_t lbase, int lsize,
1357 hammer2_io_t **diop);
1358 int hammer2_io_newnz(hammer2_dev_t *hmp, off_t lbase, int lsize,
1359 hammer2_io_t **diop);
1360 int hammer2_io_newq(hammer2_dev_t *hmp, off_t lbase, int lsize,
1361 hammer2_io_t **diop);
1362 int hammer2_io_bread(hammer2_dev_t *hmp, off_t lbase, int lsize,
1363 hammer2_io_t **diop);
1364 void hammer2_io_bawrite(hammer2_io_t **diop);
1365 void hammer2_io_bdwrite(hammer2_io_t **diop);
1366 int hammer2_io_bwrite(hammer2_io_t **diop);
1367 int hammer2_io_isdirty(hammer2_io_t *dio);
1368 void hammer2_io_setdirty(hammer2_io_t *dio);
1369 void hammer2_io_setinval(hammer2_io_t *dio, u_int bytes);
1370 void hammer2_io_brelse(hammer2_io_t **diop);
1371 void hammer2_io_bqrelse(hammer2_io_t **diop);
1376 void hammer2_xop_group_init(hammer2_pfs_t *pmp, hammer2_xop_group_t *xgrp);
1377 hammer2_xop_t *hammer2_xop_alloc(hammer2_inode_t *ip, hammer2_xop_func_t func);
1378 void hammer2_xop_helper_create(hammer2_pfs_t *pmp);
1379 void hammer2_xop_helper_cleanup(hammer2_pfs_t *pmp);
1380 void hammer2_xop_start(hammer2_xop_head_t *xop);
1381 int hammer2_xop_collect(hammer2_xop_head_t *xop);
1382 void hammer2_xop_retire(hammer2_xop_head_t *xop, uint32_t mask);
1383 int hammer2_xop_active(hammer2_xop_head_t *xop);
1384 int hammer2_xop_feed(hammer2_xop_head_t *xop, hammer2_chain_t *chain,
1385 int clindex, int error);
1388 void hammer2_xop_readdir(hammer2_xop_t *xop, int clidx);
1389 int hammer2_xop_readlink(struct vop_readlink_args *ap);
1390 int hammer2_xop_nresolve(struct vop_nresolve_args *ap);
1391 int hammer2_xop_nlookupdotdot(struct vop_nlookupdotdot_args *ap);
1392 int hammer2_xop_nmkdir(struct vop_nmkdir_args *ap);
1393 int hammer2_xop_advlock(struct vop_advlock_args *ap);
1394 int hammer2_xop_nlink(struct vop_nlink_args *ap);
1395 int hammer2_xop_ncreate(struct vop_ncreate_args *ap);
1396 int hammer2_xop_nmknod(struct vop_nmknod_args *ap);
1397 int hammer2_xop_nsymlink(struct vop_nsymlink_args *ap);
1398 int hammer2_xop_nremove(struct vop_nremove_args *ap);
1399 int hammer2_xop_nrmdir(struct vop_nrmdir_args *ap);
1400 int hammer2_xop_nrename(struct vop_nrename_args *ap);
1405 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
1406 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
1411 void hammer2_clusterctl_wakeup(kdmsg_iocom_t *iocom);
1412 void hammer2_volconf_update(hammer2_dev_t *hmp, int index);
1413 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
1414 int hammer2_vfs_sync(struct mount *mp, int waitflags);
1415 hammer2_pfs_t *hammer2_pfsalloc(hammer2_cluster_t *cluster,
1416 const hammer2_inode_data_t *ripdata,
1417 hammer2_tid_t modify_tid);
1419 void hammer2_lwinprog_ref(hammer2_pfs_t *pmp);
1420 void hammer2_lwinprog_drop(hammer2_pfs_t *pmp);
1421 void hammer2_lwinprog_wait(hammer2_pfs_t *pmp);
1426 int hammer2_freemap_alloc(hammer2_trans_t *trans, hammer2_chain_t *chain,
1428 void hammer2_freemap_adjust(hammer2_trans_t *trans, hammer2_dev_t *hmp,
1429 hammer2_blockref_t *bref, int how);
1434 int hammer2_cluster_need_resize(hammer2_cluster_t *cluster, int bytes);
1435 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
1436 const hammer2_media_data_t *hammer2_cluster_rdata(hammer2_cluster_t *cluster);
1437 const hammer2_media_data_t *hammer2_cluster_rdata_bytes(
1438 hammer2_cluster_t *cluster, size_t *bytesp);
1439 hammer2_media_data_t *hammer2_cluster_wdata(hammer2_cluster_t *cluster);
1440 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
1441 int hammer2_cluster_modified(hammer2_cluster_t *cluster);
1442 int hammer2_cluster_duplicated(hammer2_cluster_t *cluster);
1443 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
1444 void hammer2_cluster_setflush(hammer2_trans_t *trans,
1445 hammer2_cluster_t *cluster);
1446 void hammer2_cluster_setmethod_check(hammer2_trans_t *trans,
1447 hammer2_cluster_t *cluster, int check_algo);
1448 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfs_t *pmp,
1449 hammer2_trans_t *trans,
1450 hammer2_blockref_t *bref);
1451 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
1452 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1453 void hammer2_cluster_wait(hammer2_cluster_t *cluster);
1454 void hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1455 void hammer2_cluster_lock_except(hammer2_cluster_t *cluster, int idx, int how);
1456 int hammer2_cluster_check(hammer2_cluster_t *cluster, hammer2_key_t lokey,
1458 void hammer2_cluster_resolve(hammer2_cluster_t *cluster);
1459 void hammer2_cluster_forcegood(hammer2_cluster_t *cluster);
1460 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster);
1461 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1462 void hammer2_cluster_unlock_except(hammer2_cluster_t *cluster, int idx);
1463 void hammer2_cluster_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
1464 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1465 int nradix, int flags);
1466 void hammer2_cluster_modify(hammer2_trans_t *trans, hammer2_cluster_t *cluster,
1468 hammer2_inode_data_t *hammer2_cluster_modify_ip(hammer2_trans_t *trans,
1469 hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1471 void hammer2_cluster_modsync(hammer2_cluster_t *cluster);
1472 hammer2_cluster_t *hammer2_cluster_lookup_init(hammer2_cluster_t *cparent,
1474 void hammer2_cluster_lookup_done(hammer2_cluster_t *cparent);
1475 hammer2_cluster_t *hammer2_cluster_lookup(hammer2_cluster_t *cparent,
1476 hammer2_key_t *key_nextp,
1477 hammer2_key_t key_beg, hammer2_key_t key_end,
1479 hammer2_cluster_t *hammer2_cluster_next(hammer2_cluster_t *cparent,
1480 hammer2_cluster_t *cluster,
1481 hammer2_key_t *key_nextp,
1482 hammer2_key_t key_beg, hammer2_key_t key_end,
1484 void hammer2_cluster_next_single_chain(hammer2_cluster_t *cparent,
1485 hammer2_cluster_t *cluster,
1486 hammer2_key_t *key_nextp,
1487 hammer2_key_t key_beg,
1488 hammer2_key_t key_end,
1490 hammer2_cluster_t *hammer2_cluster_scan(hammer2_cluster_t *cparent,
1491 hammer2_cluster_t *cluster, int flags);
1492 int hammer2_cluster_create(hammer2_trans_t *trans, hammer2_cluster_t *cparent,
1493 hammer2_cluster_t **clusterp,
1494 hammer2_key_t key, int keybits,
1495 int type, size_t bytes, int flags);
1496 void hammer2_cluster_rename(hammer2_trans_t *trans, hammer2_blockref_t *bref,
1497 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1499 void hammer2_cluster_delete(hammer2_trans_t *trans, hammer2_cluster_t *pcluster,
1500 hammer2_cluster_t *cluster, int flags);
1501 int hammer2_cluster_snapshot(hammer2_trans_t *trans,
1502 hammer2_cluster_t *ocluster, hammer2_ioc_pfs_t *pmp);
1503 hammer2_cluster_t *hammer2_cluster_parent(hammer2_cluster_t *cluster);
1505 int hammer2_bulk_scan(hammer2_trans_t *trans, hammer2_chain_t *parent,
1506 int (*func)(hammer2_chain_t *chain, void *info),
1508 int hammer2_bulkfree_pass(hammer2_dev_t *hmp,
1509 struct hammer2_ioc_bulkfree *bfi);
1514 void hammer2_iocom_init(hammer2_dev_t *hmp);
1515 void hammer2_iocom_uninit(hammer2_dev_t *hmp);
1516 void hammer2_cluster_reconnect(hammer2_dev_t *hmp, struct file *fp);
1521 void hammer2_thr_create(hammer2_thread_t *thr, hammer2_pfs_t *pmp,
1522 const char *id, int clindex, int repidx,
1523 void (*func)(void *arg));
1524 void hammer2_thr_delete(hammer2_thread_t *thr);
1525 void hammer2_thr_remaster(hammer2_thread_t *thr);
1526 void hammer2_thr_freeze_async(hammer2_thread_t *thr);
1527 void hammer2_thr_freeze(hammer2_thread_t *thr);
1528 void hammer2_thr_unfreeze(hammer2_thread_t *thr);
1529 void hammer2_primary_sync_thread(void *arg);
1530 void hammer2_primary_xops_thread(void *arg);
1533 * hammer2_strategy.c
1535 int hammer2_vop_strategy(struct vop_strategy_args *ap);
1536 int hammer2_vop_bmap(struct vop_bmap_args *ap);
1537 void hammer2_write_thread(void *arg);
1538 void hammer2_bioq_sync(hammer2_pfs_t *pmp);
1540 #endif /* !_KERNEL */
1541 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */