2 * Copyright (c) 2011-2016 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_xxhash.h"
94 #include "hammer2_disk.h"
95 #include "hammer2_mount.h"
96 #include "hammer2_ioctl.h"
100 struct hammer2_chain;
101 struct hammer2_cluster;
102 struct hammer2_inode;
106 struct hammer2_state;
108 struct hammer2_thread;
112 * Mutex and lock shims. Hammer2 requires support for asynchronous and
113 * abortable locks, and both exclusive and shared spinlocks. Normal
114 * synchronous non-abortable locks can be substituted for spinlocks.
116 typedef mtx_t hammer2_mtx_t;
117 typedef mtx_link_t hammer2_mtx_link_t;
118 typedef mtx_state_t hammer2_mtx_state_t;
120 typedef struct spinlock hammer2_spin_t;
122 #define hammer2_mtx_ex mtx_lock_ex_quick
123 #define hammer2_mtx_sh mtx_lock_sh_quick
124 #define hammer2_mtx_sh_again mtx_lock_sh_again
125 #define hammer2_mtx_unlock mtx_unlock
126 #define hammer2_mtx_downgrade mtx_downgrade
127 #define hammer2_mtx_owned mtx_owned
128 #define hammer2_mtx_init mtx_init
129 #define hammer2_mtx_temp_release mtx_lock_temp_release
130 #define hammer2_mtx_temp_restore mtx_lock_temp_restore
131 #define hammer2_mtx_refs mtx_lockrefs
133 #define hammer2_spin_init spin_init
134 #define hammer2_spin_sh spin_lock_shared
135 #define hammer2_spin_ex spin_lock
136 #define hammer2_spin_unsh spin_unlock_shared
137 #define hammer2_spin_unex spin_unlock
139 TAILQ_HEAD(hammer2_xop_list, hammer2_xop_head);
140 TAILQ_HEAD(hammer2_chain_list, hammer2_chain);
142 typedef struct hammer2_xop_list hammer2_xop_list_t;
146 * General lock support
150 hammer2_mtx_upgrade(hammer2_mtx_t *mtx)
154 if (mtx_islocked_ex(mtx)) {
158 mtx_lock_ex_quick(mtx);
165 * The xid tracks internal transactional updates.
167 * XXX fix-me, really needs to be 64-bits
169 typedef uint32_t hammer2_xid_t;
171 #define HAMMER2_XID_MIN 0x00000000U
172 #define HAMMER2_XID_MAX 0x7FFFFFFFU
175 * The chain structure tracks a portion of the media topology from the
176 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
177 * data blocks, and freemap nodes and leafs.
179 * The chain structure utilizes a simple singly-homed topology and the
180 * chain's in-memory topology will move around as the chains do, due mainly
181 * to renames and indirect block creation.
183 * Block Table Updates
185 * Block table updates for insertions and updates are delayed until the
186 * flush. This allows us to avoid having to modify the parent chain
187 * all the way to the root.
189 * Block table deletions are performed immediately (modifying the parent
190 * in the process) because the flush code uses the chain structure to
191 * track delayed updates and the chain will be (likely) gone or moved to
192 * another location in the topology after a deletion.
194 * A prior iteration of the code tried to keep the relationship intact
195 * on deletes by doing a delete-duplicate operation on the chain, but
196 * it added way too much complexity to the codebase.
198 * Flush Synchronization
200 * The flush code must flush modified chains bottom-up. Because chain
201 * structures can shift around and are NOT topologically stable,
202 * modified chains are independently indexed for the flush. As the flush
203 * runs it modifies (or further modifies) and updates the parents,
204 * propagating the flush all the way to the volume root.
206 * Modifying front-end operations can occur during a flush but will block
207 * in two cases: (1) when the front-end tries to operate on the inode
208 * currently in the midst of being flushed and (2) if the front-end
209 * crosses an inode currently being flushed (such as during a rename).
210 * So, for example, if you rename directory "x" to "a/b/c/d/e/f/g/x" and
211 * the flusher is currently working on "a/b/c", the rename will block
212 * temporarily in order to ensure that "x" exists in one place or the
215 * Meta-data statistics are updated by the flusher. The front-end will
216 * make estimates but meta-data must be fully synchronized only during a
217 * flush in order to ensure that it remains correct across a crash.
219 * Multiple flush synchronizations can theoretically be in-flight at the
220 * same time but the implementation is not coded to handle the case and
221 * currently serializes them.
225 * Snapshots currently require the subdirectory tree being snapshotted
226 * to be flushed. The snapshot then creates a new super-root inode which
227 * copies the flushed blockdata of the directory or file that was
232 * - Note that the radix tree runs in powers of 2 only so sub-trees
233 * cannot straddle edges.
235 RB_HEAD(hammer2_chain_tree, hammer2_chain);
236 TAILQ_HEAD(h2_flush_list, hammer2_chain);
237 TAILQ_HEAD(h2_core_list, hammer2_chain);
238 TAILQ_HEAD(h2_iocb_list, hammer2_iocb);
240 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
241 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
244 * Core topology for chain (embedded in chain). Protected by a spinlock.
246 struct hammer2_chain_core {
248 struct hammer2_chain_tree rbtree; /* sub-chains */
249 int live_zero; /* blockref array opt */
250 u_int live_count; /* live (not deleted) chains in tree */
251 u_int chain_count; /* live + deleted chains under core */
252 int generation; /* generation number (inserts only) */
255 typedef struct hammer2_chain_core hammer2_chain_core_t;
257 RB_HEAD(hammer2_io_tree, hammer2_io);
260 * IOCB - IO callback (into chain, cluster, or manual request)
262 struct hammer2_iocb {
263 TAILQ_ENTRY(hammer2_iocb) entry;
264 void (*callback)(struct hammer2_iocb *iocb);
265 struct hammer2_io *dio;
266 struct hammer2_cluster *cluster;
267 struct hammer2_chain *chain;
276 typedef struct hammer2_iocb hammer2_iocb_t;
278 #define HAMMER2_IOCB_INTERLOCK 0x00000001
279 #define HAMMER2_IOCB_ONQ 0x00000002
280 #define HAMMER2_IOCB_DONE 0x00000004
281 #define HAMMER2_IOCB_INPROG 0x00000008
282 #define HAMMER2_IOCB_UNUSED10 0x00000010
283 #define HAMMER2_IOCB_QUICK 0x00010000
284 #define HAMMER2_IOCB_ZERO 0x00020000
285 #define HAMMER2_IOCB_READ 0x00040000
286 #define HAMMER2_IOCB_WAKEUP 0x00080000
289 * DIO - Management structure wrapping system buffer cache.
291 * Used for multiple purposes including concurrent management
292 * if small requests by chains into larger DIOs.
295 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
296 struct h2_iocb_list iocbq;
297 struct spinlock spin;
298 struct hammer2_dev *hmp;
300 uint64_t crc_good_mask;
304 int act; /* activity */
305 int btype; /* approximate BREF_TYPE_* */
308 typedef struct hammer2_io hammer2_io_t;
310 #define HAMMER2_DIO_INPROG 0x80000000 /* bio in progress */
311 #define HAMMER2_DIO_GOOD 0x40000000 /* dio->bp is stable */
312 #define HAMMER2_DIO_WAITING 0x20000000 /* wait for inprog clr */
313 #define HAMMER2_DIO_DIRTY 0x10000000 /* flush on last drop */
315 #define HAMMER2_DIO_MASK 0x0FFFFFFF
318 * Primary chain structure keeps track of the topology in-memory.
320 struct hammer2_chain {
322 hammer2_chain_core_t core;
323 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
324 hammer2_blockref_t bref;
325 struct hammer2_chain *parent;
326 struct hammer2_state *state; /* if active cache msg */
327 struct hammer2_dev *hmp;
328 struct hammer2_pfs *pmp; /* A PFS or super-root (spmp) */
330 hammer2_io_t *dio; /* physical data buffer */
331 u_int bytes; /* physical data size */
335 int error; /* on-lock data error state */
338 hammer2_media_data_t *data; /* data pointer shortcut */
339 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
340 TAILQ_ENTRY(hammer2_chain) lru_node; /* 0-refs LRU */
343 typedef struct hammer2_chain hammer2_chain_t;
345 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
346 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
349 * Special notes on flags:
351 * INITIAL - This flag allows a chain to be created and for storage to
352 * be allocated without having to immediately instantiate the
353 * related buffer. The data is assumed to be all-zeros. It
354 * is primarily used for indirect blocks.
356 * MODIFIED - The chain's media data has been modified. Prevents chain
357 * free on lastdrop if still in the topology.
359 * UPDATE - Chain might not be modified but parent blocktable needs
360 * an update. Prevents chain free on lastdrop if still in
363 * FICTITIOUS - Faked chain as a placeholder for an error condition. This
364 * chain is unsuitable for I/O.
366 * BMAPPED - Indicates that the chain is present in the parent blockmap.
368 * BMAPUPD - Indicates that the chain is present but needs to be updated
369 * in the parent blockmap.
371 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
372 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
373 #define HAMMER2_CHAIN_DESTROY 0x00000004
374 #define HAMMER2_CHAIN_DEDUP 0x00000008 /* used as dedup src */
375 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
376 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
377 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
378 #define HAMMER2_CHAIN_DEFERRED 0x00000080 /* flush depth defer */
379 #define HAMMER2_CHAIN_TESTEDGOOD 0x00000100 /* crc tested good */
380 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
381 #define HAMMER2_CHAIN_FICTITIOUS 0x00000400 /* unsuitable for I/O */
382 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
383 #define HAMMER2_CHAIN_DELAYED 0x00001000 /* delayed flush */
384 #define HAMMER2_CHAIN_COUNTEDBREFS 0x00002000 /* block table stats */
385 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
386 #define HAMMER2_CHAIN_ONLRU 0x00008000 /* on LRU list */
387 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
388 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
389 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
390 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
391 #define HAMMER2_CHAIN_IOINPROG 0x00100000 /* I/O interlock */
392 #define HAMMER2_CHAIN_IOSIGNAL 0x00200000 /* I/O interlock */
393 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
395 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
396 HAMMER2_CHAIN_UPDATE | \
397 HAMMER2_CHAIN_ONFLUSH)
400 * Hammer2 error codes, used by chain->error and cluster->error. The error
401 * code is typically set on-lock unless no I/O was requested, and set on
402 * I/O otherwise. If set for a cluster it generally means that the cluster
403 * code could not find a valid copy to present.
405 * IO - An I/O error occurred
406 * CHECK - I/O succeeded but did not match the check code
407 * INCOMPLETE - A cluster is not complete enough to use, or
408 * a chain cannot be loaded because its parent has an error.
410 * NOTE: API allows callers to check zero/non-zero to determine if an error
413 * NOTE: Chain's data field is usually NULL on an IO error but not necessarily
414 * NULL on other errors. Check chain->error, not chain->data.
416 #define HAMMER2_ERROR_NONE 0
417 #define HAMMER2_ERROR_IO 1 /* device I/O error */
418 #define HAMMER2_ERROR_CHECK 2 /* check code mismatch */
419 #define HAMMER2_ERROR_INCOMPLETE 3 /* incomplete cluster */
420 #define HAMMER2_ERROR_DEPTH 4 /* temporary depth limit */
423 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
426 * NOLOCK - Input and output chains are referenced only and not
427 * locked. Output chain might be temporarily locked
430 * NODATA - Asks that the chain->data not be resolved in order
433 * NODIRECT - Prevents a lookup of offset 0 in an inode from returning
434 * the inode itself if the inode is in DIRECTDATA mode
435 * (i.e. file is <= 512 bytes). Used by the synchronization
436 * code to prevent confusion.
438 * SHARED - The input chain is expected to be locked shared,
439 * and the output chain is locked shared.
441 * MATCHIND - Allows an indirect block / freemap node to be returned
442 * when the passed key range matches the radix. Remember
443 * that key_end is inclusive (e.g. {0x000,0xFFF},
444 * not {0x000,0x1000}).
446 * (Cannot be used for remote or cluster ops).
448 * ALLNODES - Allows NULL focus.
450 * ALWAYS - Always resolve the data. If ALWAYS and NODATA are both
451 * missing, bulk file data is not resolved but inodes and
452 * other meta-data will.
454 * NOUNLOCK - Used by hammer2_chain_next() to leave the lock on
455 * the input chain intact. The chain is still dropped.
456 * This allows the caller to add a reference to the chain
457 * and retain it in a locked state (used by the
458 * XOP/feed/collect code).
460 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
461 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
462 #define HAMMER2_LOOKUP_NODIRECT 0x00000004 /* no offset=0 DD */
463 #define HAMMER2_LOOKUP_SHARED 0x00000100
464 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
465 #define HAMMER2_LOOKUP_ALLNODES 0x00000400 /* allow NULL focus */
466 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
467 #define HAMMER2_LOOKUP_NOUNLOCK 0x00001000 /* leave lock intact */
470 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
472 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
473 * blocks in the INITIAL-create state.
475 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
476 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
477 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
480 * Flags passed to hammer2_chain_lock()
482 * NOTE: RDONLY is set to optimize cluster operations when *no* modifications
483 * will be made to either the cluster being locked or any underlying
484 * cluster. It allows the cluster to lock and access data for a subset
485 * of available nodes instead of all available nodes.
487 #define HAMMER2_RESOLVE_NEVER 1
488 #define HAMMER2_RESOLVE_MAYBE 2
489 #define HAMMER2_RESOLVE_ALWAYS 3
490 #define HAMMER2_RESOLVE_MASK 0x0F
492 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
493 #define HAMMER2_RESOLVE_UNUSED20 0x20
494 #define HAMMER2_RESOLVE_RDONLY 0x40 /* higher level op flag */
497 * Flags passed to hammer2_chain_delete()
499 #define HAMMER2_DELETE_PERMANENT 0x0001
502 * Flags passed to hammer2_chain_insert() or hammer2_chain_rename()
504 #define HAMMER2_INSERT_PFSROOT 0x0004
507 * Flags passed to hammer2_chain_delete_duplicate()
509 #define HAMMER2_DELDUP_RECORE 0x0001
512 * Cluster different types of storage together for allocations
514 #define HAMMER2_FREECACHE_INODE 0
515 #define HAMMER2_FREECACHE_INDIR 1
516 #define HAMMER2_FREECACHE_DATA 2
517 #define HAMMER2_FREECACHE_UNUSED3 3
518 #define HAMMER2_FREECACHE_TYPES 4
521 * hammer2_freemap_alloc() block preference
523 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
526 * BMAP read-ahead maximum parameters
528 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
529 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
532 * hammer2_freemap_adjust()
534 #define HAMMER2_FREEMAP_DORECOVER 1
535 #define HAMMER2_FREEMAP_DOMAYFREE 2
536 #define HAMMER2_FREEMAP_DOREALFREE 3
539 * HAMMER2 cluster - A set of chains representing the same entity.
541 * hammer2_cluster typically represents a temporary set of representitive
542 * chains. The one exception is that a hammer2_cluster is embedded in
543 * hammer2_inode. This embedded cluster is ONLY used to track the
544 * representitive chains and cannot be directly locked.
546 * A cluster is usually temporary (and thus per-thread) for locking purposes,
547 * allowing us to embed the asynchronous storage required for cluster
548 * operations in the cluster itself and adjust the state and status without
549 * having to worry too much about SMP issues.
551 * The exception is the cluster embedded in the hammer2_inode structure.
552 * This is used to cache the cluster state on an inode-by-inode basis.
553 * Individual hammer2_chain structures not incorporated into clusters might
554 * also stick around to cache miscellanious elements.
556 * Because the cluster is a 'working copy' and is usually subject to cluster
557 * quorum rules, it is quite possible for us to end up with an insufficient
558 * number of live chains to execute an operation. If an insufficient number
559 * of chains remain in a working copy, the operation may have to be
560 * downgraded, retried, stall until the requisit number of chains are
561 * available, or possibly even error out depending on the mount type.
563 * A cluster's focus is set when it is locked. The focus can only be set
564 * to a chain still part of the synchronized set.
566 #define HAMMER2_MAXCLUSTER 8
567 #define HAMMER2_XOPMASK_CLUSTER ((1U << HAMMER2_MAXCLUSTER) - 1)
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
573 #define HAMMER2_XOPMASK_FIFOW 0x40000000U
575 #define HAMMER2_XOPMASK_ALLDONE (HAMMER2_XOPMASK_VOP | HAMMER2_XOPMASK_CLUSTER)
577 #define HAMMER2_SPECTHREADS 1 /* sync */
579 struct hammer2_cluster_item {
580 hammer2_chain_t *chain;
586 typedef struct hammer2_cluster_item hammer2_cluster_item_t;
589 * INVALID - Invalid for focus, i.e. not part of synchronized set.
590 * Once set, this bit is sticky across operations.
592 * FEMOD - Indicates that front-end modifying operations can
593 * mess with this entry and MODSYNC will copy also
596 #define HAMMER2_CITEM_INVALID 0x00000001
597 #define HAMMER2_CITEM_FEMOD 0x00000002
598 #define HAMMER2_CITEM_NULL 0x00000004
600 struct hammer2_cluster {
601 int refs; /* track for deallocation */
603 struct hammer2_pfs *pmp;
606 int error; /* error code valid on lock */
609 hammer2_chain_t *focus; /* current focus (or mod) */
610 hammer2_cluster_item_t array[HAMMER2_MAXCLUSTER];
613 typedef struct hammer2_cluster hammer2_cluster_t;
616 * WRHARD - Hard mounts can write fully synchronized
617 * RDHARD - Hard mounts can read fully synchronized
618 * UNHARD - Unsynchronized masters present
619 * NOHARD - No masters visible
620 * WRSOFT - Soft mounts can write to at least the SOFT_MASTER
621 * RDSOFT - Soft mounts can read from at least a SOFT_SLAVE
622 * UNSOFT - Unsynchronized slaves present
623 * NOSOFT - No slaves visible
624 * RDSLAVE - slaves are accessible (possibly unsynchronized or remote).
625 * MSYNCED - All masters are fully synchronized
626 * SSYNCED - All known local slaves are fully synchronized to masters
628 * All available masters are always incorporated. All PFSs belonging to a
629 * cluster (master, slave, copy, whatever) always try to synchronize the
630 * total number of known masters in the PFSs root inode.
632 * A cluster might have access to many slaves, copies, or caches, but we
633 * have a limited number of cluster slots. Any such elements which are
634 * directly mounted from block device(s) will always be incorporated. Note
635 * that SSYNCED only applies to such elements which are directly mounted,
636 * not to any remote slaves, copies, or caches that could be available. These
637 * bits are used to monitor and drive our synchronization threads.
639 * When asking the question 'is any data accessible at all', then a simple
640 * test against (RDHARD|RDSOFT|RDSLAVE) gives you the answer. If any of
641 * these bits are set the object can be read with certain caveats:
642 * RDHARD - no caveats. RDSOFT - authoritative but might not be synchronized.
643 * and RDSLAVE - not authoritative, has some data but it could be old or
646 * When both soft and hard mounts are available, data will be read and written
647 * via the soft mount only. But all might be in the cluster because
648 * background synchronization threads still need to do their work.
650 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode struct */
651 #define HAMMER2_CLUSTER_UNUSED2 0x00000002
652 #define HAMMER2_CLUSTER_LOCKED 0x00000004 /* cluster lks not recursive */
653 #define HAMMER2_CLUSTER_WRHARD 0x00000100 /* hard-mount can write */
654 #define HAMMER2_CLUSTER_RDHARD 0x00000200 /* hard-mount can read */
655 #define HAMMER2_CLUSTER_UNHARD 0x00000400 /* unsynchronized masters */
656 #define HAMMER2_CLUSTER_NOHARD 0x00000800 /* no masters visible */
657 #define HAMMER2_CLUSTER_WRSOFT 0x00001000 /* soft-mount can write */
658 #define HAMMER2_CLUSTER_RDSOFT 0x00002000 /* soft-mount can read */
659 #define HAMMER2_CLUSTER_UNSOFT 0x00004000 /* unsynchronized slaves */
660 #define HAMMER2_CLUSTER_NOSOFT 0x00008000 /* no slaves visible */
661 #define HAMMER2_CLUSTER_MSYNCED 0x00010000 /* all masters synchronized */
662 #define HAMMER2_CLUSTER_SSYNCED 0x00020000 /* known slaves synchronized */
664 #define HAMMER2_CLUSTER_ANYDATA ( HAMMER2_CLUSTER_RDHARD | \
665 HAMMER2_CLUSTER_RDSOFT | \
666 HAMMER2_CLUSTER_RDSLAVE)
668 #define HAMMER2_CLUSTER_RDOK ( HAMMER2_CLUSTER_RDHARD | \
669 HAMMER2_CLUSTER_RDSOFT)
671 #define HAMMER2_CLUSTER_WROK ( HAMMER2_CLUSTER_WRHARD | \
672 HAMMER2_CLUSTER_WRSOFT)
674 #define HAMMER2_CLUSTER_ZFLAGS ( HAMMER2_CLUSTER_WRHARD | \
675 HAMMER2_CLUSTER_RDHARD | \
676 HAMMER2_CLUSTER_WRSOFT | \
677 HAMMER2_CLUSTER_RDSOFT | \
678 HAMMER2_CLUSTER_MSYNCED | \
679 HAMMER2_CLUSTER_SSYNCED)
682 * Helper functions (cluster must be locked for flags to be valid).
686 hammer2_cluster_rdok(hammer2_cluster_t *cluster)
688 return (cluster->flags & HAMMER2_CLUSTER_RDOK);
693 hammer2_cluster_wrok(hammer2_cluster_t *cluster)
695 return (cluster->flags & HAMMER2_CLUSTER_WROK);
698 RB_HEAD(hammer2_inode_tree, hammer2_inode);
703 * NOTE: The inode-embedded cluster is never used directly for I/O (since
704 * it may be shared). Instead it will be replicated-in and synchronized
705 * back out if changed.
707 struct hammer2_inode {
708 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
709 hammer2_mtx_t lock; /* inode lock */
710 struct hammer2_pfs *pmp; /* PFS mount */
711 struct hammer2_inode *pip; /* parent inode */
713 struct spinlock cluster_spin; /* update cluster */
714 hammer2_cluster_t cluster;
715 struct lockf advlock;
717 u_int refs; /* +vpref, +flushref */
718 uint8_t comp_heuristic;
719 hammer2_inode_meta_t meta; /* copy of meta-data */
721 hammer2_cluster_t *cluster_cache;
724 typedef struct hammer2_inode hammer2_inode_t;
727 * MODIFIED - Inode is in a modified state, ip->meta may have changes.
728 * RESIZED - Inode truncated (any) or inode extended beyond
731 #define HAMMER2_INODE_MODIFIED 0x0001
732 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
733 #define HAMMER2_INODE_RENAME_INPROG 0x0004
734 #define HAMMER2_INODE_ONRBTREE 0x0008
735 #define HAMMER2_INODE_RESIZED 0x0010 /* requires inode_fsync */
736 #define HAMMER2_INODE_ISDELETED 0x0020 /* deleted, not in ihidden */
737 #define HAMMER2_INODE_ISUNLINKED 0x0040
738 #define HAMMER2_INODE_METAGOOD 0x0080 /* inode meta-data good */
739 #define HAMMER2_INODE_ONSIDEQ 0x0100 /* on side processing queue */
741 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
742 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
746 * inode-unlink side-structure
748 struct hammer2_inode_sideq {
749 TAILQ_ENTRY(hammer2_inode_sideq) entry;
752 TAILQ_HEAD(h2_sideq_list, hammer2_inode_sideq);
754 typedef struct hammer2_inode_sideq hammer2_inode_sideq_t;
757 * Transaction management sub-structure under hammer2_pfs
759 struct hammer2_trans {
764 typedef struct hammer2_trans hammer2_trans_t;
766 #define HAMMER2_TRANS_ISFLUSH 0x80000000 /* flush code */
767 #define HAMMER2_TRANS_BUFCACHE 0x40000000 /* bio strategy */
768 #define HAMMER2_TRANS_PREFLUSH 0x20000000 /* preflush state */
769 #define HAMMER2_TRANS_FPENDING 0x10000000 /* flush pending */
770 #define HAMMER2_TRANS_WAITING 0x08000000 /* someone waiting */
771 #define HAMMER2_TRANS_MASK 0x00FFFFFF /* count mask */
773 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
774 #define HAMMER2_FREEMAP_HEUR_TYPES 8
775 #define HAMMER2_FREEMAP_HEUR_SIZE (HAMMER2_FREEMAP_HEUR_NRADIX * \
776 HAMMER2_FREEMAP_HEUR_TYPES)
778 #define HAMMER2_DEDUP_HEUR_SIZE 65536
779 #define HAMMER2_DEDUP_HEUR_MASK (HAMMER2_DEDUP_HEUR_SIZE - 1)
781 #define HAMMER2_FLUSH_TOP 0x0001
782 #define HAMMER2_FLUSH_ALL 0x0002
786 * Hammer2 support thread element.
788 * Potentially many support threads can hang off of hammer2, primarily
789 * off the hammer2_pfs structure. Typically:
791 * td x Nodes A synchronization thread for each node.
792 * td x Nodes x workers Worker threads for frontend operations.
793 * td x 1 Bioq thread for logical buffer writes.
795 * In addition, the synchronization thread(s) associated with the
796 * super-root PFS (spmp) for a node is responsible for automatic bulkfree
799 struct hammer2_thread {
800 struct hammer2_pfs *pmp;
801 hammer2_xop_list_t *xopq; /* points into pmp->xopq[] */
805 int clindex; /* cluster element index */
809 typedef struct hammer2_thread hammer2_thread_t;
811 #define HAMMER2_THREAD_UNMOUNTING 0x0001 /* unmount request */
812 #define HAMMER2_THREAD_DEV 0x0002 /* related to dev, not pfs */
813 #define HAMMER2_THREAD_WAITING 0x0004 /* thread in idle tsleep */
814 #define HAMMER2_THREAD_REMASTER 0x0008 /* remaster request */
815 #define HAMMER2_THREAD_STOP 0x0010 /* exit request */
816 #define HAMMER2_THREAD_FREEZE 0x0020 /* force idle */
817 #define HAMMER2_THREAD_FROZEN 0x0040 /* thread is frozen */
818 #define HAMMER2_THREAD_XOPQ 0x0080 /* work pending */
819 #define HAMMER2_THREAD_STOPPED 0x0100 /* thread has stopped */
820 #define HAMMER2_THREAD_UNFREEZE 0x0200
821 #define HAMMER2_THREAD_CLIENTWAIT 0x0400
823 #define HAMMER2_THREAD_WAKEUP_MASK (HAMMER2_THREAD_UNMOUNTING | \
824 HAMMER2_THREAD_REMASTER | \
825 HAMMER2_THREAD_STOP | \
826 HAMMER2_THREAD_FREEZE | \
830 * Support structure for dedup heuristic.
832 struct hammer2_dedup {
833 hammer2_off_t data_off;
839 typedef struct hammer2_dedup hammer2_dedup_t;
842 * hammer2_xop - container for VOP/XOP operation (allocated, not on stack).
844 * This structure is used to distribute a VOP operation across multiple
845 * nodes. It provides a rendezvous for concurrent node execution and
846 * can be detached from the frontend operation to allow the frontend to
849 * This structure also sequences operations on up to three inodes.
851 typedef void (*hammer2_xop_func_t)(union hammer2_xop *xop, int clidx);
853 struct hammer2_xop_fifo {
854 TAILQ_ENTRY(hammer2_xop_head) entry;
855 hammer2_chain_t *array[HAMMER2_XOPFIFO];
856 int errors[HAMMER2_XOPFIFO];
862 typedef struct hammer2_xop_fifo hammer2_xop_fifo_t;
864 #define HAMMER2_XOP_FIFO_RUN 0x0001
865 #define HAMMER2_XOP_FIFO_STALL 0x0002
867 struct hammer2_xop_head {
868 hammer2_xop_func_t func;
870 struct hammer2_inode *ip1;
871 struct hammer2_inode *ip2;
872 struct hammer2_inode *ip3;
873 uint32_t check_counter;
879 hammer2_key_t collect_key;
884 hammer2_xop_fifo_t collect[HAMMER2_MAXCLUSTER];
885 hammer2_cluster_t cluster; /* help collections */
888 typedef struct hammer2_xop_head hammer2_xop_head_t;
890 struct hammer2_xop_ipcluster {
891 hammer2_xop_head_t head;
894 struct hammer2_xop_strategy {
895 hammer2_xop_head_t head;
902 struct hammer2_xop_readdir {
903 hammer2_xop_head_t head;
907 struct hammer2_xop_nresolve {
908 hammer2_xop_head_t head;
909 hammer2_key_t lhc; /* if name is NULL used lhc */
912 struct hammer2_xop_nlink {
913 hammer2_xop_head_t head;
917 struct hammer2_xop_unlink {
918 hammer2_xop_head_t head;
923 struct hammer2_xop_nrename {
924 hammer2_xop_head_t head;
929 struct hammer2_xop_scanlhc {
930 hammer2_xop_head_t head;
934 struct hammer2_xop_scanall {
935 hammer2_xop_head_t head;
936 hammer2_key_t key_beg; /* inclusive */
937 hammer2_key_t key_end; /* inclusive */
942 struct hammer2_xop_lookup {
943 hammer2_xop_head_t head;
947 struct hammer2_xop_create {
948 hammer2_xop_head_t head;
949 hammer2_inode_meta_t meta; /* initial metadata */
954 struct hammer2_xop_destroy {
955 hammer2_xop_head_t head;
958 struct hammer2_xop_fsync {
959 hammer2_xop_head_t head;
960 hammer2_inode_meta_t meta;
963 int clear_directdata;
966 struct hammer2_xop_unlinkall {
967 hammer2_xop_head_t head;
968 hammer2_key_t key_beg;
969 hammer2_key_t key_end;
972 struct hammer2_xop_connect {
973 hammer2_xop_head_t head;
977 struct hammer2_xop_flush {
978 hammer2_xop_head_t head;
981 typedef struct hammer2_xop_readdir hammer2_xop_readdir_t;
982 typedef struct hammer2_xop_nresolve hammer2_xop_nresolve_t;
983 typedef struct hammer2_xop_nlink hammer2_xop_nlink_t;
984 typedef struct hammer2_xop_unlink hammer2_xop_unlink_t;
985 typedef struct hammer2_xop_nrename hammer2_xop_nrename_t;
986 typedef struct hammer2_xop_ipcluster hammer2_xop_ipcluster_t;
987 typedef struct hammer2_xop_strategy hammer2_xop_strategy_t;
988 typedef struct hammer2_xop_create hammer2_xop_create_t;
989 typedef struct hammer2_xop_destroy hammer2_xop_destroy_t;
990 typedef struct hammer2_xop_fsync hammer2_xop_fsync_t;
991 typedef struct hammer2_xop_unlinkall hammer2_xop_unlinkall_t;
992 typedef struct hammer2_xop_scanlhc hammer2_xop_scanlhc_t;
993 typedef struct hammer2_xop_scanall hammer2_xop_scanall_t;
994 typedef struct hammer2_xop_lookup hammer2_xop_lookup_t;
995 typedef struct hammer2_xop_connect hammer2_xop_connect_t;
996 typedef struct hammer2_xop_flush hammer2_xop_flush_t;
999 hammer2_xop_head_t head;
1000 hammer2_xop_ipcluster_t xop_ipcluster;
1001 hammer2_xop_readdir_t xop_readdir;
1002 hammer2_xop_nresolve_t xop_nresolve;
1003 hammer2_xop_nlink_t xop_nlink;
1004 hammer2_xop_unlink_t xop_unlink;
1005 hammer2_xop_nrename_t xop_nrename;
1006 hammer2_xop_strategy_t xop_strategy;
1007 hammer2_xop_create_t xop_create;
1008 hammer2_xop_destroy_t xop_destroy;
1009 hammer2_xop_fsync_t xop_fsync;
1010 hammer2_xop_unlinkall_t xop_unlinkall;
1011 hammer2_xop_scanlhc_t xop_scanlhc;
1012 hammer2_xop_scanall_t xop_scanall;
1013 hammer2_xop_lookup_t xop_lookup;
1014 hammer2_xop_flush_t xop_flush;
1015 hammer2_xop_connect_t xop_connect;
1018 typedef union hammer2_xop hammer2_xop_t;
1021 * hammer2_xop_group - Manage XOP support threads.
1023 struct hammer2_xop_group {
1024 hammer2_thread_t thrs[HAMMER2_MAXCLUSTER];
1027 typedef struct hammer2_xop_group hammer2_xop_group_t;
1030 * flags to hammer2_xop_collect()
1032 #define HAMMER2_XOP_COLLECT_NOWAIT 0x00000001
1033 #define HAMMER2_XOP_COLLECT_WAITALL 0x00000002
1036 * flags to hammer2_xop_alloc()
1038 * MODIFYING - This is a modifying transaction, allocate a mtid.
1040 #define HAMMER2_XOP_MODIFYING 0x00000001
1041 #define HAMMER2_XOP_ITERATOR 0x00000002
1044 * Global (per partition) management structure, represents a hard block
1045 * device. Typically referenced by hammer2_chain structures when applicable.
1046 * Typically not used for network-managed elements.
1048 * Note that a single hammer2_dev can be indirectly tied to multiple system
1049 * mount points. There is no direct relationship. System mounts are
1050 * per-cluster-id, not per-block-device, and a single hard mount might contain
1051 * many PFSs and those PFSs might combine together in various ways to form
1052 * the set of available clusters.
1054 struct hammer2_dev {
1055 struct vnode *devvp; /* device vnode */
1056 int ronly; /* read-only mount */
1057 int mount_count; /* number of actively mounted PFSs */
1058 TAILQ_ENTRY(hammer2_dev) mntentry; /* hammer2_mntlist */
1060 struct malloc_type *mchain;
1063 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
1064 struct spinlock io_spin; /* iotree access */
1065 struct hammer2_io_tree iotree;
1067 hammer2_chain_t vchain; /* anchor chain (topology) */
1068 hammer2_chain_t fchain; /* anchor chain (freemap) */
1069 struct spinlock list_spin;
1070 struct h2_flush_list flushq; /* flush seeds */
1071 struct hammer2_pfs *spmp; /* super-root pmp for transactions */
1072 struct lock bulklk; /* bulkfree lock */
1073 struct lock vollk; /* lockmgr lock */
1074 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR_SIZE];
1075 hammer2_dedup_t heur_dedup[HAMMER2_DEDUP_HEUR_SIZE];
1076 int volhdrno; /* last volhdrno written */
1077 int hflags; /* HMNT2 flags applicable to device */
1078 char devrepname[64]; /* for kprintf */
1079 hammer2_volume_data_t voldata;
1080 hammer2_volume_data_t volsync; /* synchronized voldata */
1083 typedef struct hammer2_dev hammer2_dev_t;
1086 * Helper functions (cluster must be locked for flags to be valid).
1090 hammer2_chain_rdok(hammer2_chain_t *chain)
1092 return (chain->error == 0);
1097 hammer2_chain_wrok(hammer2_chain_t *chain)
1099 return (chain->error == 0 && chain->hmp->ronly == 0);
1103 * Per-cluster management structure. This structure will be tied to a
1104 * system mount point if the system is mounting the PFS, but is also used
1105 * to manage clusters encountered during the super-root scan or received
1106 * via LNK_SPANs that might not be mounted.
1108 * This structure is also used to represent the super-root that hangs off
1109 * of a hard mount point. The super-root is not really a cluster element.
1110 * In this case the spmp_hmp field will be non-NULL. It's just easier to do
1111 * this than to special case super-root manipulation in the hammer2_chain*
1112 * code as being only hammer2_dev-related.
1114 * pfs_mode and pfs_nmasters are rollup fields which critically describes
1115 * how elements of the cluster act on the cluster. pfs_mode is only applicable
1116 * when a PFS is mounted by the system. pfs_nmasters is our best guess as to
1117 * how many masters have been configured for a cluster and is always
1118 * applicable. pfs_types[] is an array with 1:1 correspondance to the
1119 * iroot cluster and describes the PFS types of the nodes making up the
1122 * WARNING! Portions of this structure have deferred initialization. In
1123 * particular, if not mounted there will be no ihidden or wthread.
1124 * umounted network PFSs will also be missing iroot and numerous
1125 * other fields will not be initialized prior to mount.
1127 * Synchronization threads are chain-specific and only applicable
1128 * to local hard PFS entries. A hammer2_pfs structure may contain
1129 * more than one when multiple hard PFSs are present on the local
1130 * machine which require synchronization monitoring. Most PFSs
1131 * (such as snapshots) are 1xMASTER PFSs which do not need a
1132 * synchronization thread.
1134 * WARNING! The chains making up pfs->iroot's cluster are accounted for in
1135 * hammer2_dev->mount_count when the pfs is associated with a mount
1138 struct hammer2_pfs {
1140 TAILQ_ENTRY(hammer2_pfs) mntentry; /* hammer2_pfslist */
1142 hammer2_dev_t *spmp_hmp; /* only if super-root pmp */
1143 hammer2_dev_t *force_local; /* only if 'local' mount */
1144 hammer2_inode_t *iroot; /* PFS root inode */
1145 hammer2_inode_t *ihidden; /* PFS hidden directory */
1146 uint8_t pfs_types[HAMMER2_MAXCLUSTER];
1147 char *pfs_names[HAMMER2_MAXCLUSTER];
1148 hammer2_dev_t *pfs_hmps[HAMMER2_MAXCLUSTER];
1149 hammer2_trans_t trans;
1150 struct lock lock; /* PFS lock for certain ops */
1151 struct lock lock_nlink; /* rename and nlink lock */
1152 struct netexport export; /* nfs export */
1153 int ronly; /* read-only mount */
1154 int hflags; /* pfs-specific mount flags */
1155 struct malloc_type *minode;
1156 struct malloc_type *mmsg;
1157 struct spinlock inum_spin; /* inumber lookup */
1158 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
1159 struct spinlock lru_spin; /* inumber lookup */
1160 struct hammer2_chain_list lru_list; /* chains on LRU */
1161 int lru_count; /* #of chains on LRU */
1162 hammer2_tid_t modify_tid; /* modify transaction id */
1163 hammer2_tid_t inode_tid; /* inode allocator */
1164 uint8_t pfs_nmasters; /* total masters */
1165 uint8_t pfs_mode; /* operating mode PFSMODE */
1170 uint32_t inmem_dirty_chains;
1171 int count_lwinprog; /* logical write in prog */
1172 struct spinlock list_spin;
1173 struct h2_sideq_list sideq; /* last-close dirty/unlink */
1174 hammer2_thread_t sync_thrs[HAMMER2_MAXCLUSTER];
1175 uint32_t cluster_flags; /* cached cluster flags */
1176 int has_xop_threads;
1177 struct spinlock xop_spin; /* xop sequencer */
1178 hammer2_xop_group_t xop_groups[HAMMER2_XOPGROUPS];
1179 hammer2_xop_list_t xopq[HAMMER2_MAXCLUSTER][HAMMER2_XOPGROUPS+
1180 HAMMER2_SPECTHREADS];
1183 typedef struct hammer2_pfs hammer2_pfs_t;
1185 #define HAMMER2_LRU_LIMIT 1024 /* per pmp lru_list */
1187 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
1188 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
1190 #define HAMMER2_LWINPROG_WAITING 0x80000000
1191 #define HAMMER2_LWINPROG_WAITING0 0x40000000
1192 #define HAMMER2_LWINPROG_MASK 0x3FFFFFFF
1195 * hammer2_cluster_check
1197 #define HAMMER2_CHECK_NULL 0x00000001
1202 #define HAMMER2_BULK_ABORT 0x00000001
1207 #if defined(_KERNEL)
1209 MALLOC_DECLARE(M_HAMMER2);
1211 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
1212 #define ITOV(ip) ((ip)->vp)
1215 * Currently locked chains retain the locked buffer cache buffer for
1216 * indirect blocks, and indirect blocks can be one of two sizes. The
1217 * device buffer has to match the case to avoid deadlocking recursive
1218 * chains that might otherwise try to access different offsets within
1219 * the same device buffer.
1223 hammer2_devblkradix(int radix)
1226 if (radix <= HAMMER2_LBUFRADIX) {
1227 return (HAMMER2_LBUFRADIX);
1229 return (HAMMER2_PBUFRADIX);
1232 return (HAMMER2_PBUFRADIX);
1236 * XXX almost time to remove this. DIO uses PBUFSIZE exclusively now.
1240 hammer2_devblksize(size_t bytes)
1243 if (bytes <= HAMMER2_LBUFSIZE) {
1244 return(HAMMER2_LBUFSIZE);
1246 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
1247 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
1248 return (HAMMER2_PBUFSIZE);
1251 return (HAMMER2_PBUFSIZE);
1257 MPTOPMP(struct mount *mp)
1259 return ((hammer2_pfs_t *)mp->mnt_data);
1262 #define LOCKSTART int __nlocks = curthread->td_locks
1263 #define LOCKENTER (++curthread->td_locks)
1264 #define LOCKEXIT (--curthread->td_locks)
1265 #define LOCKSTOP KKASSERT(curthread->td_locks == __nlocks)
1267 extern struct vop_ops hammer2_vnode_vops;
1268 extern struct vop_ops hammer2_spec_vops;
1269 extern struct vop_ops hammer2_fifo_vops;
1271 extern int hammer2_debug;
1272 extern int hammer2_cluster_enable;
1273 extern int hammer2_hardlink_enable;
1274 extern int hammer2_flush_pipe;
1275 extern int hammer2_synchronous_flush;
1276 extern int hammer2_dio_count;
1277 extern long hammer2_chain_allocs;
1278 extern long hammer2_chain_frees;
1279 extern long hammer2_limit_dirty_chains;
1280 extern long hammer2_count_modified_chains;
1281 extern long hammer2_iod_file_read;
1282 extern long hammer2_iod_meta_read;
1283 extern long hammer2_iod_indr_read;
1284 extern long hammer2_iod_fmap_read;
1285 extern long hammer2_iod_volu_read;
1286 extern long hammer2_iod_file_write;
1287 extern long hammer2_iod_file_wembed;
1288 extern long hammer2_iod_file_wzero;
1289 extern long hammer2_iod_file_wdedup;
1290 extern long hammer2_iod_meta_write;
1291 extern long hammer2_iod_indr_write;
1292 extern long hammer2_iod_fmap_write;
1293 extern long hammer2_iod_volu_write;
1294 extern long hammer2_ioa_file_read;
1295 extern long hammer2_ioa_meta_read;
1296 extern long hammer2_ioa_indr_read;
1297 extern long hammer2_ioa_fmap_read;
1298 extern long hammer2_ioa_volu_read;
1299 extern long hammer2_ioa_file_write;
1300 extern long hammer2_ioa_meta_write;
1301 extern long hammer2_ioa_indr_write;
1302 extern long hammer2_ioa_fmap_write;
1303 extern long hammer2_ioa_volu_write;
1305 extern long hammer2_check_xxhash64;
1306 extern long hammer2_check_icrc32;
1308 extern struct objcache *cache_buffer_read;
1309 extern struct objcache *cache_buffer_write;
1310 extern struct objcache *cache_xops;
1315 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
1316 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
1318 int hammer2_signal_check(time_t *timep);
1319 const char *hammer2_error_str(int error);
1321 void hammer2_inode_lock(hammer2_inode_t *ip, int how);
1322 void hammer2_inode_unlock(hammer2_inode_t *ip);
1323 hammer2_chain_t *hammer2_inode_chain(hammer2_inode_t *ip, int clindex, int how);
1324 hammer2_chain_t *hammer2_inode_chain_and_parent(hammer2_inode_t *ip,
1325 int clindex, hammer2_chain_t **parentp, int how);
1326 hammer2_mtx_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
1327 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip,
1328 hammer2_mtx_state_t ostate);
1329 int hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
1330 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int);
1332 void hammer2_dev_exlock(hammer2_dev_t *hmp);
1333 void hammer2_dev_shlock(hammer2_dev_t *hmp);
1334 void hammer2_dev_unlock(hammer2_dev_t *hmp);
1336 int hammer2_get_dtype(const hammer2_inode_data_t *ipdata);
1337 int hammer2_get_vtype(uint8_t type);
1338 u_int8_t hammer2_get_obj_type(enum vtype vtype);
1339 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
1340 u_int64_t hammer2_timespec_to_time(const struct timespec *ts);
1341 u_int32_t hammer2_to_unix_xid(const uuid_t *uuid);
1342 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
1343 void hammer2_trans_manage_init(hammer2_pfs_t *pmp);
1345 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
1346 int hammer2_getradix(size_t bytes);
1348 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
1349 hammer2_key_t *lbasep, hammer2_key_t *leofp);
1350 int hammer2_calc_physical(hammer2_inode_t *ip, hammer2_key_t lbase);
1351 void hammer2_update_time(uint64_t *timep);
1352 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
1357 struct vnode *hammer2_igetv(hammer2_inode_t *ip, int *errorp);
1358 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfs_t *pmp,
1359 hammer2_tid_t inum);
1360 hammer2_inode_t *hammer2_inode_get(hammer2_pfs_t *pmp, hammer2_inode_t *dip,
1361 hammer2_cluster_t *cluster, int idx);
1362 void hammer2_inode_free(hammer2_inode_t *ip);
1363 void hammer2_inode_ref(hammer2_inode_t *ip);
1364 void hammer2_inode_drop(hammer2_inode_t *ip);
1365 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
1366 hammer2_cluster_t *cluster);
1367 void hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1369 void hammer2_inode_modify(hammer2_inode_t *ip);
1370 void hammer2_inode_run_sideq(hammer2_pfs_t *pmp);
1372 hammer2_inode_t *hammer2_inode_create(hammer2_inode_t *dip,
1373 struct vattr *vap, struct ucred *cred,
1374 const uint8_t *name, size_t name_len, hammer2_key_t lhc,
1375 hammer2_key_t inum, uint8_t type, uint8_t target_type,
1376 int flags, int *errorp);
1377 int hammer2_inode_connect(hammer2_inode_t *dip, hammer2_inode_t *ip,
1378 const char *name, size_t name_len,
1380 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
1381 hammer2_inode_t *tdip, int *errorp, int ishardlink);
1382 void hammer2_inode_chain_sync(hammer2_inode_t *ip);
1383 int hammer2_inode_unlink_finisher(hammer2_inode_t *ip, int isopen);
1384 void hammer2_inode_install_hidden(hammer2_pfs_t *pmp);
1389 void hammer2_voldata_lock(hammer2_dev_t *hmp);
1390 void hammer2_voldata_unlock(hammer2_dev_t *hmp);
1391 void hammer2_voldata_modify(hammer2_dev_t *hmp);
1392 hammer2_chain_t *hammer2_chain_alloc(hammer2_dev_t *hmp,
1394 hammer2_blockref_t *bref);
1395 void hammer2_chain_core_init(hammer2_chain_t *chain);
1396 void hammer2_chain_ref(hammer2_chain_t *chain);
1397 void hammer2_chain_drop(hammer2_chain_t *chain);
1398 void hammer2_chain_lock(hammer2_chain_t *chain, int how);
1399 void hammer2_chain_lock_downgrade(hammer2_chain_t *chain);
1400 void hammer2_chain_push_shared_lock(hammer2_chain_t *chain);
1401 void hammer2_chain_pull_shared_lock(hammer2_chain_t *chain);
1402 void hammer2_chain_load_data(hammer2_chain_t *chain);
1403 const hammer2_media_data_t *hammer2_chain_rdata(hammer2_chain_t *chain);
1404 hammer2_media_data_t *hammer2_chain_wdata(hammer2_chain_t *chain);
1405 int hammer2_chain_snapshot(hammer2_chain_t *chain, hammer2_ioc_pfs_t *pmp,
1406 hammer2_tid_t mtid);
1408 int hammer2_chain_hardlink_find(hammer2_inode_t *dip,
1409 hammer2_chain_t **parentp,
1410 hammer2_chain_t **chainp,
1412 void hammer2_chain_modify(hammer2_chain_t *chain, hammer2_tid_t mtid,
1413 hammer2_off_t dedup_off, int flags);
1414 void hammer2_chain_modify_ip(hammer2_inode_t *ip, hammer2_chain_t *chain,
1415 hammer2_tid_t mtid, int flags);
1416 void hammer2_chain_resize(hammer2_inode_t *ip, hammer2_chain_t *parent,
1417 hammer2_chain_t *chain,
1418 hammer2_tid_t mtid, hammer2_off_t dedup_off,
1419 int nradix, int flags);
1420 void hammer2_chain_unlock(hammer2_chain_t *chain);
1421 void hammer2_chain_wait(hammer2_chain_t *chain);
1422 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
1423 hammer2_blockref_t *bref);
1424 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
1425 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
1426 hammer2_chain_t *hammer2_chain_getparent(hammer2_chain_t **parentp, int how);
1427 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
1428 hammer2_key_t *key_nextp,
1429 hammer2_key_t key_beg, hammer2_key_t key_end,
1430 int *cache_indexp, int flags);
1431 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
1432 hammer2_chain_t *chain,
1433 hammer2_key_t *key_nextp,
1434 hammer2_key_t key_beg, hammer2_key_t key_end,
1435 int *cache_indexp, int flags);
1436 hammer2_blockref_t *hammer2_chain_scan(hammer2_chain_t *parent,
1437 hammer2_chain_t **chainp,
1438 hammer2_blockref_t *bref,
1439 int *firstp, int *cache_indexp, int flags);
1441 int hammer2_chain_create(hammer2_chain_t **parentp,
1442 hammer2_chain_t **chainp, hammer2_pfs_t *pmp,
1443 hammer2_key_t key, int keybits,
1444 int type, size_t bytes, hammer2_tid_t mtid,
1445 hammer2_off_t dedup_off, int flags);
1446 void hammer2_chain_rename(hammer2_blockref_t *bref,
1447 hammer2_chain_t **parentp,
1448 hammer2_chain_t *chain,
1449 hammer2_tid_t mtid, int flags);
1450 void hammer2_chain_delete(hammer2_chain_t *parent, hammer2_chain_t *chain,
1451 hammer2_tid_t mtid, int flags);
1452 void hammer2_chain_setflush(hammer2_chain_t *chain);
1453 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
1454 hammer2_blockref_t *base, int count);
1455 hammer2_chain_t *hammer2_chain_bulksnap(hammer2_chain_t *chain);
1456 void hammer2_chain_bulkdrop(hammer2_chain_t *copy);
1458 void hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata);
1459 int hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata);
1462 void hammer2_pfs_memory_wait(hammer2_pfs_t *pmp);
1463 void hammer2_pfs_memory_inc(hammer2_pfs_t *pmp);
1464 void hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp);
1466 void hammer2_base_delete(hammer2_chain_t *chain,
1467 hammer2_blockref_t *base, int count,
1468 int *cache_indexp, hammer2_chain_t *child);
1469 void hammer2_base_insert(hammer2_chain_t *chain,
1470 hammer2_blockref_t *base, int count,
1471 int *cache_indexp, hammer2_chain_t *child);
1476 void hammer2_flush(hammer2_chain_t *chain, int istop);
1477 void hammer2_flush_quick(hammer2_dev_t *hmp);
1478 void hammer2_delayed_flush(hammer2_chain_t *chain);
1483 void hammer2_trans_init(hammer2_pfs_t *pmp, uint32_t flags);
1484 hammer2_tid_t hammer2_trans_sub(hammer2_pfs_t *pmp);
1485 void hammer2_trans_clear_preflush(hammer2_pfs_t *pmp);
1486 void hammer2_trans_done(hammer2_pfs_t *pmp);
1487 hammer2_tid_t hammer2_trans_newinum(hammer2_pfs_t *pmp);
1488 void hammer2_trans_assert_strategy(hammer2_pfs_t *pmp);
1493 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
1494 int fflag, struct ucred *cred);
1499 void hammer2_io_putblk(hammer2_io_t **diop);
1500 void hammer2_io_cleanup(hammer2_dev_t *hmp, struct hammer2_io_tree *tree);
1501 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
1502 hammer2_io_t *hammer2_io_getquick(hammer2_dev_t *hmp, off_t lbase, int lsize);
1503 void hammer2_io_getblk(hammer2_dev_t *hmp, off_t lbase, int lsize,
1504 hammer2_iocb_t *iocb);
1505 void hammer2_io_complete(hammer2_iocb_t *iocb);
1506 void hammer2_io_callback(struct bio *bio);
1507 void hammer2_iocb_wait(hammer2_iocb_t *iocb);
1508 int hammer2_io_new(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1509 hammer2_io_t **diop);
1510 int hammer2_io_newnz(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1511 hammer2_io_t **diop);
1512 int hammer2_io_newq(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1513 hammer2_io_t **diop);
1514 int hammer2_io_bread(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1515 hammer2_io_t **diop);
1516 void hammer2_io_bawrite(hammer2_io_t **diop);
1517 void hammer2_io_bdwrite(hammer2_io_t **diop);
1518 int hammer2_io_bwrite(hammer2_io_t **diop);
1519 int hammer2_io_isdirty(hammer2_io_t *dio);
1520 void hammer2_io_setdirty(hammer2_io_t *dio);
1521 void hammer2_io_setinval(hammer2_io_t *dio, hammer2_off_t off, u_int bytes);
1522 void hammer2_io_brelse(hammer2_io_t **diop);
1523 void hammer2_io_bqrelse(hammer2_io_t **diop);
1524 int hammer2_io_crc_good(hammer2_chain_t *chain, uint64_t *maskp);
1525 void hammer2_io_crc_setmask(hammer2_io_t *dio, uint64_t mask);
1526 void hammer2_io_crc_clrmask(hammer2_io_t *dio, uint64_t mask);
1531 void hammer2_thr_signal(hammer2_thread_t *thr, uint32_t flags);
1532 void hammer2_thr_return(hammer2_thread_t *thr, uint32_t flags);
1533 void hammer2_thr_wait(hammer2_thread_t *thr, uint32_t flags);
1534 void hammer2_thr_wait_neg(hammer2_thread_t *thr, uint32_t flags);
1535 void hammer2_thr_create(hammer2_thread_t *thr, hammer2_pfs_t *pmp,
1536 const char *id, int clindex, int repidx,
1537 void (*func)(void *arg));
1538 void hammer2_thr_delete(hammer2_thread_t *thr);
1539 void hammer2_thr_remaster(hammer2_thread_t *thr);
1540 void hammer2_thr_freeze_async(hammer2_thread_t *thr);
1541 void hammer2_thr_freeze(hammer2_thread_t *thr);
1542 void hammer2_thr_unfreeze(hammer2_thread_t *thr);
1543 int hammer2_thr_break(hammer2_thread_t *thr);
1544 void hammer2_primary_xops_thread(void *arg);
1547 * hammer2_thread.c (XOP API)
1549 void hammer2_xop_group_init(hammer2_pfs_t *pmp, hammer2_xop_group_t *xgrp);
1550 void *hammer2_xop_alloc(hammer2_inode_t *ip, int flags);
1551 void hammer2_xop_setname(hammer2_xop_head_t *xop,
1552 const char *name, size_t name_len);
1553 void hammer2_xop_setname2(hammer2_xop_head_t *xop,
1554 const char *name, size_t name_len);
1555 void hammer2_xop_setip2(hammer2_xop_head_t *xop, hammer2_inode_t *ip2);
1556 void hammer2_xop_setip3(hammer2_xop_head_t *xop, hammer2_inode_t *ip3);
1557 void hammer2_xop_reinit(hammer2_xop_head_t *xop);
1558 void hammer2_xop_helper_create(hammer2_pfs_t *pmp);
1559 void hammer2_xop_helper_cleanup(hammer2_pfs_t *pmp);
1560 void hammer2_xop_start(hammer2_xop_head_t *xop, hammer2_xop_func_t func);
1561 void hammer2_xop_start_except(hammer2_xop_head_t *xop, hammer2_xop_func_t func,
1563 int hammer2_xop_collect(hammer2_xop_head_t *xop, int flags);
1564 void hammer2_xop_retire(hammer2_xop_head_t *xop, uint32_t mask);
1565 int hammer2_xop_active(hammer2_xop_head_t *xop);
1566 int hammer2_xop_feed(hammer2_xop_head_t *xop, hammer2_chain_t *chain,
1567 int clindex, int error);
1572 void hammer2_primary_sync_thread(void *arg);
1575 * XOP backends in hammer2_xops.c, primarily for VNOPS. Other XOP backends
1576 * may be integrated into other source files.
1578 void hammer2_xop_ipcluster(hammer2_xop_t *xop, int clidx);
1579 void hammer2_xop_readdir(hammer2_xop_t *xop, int clidx);
1580 void hammer2_xop_nresolve(hammer2_xop_t *xop, int clidx);
1581 void hammer2_xop_unlink(hammer2_xop_t *xop, int clidx);
1582 void hammer2_xop_nrename(hammer2_xop_t *xop, int clidx);
1583 void hammer2_xop_nlink(hammer2_xop_t *xop, int clidx);
1584 void hammer2_xop_scanlhc(hammer2_xop_t *xop, int clidx);
1585 void hammer2_xop_scanall(hammer2_xop_t *xop, int clidx);
1586 void hammer2_xop_lookup(hammer2_xop_t *xop, int clidx);
1587 void hammer2_inode_xop_create(hammer2_xop_t *xop, int clidx);
1588 void hammer2_inode_xop_destroy(hammer2_xop_t *xop, int clidx);
1589 void hammer2_inode_xop_chain_sync(hammer2_xop_t *xop, int clidx);
1590 void hammer2_inode_xop_unlinkall(hammer2_xop_t *xop, int clidx);
1591 void hammer2_inode_xop_connect(hammer2_xop_t *xop, int clidx);
1592 void hammer2_inode_xop_flush(hammer2_xop_t *xop, int clidx);
1597 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
1598 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
1603 void hammer2_volconf_update(hammer2_dev_t *hmp, int index);
1604 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
1605 int hammer2_vfs_sync(struct mount *mp, int waitflags);
1606 hammer2_pfs_t *hammer2_pfsalloc(hammer2_chain_t *chain,
1607 const hammer2_inode_data_t *ripdata,
1608 hammer2_tid_t modify_tid,
1609 hammer2_dev_t *force_local);
1611 void hammer2_lwinprog_ref(hammer2_pfs_t *pmp);
1612 void hammer2_lwinprog_drop(hammer2_pfs_t *pmp);
1613 void hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int pipe);
1618 int hammer2_freemap_alloc(hammer2_chain_t *chain, size_t bytes);
1619 void hammer2_freemap_adjust(hammer2_dev_t *hmp,
1620 hammer2_blockref_t *bref, int how);
1625 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
1626 const hammer2_media_data_t *hammer2_cluster_rdata(hammer2_cluster_t *cluster);
1627 hammer2_media_data_t *hammer2_cluster_wdata(hammer2_cluster_t *cluster);
1628 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
1629 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
1630 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfs_t *pmp,
1631 hammer2_blockref_t *bref);
1632 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
1633 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1634 void hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1635 int hammer2_cluster_check(hammer2_cluster_t *cluster, hammer2_key_t lokey,
1637 void hammer2_cluster_resolve(hammer2_cluster_t *cluster);
1638 void hammer2_cluster_forcegood(hammer2_cluster_t *cluster);
1639 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster);
1640 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1642 int hammer2_bulkfree_pass(hammer2_dev_t *hmp,
1643 struct hammer2_ioc_bulkfree *bfi);
1648 void hammer2_iocom_init(hammer2_dev_t *hmp);
1649 void hammer2_iocom_uninit(hammer2_dev_t *hmp);
1650 void hammer2_cluster_reconnect(hammer2_dev_t *hmp, struct file *fp);
1653 * hammer2_strategy.c
1655 int hammer2_vop_strategy(struct vop_strategy_args *ap);
1656 int hammer2_vop_bmap(struct vop_bmap_args *ap);
1657 void hammer2_write_thread(void *arg);
1658 void hammer2_bioq_sync(hammer2_pfs_t *pmp);
1659 void hammer2_dedup_clear(hammer2_dev_t *hmp);
1661 #endif /* !_KERNEL */
1662 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */