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_sh_again mtx_lock_sh_again
124 #define hammer2_mtx_unlock mtx_unlock
125 #define hammer2_mtx_downgrade mtx_downgrade
126 #define hammer2_mtx_owned mtx_owned
127 #define hammer2_mtx_init mtx_init
128 #define hammer2_mtx_temp_release mtx_lock_temp_release
129 #define hammer2_mtx_temp_restore mtx_lock_temp_restore
130 #define hammer2_mtx_refs mtx_lockrefs
132 #define hammer2_spin_init spin_init
133 #define hammer2_spin_sh spin_lock_shared
134 #define hammer2_spin_ex spin_lock
135 #define hammer2_spin_unsh spin_unlock_shared
136 #define hammer2_spin_unex spin_unlock
138 TAILQ_HEAD(hammer2_xop_list, hammer2_xop_head);
140 typedef struct hammer2_xop_list hammer2_xop_list_t;
144 * General lock support
148 hammer2_mtx_upgrade(hammer2_mtx_t *mtx)
152 if (mtx_islocked_ex(mtx)) {
156 mtx_lock_ex_quick(mtx);
163 * The xid tracks internal transactional updates.
165 * XXX fix-me, really needs to be 64-bits
167 typedef uint32_t hammer2_xid_t;
169 #define HAMMER2_XID_MIN 0x00000000U
170 #define HAMMER2_XID_MAX 0x7FFFFFFFU
173 * The chain structure tracks a portion of the media topology from the
174 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
175 * data blocks, and freemap nodes and leafs.
177 * The chain structure utilizes a simple singly-homed topology and the
178 * chain's in-memory topology will move around as the chains do, due mainly
179 * to renames and indirect block creation.
181 * Block Table Updates
183 * Block table updates for insertions and updates are delayed until the
184 * flush. This allows us to avoid having to modify the parent chain
185 * all the way to the root.
187 * Block table deletions are performed immediately (modifying the parent
188 * in the process) because the flush code uses the chain structure to
189 * track delayed updates and the chain will be (likely) gone or moved to
190 * another location in the topology after a deletion.
192 * A prior iteration of the code tried to keep the relationship intact
193 * on deletes by doing a delete-duplicate operation on the chain, but
194 * it added way too much complexity to the codebase.
196 * Flush Synchronization
198 * The flush code must flush modified chains bottom-up. Because chain
199 * structures can shift around and are NOT topologically stable,
200 * modified chains are independently indexed for the flush. As the flush
201 * runs it modifies (or further modifies) and updates the parents,
202 * propagating the flush all the way to the volume root.
204 * Modifying front-end operations can occur during a flush but will block
205 * in two cases: (1) when the front-end tries to operate on the inode
206 * currently in the midst of being flushed and (2) if the front-end
207 * crosses an inode currently being flushed (such as during a rename).
208 * So, for example, if you rename directory "x" to "a/b/c/d/e/f/g/x" and
209 * the flusher is currently working on "a/b/c", the rename will block
210 * temporarily in order to ensure that "x" exists in one place or the
213 * Meta-data statistics are updated by the flusher. The front-end will
214 * make estimates but meta-data must be fully synchronized only during a
215 * flush in order to ensure that it remains correct across a crash.
217 * Multiple flush synchronizations can theoretically be in-flight at the
218 * same time but the implementation is not coded to handle the case and
219 * currently serializes them.
223 * Snapshots currently require the subdirectory tree being snapshotted
224 * to be flushed. The snapshot then creates a new super-root inode which
225 * copies the flushed blockdata of the directory or file that was
230 * - Note that the radix tree runs in powers of 2 only so sub-trees
231 * cannot straddle edges.
233 RB_HEAD(hammer2_chain_tree, hammer2_chain);
234 TAILQ_HEAD(h2_flush_list, hammer2_chain);
235 TAILQ_HEAD(h2_core_list, hammer2_chain);
236 TAILQ_HEAD(h2_iocb_list, hammer2_iocb);
238 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
239 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
242 * Core topology for chain (embedded in chain). Protected by a spinlock.
244 struct hammer2_chain_core {
246 struct hammer2_chain_tree rbtree; /* sub-chains */
247 int live_zero; /* blockref array opt */
248 u_int live_count; /* live (not deleted) chains in tree */
249 u_int chain_count; /* live + deleted chains under core */
250 int generation; /* generation number (inserts only) */
253 typedef struct hammer2_chain_core hammer2_chain_core_t;
255 RB_HEAD(hammer2_io_tree, hammer2_io);
258 * IOCB - IO callback (into chain, cluster, or manual request)
260 struct hammer2_iocb {
261 TAILQ_ENTRY(hammer2_iocb) entry;
262 void (*callback)(struct hammer2_iocb *iocb);
263 struct hammer2_io *dio;
264 struct hammer2_cluster *cluster;
265 struct hammer2_chain *chain;
273 typedef struct hammer2_iocb hammer2_iocb_t;
275 #define HAMMER2_IOCB_INTERLOCK 0x00000001
276 #define HAMMER2_IOCB_ONQ 0x00000002
277 #define HAMMER2_IOCB_DONE 0x00000004
278 #define HAMMER2_IOCB_INPROG 0x00000008
279 #define HAMMER2_IOCB_UNUSED10 0x00000010
280 #define HAMMER2_IOCB_QUICK 0x00010000
281 #define HAMMER2_IOCB_ZERO 0x00020000
282 #define HAMMER2_IOCB_READ 0x00040000
283 #define HAMMER2_IOCB_WAKEUP 0x00080000
286 * DIO - Management structure wrapping system buffer cache.
288 * Used for multiple purposes including concurrent management
289 * if small requests by chains into larger DIOs.
292 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
293 struct h2_iocb_list iocbq;
294 struct spinlock spin;
295 struct hammer2_dev *hmp;
300 int act; /* activity */
303 typedef struct hammer2_io hammer2_io_t;
305 #define HAMMER2_DIO_INPROG 0x80000000 /* bio in progress */
306 #define HAMMER2_DIO_GOOD 0x40000000 /* dio->bp is stable */
307 #define HAMMER2_DIO_WAITING 0x20000000 /* (old) */
308 #define HAMMER2_DIO_DIRTY 0x10000000 /* flush on last drop */
310 #define HAMMER2_DIO_MASK 0x0FFFFFFF
313 * Primary chain structure keeps track of the topology in-memory.
315 struct hammer2_chain {
317 hammer2_chain_core_t core;
318 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
319 hammer2_blockref_t bref;
320 struct hammer2_chain *parent;
321 struct hammer2_state *state; /* if active cache msg */
322 struct hammer2_dev *hmp;
323 struct hammer2_pfs *pmp; /* A PFS or super-root (spmp) */
325 hammer2_io_t *dio; /* physical data buffer */
326 u_int bytes; /* physical data size */
330 int error; /* on-lock data error state */
332 hammer2_media_data_t *data; /* data pointer shortcut */
333 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
336 typedef struct hammer2_chain hammer2_chain_t;
338 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
339 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
342 * Special notes on flags:
344 * INITIAL - This flag allows a chain to be created and for storage to
345 * be allocated without having to immediately instantiate the
346 * related buffer. The data is assumed to be all-zeros. It
347 * is primarily used for indirect blocks.
349 * MODIFIED - The chain's media data has been modified. Prevents chain
350 * free on lastdrop if still in the topology.
352 * UPDATE - Chain might not be modified but parent blocktable needs
353 * an update. Prevents chain free on lastdrop if still in
356 * FICTITIOUS - Faked chain as a placeholder for an error condition. This
357 * chain is unsuitable for I/O.
359 * BMAPPED - Indicates that the chain is present in the parent blockmap.
361 * BMAPUPD - Indicates that the chain is present but needs to be updated
362 * in the parent blockmap.
364 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
365 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
366 #define HAMMER2_CHAIN_DESTROY 0x00000004
367 #define HAMMER2_CHAIN_DEDUP 0x00000008 /* used as dedup src */
368 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
369 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
370 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
371 #define HAMMER2_CHAIN_DEFERRED 0x00000080 /* flush depth defer */
372 #define HAMMER2_CHAIN_UNUSED000001000 0x00000100
373 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
374 #define HAMMER2_CHAIN_FICTITIOUS 0x00000400 /* unsuitable for I/O */
375 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
376 #define HAMMER2_CHAIN_DELAYED 0x00001000 /* delayed flush */
377 #define HAMMER2_CHAIN_COUNTEDBREFS 0x00002000 /* block table stats */
378 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
379 #define HAMMER2_CHAIN_UNUSED00008000 0x00008000
380 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
381 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
382 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
383 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
384 #define HAMMER2_CHAIN_IOINPROG 0x00100000 /* I/O interlock */
385 #define HAMMER2_CHAIN_IOSIGNAL 0x00200000 /* I/O interlock */
386 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
388 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
389 HAMMER2_CHAIN_UPDATE | \
390 HAMMER2_CHAIN_ONFLUSH)
393 * Hammer2 error codes, used by chain->error and cluster->error. The error
394 * code is typically set on-lock unless no I/O was requested, and set on
395 * I/O otherwise. If set for a cluster it generally means that the cluster
396 * code could not find a valid copy to present.
398 * IO - An I/O error occurred
399 * CHECK - I/O succeeded but did not match the check code
400 * INCOMPLETE - A cluster is not complete enough to use, or
401 * a chain cannot be loaded because its parent has an error.
403 * NOTE: API allows callers to check zero/non-zero to determine if an error
406 * NOTE: Chain's data field is usually NULL on an IO error but not necessarily
407 * NULL on other errors. Check chain->error, not chain->data.
409 #define HAMMER2_ERROR_NONE 0
410 #define HAMMER2_ERROR_IO 1 /* device I/O error */
411 #define HAMMER2_ERROR_CHECK 2 /* check code mismatch */
412 #define HAMMER2_ERROR_INCOMPLETE 3 /* incomplete cluster */
413 #define HAMMER2_ERROR_DEPTH 4 /* temporary depth limit */
416 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
419 * NOLOCK - Input and output chains are referenced only and not
420 * locked. Output chain might be temporarily locked
423 * NODATA - Asks that the chain->data not be resolved in order
426 * NODIRECT - Prevents a lookup of offset 0 in an inode from returning
427 * the inode itself if the inode is in DIRECTDATA mode
428 * (i.e. file is <= 512 bytes). Used by the synchronization
429 * code to prevent confusion.
431 * SHARED - The input chain is expected to be locked shared,
432 * and the output chain is locked shared.
434 * MATCHIND - Allows an indirect block / freemap node to be returned
435 * when the passed key range matches the radix. Remember
436 * that key_end is inclusive (e.g. {0x000,0xFFF},
437 * not {0x000,0x1000}).
439 * (Cannot be used for remote or cluster ops).
441 * ALLNODES - Allows NULL focus.
443 * ALWAYS - Always resolve the data. If ALWAYS and NODATA are both
444 * missing, bulk file data is not resolved but inodes and
445 * other meta-data will.
447 * NOUNLOCK - Used by hammer2_chain_next() to leave the lock on
448 * the input chain intact. The chain is still dropped.
449 * This allows the caller to add a reference to the chain
450 * and retain it in a locked state (used by the
451 * XOP/feed/collect code).
453 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
454 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
455 #define HAMMER2_LOOKUP_NODIRECT 0x00000004 /* no offset=0 DD */
456 #define HAMMER2_LOOKUP_SHARED 0x00000100
457 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
458 #define HAMMER2_LOOKUP_ALLNODES 0x00000400 /* allow NULL focus */
459 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
460 #define HAMMER2_LOOKUP_NOUNLOCK 0x00001000 /* leave lock intact */
463 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
465 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
466 * blocks in the INITIAL-create state.
468 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
469 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
470 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
473 * Flags passed to hammer2_chain_lock()
475 * NOTE: RDONLY is set to optimize cluster operations when *no* modifications
476 * will be made to either the cluster being locked or any underlying
477 * cluster. It allows the cluster to lock and access data for a subset
478 * of available nodes instead of all available nodes.
480 #define HAMMER2_RESOLVE_NEVER 1
481 #define HAMMER2_RESOLVE_MAYBE 2
482 #define HAMMER2_RESOLVE_ALWAYS 3
483 #define HAMMER2_RESOLVE_MASK 0x0F
485 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
486 #define HAMMER2_RESOLVE_UNUSED20 0x20
487 #define HAMMER2_RESOLVE_RDONLY 0x40 /* higher level op flag */
490 * Flags passed to hammer2_chain_delete()
492 #define HAMMER2_DELETE_PERMANENT 0x0001
495 * Flags passed to hammer2_chain_insert() or hammer2_chain_rename()
497 #define HAMMER2_INSERT_PFSROOT 0x0004
500 * Flags passed to hammer2_chain_delete_duplicate()
502 #define HAMMER2_DELDUP_RECORE 0x0001
505 * Cluster different types of storage together for allocations
507 #define HAMMER2_FREECACHE_INODE 0
508 #define HAMMER2_FREECACHE_INDIR 1
509 #define HAMMER2_FREECACHE_DATA 2
510 #define HAMMER2_FREECACHE_UNUSED3 3
511 #define HAMMER2_FREECACHE_TYPES 4
514 * hammer2_freemap_alloc() block preference
516 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
519 * BMAP read-ahead maximum parameters
521 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
522 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
525 * hammer2_freemap_adjust()
527 #define HAMMER2_FREEMAP_DORECOVER 1
528 #define HAMMER2_FREEMAP_DOMAYFREE 2
529 #define HAMMER2_FREEMAP_DOREALFREE 3
532 * HAMMER2 cluster - A set of chains representing the same entity.
534 * hammer2_cluster typically represents a temporary set of representitive
535 * chains. The one exception is that a hammer2_cluster is embedded in
536 * hammer2_inode. This embedded cluster is ONLY used to track the
537 * representitive chains and cannot be directly locked.
539 * A cluster is usually temporary (and thus per-thread) for locking purposes,
540 * allowing us to embed the asynchronous storage required for cluster
541 * operations in the cluster itself and adjust the state and status without
542 * having to worry too much about SMP issues.
544 * The exception is the cluster embedded in the hammer2_inode structure.
545 * This is used to cache the cluster state on an inode-by-inode basis.
546 * Individual hammer2_chain structures not incorporated into clusters might
547 * also stick around to cache miscellanious elements.
549 * Because the cluster is a 'working copy' and is usually subject to cluster
550 * quorum rules, it is quite possible for us to end up with an insufficient
551 * number of live chains to execute an operation. If an insufficient number
552 * of chains remain in a working copy, the operation may have to be
553 * downgraded, retried, stall until the requisit number of chains are
554 * available, or possibly even error out depending on the mount type.
556 * A cluster's focus is set when it is locked. The focus can only be set
557 * to a chain still part of the synchronized set.
559 #define HAMMER2_MAXCLUSTER 8
560 #define HAMMER2_XOPFIFO 16
561 #define HAMMER2_XOPFIFO_MASK (HAMMER2_XOPFIFO - 1)
562 #define HAMMER2_XOPGROUPS 16
563 #define HAMMER2_XOPGROUPS_MASK (HAMMER2_XOPGROUPS - 1)
564 #define HAMMER2_XOPMASK_VOP 0x80000000U
566 struct hammer2_cluster_item {
567 hammer2_chain_t *chain;
573 typedef struct hammer2_cluster_item hammer2_cluster_item_t;
576 * INVALID - Invalid for focus, i.e. not part of synchronized set.
577 * Once set, this bit is sticky across operations.
579 * FEMOD - Indicates that front-end modifying operations can
580 * mess with this entry and MODSYNC will copy also
583 #define HAMMER2_CITEM_INVALID 0x00000001
584 #define HAMMER2_CITEM_FEMOD 0x00000002
585 #define HAMMER2_CITEM_NULL 0x00000004
587 struct hammer2_cluster {
588 int refs; /* track for deallocation */
590 struct hammer2_pfs *pmp;
593 int error; /* error code valid on lock */
596 hammer2_chain_t *focus; /* current focus (or mod) */
597 hammer2_cluster_item_t array[HAMMER2_MAXCLUSTER];
600 typedef struct hammer2_cluster hammer2_cluster_t;
603 * WRHARD - Hard mounts can write fully synchronized
604 * RDHARD - Hard mounts can read fully synchronized
605 * UNHARD - Unsynchronized masters present
606 * NOHARD - No masters visible
607 * WRSOFT - Soft mounts can write to at least the SOFT_MASTER
608 * RDSOFT - Soft mounts can read from at least a SOFT_SLAVE
609 * UNSOFT - Unsynchronized slaves present
610 * NOSOFT - No slaves visible
611 * RDSLAVE - slaves are accessible (possibly unsynchronized or remote).
612 * MSYNCED - All masters are fully synchronized
613 * SSYNCED - All known local slaves are fully synchronized to masters
615 * All available masters are always incorporated. All PFSs belonging to a
616 * cluster (master, slave, copy, whatever) always try to synchronize the
617 * total number of known masters in the PFSs root inode.
619 * A cluster might have access to many slaves, copies, or caches, but we
620 * have a limited number of cluster slots. Any such elements which are
621 * directly mounted from block device(s) will always be incorporated. Note
622 * that SSYNCED only applies to such elements which are directly mounted,
623 * not to any remote slaves, copies, or caches that could be available. These
624 * bits are used to monitor and drive our synchronization threads.
626 * When asking the question 'is any data accessible at all', then a simple
627 * test against (RDHARD|RDSOFT|RDSLAVE) gives you the answer. If any of
628 * these bits are set the object can be read with certain caveats:
629 * RDHARD - no caveats. RDSOFT - authoritative but might not be synchronized.
630 * and RDSLAVE - not authoritative, has some data but it could be old or
633 * When both soft and hard mounts are available, data will be read and written
634 * via the soft mount only. But all might be in the cluster because
635 * background synchronization threads still need to do their work.
637 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode struct */
638 #define HAMMER2_CLUSTER_UNUSED2 0x00000002
639 #define HAMMER2_CLUSTER_LOCKED 0x00000004 /* cluster lks not recursive */
640 #define HAMMER2_CLUSTER_WRHARD 0x00000100 /* hard-mount can write */
641 #define HAMMER2_CLUSTER_RDHARD 0x00000200 /* hard-mount can read */
642 #define HAMMER2_CLUSTER_UNHARD 0x00000400 /* unsynchronized masters */
643 #define HAMMER2_CLUSTER_NOHARD 0x00000800 /* no masters visible */
644 #define HAMMER2_CLUSTER_WRSOFT 0x00001000 /* soft-mount can write */
645 #define HAMMER2_CLUSTER_RDSOFT 0x00002000 /* soft-mount can read */
646 #define HAMMER2_CLUSTER_UNSOFT 0x00004000 /* unsynchronized slaves */
647 #define HAMMER2_CLUSTER_NOSOFT 0x00008000 /* no slaves visible */
648 #define HAMMER2_CLUSTER_MSYNCED 0x00010000 /* all masters synchronized */
649 #define HAMMER2_CLUSTER_SSYNCED 0x00020000 /* known slaves synchronized */
651 #define HAMMER2_CLUSTER_ANYDATA ( HAMMER2_CLUSTER_RDHARD | \
652 HAMMER2_CLUSTER_RDSOFT | \
653 HAMMER2_CLUSTER_RDSLAVE)
655 #define HAMMER2_CLUSTER_RDOK ( HAMMER2_CLUSTER_RDHARD | \
656 HAMMER2_CLUSTER_RDSOFT)
658 #define HAMMER2_CLUSTER_WROK ( HAMMER2_CLUSTER_WRHARD | \
659 HAMMER2_CLUSTER_WRSOFT)
661 #define HAMMER2_CLUSTER_ZFLAGS ( HAMMER2_CLUSTER_WRHARD | \
662 HAMMER2_CLUSTER_RDHARD | \
663 HAMMER2_CLUSTER_WRSOFT | \
664 HAMMER2_CLUSTER_RDSOFT | \
665 HAMMER2_CLUSTER_MSYNCED | \
666 HAMMER2_CLUSTER_SSYNCED)
669 * Helper functions (cluster must be locked for flags to be valid).
673 hammer2_cluster_rdok(hammer2_cluster_t *cluster)
675 return (cluster->flags & HAMMER2_CLUSTER_RDOK);
680 hammer2_cluster_wrok(hammer2_cluster_t *cluster)
682 return (cluster->flags & HAMMER2_CLUSTER_WROK);
685 RB_HEAD(hammer2_inode_tree, hammer2_inode);
690 * NOTE: The inode-embedded cluster is never used directly for I/O (since
691 * it may be shared). Instead it will be replicated-in and synchronized
692 * back out if changed.
694 struct hammer2_inode {
695 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
696 hammer2_mtx_t lock; /* inode lock */
697 struct hammer2_pfs *pmp; /* PFS mount */
698 struct hammer2_inode *pip; /* parent inode */
700 struct spinlock cluster_spin; /* update cluster */
701 hammer2_cluster_t cluster;
702 struct lockf advlock;
704 u_int refs; /* +vpref, +flushref */
705 uint8_t comp_heuristic;
706 hammer2_inode_meta_t meta; /* copy of meta-data */
710 typedef struct hammer2_inode hammer2_inode_t;
713 * MODIFIED - Inode is in a modified state, ip->meta may have changes.
714 * RESIZED - Inode truncated (any) or inode extended beyond
717 #define HAMMER2_INODE_MODIFIED 0x0001
718 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
719 #define HAMMER2_INODE_RENAME_INPROG 0x0004
720 #define HAMMER2_INODE_ONRBTREE 0x0008
721 #define HAMMER2_INODE_RESIZED 0x0010 /* requires inode_fsync */
722 #define HAMMER2_INODE_ISDELETED 0x0020 /* deleted, not in ihidden */
723 #define HAMMER2_INODE_ISUNLINKED 0x0040
724 #define HAMMER2_INODE_METAGOOD 0x0080 /* inode meta-data good */
725 #define HAMMER2_INODE_ONSIDEQ 0x0100 /* on side processing queue */
727 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
728 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
732 * inode-unlink side-structure
734 struct hammer2_inode_sideq {
735 TAILQ_ENTRY(hammer2_inode_sideq) entry;
738 TAILQ_HEAD(h2_sideq_list, hammer2_inode_sideq);
740 typedef struct hammer2_inode_sideq hammer2_inode_sideq_t;
743 * Transaction management sub-structure under hammer2_pfs
745 struct hammer2_trans {
750 typedef struct hammer2_trans hammer2_trans_t;
752 #define HAMMER2_TRANS_ISFLUSH 0x80000000 /* flush code */
753 #define HAMMER2_TRANS_BUFCACHE 0x40000000 /* bio strategy */
754 #define HAMMER2_TRANS_PREFLUSH 0x20000000 /* preflush state */
755 #define HAMMER2_TRANS_FPENDING 0x10000000 /* flush pending */
756 #define HAMMER2_TRANS_WAITING 0x08000000 /* someone waiting */
757 #define HAMMER2_TRANS_MASK 0x00FFFFFF /* count mask */
759 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
760 #define HAMMER2_FREEMAP_HEUR_TYPES 8
761 #define HAMMER2_FREEMAP_HEUR_SIZE (HAMMER2_FREEMAP_HEUR_NRADIX * \
762 HAMMER2_FREEMAP_HEUR_TYPES)
764 #define HAMMER2_DEDUP_HEUR_SIZE 65536
765 #define HAMMER2_DEDUP_HEUR_MASK (HAMMER2_DEDUP_HEUR_SIZE - 1)
767 #define HAMMER2_FLUSH_TOP 0x0001
768 #define HAMMER2_FLUSH_ALL 0x0002
772 * Hammer2 support thread element.
774 * Potentially many support threads can hang off of hammer2, primarily
775 * off the hammer2_pfs structure. Typically:
777 * td x Nodes A synchronization thread for each node.
778 * td x Nodes x workers Worker threads for frontend operations.
779 * td x 1 Bioq thread for logical buffer writes.
781 * In addition, the synchronization thread(s) associated with the
782 * super-root PFS (spmp) for a node is responsible for automatic bulkfree
785 struct hammer2_thread {
786 struct hammer2_pfs *pmp;
787 hammer2_xop_list_t *xopq; /* points into pmp->xopq[] */
791 int clindex; /* cluster element index */
793 struct lock lk; /* thread control lock */
796 typedef struct hammer2_thread hammer2_thread_t;
798 #define HAMMER2_THREAD_UNMOUNTING 0x0001 /* unmount request */
799 #define HAMMER2_THREAD_DEV 0x0002 /* related to dev, not pfs */
800 #define HAMMER2_THREAD_UNUSED04 0x0004
801 #define HAMMER2_THREAD_REMASTER 0x0008 /* remaster request */
802 #define HAMMER2_THREAD_STOP 0x0010 /* exit request */
803 #define HAMMER2_THREAD_FREEZE 0x0020 /* force idle */
804 #define HAMMER2_THREAD_FROZEN 0x0040 /* restart */
807 * Support structure for dedup heuristic.
809 struct hammer2_dedup {
810 hammer2_off_t data_off;
815 typedef struct hammer2_dedup hammer2_dedup_t;
818 * hammer2_xop - container for VOP/XOP operation (allocated, not on stack).
820 * This structure is used to distribute a VOP operation across multiple
821 * nodes. It provides a rendezvous for concurrent node execution and
822 * can be detached from the frontend operation to allow the frontend to
825 * This structure also sequences operations on up to three inodes.
827 typedef void (*hammer2_xop_func_t)(union hammer2_xop *xop, int clidx);
829 struct hammer2_xop_fifo {
830 TAILQ_ENTRY(hammer2_xop_head) entry;
831 hammer2_chain_t *array[HAMMER2_XOPFIFO];
832 int errors[HAMMER2_XOPFIFO];
838 typedef struct hammer2_xop_fifo hammer2_xop_fifo_t;
840 #define HAMMER2_XOP_FIFO_RUN 0x0001
841 #define HAMMER2_XOP_FIFO_STALL 0x0002
843 struct hammer2_xop_head {
844 hammer2_xop_func_t func;
846 struct hammer2_inode *ip1;
847 struct hammer2_inode *ip2;
848 struct hammer2_inode *ip3;
849 uint32_t check_counter;
854 hammer2_key_t collect_key;
859 hammer2_xop_fifo_t collect[HAMMER2_MAXCLUSTER];
860 hammer2_cluster_t cluster; /* help collections */
863 typedef struct hammer2_xop_head hammer2_xop_head_t;
865 struct hammer2_xop_ipcluster {
866 hammer2_xop_head_t head;
869 struct hammer2_xop_strategy {
870 hammer2_xop_head_t head;
877 struct hammer2_xop_readdir {
878 hammer2_xop_head_t head;
882 struct hammer2_xop_nresolve {
883 hammer2_xop_head_t head;
884 hammer2_key_t lhc; /* if name is NULL used lhc */
887 struct hammer2_xop_nlink {
888 hammer2_xop_head_t head;
892 struct hammer2_xop_unlink {
893 hammer2_xop_head_t head;
898 struct hammer2_xop_nrename {
899 hammer2_xop_head_t head;
904 struct hammer2_xop_scanlhc {
905 hammer2_xop_head_t head;
909 struct hammer2_xop_scanall {
910 hammer2_xop_head_t head;
911 hammer2_key_t key_beg; /* inclusive */
912 hammer2_key_t key_end; /* inclusive */
915 struct hammer2_xop_lookup {
916 hammer2_xop_head_t head;
920 struct hammer2_xop_create {
921 hammer2_xop_head_t head;
922 hammer2_inode_meta_t meta; /* initial metadata */
927 struct hammer2_xop_destroy {
928 hammer2_xop_head_t head;
931 struct hammer2_xop_fsync {
932 hammer2_xop_head_t head;
933 hammer2_inode_meta_t meta;
936 int clear_directdata;
939 struct hammer2_xop_unlinkall {
940 hammer2_xop_head_t head;
941 hammer2_key_t key_beg;
942 hammer2_key_t key_end;
945 struct hammer2_xop_connect {
946 hammer2_xop_head_t head;
950 struct hammer2_xop_flush {
951 hammer2_xop_head_t head;
954 typedef struct hammer2_xop_readdir hammer2_xop_readdir_t;
955 typedef struct hammer2_xop_nresolve hammer2_xop_nresolve_t;
956 typedef struct hammer2_xop_nlink hammer2_xop_nlink_t;
957 typedef struct hammer2_xop_unlink hammer2_xop_unlink_t;
958 typedef struct hammer2_xop_nrename hammer2_xop_nrename_t;
959 typedef struct hammer2_xop_ipcluster hammer2_xop_ipcluster_t;
960 typedef struct hammer2_xop_strategy hammer2_xop_strategy_t;
961 typedef struct hammer2_xop_create hammer2_xop_create_t;
962 typedef struct hammer2_xop_destroy hammer2_xop_destroy_t;
963 typedef struct hammer2_xop_fsync hammer2_xop_fsync_t;
964 typedef struct hammer2_xop_unlinkall hammer2_xop_unlinkall_t;
965 typedef struct hammer2_xop_scanlhc hammer2_xop_scanlhc_t;
966 typedef struct hammer2_xop_scanall hammer2_xop_scanall_t;
967 typedef struct hammer2_xop_lookup hammer2_xop_lookup_t;
968 typedef struct hammer2_xop_connect hammer2_xop_connect_t;
969 typedef struct hammer2_xop_flush hammer2_xop_flush_t;
972 hammer2_xop_head_t head;
973 hammer2_xop_ipcluster_t xop_ipcluster;
974 hammer2_xop_readdir_t xop_readdir;
975 hammer2_xop_nresolve_t xop_nresolve;
976 hammer2_xop_nlink_t xop_nlink;
977 hammer2_xop_unlink_t xop_unlink;
978 hammer2_xop_nrename_t xop_nrename;
979 hammer2_xop_strategy_t xop_strategy;
980 hammer2_xop_create_t xop_create;
981 hammer2_xop_destroy_t xop_destroy;
982 hammer2_xop_fsync_t xop_fsync;
983 hammer2_xop_unlinkall_t xop_unlinkall;
984 hammer2_xop_scanlhc_t xop_scanlhc;
985 hammer2_xop_scanall_t xop_scanall;
986 hammer2_xop_lookup_t xop_lookup;
987 hammer2_xop_flush_t xop_flush;
988 hammer2_xop_connect_t xop_connect;
991 typedef union hammer2_xop hammer2_xop_t;
994 * hammer2_xop_group - Manage XOP support threads.
996 struct hammer2_xop_group {
997 hammer2_thread_t thrs[HAMMER2_MAXCLUSTER];
1000 typedef struct hammer2_xop_group hammer2_xop_group_t;
1003 * flags to hammer2_xop_collect()
1005 #define HAMMER2_XOP_COLLECT_NOWAIT 0x00000001
1006 #define HAMMER2_XOP_COLLECT_WAITALL 0x00000002
1009 * flags to hammer2_xop_alloc()
1011 * MODIFYING - This is a modifying transaction, allocate a mtid.
1013 #define HAMMER2_XOP_MODIFYING 0x00000001
1016 * Global (per partition) management structure, represents a hard block
1017 * device. Typically referenced by hammer2_chain structures when applicable.
1018 * Typically not used for network-managed elements.
1020 * Note that a single hammer2_dev can be indirectly tied to multiple system
1021 * mount points. There is no direct relationship. System mounts are
1022 * per-cluster-id, not per-block-device, and a single hard mount might contain
1023 * many PFSs and those PFSs might combine together in various ways to form
1024 * the set of available clusters.
1026 struct hammer2_dev {
1027 struct vnode *devvp; /* device vnode */
1028 int ronly; /* read-only mount */
1029 int mount_count; /* number of actively mounted PFSs */
1030 TAILQ_ENTRY(hammer2_dev) mntentry; /* hammer2_mntlist */
1032 struct malloc_type *mchain;
1035 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
1036 struct spinlock io_spin; /* iotree access */
1037 struct hammer2_io_tree iotree;
1039 hammer2_chain_t vchain; /* anchor chain (topology) */
1040 hammer2_chain_t fchain; /* anchor chain (freemap) */
1041 struct spinlock list_spin;
1042 struct h2_flush_list flushq; /* flush seeds */
1043 struct hammer2_pfs *spmp; /* super-root pmp for transactions */
1044 struct lock bulklk; /* bulkfree lock */
1045 struct lock vollk; /* lockmgr lock */
1046 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR_SIZE];
1047 hammer2_dedup_t heur_dedup[HAMMER2_DEDUP_HEUR_SIZE];
1048 int volhdrno; /* last volhdrno written */
1049 char devrepname[64]; /* for kprintf */
1050 hammer2_volume_data_t voldata;
1051 hammer2_volume_data_t volsync; /* synchronized voldata */
1054 typedef struct hammer2_dev hammer2_dev_t;
1057 * Helper functions (cluster must be locked for flags to be valid).
1061 hammer2_chain_rdok(hammer2_chain_t *chain)
1063 return (chain->error == 0);
1068 hammer2_chain_wrok(hammer2_chain_t *chain)
1070 return (chain->error == 0 && chain->hmp->ronly == 0);
1074 * Per-cluster management structure. This structure will be tied to a
1075 * system mount point if the system is mounting the PFS, but is also used
1076 * to manage clusters encountered during the super-root scan or received
1077 * via LNK_SPANs that might not be mounted.
1079 * This structure is also used to represent the super-root that hangs off
1080 * of a hard mount point. The super-root is not really a cluster element.
1081 * In this case the spmp_hmp field will be non-NULL. It's just easier to do
1082 * this than to special case super-root manipulation in the hammer2_chain*
1083 * code as being only hammer2_dev-related.
1085 * pfs_mode and pfs_nmasters are rollup fields which critically describes
1086 * how elements of the cluster act on the cluster. pfs_mode is only applicable
1087 * when a PFS is mounted by the system. pfs_nmasters is our best guess as to
1088 * how many masters have been configured for a cluster and is always
1089 * applicable. pfs_types[] is an array with 1:1 correspondance to the
1090 * iroot cluster and describes the PFS types of the nodes making up the
1093 * WARNING! Portions of this structure have deferred initialization. In
1094 * particular, if not mounted there will be no ihidden or wthread.
1095 * umounted network PFSs will also be missing iroot and numerous
1096 * other fields will not be initialized prior to mount.
1098 * Synchronization threads are chain-specific and only applicable
1099 * to local hard PFS entries. A hammer2_pfs structure may contain
1100 * more than one when multiple hard PFSs are present on the local
1101 * machine which require synchronization monitoring. Most PFSs
1102 * (such as snapshots) are 1xMASTER PFSs which do not need a
1103 * synchronization thread.
1105 * WARNING! The chains making up pfs->iroot's cluster are accounted for in
1106 * hammer2_dev->mount_count when the pfs is associated with a mount
1109 struct hammer2_pfs {
1111 TAILQ_ENTRY(hammer2_pfs) mntentry; /* hammer2_pfslist */
1113 hammer2_dev_t *spmp_hmp; /* only if super-root pmp */
1114 hammer2_inode_t *iroot; /* PFS root inode */
1115 hammer2_inode_t *ihidden; /* PFS hidden directory */
1116 uint8_t pfs_types[HAMMER2_MAXCLUSTER];
1117 char *pfs_names[HAMMER2_MAXCLUSTER];
1118 hammer2_dev_t *pfs_hmps[HAMMER2_MAXCLUSTER];
1119 hammer2_trans_t trans;
1120 struct lock lock; /* PFS lock for certain ops */
1121 struct lock lock_nlink; /* rename and nlink lock */
1122 struct netexport export; /* nfs export */
1123 int ronly; /* read-only mount */
1124 struct malloc_type *minode;
1125 struct malloc_type *mmsg;
1126 struct spinlock inum_spin; /* inumber lookup */
1127 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
1128 hammer2_tid_t modify_tid; /* modify transaction id */
1129 hammer2_tid_t inode_tid; /* inode allocator */
1130 uint8_t pfs_nmasters; /* total masters */
1131 uint8_t pfs_mode; /* operating mode PFSMODE */
1136 uint32_t inmem_dirty_chains;
1137 int count_lwinprog; /* logical write in prog */
1138 struct spinlock list_spin;
1139 struct h2_sideq_list sideq; /* last-close dirty/unlink */
1140 hammer2_thread_t sync_thrs[HAMMER2_MAXCLUSTER];
1141 uint32_t cluster_flags; /* cached cluster flags */
1142 int has_xop_threads;
1143 struct spinlock xop_spin; /* xop sequencer */
1144 hammer2_xop_group_t xop_groups[HAMMER2_XOPGROUPS];
1145 hammer2_xop_list_t xopq[HAMMER2_MAXCLUSTER];
1148 typedef struct hammer2_pfs hammer2_pfs_t;
1150 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
1151 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
1153 #define HAMMER2_LWINPROG_WAITING 0x80000000
1154 #define HAMMER2_LWINPROG_WAITING0 0x40000000
1155 #define HAMMER2_LWINPROG_MASK 0x3FFFFFFF
1158 * hammer2_cluster_check
1160 #define HAMMER2_CHECK_NULL 0x00000001
1165 #define HAMMER2_BULK_ABORT 0x00000001
1170 #if defined(_KERNEL)
1172 MALLOC_DECLARE(M_HAMMER2);
1174 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
1175 #define ITOV(ip) ((ip)->vp)
1178 * Currently locked chains retain the locked buffer cache buffer for
1179 * indirect blocks, and indirect blocks can be one of two sizes. The
1180 * device buffer has to match the case to avoid deadlocking recursive
1181 * chains that might otherwise try to access different offsets within
1182 * the same device buffer.
1186 hammer2_devblkradix(int radix)
1189 if (radix <= HAMMER2_LBUFRADIX) {
1190 return (HAMMER2_LBUFRADIX);
1192 return (HAMMER2_PBUFRADIX);
1195 return (HAMMER2_PBUFRADIX);
1199 * XXX almost time to remove this. DIO uses PBUFSIZE exclusively now.
1203 hammer2_devblksize(size_t bytes)
1206 if (bytes <= HAMMER2_LBUFSIZE) {
1207 return(HAMMER2_LBUFSIZE);
1209 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
1210 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
1211 return (HAMMER2_PBUFSIZE);
1214 return (HAMMER2_PBUFSIZE);
1220 MPTOPMP(struct mount *mp)
1222 return ((hammer2_pfs_t *)mp->mnt_data);
1225 #define LOCKSTART int __nlocks = curthread->td_locks
1226 #define LOCKENTER (++curthread->td_locks)
1227 #define LOCKEXIT (--curthread->td_locks)
1228 #define LOCKSTOP KKASSERT(curthread->td_locks == __nlocks)
1230 extern struct vop_ops hammer2_vnode_vops;
1231 extern struct vop_ops hammer2_spec_vops;
1232 extern struct vop_ops hammer2_fifo_vops;
1234 extern int hammer2_debug;
1235 extern int hammer2_cluster_enable;
1236 extern int hammer2_hardlink_enable;
1237 extern int hammer2_flush_pipe;
1238 extern int hammer2_synchronous_flush;
1239 extern int hammer2_dio_count;
1240 extern long hammer2_limit_dirty_chains;
1241 extern long hammer2_count_modified_chains;
1242 extern long hammer2_iod_file_read;
1243 extern long hammer2_iod_meta_read;
1244 extern long hammer2_iod_indr_read;
1245 extern long hammer2_iod_fmap_read;
1246 extern long hammer2_iod_volu_read;
1247 extern long hammer2_iod_file_write;
1248 extern long hammer2_iod_file_wembed;
1249 extern long hammer2_iod_file_wzero;
1250 extern long hammer2_iod_file_wdedup;
1251 extern long hammer2_iod_meta_write;
1252 extern long hammer2_iod_indr_write;
1253 extern long hammer2_iod_fmap_write;
1254 extern long hammer2_iod_volu_write;
1255 extern long hammer2_ioa_file_read;
1256 extern long hammer2_ioa_meta_read;
1257 extern long hammer2_ioa_indr_read;
1258 extern long hammer2_ioa_fmap_read;
1259 extern long hammer2_ioa_volu_read;
1260 extern long hammer2_ioa_file_write;
1261 extern long hammer2_ioa_meta_write;
1262 extern long hammer2_ioa_indr_write;
1263 extern long hammer2_ioa_fmap_write;
1264 extern long hammer2_ioa_volu_write;
1266 extern struct objcache *cache_buffer_read;
1267 extern struct objcache *cache_buffer_write;
1268 extern struct objcache *cache_xops;
1273 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
1274 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
1276 int hammer2_signal_check(time_t *timep);
1277 const char *hammer2_error_str(int error);
1279 void hammer2_inode_lock(hammer2_inode_t *ip, int how);
1280 void hammer2_inode_unlock(hammer2_inode_t *ip);
1281 hammer2_chain_t *hammer2_inode_chain(hammer2_inode_t *ip, int clindex, int how);
1282 hammer2_chain_t *hammer2_inode_chain_and_parent(hammer2_inode_t *ip,
1283 int clindex, hammer2_chain_t **parentp, int how);
1284 hammer2_mtx_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
1285 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip,
1286 hammer2_mtx_state_t ostate);
1287 int hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
1288 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int);
1290 void hammer2_dev_exlock(hammer2_dev_t *hmp);
1291 void hammer2_dev_shlock(hammer2_dev_t *hmp);
1292 void hammer2_dev_unlock(hammer2_dev_t *hmp);
1294 int hammer2_get_dtype(const hammer2_inode_data_t *ipdata);
1295 int hammer2_get_vtype(uint8_t type);
1296 u_int8_t hammer2_get_obj_type(enum vtype vtype);
1297 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1298 u_int64_t hammer2_timespec_to_time(const struct timespec *ts);
1299 u_int32_t hammer2_to_unix_xid(const uuid_t *uuid);
1300 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1301 void hammer2_trans_manage_init(hammer2_pfs_t *pmp);
1303 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
1304 int hammer2_getradix(size_t bytes);
1306 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
1307 hammer2_key_t *lbasep, hammer2_key_t *leofp);
1308 int hammer2_calc_physical(hammer2_inode_t *ip, hammer2_key_t lbase);
1309 void hammer2_update_time(uint64_t *timep);
1310 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
1315 struct vnode *hammer2_igetv(hammer2_inode_t *ip, int *errorp);
1316 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfs_t *pmp,
1317 hammer2_tid_t inum);
1318 hammer2_inode_t *hammer2_inode_get(hammer2_pfs_t *pmp, hammer2_inode_t *dip,
1319 hammer2_cluster_t *cluster, int idx);
1320 void hammer2_inode_free(hammer2_inode_t *ip);
1321 void hammer2_inode_ref(hammer2_inode_t *ip);
1322 void hammer2_inode_drop(hammer2_inode_t *ip);
1323 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
1324 hammer2_cluster_t *cluster);
1325 void hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1327 void hammer2_inode_modify(hammer2_inode_t *ip);
1328 void hammer2_inode_run_sideq(hammer2_pfs_t *pmp);
1330 hammer2_inode_t *hammer2_inode_create(hammer2_inode_t *dip,
1331 struct vattr *vap, struct ucred *cred,
1332 const uint8_t *name, size_t name_len, hammer2_key_t lhc,
1333 hammer2_key_t inum, uint8_t type, uint8_t target_type,
1334 int flags, int *errorp);
1335 int hammer2_inode_connect(hammer2_inode_t *dip, hammer2_inode_t *ip,
1336 const char *name, size_t name_len,
1338 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
1339 hammer2_inode_t *tdip, int *errorp, int ishardlink);
1340 void hammer2_inode_chain_sync(hammer2_inode_t *ip);
1341 int hammer2_inode_unlink_finisher(hammer2_inode_t *ip, int isopen);
1342 void hammer2_inode_install_hidden(hammer2_pfs_t *pmp);
1347 void hammer2_voldata_lock(hammer2_dev_t *hmp);
1348 void hammer2_voldata_unlock(hammer2_dev_t *hmp);
1349 void hammer2_voldata_modify(hammer2_dev_t *hmp);
1350 hammer2_chain_t *hammer2_chain_alloc(hammer2_dev_t *hmp,
1352 hammer2_blockref_t *bref);
1353 void hammer2_chain_core_init(hammer2_chain_t *chain);
1354 void hammer2_chain_ref(hammer2_chain_t *chain);
1355 void hammer2_chain_drop(hammer2_chain_t *chain);
1356 void hammer2_chain_lock(hammer2_chain_t *chain, int how);
1357 void hammer2_chain_lock_downgrade(hammer2_chain_t *chain);
1358 void hammer2_chain_push_shared_lock(hammer2_chain_t *chain);
1359 void hammer2_chain_pull_shared_lock(hammer2_chain_t *chain);
1360 void hammer2_chain_load_data(hammer2_chain_t *chain);
1361 const hammer2_media_data_t *hammer2_chain_rdata(hammer2_chain_t *chain);
1362 hammer2_media_data_t *hammer2_chain_wdata(hammer2_chain_t *chain);
1363 int hammer2_chain_snapshot(hammer2_chain_t *chain, hammer2_ioc_pfs_t *pmp,
1364 hammer2_tid_t mtid);
1366 int hammer2_chain_hardlink_find(hammer2_inode_t *dip,
1367 hammer2_chain_t **parentp,
1368 hammer2_chain_t **chainp,
1370 void hammer2_chain_modify(hammer2_chain_t *chain, hammer2_tid_t mtid,
1371 hammer2_off_t dedup_off, int flags);
1372 void hammer2_chain_modify_ip(hammer2_inode_t *ip, hammer2_chain_t *chain,
1373 hammer2_tid_t mtid, int flags);
1374 void hammer2_chain_resize(hammer2_inode_t *ip, hammer2_chain_t *parent,
1375 hammer2_chain_t *chain,
1376 hammer2_tid_t mtid, hammer2_off_t dedup_off,
1377 int nradix, int flags);
1378 void hammer2_chain_unlock(hammer2_chain_t *chain);
1379 void hammer2_chain_wait(hammer2_chain_t *chain);
1380 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
1381 hammer2_blockref_t *bref);
1382 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
1383 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
1384 hammer2_chain_t *hammer2_chain_getparent(hammer2_chain_t **parentp, int how);
1385 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
1386 hammer2_key_t *key_nextp,
1387 hammer2_key_t key_beg, hammer2_key_t key_end,
1388 int *cache_indexp, int flags);
1389 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
1390 hammer2_chain_t *chain,
1391 hammer2_key_t *key_nextp,
1392 hammer2_key_t key_beg, hammer2_key_t key_end,
1393 int *cache_indexp, int flags);
1394 hammer2_blockref_t *hammer2_chain_scan(hammer2_chain_t *parent,
1395 hammer2_chain_t **chainp,
1396 hammer2_blockref_t *bref,
1397 int *firstp, int *cache_indexp, int flags);
1399 int hammer2_chain_create(hammer2_chain_t **parentp,
1400 hammer2_chain_t **chainp, hammer2_pfs_t *pmp,
1401 hammer2_key_t key, int keybits,
1402 int type, size_t bytes, hammer2_tid_t mtid,
1403 hammer2_off_t dedup_off, int flags);
1404 void hammer2_chain_rename(hammer2_blockref_t *bref,
1405 hammer2_chain_t **parentp,
1406 hammer2_chain_t *chain,
1407 hammer2_tid_t mtid, int flags);
1408 void hammer2_chain_delete(hammer2_chain_t *parent, hammer2_chain_t *chain,
1409 hammer2_tid_t mtid, int flags);
1410 void hammer2_chain_setflush(hammer2_chain_t *chain);
1411 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
1412 hammer2_blockref_t *base, int count);
1413 hammer2_chain_t *hammer2_chain_bulksnap(hammer2_chain_t *chain);
1414 void hammer2_chain_bulkdrop(hammer2_chain_t *copy);
1416 void hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata);
1417 int hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata);
1420 void hammer2_pfs_memory_wait(hammer2_pfs_t *pmp);
1421 void hammer2_pfs_memory_inc(hammer2_pfs_t *pmp);
1422 void hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp);
1424 void hammer2_base_delete(hammer2_chain_t *chain,
1425 hammer2_blockref_t *base, int count,
1426 int *cache_indexp, hammer2_chain_t *child);
1427 void hammer2_base_insert(hammer2_chain_t *chain,
1428 hammer2_blockref_t *base, int count,
1429 int *cache_indexp, hammer2_chain_t *child);
1434 void hammer2_flush(hammer2_chain_t *chain, int istop);
1435 void hammer2_flush_quick(hammer2_dev_t *hmp);
1436 void hammer2_delayed_flush(hammer2_chain_t *chain);
1441 void hammer2_trans_init(hammer2_pfs_t *pmp, uint32_t flags);
1442 hammer2_tid_t hammer2_trans_sub(hammer2_pfs_t *pmp);
1443 void hammer2_trans_clear_preflush(hammer2_pfs_t *pmp);
1444 void hammer2_trans_done(hammer2_pfs_t *pmp);
1445 hammer2_tid_t hammer2_trans_newinum(hammer2_pfs_t *pmp);
1446 void hammer2_trans_assert_strategy(hammer2_pfs_t *pmp);
1451 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
1452 int fflag, struct ucred *cred);
1457 void hammer2_io_putblk(hammer2_io_t **diop);
1458 void hammer2_io_cleanup(hammer2_dev_t *hmp, struct hammer2_io_tree *tree);
1459 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
1460 hammer2_io_t *hammer2_io_getquick(hammer2_dev_t *hmp, off_t lbase, int lsize);
1461 void hammer2_io_getblk(hammer2_dev_t *hmp, off_t lbase, int lsize,
1462 hammer2_iocb_t *iocb);
1463 void hammer2_io_complete(hammer2_iocb_t *iocb);
1464 void hammer2_io_callback(struct bio *bio);
1465 void hammer2_iocb_wait(hammer2_iocb_t *iocb);
1466 int hammer2_io_new(hammer2_dev_t *hmp, off_t lbase, int lsize,
1467 hammer2_io_t **diop);
1468 int hammer2_io_newnz(hammer2_dev_t *hmp, off_t lbase, int lsize,
1469 hammer2_io_t **diop);
1470 int hammer2_io_newq(hammer2_dev_t *hmp, off_t lbase, int lsize,
1471 hammer2_io_t **diop);
1472 int hammer2_io_bread(hammer2_dev_t *hmp, off_t lbase, int lsize,
1473 hammer2_io_t **diop);
1474 void hammer2_io_bawrite(hammer2_io_t **diop);
1475 void hammer2_io_bdwrite(hammer2_io_t **diop);
1476 int hammer2_io_bwrite(hammer2_io_t **diop);
1477 int hammer2_io_isdirty(hammer2_io_t *dio);
1478 void hammer2_io_setdirty(hammer2_io_t *dio);
1479 void hammer2_io_setinval(hammer2_io_t *dio, u_int bytes);
1480 void hammer2_io_brelse(hammer2_io_t **diop);
1481 void hammer2_io_bqrelse(hammer2_io_t **diop);
1486 void hammer2_thr_create(hammer2_thread_t *thr, hammer2_pfs_t *pmp,
1487 const char *id, int clindex, int repidx,
1488 void (*func)(void *arg));
1489 void hammer2_thr_delete(hammer2_thread_t *thr);
1490 void hammer2_thr_remaster(hammer2_thread_t *thr);
1491 void hammer2_thr_freeze_async(hammer2_thread_t *thr);
1492 void hammer2_thr_freeze(hammer2_thread_t *thr);
1493 void hammer2_thr_unfreeze(hammer2_thread_t *thr);
1494 int hammer2_thr_break(hammer2_thread_t *thr);
1495 void hammer2_primary_xops_thread(void *arg);
1498 * hammer2_thread.c (XOP API)
1500 void hammer2_xop_group_init(hammer2_pfs_t *pmp, hammer2_xop_group_t *xgrp);
1501 void *hammer2_xop_alloc(hammer2_inode_t *ip, int flags);
1502 void hammer2_xop_setname(hammer2_xop_head_t *xop,
1503 const char *name, size_t name_len);
1504 void hammer2_xop_setname2(hammer2_xop_head_t *xop,
1505 const char *name, size_t name_len);
1506 void hammer2_xop_setip2(hammer2_xop_head_t *xop, hammer2_inode_t *ip2);
1507 void hammer2_xop_setip3(hammer2_xop_head_t *xop, hammer2_inode_t *ip3);
1508 void hammer2_xop_reinit(hammer2_xop_head_t *xop);
1509 void hammer2_xop_helper_create(hammer2_pfs_t *pmp);
1510 void hammer2_xop_helper_cleanup(hammer2_pfs_t *pmp);
1511 void hammer2_xop_start(hammer2_xop_head_t *xop, hammer2_xop_func_t func);
1512 void hammer2_xop_start_except(hammer2_xop_head_t *xop, hammer2_xop_func_t func,
1514 int hammer2_xop_collect(hammer2_xop_head_t *xop, int flags);
1515 void hammer2_xop_retire(hammer2_xop_head_t *xop, uint32_t mask);
1516 int hammer2_xop_active(hammer2_xop_head_t *xop);
1517 int hammer2_xop_feed(hammer2_xop_head_t *xop, hammer2_chain_t *chain,
1518 int clindex, int error);
1523 void hammer2_primary_sync_thread(void *arg);
1526 * XOP backends in hammer2_xops.c, primarily for VNOPS. Other XOP backends
1527 * may be integrated into other source files.
1529 void hammer2_xop_ipcluster(hammer2_xop_t *xop, int clidx);
1530 void hammer2_xop_readdir(hammer2_xop_t *xop, int clidx);
1531 void hammer2_xop_nresolve(hammer2_xop_t *xop, int clidx);
1532 void hammer2_xop_unlink(hammer2_xop_t *xop, int clidx);
1533 void hammer2_xop_nrename(hammer2_xop_t *xop, int clidx);
1534 void hammer2_xop_nlink(hammer2_xop_t *xop, int clidx);
1535 void hammer2_xop_scanlhc(hammer2_xop_t *xop, int clidx);
1536 void hammer2_xop_scanall(hammer2_xop_t *xop, int clidx);
1537 void hammer2_xop_lookup(hammer2_xop_t *xop, int clidx);
1538 void hammer2_inode_xop_create(hammer2_xop_t *xop, int clidx);
1539 void hammer2_inode_xop_destroy(hammer2_xop_t *xop, int clidx);
1540 void hammer2_inode_xop_chain_sync(hammer2_xop_t *xop, int clidx);
1541 void hammer2_inode_xop_unlinkall(hammer2_xop_t *xop, int clidx);
1542 void hammer2_inode_xop_connect(hammer2_xop_t *xop, int clidx);
1543 void hammer2_inode_xop_flush(hammer2_xop_t *xop, int clidx);
1548 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
1549 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
1554 void hammer2_volconf_update(hammer2_dev_t *hmp, int index);
1555 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
1556 int hammer2_vfs_sync(struct mount *mp, int waitflags);
1557 hammer2_pfs_t *hammer2_pfsalloc(hammer2_chain_t *chain,
1558 const hammer2_inode_data_t *ripdata,
1559 hammer2_tid_t modify_tid);
1561 void hammer2_lwinprog_ref(hammer2_pfs_t *pmp);
1562 void hammer2_lwinprog_drop(hammer2_pfs_t *pmp);
1563 void hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int pipe);
1568 int hammer2_freemap_alloc(hammer2_chain_t *chain, size_t bytes);
1569 void hammer2_freemap_adjust(hammer2_dev_t *hmp,
1570 hammer2_blockref_t *bref, int how);
1575 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
1576 const hammer2_media_data_t *hammer2_cluster_rdata(hammer2_cluster_t *cluster);
1577 hammer2_media_data_t *hammer2_cluster_wdata(hammer2_cluster_t *cluster);
1578 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
1579 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
1580 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfs_t *pmp,
1581 hammer2_blockref_t *bref);
1582 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
1583 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1584 void hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1585 int hammer2_cluster_check(hammer2_cluster_t *cluster, hammer2_key_t lokey,
1587 void hammer2_cluster_resolve(hammer2_cluster_t *cluster);
1588 void hammer2_cluster_forcegood(hammer2_cluster_t *cluster);
1589 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster);
1590 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1592 int hammer2_bulkfree_pass(hammer2_dev_t *hmp,
1593 struct hammer2_ioc_bulkfree *bfi);
1598 void hammer2_iocom_init(hammer2_dev_t *hmp);
1599 void hammer2_iocom_uninit(hammer2_dev_t *hmp);
1600 void hammer2_cluster_reconnect(hammer2_dev_t *hmp, struct file *fp);
1603 * hammer2_strategy.c
1605 int hammer2_vop_strategy(struct vop_strategy_args *ap);
1606 int hammer2_vop_bmap(struct vop_bmap_args *ap);
1607 void hammer2_write_thread(void *arg);
1608 void hammer2_bioq_sync(hammer2_pfs_t *pmp);
1609 void hammer2_dedup_clear(hammer2_dev_t *hmp);
1611 #endif /* !_KERNEL */
1612 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */