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_io_t *dio; /* physical data buffer */
337 u_int bytes; /* physical data size */
341 int error; /* on-lock data error state */
343 hammer2_media_data_t *data; /* data pointer shortcut */
344 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
347 typedef struct hammer2_chain hammer2_chain_t;
349 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
350 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
353 * Special notes on flags:
355 * INITIAL - This flag allows a chain to be created and for storage to
356 * be allocated without having to immediately instantiate the
357 * related buffer. The data is assumed to be all-zeros. It
358 * is primarily used for indirect blocks.
360 * MODIFIED - The chain's media data has been modified.
362 * UPDATE - Chain might not be modified but parent blocktable needs update
364 * FICTITIOUS - Faked chain as a placeholder for an error condition. This
365 * chain is unsuitable for I/O.
367 * BMAPPED - Indicates that the chain is present in the parent blockmap.
369 * BMAPUPD - Indicates that the chain is present but needs to be updated
370 * in the parent blockmap.
372 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
373 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
374 #define HAMMER2_CHAIN_DESTROY 0x00000004
375 #define HAMMER2_CHAIN_UNUSED0008 0x00000008
376 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
377 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
378 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
379 #define HAMMER2_CHAIN_DEFERRED 0x00000080 /* flush depth defer */
380 #define HAMMER2_CHAIN_IOFLUSH 0x00000100 /* bawrite on put */
381 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
382 #define HAMMER2_CHAIN_FICTITIOUS 0x00000400 /* unsuitable for I/O */
383 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
384 #define HAMMER2_CHAIN_DELAYED 0x00001000 /* delayed flush */
385 #define HAMMER2_CHAIN_COUNTEDBREFS 0x00002000 /* block table stats */
386 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
387 #define HAMMER2_CHAIN_UNUSED00008000 0x00008000
388 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
389 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
390 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
391 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
392 #define HAMMER2_CHAIN_IOINPROG 0x00100000 /* I/O interlock */
393 #define HAMMER2_CHAIN_IOSIGNAL 0x00200000 /* I/O interlock */
394 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
396 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
397 HAMMER2_CHAIN_UPDATE | \
398 HAMMER2_CHAIN_ONFLUSH)
401 * Hammer2 error codes, used by chain->error and cluster->error. The error
402 * code is typically set on-lock unless no I/O was requested, and set on
403 * I/O otherwise. If set for a cluster it generally means that the cluster
404 * code could not find a valid copy to present.
406 * IO - An I/O error occurred
407 * CHECK - I/O succeeded but did not match the check code
408 * INCOMPLETE - A cluster is not complete enough to use, or
409 * a chain cannot be loaded because its parent has an error.
411 * NOTE: API allows callers to check zero/non-zero to determine if an error
414 * NOTE: Chain's data field is usually NULL on an IO error but not necessarily
415 * NULL on other errors. Check chain->error, not chain->data.
417 #define HAMMER2_ERROR_NONE 0
418 #define HAMMER2_ERROR_IO 1 /* device I/O error */
419 #define HAMMER2_ERROR_CHECK 2 /* check code mismatch */
420 #define HAMMER2_ERROR_INCOMPLETE 3 /* incomplete cluster */
421 #define HAMMER2_ERROR_DEPTH 4 /* temporary depth limit */
424 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
427 * NOLOCK - Input and output chains are referenced only and not
428 * locked. Output chain might be temporarily locked
431 * NODATA - Asks that the chain->data not be resolved in order
434 * NODIRECT - Prevents a lookup of offset 0 in an inode from returning
435 * the inode itself if the inode is in DIRECTDATA mode
436 * (i.e. file is <= 512 bytes). Used by the synchronization
437 * code to prevent confusion.
439 * SHARED - The input chain is expected to be locked shared,
440 * and the output chain is locked shared.
442 * MATCHIND - Allows an indirect block / freemap node to be returned
443 * when the passed key range matches the radix. Remember
444 * that key_end is inclusive (e.g. {0x000,0xFFF},
445 * not {0x000,0x1000}).
447 * (Cannot be used for remote or cluster ops).
449 * ALLNODES - Allows NULL focus.
451 * ALWAYS - Always resolve the data. If ALWAYS and NODATA are both
452 * missing, bulk file data is not resolved but inodes and
453 * other meta-data will.
455 * NOUNLOCK - Used by hammer2_chain_next() to leave the lock on
456 * the input chain intact. The chain is still dropped.
457 * This allows the caller to add a reference to the chain
458 * and retain it in a locked state (used by the
459 * XOP/feed/collect code).
461 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
462 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
463 #define HAMMER2_LOOKUP_NODIRECT 0x00000004 /* no offset=0 DD */
464 #define HAMMER2_LOOKUP_SHARED 0x00000100
465 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
466 #define HAMMER2_LOOKUP_ALLNODES 0x00000400 /* allow NULL focus */
467 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
468 #define HAMMER2_LOOKUP_NOUNLOCK 0x00001000 /* leave lock intact */
471 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
473 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
474 * blocks in the INITIAL-create state.
476 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
477 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
478 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
481 * Flags passed to hammer2_chain_lock()
483 * NOTE: RDONLY is set to optimize cluster operations when *no* modifications
484 * will be made to either the cluster being locked or any underlying
485 * cluster. It allows the cluster to lock and access data for a subset
486 * of available nodes instead of all available nodes.
488 #define HAMMER2_RESOLVE_NEVER 1
489 #define HAMMER2_RESOLVE_MAYBE 2
490 #define HAMMER2_RESOLVE_ALWAYS 3
491 #define HAMMER2_RESOLVE_MASK 0x0F
493 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
494 #define HAMMER2_RESOLVE_UNUSED20 0x20
495 #define HAMMER2_RESOLVE_RDONLY 0x40 /* higher level op flag */
498 * Flags passed to hammer2_chain_delete()
500 #define HAMMER2_DELETE_PERMANENT 0x0001
503 * Flags passed to hammer2_chain_insert() or hammer2_chain_rename()
505 #define HAMMER2_INSERT_PFSROOT 0x0004
508 * Flags passed to hammer2_chain_delete_duplicate()
510 #define HAMMER2_DELDUP_RECORE 0x0001
513 * Cluster different types of storage together for allocations
515 #define HAMMER2_FREECACHE_INODE 0
516 #define HAMMER2_FREECACHE_INDIR 1
517 #define HAMMER2_FREECACHE_DATA 2
518 #define HAMMER2_FREECACHE_UNUSED3 3
519 #define HAMMER2_FREECACHE_TYPES 4
522 * hammer2_freemap_alloc() block preference
524 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
527 * BMAP read-ahead maximum parameters
529 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
530 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
533 * hammer2_freemap_adjust()
535 #define HAMMER2_FREEMAP_DORECOVER 1
536 #define HAMMER2_FREEMAP_DOMAYFREE 2
537 #define HAMMER2_FREEMAP_DOREALFREE 3
540 * HAMMER2 cluster - A set of chains representing the same entity.
542 * hammer2_cluster typically represents a temporary set of representitive
543 * chains. The one exception is that a hammer2_cluster is embedded in
544 * hammer2_inode. This embedded cluster is ONLY used to track the
545 * representitive chains and cannot be directly locked.
547 * A cluster is usually temporary (and thus per-thread) for locking purposes,
548 * allowing us to embed the asynchronous storage required for cluster
549 * operations in the cluster itself and adjust the state and status without
550 * having to worry too much about SMP issues.
552 * The exception is the cluster embedded in the hammer2_inode structure.
553 * This is used to cache the cluster state on an inode-by-inode basis.
554 * Individual hammer2_chain structures not incorporated into clusters might
555 * also stick around to cache miscellanious elements.
557 * Because the cluster is a 'working copy' and is usually subject to cluster
558 * quorum rules, it is quite possible for us to end up with an insufficient
559 * number of live chains to execute an operation. If an insufficient number
560 * of chains remain in a working copy, the operation may have to be
561 * downgraded, retried, stall until the requisit number of chains are
562 * available, or possibly even error out depending on the mount type.
564 * A cluster's focus is set when it is locked. The focus can only be set
565 * to a chain still part of the synchronized set.
567 #define HAMMER2_MAXCLUSTER 8
568 #define HAMMER2_XOPFIFO 16
569 #define HAMMER2_XOPFIFO_MASK (HAMMER2_XOPFIFO - 1)
570 #define HAMMER2_XOPGROUPS 16
571 #define HAMMER2_XOPGROUPS_MASK (HAMMER2_XOPGROUPS - 1)
572 #define HAMMER2_XOPMASK_VOP 0x80000000U
574 struct hammer2_cluster_item {
575 hammer2_chain_t *chain;
580 typedef struct hammer2_cluster_item hammer2_cluster_item_t;
583 * INVALID - Invalid for focus, i.e. not part of synchronized set.
584 * Once set, this bit is sticky across operations.
586 * FEMOD - Indicates that front-end modifying operations can
587 * mess with this entry and MODSYNC will copy also
590 #define HAMMER2_CITEM_INVALID 0x00000001
591 #define HAMMER2_CITEM_FEMOD 0x00000002
592 #define HAMMER2_CITEM_NULL 0x00000004
594 struct hammer2_cluster {
595 int refs; /* track for deallocation */
597 struct hammer2_pfs *pmp;
600 int error; /* error code valid on lock */
603 hammer2_chain_t *focus; /* current focus (or mod) */
604 hammer2_cluster_item_t array[HAMMER2_MAXCLUSTER];
607 typedef struct hammer2_cluster hammer2_cluster_t;
610 * WRHARD - Hard mounts can write fully synchronized
611 * RDHARD - Hard mounts can read fully synchronized
612 * UNHARD - Unsynchronized masters present
613 * NOHARD - No masters visible
614 * WRSOFT - Soft mounts can write to at least the SOFT_MASTER
615 * RDSOFT - Soft mounts can read from at least a SOFT_SLAVE
616 * UNSOFT - Unsynchronized slaves present
617 * NOSOFT - No slaves visible
618 * RDSLAVE - slaves are accessible (possibly unsynchronized or remote).
619 * MSYNCED - All masters are fully synchronized
620 * SSYNCED - All known local slaves are fully synchronized to masters
622 * All available masters are always incorporated. All PFSs belonging to a
623 * cluster (master, slave, copy, whatever) always try to synchronize the
624 * total number of known masters in the PFSs root inode.
626 * A cluster might have access to many slaves, copies, or caches, but we
627 * have a limited number of cluster slots. Any such elements which are
628 * directly mounted from block device(s) will always be incorporated. Note
629 * that SSYNCED only applies to such elements which are directly mounted,
630 * not to any remote slaves, copies, or caches that could be available. These
631 * bits are used to monitor and drive our synchronization threads.
633 * When asking the question 'is any data accessible at all', then a simple
634 * test against (RDHARD|RDSOFT|RDSLAVE) gives you the answer. If any of
635 * these bits are set the object can be read with certain caveats:
636 * RDHARD - no caveats. RDSOFT - authoritative but might not be synchronized.
637 * and RDSLAVE - not authoritative, has some data but it could be old or
640 * When both soft and hard mounts are available, data will be read and written
641 * via the soft mount only. But all might be in the cluster because
642 * background synchronization threads still need to do their work.
644 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode struct */
645 #define HAMMER2_CLUSTER_UNUSED2 0x00000002
646 #define HAMMER2_CLUSTER_LOCKED 0x00000004 /* cluster lks not recursive */
647 #define HAMMER2_CLUSTER_WRHARD 0x00000100 /* hard-mount can write */
648 #define HAMMER2_CLUSTER_RDHARD 0x00000200 /* hard-mount can read */
649 #define HAMMER2_CLUSTER_UNHARD 0x00000400 /* unsynchronized masters */
650 #define HAMMER2_CLUSTER_NOHARD 0x00000800 /* no masters visible */
651 #define HAMMER2_CLUSTER_WRSOFT 0x00001000 /* soft-mount can write */
652 #define HAMMER2_CLUSTER_RDSOFT 0x00002000 /* soft-mount can read */
653 #define HAMMER2_CLUSTER_UNSOFT 0x00004000 /* unsynchronized slaves */
654 #define HAMMER2_CLUSTER_NOSOFT 0x00008000 /* no slaves visible */
655 #define HAMMER2_CLUSTER_MSYNCED 0x00010000 /* all masters synchronized */
656 #define HAMMER2_CLUSTER_SSYNCED 0x00020000 /* known slaves synchronized */
658 #define HAMMER2_CLUSTER_ANYDATA ( HAMMER2_CLUSTER_RDHARD | \
659 HAMMER2_CLUSTER_RDSOFT | \
660 HAMMER2_CLUSTER_RDSLAVE)
662 #define HAMMER2_CLUSTER_RDOK ( HAMMER2_CLUSTER_RDHARD | \
663 HAMMER2_CLUSTER_RDSOFT)
665 #define HAMMER2_CLUSTER_WROK ( HAMMER2_CLUSTER_WRHARD | \
666 HAMMER2_CLUSTER_WRSOFT)
668 #define HAMMER2_CLUSTER_ZFLAGS ( HAMMER2_CLUSTER_WRHARD | \
669 HAMMER2_CLUSTER_RDHARD | \
670 HAMMER2_CLUSTER_WRSOFT | \
671 HAMMER2_CLUSTER_RDSOFT | \
672 HAMMER2_CLUSTER_MSYNCED | \
673 HAMMER2_CLUSTER_SSYNCED)
676 * Helper functions (cluster must be locked for flags to be valid).
680 hammer2_cluster_rdok(hammer2_cluster_t *cluster)
682 return (cluster->flags & HAMMER2_CLUSTER_RDOK);
687 hammer2_cluster_wrok(hammer2_cluster_t *cluster)
689 return (cluster->flags & HAMMER2_CLUSTER_WROK);
692 RB_HEAD(hammer2_inode_tree, hammer2_inode);
697 * NOTE: The inode-embedded cluster is never used directly for I/O (since
698 * it may be shared). Instead it will be replicated-in and synchronized
699 * back out if changed.
701 struct hammer2_inode {
702 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
703 hammer2_mtx_t lock; /* inode lock */
704 struct hammer2_pfs *pmp; /* PFS mount */
705 struct hammer2_inode *pip; /* parent inode */
707 struct spinlock cluster_spin; /* update cluster */
708 hammer2_cluster_t cluster;
709 struct lockf advlock;
711 u_int refs; /* +vpref, +flushref */
712 uint8_t comp_heuristic;
713 hammer2_inode_meta_t meta; /* copy of meta-data */
717 typedef struct hammer2_inode hammer2_inode_t;
720 * MODIFIED - Inode is in a modified state, ip->meta may have changes.
721 * RESIZED - Inode truncated (any) or inode extended beyond
724 #define HAMMER2_INODE_MODIFIED 0x0001
725 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
726 #define HAMMER2_INODE_RENAME_INPROG 0x0004
727 #define HAMMER2_INODE_ONRBTREE 0x0008
728 #define HAMMER2_INODE_RESIZED 0x0010 /* requires inode_fsync */
729 #define HAMMER2_INODE_ISDELETED 0x0020 /* deleted, not in ihidden */
730 #define HAMMER2_INODE_ISUNLINKED 0x0040
731 #define HAMMER2_INODE_METAGOOD 0x0080 /* inode meta-data good */
733 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
734 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
738 * inode-unlink side-structure
740 struct hammer2_inode_unlink {
741 TAILQ_ENTRY(hammer2_inode_unlink) entry;
744 TAILQ_HEAD(h2_unlk_list, hammer2_inode_unlink);
746 typedef struct hammer2_inode_unlink hammer2_inode_unlink_t;
749 * Transaction management sub-structure under hammer2_pfs
751 struct hammer2_trans {
756 typedef struct hammer2_trans hammer2_trans_t;
758 #define HAMMER2_TRANS_ISFLUSH 0x80000000 /* flush code */
759 #define HAMMER2_TRANS_BUFCACHE 0x40000000 /* bio strategy */
760 #define HAMMER2_TRANS_PREFLUSH 0x20000000 /* preflush state */
761 #define HAMMER2_TRANS_FPENDING 0x10000000 /* flush pending */
762 #define HAMMER2_TRANS_WAITING 0x08000000 /* someone waiting */
763 #define HAMMER2_TRANS_MASK 0x00FFFFFF /* count mask */
765 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
766 #define HAMMER2_FREEMAP_HEUR_TYPES 8
767 #define HAMMER2_FREEMAP_HEUR (HAMMER2_FREEMAP_HEUR_NRADIX * \
768 HAMMER2_FREEMAP_HEUR_TYPES)
770 #define HAMMER2_FLUSH_TOP 0x0001
771 #define HAMMER2_FLUSH_ALL 0x0002
775 * Hammer2 support thread element.
777 * Potentially many support threads can hang off of hammer2, primarily
778 * off the hammer2_pfs structure. Typically:
780 * td x Nodes A synchronization thread for each node.
781 * td x Nodes x workers Worker threads for frontend operations.
782 * td x 1 Bioq thread for logical buffer writes.
784 * In addition, the synchronization thread(s) associated with the
785 * super-root PFS (spmp) for a node is responsible for automatic bulkfree
788 struct hammer2_thread {
789 struct hammer2_pfs *pmp;
790 hammer2_xop_list_t *xopq; /* points into pmp->xopq[] */
794 int clindex; /* cluster element index */
796 struct lock lk; /* thread control lock */
799 typedef struct hammer2_thread hammer2_thread_t;
801 #define HAMMER2_THREAD_UNMOUNTING 0x0001 /* unmount request */
802 #define HAMMER2_THREAD_DEV 0x0002 /* related to dev, not pfs */
803 #define HAMMER2_THREAD_UNUSED04 0x0004
804 #define HAMMER2_THREAD_REMASTER 0x0008 /* remaster request */
805 #define HAMMER2_THREAD_STOP 0x0010 /* exit request */
806 #define HAMMER2_THREAD_FREEZE 0x0020 /* force idle */
807 #define HAMMER2_THREAD_FROZEN 0x0040 /* restart */
811 * hammer2_xop - container for VOP/XOP operation (allocated, not on stack).
813 * This structure is used to distribute a VOP operation across multiple
814 * nodes. It provides a rendezvous for concurrent node execution and
815 * can be detached from the frontend operation to allow the frontend to
818 * This structure also sequences operations on up to three inodes.
820 typedef void (*hammer2_xop_func_t)(union hammer2_xop *xop, int clidx);
822 typedef struct hammer2_xop_fifo {
823 TAILQ_ENTRY(hammer2_xop_head) entry;
824 hammer2_chain_t *array[HAMMER2_XOPFIFO];
825 int errors[HAMMER2_XOPFIFO];
829 } hammer2_xop_fifo_t;
831 #define HAMMER2_XOP_FIFO_RUN 0x0001
833 struct hammer2_xop_head {
834 hammer2_xop_func_t func;
836 struct hammer2_inode *ip1;
837 struct hammer2_inode *ip2;
838 struct hammer2_inode *ip3;
839 uint32_t check_counter;
844 hammer2_key_t collect_key;
849 hammer2_xop_fifo_t collect[HAMMER2_MAXCLUSTER];
850 hammer2_cluster_t cluster; /* help collections */
853 typedef struct hammer2_xop_head hammer2_xop_head_t;
855 struct hammer2_xop_ipcluster {
856 hammer2_xop_head_t head;
859 struct hammer2_xop_strategy {
860 hammer2_xop_head_t head;
867 struct hammer2_xop_readdir {
868 hammer2_xop_head_t head;
872 struct hammer2_xop_nresolve {
873 hammer2_xop_head_t head;
874 hammer2_key_t lhc; /* if name is NULL used lhc */
877 struct hammer2_xop_nlink {
878 hammer2_xop_head_t head;
881 struct hammer2_xop_unlink {
882 hammer2_xop_head_t head;
887 struct hammer2_xop_nrename {
888 hammer2_xop_head_t head;
893 struct hammer2_xop_scanlhc {
894 hammer2_xop_head_t head;
898 struct hammer2_xop_scanall {
899 hammer2_xop_head_t head;
900 hammer2_key_t key_beg; /* inclusive */
901 hammer2_key_t key_end; /* inclusive */
904 struct hammer2_xop_lookup {
905 hammer2_xop_head_t head;
909 struct hammer2_xop_create {
910 hammer2_xop_head_t head;
911 hammer2_inode_meta_t meta; /* initial metadata */
916 struct hammer2_xop_destroy {
917 hammer2_xop_head_t head;
920 struct hammer2_xop_fsync {
921 hammer2_xop_head_t head;
922 hammer2_inode_meta_t meta;
925 int clear_directdata;
928 struct hammer2_xop_unlinkall {
929 hammer2_xop_head_t head;
930 hammer2_key_t key_beg;
931 hammer2_key_t key_end;
934 struct hammer2_xop_connect {
935 hammer2_xop_head_t head;
939 struct hammer2_xop_flush {
940 hammer2_xop_head_t head;
943 typedef struct hammer2_xop_readdir hammer2_xop_readdir_t;
944 typedef struct hammer2_xop_nresolve hammer2_xop_nresolve_t;
945 typedef struct hammer2_xop_nlink hammer2_xop_nlink_t;
946 typedef struct hammer2_xop_unlink hammer2_xop_unlink_t;
947 typedef struct hammer2_xop_nrename hammer2_xop_nrename_t;
948 typedef struct hammer2_xop_ipcluster hammer2_xop_ipcluster_t;
949 typedef struct hammer2_xop_strategy hammer2_xop_strategy_t;
950 typedef struct hammer2_xop_create hammer2_xop_create_t;
951 typedef struct hammer2_xop_destroy hammer2_xop_destroy_t;
952 typedef struct hammer2_xop_fsync hammer2_xop_fsync_t;
953 typedef struct hammer2_xop_unlinkall hammer2_xop_unlinkall_t;
954 typedef struct hammer2_xop_scanlhc hammer2_xop_scanlhc_t;
955 typedef struct hammer2_xop_scanall hammer2_xop_scanall_t;
956 typedef struct hammer2_xop_lookup hammer2_xop_lookup_t;
957 typedef struct hammer2_xop_connect hammer2_xop_connect_t;
958 typedef struct hammer2_xop_flush hammer2_xop_flush_t;
961 hammer2_xop_head_t head;
962 hammer2_xop_ipcluster_t xop_ipcluster;
963 hammer2_xop_readdir_t xop_readdir;
964 hammer2_xop_nresolve_t xop_nresolve;
965 hammer2_xop_nlink_t xop_nlink;
966 hammer2_xop_unlink_t xop_unlink;
967 hammer2_xop_nrename_t xop_nrename;
968 hammer2_xop_strategy_t xop_strategy;
969 hammer2_xop_create_t xop_create;
970 hammer2_xop_destroy_t xop_destroy;
971 hammer2_xop_fsync_t xop_fsync;
972 hammer2_xop_unlinkall_t xop_unlinkall;
973 hammer2_xop_scanlhc_t xop_scanlhc;
974 hammer2_xop_scanall_t xop_scanall;
975 hammer2_xop_lookup_t xop_lookup;
976 hammer2_xop_flush_t xop_flush;
977 hammer2_xop_connect_t xop_connect;
980 typedef union hammer2_xop hammer2_xop_t;
983 * hammer2_xop_group - Manage XOP support threads.
985 struct hammer2_xop_group {
986 hammer2_thread_t thrs[HAMMER2_MAXCLUSTER];
989 typedef struct hammer2_xop_group hammer2_xop_group_t;
992 * flags to hammer2_xop_collect()
994 #define HAMMER2_XOP_COLLECT_NOWAIT 0x00000001
995 #define HAMMER2_XOP_COLLECT_WAITALL 0x00000002
998 * flags to hammer2_xop_alloc()
1000 * MODIFYING - This is a modifying transaction, allocate a mtid.
1002 #define HAMMER2_XOP_MODIFYING 0x00000001
1005 * Global (per partition) management structure, represents a hard block
1006 * device. Typically referenced by hammer2_chain structures when applicable.
1007 * Typically not used for network-managed elements.
1009 * Note that a single hammer2_dev can be indirectly tied to multiple system
1010 * mount points. There is no direct relationship. System mounts are
1011 * per-cluster-id, not per-block-device, and a single hard mount might contain
1012 * many PFSs and those PFSs might combine together in various ways to form
1013 * the set of available clusters.
1015 struct hammer2_dev {
1016 struct vnode *devvp; /* device vnode */
1017 int ronly; /* read-only mount */
1018 int mount_count; /* number of actively mounted PFSs */
1019 TAILQ_ENTRY(hammer2_dev) mntentry; /* hammer2_mntlist */
1021 struct malloc_type *mchain;
1024 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
1025 struct spinlock io_spin; /* iotree access */
1026 struct hammer2_io_tree iotree;
1028 hammer2_chain_t vchain; /* anchor chain (topology) */
1029 hammer2_chain_t fchain; /* anchor chain (freemap) */
1030 struct spinlock list_spin;
1031 struct h2_flush_list flushq; /* flush seeds */
1032 struct hammer2_pfs *spmp; /* super-root pmp for transactions */
1033 struct lock bulklk; /* bulkfree lock */
1034 struct lock vollk; /* lockmgr lock */
1035 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR];
1036 int volhdrno; /* last volhdrno written */
1037 char devrepname[64]; /* for kprintf */
1038 hammer2_volume_data_t voldata;
1039 hammer2_volume_data_t volsync; /* synchronized voldata */
1042 typedef struct hammer2_dev hammer2_dev_t;
1045 * Helper functions (cluster must be locked for flags to be valid).
1049 hammer2_chain_rdok(hammer2_chain_t *chain)
1051 return (chain->error == 0);
1056 hammer2_chain_wrok(hammer2_chain_t *chain)
1058 return (chain->error == 0 && chain->hmp->ronly == 0);
1062 * Per-cluster management structure. This structure will be tied to a
1063 * system mount point if the system is mounting the PFS, but is also used
1064 * to manage clusters encountered during the super-root scan or received
1065 * via LNK_SPANs that might not be mounted.
1067 * This structure is also used to represent the super-root that hangs off
1068 * of a hard mount point. The super-root is not really a cluster element.
1069 * In this case the spmp_hmp field will be non-NULL. It's just easier to do
1070 * this than to special case super-root manipulation in the hammer2_chain*
1071 * code as being only hammer2_dev-related.
1073 * pfs_mode and pfs_nmasters are rollup fields which critically describes
1074 * how elements of the cluster act on the cluster. pfs_mode is only applicable
1075 * when a PFS is mounted by the system. pfs_nmasters is our best guess as to
1076 * how many masters have been configured for a cluster and is always
1077 * applicable. pfs_types[] is an array with 1:1 correspondance to the
1078 * iroot cluster and describes the PFS types of the nodes making up the
1081 * WARNING! Portions of this structure have deferred initialization. In
1082 * particular, if not mounted there will be no ihidden or wthread.
1083 * umounted network PFSs will also be missing iroot and numerous
1084 * other fields will not be initialized prior to mount.
1086 * Synchronization threads are chain-specific and only applicable
1087 * to local hard PFS entries. A hammer2_pfs structure may contain
1088 * more than one when multiple hard PFSs are present on the local
1089 * machine which require synchronization monitoring. Most PFSs
1090 * (such as snapshots) are 1xMASTER PFSs which do not need a
1091 * synchronization thread.
1093 * WARNING! The chains making up pfs->iroot's cluster are accounted for in
1094 * hammer2_dev->mount_count when the pfs is associated with a mount
1097 struct hammer2_pfs {
1099 TAILQ_ENTRY(hammer2_pfs) mntentry; /* hammer2_pfslist */
1101 hammer2_dev_t *spmp_hmp; /* only if super-root pmp */
1102 hammer2_inode_t *iroot; /* PFS root inode */
1103 hammer2_inode_t *ihidden; /* PFS hidden directory */
1104 uint8_t pfs_types[HAMMER2_MAXCLUSTER];
1105 char *pfs_names[HAMMER2_MAXCLUSTER];
1106 hammer2_dev_t *pfs_hmps[HAMMER2_MAXCLUSTER];
1107 hammer2_trans_t trans;
1108 struct lock lock; /* PFS lock for certain ops */
1109 struct netexport export; /* nfs export */
1110 int ronly; /* read-only mount */
1111 struct malloc_type *minode;
1112 struct malloc_type *mmsg;
1113 struct spinlock inum_spin; /* inumber lookup */
1114 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
1115 hammer2_tid_t modify_tid; /* modify transaction id */
1116 hammer2_tid_t inode_tid; /* inode allocator */
1117 uint8_t pfs_nmasters; /* total masters */
1118 uint8_t pfs_mode; /* operating mode PFSMODE */
1123 uint32_t inmem_dirty_chains;
1124 int count_lwinprog; /* logical write in prog */
1125 struct spinlock list_spin;
1126 struct h2_unlk_list unlinkq; /* last-close unlink */
1127 hammer2_thread_t sync_thrs[HAMMER2_MAXCLUSTER];
1128 uint32_t cluster_flags; /* cached cluster flags */
1129 int has_xop_threads;
1130 struct spinlock xop_spin; /* xop sequencer */
1131 hammer2_xop_group_t xop_groups[HAMMER2_XOPGROUPS];
1132 hammer2_xop_list_t xopq[HAMMER2_MAXCLUSTER];
1135 typedef struct hammer2_pfs hammer2_pfs_t;
1137 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
1138 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
1140 #define HAMMER2_LWINPROG_WAITING 0x80000000
1141 #define HAMMER2_LWINPROG_WAITING0 0x40000000
1142 #define HAMMER2_LWINPROG_MASK 0x3FFFFFFF
1145 * hammer2_cluster_check
1147 #define HAMMER2_CHECK_NULL 0x00000001
1152 #define HAMMER2_BULK_ABORT 0x00000001
1157 #if defined(_KERNEL)
1159 MALLOC_DECLARE(M_HAMMER2);
1161 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
1162 #define ITOV(ip) ((ip)->vp)
1165 * Currently locked chains retain the locked buffer cache buffer for
1166 * indirect blocks, and indirect blocks can be one of two sizes. The
1167 * device buffer has to match the case to avoid deadlocking recursive
1168 * chains that might otherwise try to access different offsets within
1169 * the same device buffer.
1173 hammer2_devblkradix(int radix)
1176 if (radix <= HAMMER2_LBUFRADIX) {
1177 return (HAMMER2_LBUFRADIX);
1179 return (HAMMER2_PBUFRADIX);
1182 return (HAMMER2_PBUFRADIX);
1186 * XXX almost time to remove this. DIO uses PBUFSIZE exclusively now.
1190 hammer2_devblksize(size_t bytes)
1193 if (bytes <= HAMMER2_LBUFSIZE) {
1194 return(HAMMER2_LBUFSIZE);
1196 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
1197 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
1198 return (HAMMER2_PBUFSIZE);
1201 return (HAMMER2_PBUFSIZE);
1207 MPTOPMP(struct mount *mp)
1209 return ((hammer2_pfs_t *)mp->mnt_data);
1212 #define LOCKSTART int __nlocks = curthread->td_locks
1213 #define LOCKENTER (++curthread->td_locks)
1214 #define LOCKEXIT (--curthread->td_locks)
1215 #define LOCKSTOP KKASSERT(curthread->td_locks == __nlocks)
1217 extern struct vop_ops hammer2_vnode_vops;
1218 extern struct vop_ops hammer2_spec_vops;
1219 extern struct vop_ops hammer2_fifo_vops;
1221 extern int hammer2_debug;
1222 extern int hammer2_cluster_enable;
1223 extern int hammer2_hardlink_enable;
1224 extern int hammer2_flush_pipe;
1225 extern int hammer2_synchronous_flush;
1226 extern int hammer2_dio_count;
1227 extern long hammer2_limit_dirty_chains;
1228 extern long hammer2_iod_file_read;
1229 extern long hammer2_iod_meta_read;
1230 extern long hammer2_iod_indr_read;
1231 extern long hammer2_iod_fmap_read;
1232 extern long hammer2_iod_volu_read;
1233 extern long hammer2_iod_file_write;
1234 extern long hammer2_iod_meta_write;
1235 extern long hammer2_iod_indr_write;
1236 extern long hammer2_iod_fmap_write;
1237 extern long hammer2_iod_volu_write;
1238 extern long hammer2_ioa_file_read;
1239 extern long hammer2_ioa_meta_read;
1240 extern long hammer2_ioa_indr_read;
1241 extern long hammer2_ioa_fmap_read;
1242 extern long hammer2_ioa_volu_read;
1243 extern long hammer2_ioa_file_write;
1244 extern long hammer2_ioa_meta_write;
1245 extern long hammer2_ioa_indr_write;
1246 extern long hammer2_ioa_fmap_write;
1247 extern long hammer2_ioa_volu_write;
1249 extern struct objcache *cache_buffer_read;
1250 extern struct objcache *cache_buffer_write;
1251 extern struct objcache *cache_xops;
1256 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
1257 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
1259 int hammer2_signal_check(time_t *timep);
1260 const char *hammer2_error_str(int error);
1262 void hammer2_inode_lock(hammer2_inode_t *ip, int how);
1263 void hammer2_inode_unlock(hammer2_inode_t *ip);
1264 hammer2_chain_t *hammer2_inode_chain(hammer2_inode_t *ip, int clindex, int how);
1265 hammer2_chain_t *hammer2_inode_chain_and_parent(hammer2_inode_t *ip,
1266 int clindex, hammer2_chain_t **parentp, int how);
1267 hammer2_mtx_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
1268 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip,
1269 hammer2_mtx_state_t ostate);
1270 int hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
1271 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int);
1273 void hammer2_dev_exlock(hammer2_dev_t *hmp);
1274 void hammer2_dev_shlock(hammer2_dev_t *hmp);
1275 void hammer2_dev_unlock(hammer2_dev_t *hmp);
1277 int hammer2_get_dtype(const hammer2_inode_data_t *ipdata);
1278 int hammer2_get_vtype(uint8_t type);
1279 u_int8_t hammer2_get_obj_type(enum vtype vtype);
1280 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1281 u_int64_t hammer2_timespec_to_time(const struct timespec *ts);
1282 u_int32_t hammer2_to_unix_xid(const uuid_t *uuid);
1283 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1284 void hammer2_trans_manage_init(hammer2_pfs_t *pmp);
1286 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
1287 int hammer2_getradix(size_t bytes);
1289 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
1290 hammer2_key_t *lbasep, hammer2_key_t *leofp);
1291 int hammer2_calc_physical(hammer2_inode_t *ip, hammer2_key_t lbase);
1292 void hammer2_update_time(uint64_t *timep);
1293 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
1298 struct vnode *hammer2_igetv(hammer2_inode_t *ip, int *errorp);
1299 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfs_t *pmp,
1300 hammer2_tid_t inum);
1301 hammer2_inode_t *hammer2_inode_get(hammer2_pfs_t *pmp, hammer2_inode_t *dip,
1302 hammer2_cluster_t *cluster, int idx);
1303 void hammer2_inode_free(hammer2_inode_t *ip);
1304 void hammer2_inode_ref(hammer2_inode_t *ip);
1305 void hammer2_inode_drop(hammer2_inode_t *ip);
1306 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
1307 hammer2_cluster_t *cluster);
1308 void hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1310 void hammer2_inode_modify(hammer2_inode_t *ip);
1311 void hammer2_inode_run_unlinkq(hammer2_pfs_t *pmp);
1313 hammer2_inode_t *hammer2_inode_create(hammer2_inode_t *dip,
1314 struct vattr *vap, struct ucred *cred,
1315 const uint8_t *name, size_t name_len, hammer2_key_t lhc,
1316 hammer2_key_t inum, uint8_t type, uint8_t target_type,
1317 int flags, int *errorp);
1318 int hammer2_inode_connect(hammer2_inode_t *dip, hammer2_inode_t *ip,
1319 const char *name, size_t name_len,
1321 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
1322 hammer2_inode_t *tdip);
1323 void hammer2_inode_chain_sync(hammer2_inode_t *ip);
1324 int hammer2_inode_unlink_finisher(hammer2_inode_t *ip, int isopen);
1325 void hammer2_inode_install_hidden(hammer2_pfs_t *pmp);
1330 void hammer2_voldata_lock(hammer2_dev_t *hmp);
1331 void hammer2_voldata_unlock(hammer2_dev_t *hmp);
1332 void hammer2_voldata_modify(hammer2_dev_t *hmp);
1333 hammer2_chain_t *hammer2_chain_alloc(hammer2_dev_t *hmp,
1335 hammer2_blockref_t *bref);
1336 void hammer2_chain_core_init(hammer2_chain_t *chain);
1337 void hammer2_chain_ref(hammer2_chain_t *chain);
1338 void hammer2_chain_drop(hammer2_chain_t *chain);
1339 void hammer2_chain_lock(hammer2_chain_t *chain, int how);
1340 void hammer2_chain_load_data(hammer2_chain_t *chain);
1341 const hammer2_media_data_t *hammer2_chain_rdata(hammer2_chain_t *chain);
1342 hammer2_media_data_t *hammer2_chain_wdata(hammer2_chain_t *chain);
1343 int hammer2_chain_snapshot(hammer2_chain_t *chain, hammer2_ioc_pfs_t *pmp,
1344 hammer2_tid_t mtid);
1346 int hammer2_chain_hardlink_find(hammer2_inode_t *dip,
1347 hammer2_chain_t **parentp,
1348 hammer2_chain_t **chainp,
1350 void hammer2_chain_modify(hammer2_chain_t *chain,
1351 hammer2_tid_t mtid, int flags);
1352 void hammer2_chain_modify_ip(hammer2_inode_t *ip, hammer2_chain_t *chain,
1353 hammer2_tid_t mtid, int flags);
1354 void hammer2_chain_resize(hammer2_inode_t *ip, hammer2_chain_t *parent,
1355 hammer2_chain_t *chain,
1356 hammer2_tid_t mtid, int nradix, int flags);
1357 void hammer2_chain_unlock(hammer2_chain_t *chain);
1358 void hammer2_chain_wait(hammer2_chain_t *chain);
1359 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
1360 hammer2_blockref_t *bref);
1361 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
1362 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
1363 hammer2_chain_t *hammer2_chain_getparent(hammer2_chain_t **parentp, int how);
1364 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
1365 hammer2_key_t *key_nextp,
1366 hammer2_key_t key_beg, hammer2_key_t key_end,
1367 int *cache_indexp, int flags);
1368 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
1369 hammer2_chain_t *chain,
1370 hammer2_key_t *key_nextp,
1371 hammer2_key_t key_beg, hammer2_key_t key_end,
1372 int *cache_indexp, int flags);
1373 hammer2_chain_t *hammer2_chain_scan(hammer2_chain_t *parent,
1374 hammer2_chain_t *chain,
1375 int *cache_indexp, int flags);
1377 int hammer2_chain_create(hammer2_chain_t **parentp,
1378 hammer2_chain_t **chainp, hammer2_pfs_t *pmp,
1379 hammer2_key_t key, int keybits,
1380 int type, size_t bytes,
1381 hammer2_tid_t mtid, int flags);
1382 void hammer2_chain_rename(hammer2_blockref_t *bref,
1383 hammer2_chain_t **parentp,
1384 hammer2_chain_t *chain,
1385 hammer2_tid_t mtid, int flags);
1386 void hammer2_chain_delete(hammer2_chain_t *parent, hammer2_chain_t *chain,
1387 hammer2_tid_t mtid, int flags);
1388 void hammer2_chain_setflush(hammer2_chain_t *chain);
1389 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
1390 hammer2_blockref_t *base, int count);
1391 hammer2_chain_t *hammer2_chain_bulksnap(hammer2_chain_t *chain);
1392 void hammer2_chain_bulkdrop(hammer2_chain_t *copy);
1394 void hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata);
1395 int hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata);
1398 void hammer2_pfs_memory_wait(hammer2_pfs_t *pmp);
1399 void hammer2_pfs_memory_inc(hammer2_pfs_t *pmp);
1400 void hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp);
1402 void hammer2_base_delete(hammer2_chain_t *chain,
1403 hammer2_blockref_t *base, int count,
1404 int *cache_indexp, hammer2_chain_t *child);
1405 void hammer2_base_insert(hammer2_chain_t *chain,
1406 hammer2_blockref_t *base, int count,
1407 int *cache_indexp, hammer2_chain_t *child);
1412 void hammer2_flush(hammer2_chain_t *chain, int istop);
1413 void hammer2_flush_quick(hammer2_dev_t *hmp);
1414 void hammer2_delayed_flush(hammer2_chain_t *chain);
1419 void hammer2_trans_init(hammer2_pfs_t *pmp, uint32_t flags);
1420 hammer2_tid_t hammer2_trans_sub(hammer2_pfs_t *pmp);
1421 void hammer2_trans_clear_preflush(hammer2_pfs_t *pmp);
1422 void hammer2_trans_done(hammer2_pfs_t *pmp);
1423 hammer2_tid_t hammer2_trans_newinum(hammer2_pfs_t *pmp);
1424 void hammer2_trans_assert_strategy(hammer2_pfs_t *pmp);
1429 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
1430 int fflag, struct ucred *cred);
1435 void hammer2_io_putblk(hammer2_io_t **diop);
1436 void hammer2_io_cleanup(hammer2_dev_t *hmp, struct hammer2_io_tree *tree);
1437 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
1438 void hammer2_io_getblk(hammer2_dev_t *hmp, off_t lbase, int lsize,
1439 hammer2_iocb_t *iocb);
1440 void hammer2_io_complete(hammer2_iocb_t *iocb);
1441 void hammer2_io_callback(struct bio *bio);
1442 void hammer2_iocb_wait(hammer2_iocb_t *iocb);
1443 int hammer2_io_new(hammer2_dev_t *hmp, off_t lbase, int lsize,
1444 hammer2_io_t **diop);
1445 int hammer2_io_newnz(hammer2_dev_t *hmp, off_t lbase, int lsize,
1446 hammer2_io_t **diop);
1447 int hammer2_io_newq(hammer2_dev_t *hmp, off_t lbase, int lsize,
1448 hammer2_io_t **diop);
1449 int hammer2_io_bread(hammer2_dev_t *hmp, off_t lbase, int lsize,
1450 hammer2_io_t **diop);
1451 void hammer2_io_bawrite(hammer2_io_t **diop);
1452 void hammer2_io_bdwrite(hammer2_io_t **diop);
1453 int hammer2_io_bwrite(hammer2_io_t **diop);
1454 int hammer2_io_isdirty(hammer2_io_t *dio);
1455 void hammer2_io_setdirty(hammer2_io_t *dio);
1456 void hammer2_io_setinval(hammer2_io_t *dio, u_int bytes);
1457 void hammer2_io_brelse(hammer2_io_t **diop);
1458 void hammer2_io_bqrelse(hammer2_io_t **diop);
1461 * XOP API in hammer2_thread.c
1463 void hammer2_xop_group_init(hammer2_pfs_t *pmp, hammer2_xop_group_t *xgrp);
1464 void *hammer2_xop_alloc(hammer2_inode_t *ip, int flags);
1465 void hammer2_xop_setname(hammer2_xop_head_t *xop,
1466 const char *name, size_t name_len);
1467 void hammer2_xop_setname2(hammer2_xop_head_t *xop,
1468 const char *name, size_t name_len);
1469 void hammer2_xop_setip2(hammer2_xop_head_t *xop, hammer2_inode_t *ip2);
1470 void hammer2_xop_setip3(hammer2_xop_head_t *xop, hammer2_inode_t *ip3);
1471 void hammer2_xop_reinit(hammer2_xop_head_t *xop);
1472 void hammer2_xop_helper_create(hammer2_pfs_t *pmp);
1473 void hammer2_xop_helper_cleanup(hammer2_pfs_t *pmp);
1474 void hammer2_xop_start(hammer2_xop_head_t *xop, hammer2_xop_func_t func);
1475 void hammer2_xop_start_except(hammer2_xop_head_t *xop, hammer2_xop_func_t func,
1477 int hammer2_xop_collect(hammer2_xop_head_t *xop, int flags);
1478 void hammer2_xop_retire(hammer2_xop_head_t *xop, uint32_t mask);
1479 int hammer2_xop_active(hammer2_xop_head_t *xop);
1480 int hammer2_xop_feed(hammer2_xop_head_t *xop, hammer2_chain_t *chain,
1481 int clindex, int error);
1484 * XOP backends in hammer2_xops.c
1486 void hammer2_xop_ipcluster(hammer2_xop_t *xop, int clidx);
1487 void hammer2_xop_readdir(hammer2_xop_t *xop, int clidx);
1488 void hammer2_xop_nresolve(hammer2_xop_t *xop, int clidx);
1489 void hammer2_xop_unlink(hammer2_xop_t *xop, int clidx);
1490 void hammer2_xop_nrename(hammer2_xop_t *xop, int clidx);
1491 void hammer2_xop_nlink(hammer2_xop_t *xop, int clidx);
1492 void hammer2_xop_scanlhc(hammer2_xop_t *xop, int clidx);
1493 void hammer2_xop_scanall(hammer2_xop_t *xop, int clidx);
1494 void hammer2_xop_lookup(hammer2_xop_t *xop, int clidx);
1495 void hammer2_inode_xop_create(hammer2_xop_t *xop, int clidx);
1496 void hammer2_inode_xop_destroy(hammer2_xop_t *xop, int clidx);
1497 void hammer2_inode_xop_chain_sync(hammer2_xop_t *xop, int clidx);
1498 void hammer2_inode_xop_unlinkall(hammer2_xop_t *xop, int clidx);
1499 void hammer2_inode_xop_connect(hammer2_xop_t *xop, int clidx);
1500 void hammer2_inode_xop_flush(hammer2_xop_t *xop, int clidx);
1505 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
1506 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
1511 void hammer2_volconf_update(hammer2_dev_t *hmp, int index);
1512 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
1513 int hammer2_vfs_sync(struct mount *mp, int waitflags);
1514 hammer2_pfs_t *hammer2_pfsalloc(hammer2_chain_t *chain,
1515 const hammer2_inode_data_t *ripdata,
1516 hammer2_tid_t modify_tid);
1518 void hammer2_lwinprog_ref(hammer2_pfs_t *pmp);
1519 void hammer2_lwinprog_drop(hammer2_pfs_t *pmp);
1520 void hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int pipe);
1525 int hammer2_freemap_alloc(hammer2_chain_t *chain, size_t bytes);
1526 void hammer2_freemap_adjust(hammer2_dev_t *hmp,
1527 hammer2_blockref_t *bref, int how);
1532 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
1533 const hammer2_media_data_t *hammer2_cluster_rdata(hammer2_cluster_t *cluster);
1534 hammer2_media_data_t *hammer2_cluster_wdata(hammer2_cluster_t *cluster);
1535 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
1536 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
1537 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfs_t *pmp,
1538 hammer2_blockref_t *bref);
1539 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
1540 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1541 void hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1542 int hammer2_cluster_check(hammer2_cluster_t *cluster, hammer2_key_t lokey,
1544 void hammer2_cluster_resolve(hammer2_cluster_t *cluster);
1545 void hammer2_cluster_forcegood(hammer2_cluster_t *cluster);
1546 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster);
1547 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1549 int hammer2_bulk_scan(hammer2_chain_t *parent,
1550 int (*func)(hammer2_chain_t *chain, void *info),
1552 int hammer2_bulkfree_pass(hammer2_dev_t *hmp,
1553 struct hammer2_ioc_bulkfree *bfi);
1558 void hammer2_iocom_init(hammer2_dev_t *hmp);
1559 void hammer2_iocom_uninit(hammer2_dev_t *hmp);
1560 void hammer2_cluster_reconnect(hammer2_dev_t *hmp, struct file *fp);
1565 void hammer2_thr_create(hammer2_thread_t *thr, hammer2_pfs_t *pmp,
1566 const char *id, int clindex, int repidx,
1567 void (*func)(void *arg));
1568 void hammer2_thr_delete(hammer2_thread_t *thr);
1569 void hammer2_thr_remaster(hammer2_thread_t *thr);
1570 void hammer2_thr_freeze_async(hammer2_thread_t *thr);
1571 void hammer2_thr_freeze(hammer2_thread_t *thr);
1572 void hammer2_thr_unfreeze(hammer2_thread_t *thr);
1573 void hammer2_primary_sync_thread(void *arg);
1574 void hammer2_primary_xops_thread(void *arg);
1577 * hammer2_strategy.c
1579 int hammer2_vop_strategy(struct vop_strategy_args *ap);
1580 int hammer2_vop_bmap(struct vop_bmap_args *ap);
1581 void hammer2_write_thread(void *arg);
1582 void hammer2_bioq_sync(hammer2_pfs_t *pmp);
1584 #endif /* !_KERNEL */
1585 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */