2 * Copyright (c) 2011-2018 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_
66 #include <sys/param.h>
68 #include <sys/types.h>
70 #include <sys/kernel.h>
74 #include <sys/systm.h>
77 #include <sys/malloc.h>
78 #include <sys/mount.h>
79 #include <sys/vnode.h>
81 #include <sys/mountctl.h>
84 #include <sys/thread.h>
85 #include <sys/globaldata.h>
86 #include <sys/lockf.h>
88 #include <sys/queue.h>
89 #include <sys/limits.h>
91 #include <sys/mutex.h>
93 #include <sys/kern_syscall.h>
97 #include <sys/signal2.h>
99 #include <sys/mutex2.h>
102 #include "hammer2_xxhash.h"
103 #include "hammer2_disk.h"
104 #include "hammer2_mount.h"
105 #include "hammer2_ioctl.h"
108 struct hammer2_chain;
109 struct hammer2_cluster;
110 struct hammer2_inode;
111 struct hammer2_depend;
116 struct hammer2_thread;
120 * Mutex and lock shims. Hammer2 requires support for asynchronous and
121 * abortable locks, and both exclusive and shared spinlocks. Normal
122 * synchronous non-abortable locks can be substituted for spinlocks.
124 typedef mtx_t hammer2_mtx_t;
125 typedef mtx_link_t hammer2_mtx_link_t;
126 typedef mtx_state_t hammer2_mtx_state_t;
128 typedef struct spinlock hammer2_spin_t;
130 #define hammer2_mtx_ex mtx_lock_ex_quick
131 #define hammer2_mtx_ex_try mtx_lock_ex_try
132 #define hammer2_mtx_sh mtx_lock_sh_quick
133 #define hammer2_mtx_sh_again mtx_lock_sh_again
134 #define hammer2_mtx_sh_try mtx_lock_sh_try
135 #define hammer2_mtx_unlock mtx_unlock
136 #define hammer2_mtx_downgrade mtx_downgrade
137 #define hammer2_mtx_owned mtx_owned
138 #define hammer2_mtx_init mtx_init
139 #define hammer2_mtx_temp_release mtx_lock_temp_release
140 #define hammer2_mtx_temp_restore mtx_lock_temp_restore
141 #define hammer2_mtx_refs mtx_lockrefs
143 #define hammer2_spin_init spin_init
144 #define hammer2_spin_sh spin_lock_shared
145 #define hammer2_spin_ex spin_lock
146 #define hammer2_spin_unsh spin_unlock_shared
147 #define hammer2_spin_unex spin_unlock
149 TAILQ_HEAD(hammer2_xop_list, hammer2_xop_head);
150 TAILQ_HEAD(hammer2_chain_list, hammer2_chain);
152 typedef struct hammer2_xop_list hammer2_xop_list_t;
156 * General lock support
160 hammer2_mtx_upgrade_try(hammer2_mtx_t *mtx)
162 return mtx_upgrade_try(mtx);
168 * The xid tracks internal transactional updates.
170 * XXX fix-me, really needs to be 64-bits
172 typedef uint32_t hammer2_xid_t;
174 #define HAMMER2_XID_MIN 0x00000000U
175 #define HAMMER2_XID_MAX 0x7FFFFFFFU
178 * Cap the dynamic calculation for the maximum number of dirty
179 * chains and dirty inodes allowed.
181 #define HAMMER2_LIMIT_DIRTY_CHAINS (1024*1024)
182 #define HAMMER2_LIMIT_DIRTY_INODES (65536)
185 * The chain structure tracks a portion of the media topology from the
186 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
187 * data blocks, and freemap nodes and leafs.
189 * The chain structure utilizes a simple singly-homed topology and the
190 * chain's in-memory topology will move around as the chains do, due mainly
191 * to renames and indirect block creation.
193 * Block Table Updates
195 * Block table updates for insertions and updates are delayed until the
196 * flush. This allows us to avoid having to modify the parent chain
197 * all the way to the root.
199 * Block table deletions are performed immediately (modifying the parent
200 * in the process) because the flush code uses the chain structure to
201 * track delayed updates and the chain will be (likely) gone or moved to
202 * another location in the topology after a deletion.
204 * A prior iteration of the code tried to keep the relationship intact
205 * on deletes by doing a delete-duplicate operation on the chain, but
206 * it added way too much complexity to the codebase.
208 * Flush Synchronization
210 * The flush code must flush modified chains bottom-up. Because chain
211 * structures can shift around and are NOT topologically stable,
212 * modified chains are independently indexed for the flush. As the flush
213 * runs it modifies (or further modifies) and updates the parents,
214 * propagating the flush all the way to the volume root.
216 * Modifying front-end operations can occur during a flush but will block
217 * in two cases: (1) when the front-end tries to operate on the inode
218 * currently in the midst of being flushed and (2) if the front-end
219 * crosses an inode currently being flushed (such as during a rename).
220 * So, for example, if you rename directory "x" to "a/b/c/d/e/f/g/x" and
221 * the flusher is currently working on "a/b/c", the rename will block
222 * temporarily in order to ensure that "x" exists in one place or the
225 * Meta-data statistics are updated by the flusher. The front-end will
226 * make estimates but meta-data must be fully synchronized only during a
227 * flush in order to ensure that it remains correct across a crash.
229 * Multiple flush synchronizations can theoretically be in-flight at the
230 * same time but the implementation is not coded to handle the case and
231 * currently serializes them.
235 * Snapshots currently require the subdirectory tree being snapshotted
236 * to be flushed. The snapshot then creates a new super-root inode which
237 * copies the flushed blockdata of the directory or file that was
242 * - Note that the radix tree runs in powers of 2 only so sub-trees
243 * cannot straddle edges.
245 RB_HEAD(hammer2_chain_tree, hammer2_chain);
246 TAILQ_HEAD(h2_flush_list, hammer2_chain);
247 TAILQ_HEAD(h2_core_list, hammer2_chain);
249 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
250 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
252 struct hammer2_reptrack {
254 struct hammer2_reptrack *next;
255 struct hammer2_chain *chain;
259 * Core topology for chain (embedded in chain). Protected by a spinlock.
261 struct hammer2_chain_core {
263 struct hammer2_reptrack *reptrack;
264 struct hammer2_chain_tree rbtree; /* sub-chains */
265 int live_zero; /* blockref array opt */
266 u_int live_count; /* live (not deleted) chains in tree */
267 u_int chain_count; /* live + deleted chains under core */
268 int generation; /* generation number (inserts only) */
271 typedef struct hammer2_chain_core hammer2_chain_core_t;
273 RB_HEAD(hammer2_io_tree, hammer2_io);
276 * DIO - Management structure wrapping system buffer cache.
278 * HAMMER2 uses an I/O abstraction that allows it to cache and manipulate
279 * fixed-sized filesystem buffers frontend by variable-sized hammer2_chain
282 /* #define HAMMER2_IO_DEBUG */
284 #ifdef HAMMER2_IO_DEBUG
285 #define HAMMER2_IO_DEBUG_ARGS , const char *file, int line
286 #define HAMMER2_IO_DEBUG_CALL , file, line
287 #define HAMMER2_IO_DEBUG_COUNT 2048
288 #define HAMMER2_IO_DEBUG_MASK (HAMMER2_IO_DEBUG_COUNT - 1)
290 #define HAMMER2_IO_DEBUG_ARGS
291 #define HAMMER2_IO_DEBUG_CALL
295 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
296 struct hammer2_dev *hmp;
301 int act; /* activity */
302 int btype; /* approximate BREF_TYPE_* */
305 #ifdef HAMMER2_IO_DEBUG
310 uint64_t dedup_valid; /* valid for dedup operation */
311 uint64_t dedup_alloc; /* allocated / de-dupable */
312 #ifdef HAMMER2_IO_DEBUG
313 const char *debug_file[HAMMER2_IO_DEBUG_COUNT];
314 void *debug_td[HAMMER2_IO_DEBUG_COUNT];
315 int debug_line[HAMMER2_IO_DEBUG_COUNT];
316 uint64_t debug_refs[HAMMER2_IO_DEBUG_COUNT];
320 typedef struct hammer2_io hammer2_io_t;
322 #define HAMMER2_DIO_INPROG 0x8000000000000000LLU /* bio in progress */
323 #define HAMMER2_DIO_GOOD 0x4000000000000000LLU /* dio->bp is stable */
324 #define HAMMER2_DIO_WAITING 0x2000000000000000LLU /* wait on INPROG */
325 #define HAMMER2_DIO_DIRTY 0x1000000000000000LLU /* flush last drop */
326 #define HAMMER2_DIO_FLUSH 0x0800000000000000LLU /* immediate flush */
328 #define HAMMER2_DIO_MASK 0x00FFFFFFFFFFFFFFLLU
331 * Primary chain structure keeps track of the topology in-memory.
333 struct hammer2_chain {
335 hammer2_chain_core_t core;
336 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
337 hammer2_blockref_t bref;
338 struct hammer2_chain *parent;
339 struct hammer2_dev *hmp;
340 struct hammer2_pfs *pmp; /* A PFS or super-root (spmp) */
342 struct lock diolk; /* xop focus interlock */
343 hammer2_io_t *dio; /* physical data buffer */
344 hammer2_media_data_t *data; /* data pointer shortcut */
345 u_int bytes; /* physical data size */
349 int error; /* on-lock data error state */
350 int cache_index; /* heur speeds up lookup */
352 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
353 TAILQ_ENTRY(hammer2_chain) lru_node; /* 0-refs LRU */
356 typedef struct hammer2_chain hammer2_chain_t;
358 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
359 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
362 * Special notes on flags:
364 * INITIAL - This flag allows a chain to be created and for storage to
365 * be allocated without having to immediately instantiate the
366 * related buffer. The data is assumed to be all-zeros. It
367 * is primarily used for indirect blocks.
369 * MODIFIED - The chain's media data has been modified. Prevents chain
370 * free on lastdrop if still in the topology.
372 * UPDATE - Chain might not be modified but parent blocktable needs
373 * an update. Prevents chain free on lastdrop if still in
376 * FICTITIOUS - Faked chain as a placeholder for an error condition. This
377 * chain is unsuitable for I/O.
379 * BMAPPED - Indicates that the chain is present in the parent blockmap.
381 * BMAPUPD - Indicates that the chain is present but needs to be updated
382 * in the parent blockmap.
384 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
385 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
386 #define HAMMER2_CHAIN_DESTROY 0x00000004
387 #define HAMMER2_CHAIN_DEDUPABLE 0x00000008 /* registered w/dedup */
388 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
389 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
390 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
391 #define HAMMER2_CHAIN_UNUSED0080 0x00000080
392 #define HAMMER2_CHAIN_TESTEDGOOD 0x00000100 /* crc tested good */
393 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
394 #define HAMMER2_CHAIN_FICTITIOUS 0x00000400 /* unsuitable for I/O */
395 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
396 #define HAMMER2_CHAIN_UNUSED1000 0x00001000
397 #define HAMMER2_CHAIN_COUNTEDBREFS 0x00002000 /* block table stats */
398 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
399 #define HAMMER2_CHAIN_ONLRU 0x00008000 /* on LRU list */
400 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
401 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
402 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
403 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
404 #define HAMMER2_CHAIN_IOINPROG 0x00100000 /* I/O interlock */
405 #define HAMMER2_CHAIN_IOSIGNAL 0x00200000 /* I/O interlock */
406 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
407 #define HAMMER2_CHAIN_HINT_LEAF_COUNT 0x00800000 /* redo leaf count */
408 #define HAMMER2_CHAIN_LRUHINT 0x01000000 /* was reused */
410 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
411 HAMMER2_CHAIN_UPDATE | \
412 HAMMER2_CHAIN_ONFLUSH | \
413 HAMMER2_CHAIN_DESTROY)
416 * Hammer2 error codes, used by chain->error and cluster->error. The error
417 * code is typically set on-lock unless no I/O was requested, and set on
418 * I/O otherwise. If set for a cluster it generally means that the cluster
419 * code could not find a valid copy to present.
421 * All H2 error codes are flags and can be accumulated by ORing them
424 * IO - An I/O error occurred
425 * CHECK - I/O succeeded but did not match the check code
426 * INCOMPLETE - A cluster is not complete enough to use, or
427 * a chain cannot be loaded because its parent has an error.
429 * NOTE: API allows callers to check zero/non-zero to determine if an error
432 * NOTE: Chain's data field is usually NULL on an IO error but not necessarily
433 * NULL on other errors. Check chain->error, not chain->data.
435 #define HAMMER2_ERROR_NONE 0 /* no error (must be 0) */
436 #define HAMMER2_ERROR_EIO 0x00000001 /* device I/O error */
437 #define HAMMER2_ERROR_CHECK 0x00000002 /* check code error */
438 #define HAMMER2_ERROR_INCOMPLETE 0x00000004 /* incomplete cluster */
439 #define HAMMER2_ERROR_DEPTH 0x00000008 /* tmp depth limit */
440 #define HAMMER2_ERROR_BADBREF 0x00000010 /* illegal bref */
441 #define HAMMER2_ERROR_ENOSPC 0x00000020 /* allocation failure */
442 #define HAMMER2_ERROR_ENOENT 0x00000040 /* entry not found */
443 #define HAMMER2_ERROR_ENOTEMPTY 0x00000080 /* dir not empty */
444 #define HAMMER2_ERROR_EAGAIN 0x00000100 /* retry */
445 #define HAMMER2_ERROR_ENOTDIR 0x00000200 /* not directory */
446 #define HAMMER2_ERROR_EISDIR 0x00000400 /* is directory */
447 #define HAMMER2_ERROR_EINPROGRESS 0x00000800 /* already running */
448 #define HAMMER2_ERROR_ABORTED 0x00001000 /* aborted operation */
449 #define HAMMER2_ERROR_EOF 0x00002000 /* end of scan */
450 #define HAMMER2_ERROR_EINVAL 0x00004000 /* catch-all */
451 #define HAMMER2_ERROR_EEXIST 0x00008000 /* entry exists */
452 #define HAMMER2_ERROR_EDEADLK 0x00010000
453 #define HAMMER2_ERROR_ESRCH 0x00020000
454 #define HAMMER2_ERROR_ETIMEDOUT 0x00040000
457 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
460 * NODATA - Asks that the chain->data not be resolved in order
463 * NODIRECT - Prevents a lookup of offset 0 in an inode from returning
464 * the inode itself if the inode is in DIRECTDATA mode
465 * (i.e. file is <= 512 bytes). Used by the synchronization
466 * code to prevent confusion.
468 * SHARED - The input chain is expected to be locked shared,
469 * and the output chain is locked shared.
471 * MATCHIND - Allows an indirect block / freemap node to be returned
472 * when the passed key range matches the radix. Remember
473 * that key_end is inclusive (e.g. {0x000,0xFFF},
474 * not {0x000,0x1000}).
476 * (Cannot be used for remote or cluster ops).
478 * ALLNODES - Allows NULL focus.
480 * ALWAYS - Always resolve the data. If ALWAYS and NODATA are both
481 * missing, bulk file data is not resolved but inodes and
482 * other meta-data will.
484 #define HAMMER2_LOOKUP_UNUSED0001 0x00000001
485 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
486 #define HAMMER2_LOOKUP_NODIRECT 0x00000004 /* no offset=0 DD */
487 #define HAMMER2_LOOKUP_SHARED 0x00000100
488 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
489 #define HAMMER2_LOOKUP_ALLNODES 0x00000400 /* allow NULL focus */
490 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
491 #define HAMMER2_LOOKUP_UNUSED1000 0x00001000
494 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
496 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
497 * blocks in the INITIAL-create state.
499 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
500 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
501 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
504 * Flags passed to hammer2_chain_lock()
506 * NOTE: RDONLY is set to optimize cluster operations when *no* modifications
507 * will be made to either the cluster being locked or any underlying
508 * cluster. It allows the cluster to lock and access data for a subset
509 * of available nodes instead of all available nodes.
511 * NOTE: NONBLOCK is only used for hammer2_chain_repparent() and getparent(),
512 * other functions (e.g. hammer2_chain_lookup(), etc) can't handle its
515 #define HAMMER2_RESOLVE_NEVER 1
516 #define HAMMER2_RESOLVE_MAYBE 2
517 #define HAMMER2_RESOLVE_ALWAYS 3
518 #define HAMMER2_RESOLVE_MASK 0x0F
520 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
521 #define HAMMER2_RESOLVE_LOCKAGAIN 0x20 /* another shared lock */
522 #define HAMMER2_RESOLVE_UNUSED40 0x40
523 #define HAMMER2_RESOLVE_NONBLOCK 0x80 /* non-blocking */
526 * Flags passed to hammer2_chain_delete()
528 #define HAMMER2_DELETE_PERMANENT 0x0001
531 * Flags passed to hammer2_chain_insert() or hammer2_chain_rename()
532 * or hammer2_chain_create().
534 #define HAMMER2_INSERT_PFSROOT 0x0004
535 #define HAMMER2_INSERT_SAMEPARENT 0x0008
538 * Flags passed to hammer2_chain_delete_duplicate()
540 #define HAMMER2_DELDUP_RECORE 0x0001
543 * Cluster different types of storage together for allocations
545 #define HAMMER2_FREECACHE_INODE 0
546 #define HAMMER2_FREECACHE_INDIR 1
547 #define HAMMER2_FREECACHE_DATA 2
548 #define HAMMER2_FREECACHE_UNUSED3 3
549 #define HAMMER2_FREECACHE_TYPES 4
552 * hammer2_freemap_alloc() block preference
554 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
557 * BMAP read-ahead maximum parameters
559 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
560 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
563 * hammer2_freemap_adjust()
565 #define HAMMER2_FREEMAP_DORECOVER 1
566 #define HAMMER2_FREEMAP_DOMAYFREE 2
567 #define HAMMER2_FREEMAP_DOREALFREE 3
570 * HAMMER2 cluster - A set of chains representing the same entity.
572 * hammer2_cluster typically represents a temporary set of representitive
573 * chains. The one exception is that a hammer2_cluster is embedded in
574 * hammer2_inode. This embedded cluster is ONLY used to track the
575 * representitive chains and cannot be directly locked.
577 * A cluster is usually temporary (and thus per-thread) for locking purposes,
578 * allowing us to embed the asynchronous storage required for cluster
579 * operations in the cluster itself and adjust the state and status without
580 * having to worry too much about SMP issues.
582 * The exception is the cluster embedded in the hammer2_inode structure.
583 * This is used to cache the cluster state on an inode-by-inode basis.
584 * Individual hammer2_chain structures not incorporated into clusters might
585 * also stick around to cache miscellanious elements.
587 * Because the cluster is a 'working copy' and is usually subject to cluster
588 * quorum rules, it is quite possible for us to end up with an insufficient
589 * number of live chains to execute an operation. If an insufficient number
590 * of chains remain in a working copy, the operation may have to be
591 * downgraded, retried, stall until the requisit number of chains are
592 * available, or possibly even error out depending on the mount type.
594 * A cluster's focus is set when it is locked. The focus can only be set
595 * to a chain still part of the synchronized set.
597 #define HAMMER2_XOPFIFO 16
598 #define HAMMER2_XOPFIFO_MASK (HAMMER2_XOPFIFO - 1)
599 #define HAMMER2_XOPGROUPS 32
600 #define HAMMER2_XOPGROUPS_MASK (HAMMER2_XOPGROUPS - 1)
602 #define HAMMER2_MAXCLUSTER 8
603 #define HAMMER2_XOPMASK_CLUSTER (uint64_t)((1LLU << HAMMER2_MAXCLUSTER) - 1)
604 #define HAMMER2_XOPMASK_VOP (uint64_t)0x0000000080000000LLU
605 #define HAMMER2_XOPMASK_FIFOW (uint64_t)0x0000000040000000LLU
606 #define HAMMER2_XOPMASK_WAIT (uint64_t)0x0000000020000000LLU
607 #define HAMMER2_XOPMASK_FEED (uint64_t)0x0000000100000000LLU
609 #define HAMMER2_XOPMASK_ALLDONE (HAMMER2_XOPMASK_VOP | HAMMER2_XOPMASK_CLUSTER)
611 #define HAMMER2_SPECTHREADS 1 /* sync */
613 struct hammer2_cluster_item {
614 hammer2_chain_t *chain;
619 typedef struct hammer2_cluster_item hammer2_cluster_item_t;
622 * INVALID - Invalid for focus, i.e. not part of synchronized set.
623 * Once set, this bit is sticky across operations.
625 * FEMOD - Indicates that front-end modifying operations can
626 * mess with this entry and MODSYNC will copy also
629 #define HAMMER2_CITEM_INVALID 0x00000001
630 #define HAMMER2_CITEM_FEMOD 0x00000002
631 #define HAMMER2_CITEM_NULL 0x00000004
633 struct hammer2_cluster {
634 int refs; /* track for deallocation */
636 struct hammer2_pfs *pmp;
639 int error; /* error code valid on lock */
641 hammer2_chain_t *focus; /* current focus (or mod) */
642 hammer2_cluster_item_t array[HAMMER2_MAXCLUSTER];
645 typedef struct hammer2_cluster hammer2_cluster_t;
648 * WRHARD - Hard mounts can write fully synchronized
649 * RDHARD - Hard mounts can read fully synchronized
650 * UNHARD - Unsynchronized masters present
651 * NOHARD - No masters visible
652 * WRSOFT - Soft mounts can write to at least the SOFT_MASTER
653 * RDSOFT - Soft mounts can read from at least a SOFT_SLAVE
654 * UNSOFT - Unsynchronized slaves present
655 * NOSOFT - No slaves visible
656 * RDSLAVE - slaves are accessible (possibly unsynchronized or remote).
657 * MSYNCED - All masters are fully synchronized
658 * SSYNCED - All known local slaves are fully synchronized to masters
660 * All available masters are always incorporated. All PFSs belonging to a
661 * cluster (master, slave, copy, whatever) always try to synchronize the
662 * total number of known masters in the PFSs root inode.
664 * A cluster might have access to many slaves, copies, or caches, but we
665 * have a limited number of cluster slots. Any such elements which are
666 * directly mounted from block device(s) will always be incorporated. Note
667 * that SSYNCED only applies to such elements which are directly mounted,
668 * not to any remote slaves, copies, or caches that could be available. These
669 * bits are used to monitor and drive our synchronization threads.
671 * When asking the question 'is any data accessible at all', then a simple
672 * test against (RDHARD|RDSOFT|RDSLAVE) gives you the answer. If any of
673 * these bits are set the object can be read with certain caveats:
674 * RDHARD - no caveats. RDSOFT - authoritative but might not be synchronized.
675 * and RDSLAVE - not authoritative, has some data but it could be old or
678 * When both soft and hard mounts are available, data will be read and written
679 * via the soft mount only. But all might be in the cluster because
680 * background synchronization threads still need to do their work.
682 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode struct */
683 #define HAMMER2_CLUSTER_UNUSED2 0x00000002
684 #define HAMMER2_CLUSTER_LOCKED 0x00000004 /* cluster lks not recursive */
685 #define HAMMER2_CLUSTER_WRHARD 0x00000100 /* hard-mount can write */
686 #define HAMMER2_CLUSTER_RDHARD 0x00000200 /* hard-mount can read */
687 #define HAMMER2_CLUSTER_UNHARD 0x00000400 /* unsynchronized masters */
688 #define HAMMER2_CLUSTER_NOHARD 0x00000800 /* no masters visible */
689 #define HAMMER2_CLUSTER_WRSOFT 0x00001000 /* soft-mount can write */
690 #define HAMMER2_CLUSTER_RDSOFT 0x00002000 /* soft-mount can read */
691 #define HAMMER2_CLUSTER_UNSOFT 0x00004000 /* unsynchronized slaves */
692 #define HAMMER2_CLUSTER_NOSOFT 0x00008000 /* no slaves visible */
693 #define HAMMER2_CLUSTER_MSYNCED 0x00010000 /* all masters synchronized */
694 #define HAMMER2_CLUSTER_SSYNCED 0x00020000 /* known slaves synchronized */
696 #define HAMMER2_CLUSTER_ANYDATA ( HAMMER2_CLUSTER_RDHARD | \
697 HAMMER2_CLUSTER_RDSOFT | \
698 HAMMER2_CLUSTER_RDSLAVE)
700 #define HAMMER2_CLUSTER_RDOK ( HAMMER2_CLUSTER_RDHARD | \
701 HAMMER2_CLUSTER_RDSOFT)
703 #define HAMMER2_CLUSTER_WROK ( HAMMER2_CLUSTER_WRHARD | \
704 HAMMER2_CLUSTER_WRSOFT)
706 #define HAMMER2_CLUSTER_ZFLAGS ( HAMMER2_CLUSTER_WRHARD | \
707 HAMMER2_CLUSTER_RDHARD | \
708 HAMMER2_CLUSTER_WRSOFT | \
709 HAMMER2_CLUSTER_RDSOFT | \
710 HAMMER2_CLUSTER_MSYNCED | \
711 HAMMER2_CLUSTER_SSYNCED)
714 * Helper functions (cluster must be locked for flags to be valid).
718 hammer2_cluster_rdok(hammer2_cluster_t *cluster)
720 return (cluster->flags & HAMMER2_CLUSTER_RDOK);
725 hammer2_cluster_wrok(hammer2_cluster_t *cluster)
727 return (cluster->flags & HAMMER2_CLUSTER_WROK);
730 RB_HEAD(hammer2_inode_tree, hammer2_inode); /* ip->rbnode */
731 TAILQ_HEAD(inoq_head, hammer2_inode); /* ip->entry */
732 TAILQ_HEAD(depq_head, hammer2_depend); /* depend->entry */
734 struct hammer2_depend {
735 TAILQ_ENTRY(hammer2_depend) entry;
736 struct inoq_head sideq;
742 typedef struct hammer2_depend hammer2_depend_t;
747 * NOTE: The inode-embedded cluster is never used directly for I/O (since
748 * it may be shared). Instead it will be replicated-in and synchronized
749 * back out if changed.
751 struct hammer2_inode {
752 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
753 TAILQ_ENTRY(hammer2_inode) entry; /* SYNCQ/SIDEQ */
754 hammer2_depend_t *depend; /* non-NULL if SIDEQ */
755 hammer2_depend_t depend_static; /* (in-place allocation) */
756 hammer2_mtx_t lock; /* inode lock */
757 hammer2_mtx_t truncate_lock; /* prevent truncates */
758 struct hammer2_pfs *pmp; /* PFS mount */
760 struct spinlock cluster_spin; /* update cluster */
761 hammer2_cluster_t cluster;
762 struct lockf advlock;
764 u_int refs; /* +vpref, +flushref */
765 uint8_t comp_heuristic;
766 hammer2_inode_meta_t meta; /* copy of meta-data */
770 typedef struct hammer2_inode hammer2_inode_t;
773 * MODIFIED - Inode is in a modified state, ip->meta may have changes.
774 * RESIZED - Inode truncated (any) or inode extended beyond
777 * SYNCQ - Inode is included in the current filesystem sync. The
778 * DELETING and CREATING flags will be acted upon.
780 * SIDEQ - Inode has likely been disconnected from the vnode topology
781 * and so is not visible to the vnode-based filesystem syncer
782 * code, but is dirty and must be included in the next
783 * filesystem sync. These inodes are moved to the SYNCQ at
784 * the time the sync occurs.
786 * Inodes are not placed on this queue simply because they have
787 * become dirty, if a vnode is attached.
789 * DELETING - Inode is flagged for deletion during the next filesystem
790 * sync. That is, the inode's chain is currently connected
791 * and must be deleting during the current or next fs sync.
793 * CREATING - Inode is flagged for creation during the next filesystem
794 * sync. That is, the inode's chain topology exists (so
795 * kernel buffer flushes can occur), but is currently
796 * disconnected and must be inserted during the current or
797 * next fs sync. If the DELETING flag is also set, the
798 * topology can be thrown away instead.
800 * If an inode that is already part of the current filesystem sync is
801 * modified by the frontend, including by buffer flushes, the inode lock
802 * code detects the SYNCQ flag and moves the inode to the head of the
803 * flush-in-progress, then blocks until the flush has gotten past it.
805 #define HAMMER2_INODE_MODIFIED 0x0001
806 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
807 #define HAMMER2_INODE_RENAME_INPROG 0x0004
808 #define HAMMER2_INODE_ONRBTREE 0x0008
809 #define HAMMER2_INODE_RESIZED 0x0010 /* requires inode_fsync */
810 #define HAMMER2_INODE_UNUSED0020 0x0020
811 #define HAMMER2_INODE_ISUNLINKED 0x0040
812 #define HAMMER2_INODE_METAGOOD 0x0080 /* inode meta-data good */
813 #define HAMMER2_INODE_SIDEQ 0x0100 /* on side processing queue */
814 #define HAMMER2_INODE_NOSIDEQ 0x0200 /* disable sideq operation */
815 #define HAMMER2_INODE_DIRTYDATA 0x0400 /* interlocks inode flush */
816 #define HAMMER2_INODE_SYNCQ 0x0800 /* sync interlock, sequenced */
817 #define HAMMER2_INODE_DELETING 0x1000 /* sync interlock, chain topo */
818 #define HAMMER2_INODE_CREATING 0x2000 /* sync interlock, chain topo */
819 #define HAMMER2_INODE_SYNCQ_WAKEUP 0x4000 /* sync interlock wakeup */
820 #define HAMMER2_INODE_SYNCQ_PASS2 0x8000 /* force retry delay */
822 #define HAMMER2_INODE_DIRTY (HAMMER2_INODE_MODIFIED | \
823 HAMMER2_INODE_DIRTYDATA | \
824 HAMMER2_INODE_DELETING | \
825 HAMMER2_INODE_CREATING)
827 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
828 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
832 * Transaction management sub-structure under hammer2_pfs
834 struct hammer2_trans {
839 typedef struct hammer2_trans hammer2_trans_t;
841 #define HAMMER2_TRANS_ISFLUSH 0x80000000 /* flush code */
842 #define HAMMER2_TRANS_BUFCACHE 0x40000000 /* bio strategy */
843 #define HAMMER2_TRANS_SIDEQ 0x20000000 /* run sideq */
844 #define HAMMER2_TRANS_UNUSED10 0x10000000
845 #define HAMMER2_TRANS_WAITING 0x08000000 /* someone waiting */
846 #define HAMMER2_TRANS_RESCAN 0x04000000 /* rescan sideq */
847 #define HAMMER2_TRANS_MASK 0x00FFFFFF /* count mask */
849 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
850 #define HAMMER2_FREEMAP_HEUR_TYPES 8
851 #define HAMMER2_FREEMAP_HEUR_SIZE (HAMMER2_FREEMAP_HEUR_NRADIX * \
852 HAMMER2_FREEMAP_HEUR_TYPES)
854 #define HAMMER2_DEDUP_HEUR_SIZE (65536 * 4)
855 #define HAMMER2_DEDUP_HEUR_MASK (HAMMER2_DEDUP_HEUR_SIZE - 1)
857 #define HAMMER2_FLUSH_TOP 0x0001
858 #define HAMMER2_FLUSH_ALL 0x0002
859 #define HAMMER2_FLUSH_INODE_STOP 0x0004 /* stop at sub-inode */
860 #define HAMMER2_FLUSH_FSSYNC 0x0008 /* part of filesystem sync */
864 * Hammer2 support thread element.
866 * Potentially many support threads can hang off of hammer2, primarily
867 * off the hammer2_pfs structure. Typically:
869 * td x Nodes A synchronization thread for each node.
870 * td x Nodes x workers Worker threads for frontend operations.
871 * td x 1 Bioq thread for logical buffer writes.
873 * In addition, the synchronization thread(s) associated with the
874 * super-root PFS (spmp) for a node is responsible for automatic bulkfree
877 struct hammer2_thread {
878 struct hammer2_pfs *pmp;
879 struct hammer2_dev *hmp;
880 hammer2_xop_list_t xopq;
884 int clindex; /* cluster element index */
886 char *scratch; /* MAXPHYS */
889 typedef struct hammer2_thread hammer2_thread_t;
891 #define HAMMER2_THREAD_UNMOUNTING 0x0001 /* unmount request */
892 #define HAMMER2_THREAD_DEV 0x0002 /* related to dev, not pfs */
893 #define HAMMER2_THREAD_WAITING 0x0004 /* thread in idle tsleep */
894 #define HAMMER2_THREAD_REMASTER 0x0008 /* remaster request */
895 #define HAMMER2_THREAD_STOP 0x0010 /* exit request */
896 #define HAMMER2_THREAD_FREEZE 0x0020 /* force idle */
897 #define HAMMER2_THREAD_FROZEN 0x0040 /* thread is frozen */
898 #define HAMMER2_THREAD_XOPQ 0x0080 /* work pending */
899 #define HAMMER2_THREAD_STOPPED 0x0100 /* thread has stopped */
900 #define HAMMER2_THREAD_UNFREEZE 0x0200
902 #define HAMMER2_THREAD_WAKEUP_MASK (HAMMER2_THREAD_UNMOUNTING | \
903 HAMMER2_THREAD_REMASTER | \
904 HAMMER2_THREAD_STOP | \
905 HAMMER2_THREAD_FREEZE | \
909 * Support structure for dedup heuristic.
911 struct hammer2_dedup {
912 hammer2_off_t data_off;
918 typedef struct hammer2_dedup hammer2_dedup_t;
921 * hammer2_xop - container for VOP/XOP operation (allocated, not on stack).
923 * This structure is used to distribute a VOP operation across multiple
924 * nodes. It provides a rendezvous for concurrent node execution and
925 * can be detached from the frontend operation to allow the frontend to
928 * This structure also sequences operations on up to three inodes.
930 typedef void (*hammer2_xop_func_t)(union hammer2_xop *xop, void *scratch,
933 struct hammer2_xop_desc {
934 hammer2_xop_func_t storage_func; /* local storage function */
935 hammer2_xop_func_t dmsg_dispatch; /* dmsg dispatch function */
936 hammer2_xop_func_t dmsg_process; /* dmsg processing function */
940 typedef struct hammer2_xop_desc hammer2_xop_desc_t;
942 struct hammer2_xop_fifo {
943 TAILQ_ENTRY(hammer2_xop_head) entry;
944 hammer2_chain_t *array[HAMMER2_XOPFIFO];
945 int errors[HAMMER2_XOPFIFO];
949 hammer2_thread_t *thr;
952 typedef struct hammer2_xop_fifo hammer2_xop_fifo_t;
954 #define HAMMER2_XOP_FIFO_RUN 0x0001
955 #define HAMMER2_XOP_FIFO_STALL 0x0002
957 struct hammer2_xop_head {
958 hammer2_xop_desc_t *desc;
960 struct hammer2_inode *ip1;
961 struct hammer2_inode *ip2;
962 struct hammer2_inode *ip3;
968 hammer2_key_t collect_key;
973 hammer2_xop_fifo_t collect[HAMMER2_MAXCLUSTER];
974 hammer2_cluster_t cluster; /* help collections */
975 hammer2_io_t *focus_dio;
978 typedef struct hammer2_xop_head hammer2_xop_head_t;
980 struct hammer2_xop_ipcluster {
981 hammer2_xop_head_t head;
984 struct hammer2_xop_strategy {
985 hammer2_xop_head_t head;
992 struct hammer2_xop_readdir {
993 hammer2_xop_head_t head;
997 struct hammer2_xop_nresolve {
998 hammer2_xop_head_t head;
999 hammer2_key_t lhc; /* if name is NULL used lhc */
1002 struct hammer2_xop_unlink {
1003 hammer2_xop_head_t head;
1008 #define H2DOPERM_PERMANENT 0x01
1009 #define H2DOPERM_FORCE 0x02
1010 #define H2DOPERM_IGNINO 0x04
1012 struct hammer2_xop_nrename {
1013 hammer2_xop_head_t head;
1018 struct hammer2_xop_scanlhc {
1019 hammer2_xop_head_t head;
1023 struct hammer2_xop_scanall {
1024 hammer2_xop_head_t head;
1025 hammer2_key_t key_beg; /* inclusive */
1026 hammer2_key_t key_end; /* inclusive */
1031 struct hammer2_xop_lookup {
1032 hammer2_xop_head_t head;
1036 struct hammer2_xop_mkdirent {
1037 hammer2_xop_head_t head;
1038 hammer2_dirent_head_t dirent;
1042 struct hammer2_xop_create {
1043 hammer2_xop_head_t head;
1044 hammer2_inode_meta_t meta; /* initial metadata */
1049 struct hammer2_xop_destroy {
1050 hammer2_xop_head_t head;
1053 struct hammer2_xop_fsync {
1054 hammer2_xop_head_t head;
1055 hammer2_inode_meta_t meta;
1056 hammer2_off_t osize;
1058 int clear_directdata;
1061 struct hammer2_xop_unlinkall {
1062 hammer2_xop_head_t head;
1063 hammer2_key_t key_beg;
1064 hammer2_key_t key_end;
1067 struct hammer2_xop_connect {
1068 hammer2_xop_head_t head;
1072 struct hammer2_xop_flush {
1073 hammer2_xop_head_t head;
1076 typedef struct hammer2_xop_readdir hammer2_xop_readdir_t;
1077 typedef struct hammer2_xop_nresolve hammer2_xop_nresolve_t;
1078 typedef struct hammer2_xop_unlink hammer2_xop_unlink_t;
1079 typedef struct hammer2_xop_nrename hammer2_xop_nrename_t;
1080 typedef struct hammer2_xop_ipcluster hammer2_xop_ipcluster_t;
1081 typedef struct hammer2_xop_strategy hammer2_xop_strategy_t;
1082 typedef struct hammer2_xop_mkdirent hammer2_xop_mkdirent_t;
1083 typedef struct hammer2_xop_create hammer2_xop_create_t;
1084 typedef struct hammer2_xop_destroy hammer2_xop_destroy_t;
1085 typedef struct hammer2_xop_fsync hammer2_xop_fsync_t;
1086 typedef struct hammer2_xop_unlinkall hammer2_xop_unlinkall_t;
1087 typedef struct hammer2_xop_scanlhc hammer2_xop_scanlhc_t;
1088 typedef struct hammer2_xop_scanall hammer2_xop_scanall_t;
1089 typedef struct hammer2_xop_lookup hammer2_xop_lookup_t;
1090 typedef struct hammer2_xop_connect hammer2_xop_connect_t;
1091 typedef struct hammer2_xop_flush hammer2_xop_flush_t;
1094 hammer2_xop_head_t head;
1095 hammer2_xop_ipcluster_t xop_ipcluster;
1096 hammer2_xop_readdir_t xop_readdir;
1097 hammer2_xop_nresolve_t xop_nresolve;
1098 hammer2_xop_unlink_t xop_unlink;
1099 hammer2_xop_nrename_t xop_nrename;
1100 hammer2_xop_strategy_t xop_strategy;
1101 hammer2_xop_mkdirent_t xop_mkdirent;
1102 hammer2_xop_create_t xop_create;
1103 hammer2_xop_destroy_t xop_destroy;
1104 hammer2_xop_fsync_t xop_fsync;
1105 hammer2_xop_unlinkall_t xop_unlinkall;
1106 hammer2_xop_scanlhc_t xop_scanlhc;
1107 hammer2_xop_scanall_t xop_scanall;
1108 hammer2_xop_lookup_t xop_lookup;
1109 hammer2_xop_flush_t xop_flush;
1110 hammer2_xop_connect_t xop_connect;
1113 typedef union hammer2_xop hammer2_xop_t;
1116 * hammer2_xop_group - Manage XOP support threads.
1118 struct hammer2_xop_group {
1119 hammer2_thread_t thrs[HAMMER2_MAXCLUSTER];
1122 typedef struct hammer2_xop_group hammer2_xop_group_t;
1125 * flags to hammer2_xop_collect()
1127 #define HAMMER2_XOP_COLLECT_NOWAIT 0x00000001
1128 #define HAMMER2_XOP_COLLECT_WAITALL 0x00000002
1131 * flags to hammer2_xop_alloc()
1133 * MODIFYING - This is a modifying transaction, allocate a mtid.
1134 * RECURSE - Recurse top-level inode (for root flushes)
1136 #define HAMMER2_XOP_MODIFYING 0x00000001
1137 #define HAMMER2_XOP_STRATEGY 0x00000002
1138 #define HAMMER2_XOP_INODE_STOP 0x00000004
1139 #define HAMMER2_XOP_VOLHDR 0x00000008
1140 #define HAMMER2_XOP_FSSYNC 0x00000010
1141 #define HAMMER2_XOP_IROOT 0x00000020
1144 * Global (per partition) management structure, represents a hard block
1145 * device. Typically referenced by hammer2_chain structures when applicable.
1146 * Typically not used for network-managed elements.
1148 * Note that a single hammer2_dev can be indirectly tied to multiple system
1149 * mount points. There is no direct relationship. System mounts are
1150 * per-cluster-id, not per-block-device, and a single hard mount might contain
1151 * many PFSs and those PFSs might combine together in various ways to form
1152 * the set of available clusters.
1154 struct hammer2_dev {
1155 struct vnode *devvp; /* device vnode */
1156 int ronly; /* read-only mount */
1157 int mount_count; /* number of actively mounted PFSs */
1158 TAILQ_ENTRY(hammer2_dev) mntentry; /* hammer2_mntlist */
1160 struct malloc_type *mchain;
1163 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
1164 struct spinlock io_spin; /* iotree, iolruq access */
1165 struct hammer2_io_tree iotree;
1167 int freemap_relaxed;
1168 hammer2_chain_t vchain; /* anchor chain (topology) */
1169 hammer2_chain_t fchain; /* anchor chain (freemap) */
1170 struct spinlock list_spin;
1171 struct hammer2_pfs *spmp; /* super-root pmp for transactions */
1172 struct lock vollk; /* lockmgr lock */
1173 struct lock bulklk; /* bulkfree operation lock */
1174 struct lock bflock; /* bulk-free manual function lock */
1175 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR_SIZE];
1176 hammer2_dedup_t heur_dedup[HAMMER2_DEDUP_HEUR_SIZE];
1177 int volhdrno; /* last volhdrno written */
1178 uint32_t hflags; /* HMNT2 flags applicable to device */
1179 hammer2_off_t free_reserved; /* nominal free reserved */
1180 hammer2_thread_t bfthr; /* bulk-free thread */
1181 char devrepname[64]; /* for kprintf */
1182 hammer2_ioc_bulkfree_t bflast; /* stats for last bulkfree run */
1183 hammer2_volume_data_t voldata;
1184 hammer2_volume_data_t volsync; /* synchronized voldata */
1187 typedef struct hammer2_dev hammer2_dev_t;
1190 * Helper functions (cluster must be locked for flags to be valid).
1194 hammer2_chain_rdok(hammer2_chain_t *chain)
1196 return (chain->error == 0);
1201 hammer2_chain_wrok(hammer2_chain_t *chain)
1203 return (chain->error == 0 && chain->hmp->ronly == 0);
1207 * Per-cluster management structure. This structure will be tied to a
1208 * system mount point if the system is mounting the PFS, but is also used
1209 * to manage clusters encountered during the super-root scan or received
1210 * via LNK_SPANs that might not be mounted.
1212 * This structure is also used to represent the super-root that hangs off
1213 * of a hard mount point. The super-root is not really a cluster element.
1214 * In this case the spmp_hmp field will be non-NULL. It's just easier to do
1215 * this than to special case super-root manipulation in the hammer2_chain*
1216 * code as being only hammer2_dev-related.
1218 * pfs_mode and pfs_nmasters are rollup fields which critically describes
1219 * how elements of the cluster act on the cluster. pfs_mode is only applicable
1220 * when a PFS is mounted by the system. pfs_nmasters is our best guess as to
1221 * how many masters have been configured for a cluster and is always
1222 * applicable. pfs_types[] is an array with 1:1 correspondance to the
1223 * iroot cluster and describes the PFS types of the nodes making up the
1226 * WARNING! Portions of this structure have deferred initialization. In
1227 * particular, if not mounted there will be no wthread.
1228 * umounted network PFSs will also be missing iroot and numerous
1229 * other fields will not be initialized prior to mount.
1231 * Synchronization threads are chain-specific and only applicable
1232 * to local hard PFS entries. A hammer2_pfs structure may contain
1233 * more than one when multiple hard PFSs are present on the local
1234 * machine which require synchronization monitoring. Most PFSs
1235 * (such as snapshots) are 1xMASTER PFSs which do not need a
1236 * synchronization thread.
1238 * WARNING! The chains making up pfs->iroot's cluster are accounted for in
1239 * hammer2_dev->mount_count when the pfs is associated with a mount
1242 struct hammer2_pfs {
1244 TAILQ_ENTRY(hammer2_pfs) mntentry; /* hammer2_pfslist */
1246 hammer2_dev_t *spmp_hmp; /* only if super-root pmp */
1247 hammer2_dev_t *force_local; /* only if 'local' mount */
1248 hammer2_inode_t *iroot; /* PFS root inode */
1249 uint8_t pfs_types[HAMMER2_MAXCLUSTER];
1250 char *pfs_names[HAMMER2_MAXCLUSTER];
1251 hammer2_dev_t *pfs_hmps[HAMMER2_MAXCLUSTER];
1252 hammer2_blockset_t pfs_iroot_blocksets[HAMMER2_MAXCLUSTER];
1253 hammer2_trans_t trans;
1254 struct lock lock; /* PFS lock for certain ops */
1255 struct lock lock_nlink; /* rename and nlink lock */
1256 struct netexport export; /* nfs export */
1258 int ronly; /* read-only mount */
1259 int hflags; /* pfs-specific mount flags */
1260 struct malloc_type *minode;
1261 struct malloc_type *mmsg;
1262 struct spinlock inum_spin; /* inumber lookup */
1263 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
1264 long inum_count; /* #of inodes in inum_tree */
1265 struct spinlock lru_spin; /* inumber lookup */
1266 struct hammer2_chain_list lru_list; /* basis for LRU tests */
1267 int lru_count; /* #of chains on LRU */
1269 hammer2_tid_t modify_tid; /* modify transaction id */
1270 hammer2_tid_t inode_tid; /* inode allocator */
1271 uint8_t pfs_nmasters; /* total masters */
1272 uint8_t pfs_mode; /* operating mode PFSMODE */
1275 int free_ticks; /* free_* calculations */
1277 hammer2_off_t free_reserved;
1278 hammer2_off_t free_nominal;
1279 uint32_t inmem_dirty_chains;
1280 int count_lwinprog; /* logical write in prog */
1281 struct spinlock list_spin;
1282 struct inoq_head syncq; /* SYNCQ flagged inodes */
1283 struct depq_head depq; /* SIDEQ flagged inodes */
1284 long sideq_count; /* total inodes on depq */
1285 hammer2_thread_t sync_thrs[HAMMER2_MAXCLUSTER];
1286 uint32_t cluster_flags; /* cached cluster flags */
1287 int has_xop_threads;
1288 struct spinlock xop_spin; /* xop sequencer */
1289 hammer2_xop_group_t xop_groups[HAMMER2_XOPGROUPS];
1292 typedef struct hammer2_pfs hammer2_pfs_t;
1294 TAILQ_HEAD(hammer2_pfslist, hammer2_pfs);
1296 #define HAMMER2_PMPF_SPMP 0x00000001
1299 * NOTE: The LRU list contains at least all the chains with refs == 0
1300 * that can be recycled, and may contain additional chains which
1303 #define HAMMER2_LRU_LIMIT 4096
1305 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
1306 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
1308 #define HAMMER2_LWINPROG_WAITING 0x80000000
1309 #define HAMMER2_LWINPROG_WAITING0 0x40000000
1310 #define HAMMER2_LWINPROG_MASK 0x3FFFFFFF
1313 * hammer2_cluster_check
1315 #define HAMMER2_CHECK_NULL 0x00000001
1320 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
1321 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
1324 #if defined(_KERNEL)
1326 MALLOC_DECLARE(M_HAMMER2);
1328 #define ITOV(ip) ((ip)->vp)
1331 * Currently locked chains retain the locked buffer cache buffer for
1332 * indirect blocks, and indirect blocks can be one of two sizes. The
1333 * device buffer has to match the case to avoid deadlocking recursive
1334 * chains that might otherwise try to access different offsets within
1335 * the same device buffer.
1339 hammer2_devblkradix(int radix)
1342 if (radix <= HAMMER2_LBUFRADIX) {
1343 return (HAMMER2_LBUFRADIX);
1345 return (HAMMER2_PBUFRADIX);
1348 return (HAMMER2_PBUFRADIX);
1352 * XXX almost time to remove this. DIO uses PBUFSIZE exclusively now.
1356 hammer2_devblksize(size_t bytes)
1359 if (bytes <= HAMMER2_LBUFSIZE) {
1360 return(HAMMER2_LBUFSIZE);
1362 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
1363 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
1364 return (HAMMER2_PBUFSIZE);
1367 return (HAMMER2_PBUFSIZE);
1373 MPTOPMP(struct mount *mp)
1375 return ((hammer2_pfs_t *)mp->mnt_data);
1378 #define HAMMER2_DEDUP_FRAG (HAMMER2_PBUFSIZE / 64)
1379 #define HAMMER2_DEDUP_FRAGRADIX (HAMMER2_PBUFRADIX - 6)
1383 hammer2_dedup_mask(hammer2_io_t *dio, hammer2_off_t data_off, u_int bytes)
1389 bbeg = (int)((data_off & ~HAMMER2_OFF_MASK_RADIX) - dio->pbase) >>
1390 HAMMER2_DEDUP_FRAGRADIX;
1391 bits = (int)((bytes + (HAMMER2_DEDUP_FRAG - 1)) >>
1392 HAMMER2_DEDUP_FRAGRADIX);
1393 mask = ((uint64_t)1 << bbeg) - 1;
1394 if (bbeg + bits == 64)
1395 mask = (uint64_t)-1;
1397 mask = ((uint64_t)1 << (bbeg + bits)) - 1;
1399 mask &= ~(((uint64_t)1 << bbeg) - 1);
1406 hammer2_error_to_errno(int error)
1409 if (error & HAMMER2_ERROR_EIO)
1411 else if (error & HAMMER2_ERROR_CHECK)
1413 else if (error & HAMMER2_ERROR_ABORTED)
1415 else if (error & HAMMER2_ERROR_BADBREF)
1417 else if (error & HAMMER2_ERROR_ENOSPC)
1419 else if (error & HAMMER2_ERROR_ENOENT)
1421 else if (error & HAMMER2_ERROR_ENOTEMPTY)
1423 else if (error & HAMMER2_ERROR_EAGAIN)
1425 else if (error & HAMMER2_ERROR_ENOTDIR)
1427 else if (error & HAMMER2_ERROR_EISDIR)
1429 else if (error & HAMMER2_ERROR_EINPROGRESS)
1430 error = EINPROGRESS;
1431 else if (error & HAMMER2_ERROR_EEXIST)
1441 hammer2_errno_to_error(int error)
1447 return HAMMER2_ERROR_EIO;
1450 return HAMMER2_ERROR_EINVAL;
1455 extern struct vop_ops hammer2_vnode_vops;
1456 extern struct vop_ops hammer2_spec_vops;
1457 extern struct vop_ops hammer2_fifo_vops;
1458 extern struct hammer2_pfslist hammer2_pfslist;
1459 extern struct lock hammer2_mntlk;
1462 extern int hammer2_debug;
1463 extern long hammer2_debug_inode;
1464 extern int hammer2_cluster_meta_read;
1465 extern int hammer2_cluster_data_read;
1466 extern int hammer2_cluster_write;
1467 extern int hammer2_dedup_enable;
1468 extern int hammer2_always_compress;
1469 extern int hammer2_inval_enable;
1470 extern int hammer2_flush_pipe;
1471 extern int hammer2_dio_count;
1472 extern int hammer2_dio_limit;
1473 extern int hammer2_bulkfree_tps;
1474 extern int hammer2_worker_rmask;
1475 extern long hammer2_chain_allocs;
1476 extern long hammer2_chain_frees;
1477 extern long hammer2_limit_dirty_chains;
1478 extern long hammer2_limit_dirty_inodes;
1479 extern long hammer2_count_modified_chains;
1480 extern long hammer2_iod_invals;
1481 extern long hammer2_iod_file_read;
1482 extern long hammer2_iod_meta_read;
1483 extern long hammer2_iod_indr_read;
1484 extern long hammer2_iod_fmap_read;
1485 extern long hammer2_iod_volu_read;
1486 extern long hammer2_iod_file_write;
1487 extern long hammer2_iod_file_wembed;
1488 extern long hammer2_iod_file_wzero;
1489 extern long hammer2_iod_file_wdedup;
1490 extern long hammer2_iod_meta_write;
1491 extern long hammer2_iod_indr_write;
1492 extern long hammer2_iod_fmap_write;
1493 extern long hammer2_iod_volu_write;
1495 extern long hammer2_process_xxhash64;
1496 extern long hammer2_process_icrc32;
1498 extern struct objcache *cache_buffer_read;
1499 extern struct objcache *cache_buffer_write;
1500 extern struct objcache *cache_xops;
1505 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
1506 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
1508 int hammer2_signal_check(time_t *timep);
1509 const char *hammer2_error_str(int error);
1510 const char *hammer2_bref_type_str(hammer2_blockref_t *bref);
1512 void hammer2_inode_delayed_sideq(hammer2_inode_t *ip);
1513 void hammer2_inode_lock(hammer2_inode_t *ip, int how);
1514 void hammer2_inode_lock4(hammer2_inode_t *ip1, hammer2_inode_t *ip2,
1515 hammer2_inode_t *ip3, hammer2_inode_t *ip4);
1516 void hammer2_inode_unlock(hammer2_inode_t *ip);
1517 void hammer2_inode_depend(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
1518 hammer2_chain_t *hammer2_inode_chain(hammer2_inode_t *ip, int clindex, int how);
1519 hammer2_chain_t *hammer2_inode_chain_and_parent(hammer2_inode_t *ip,
1520 int clindex, hammer2_chain_t **parentp, int how);
1521 hammer2_mtx_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
1522 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip,
1523 hammer2_mtx_state_t ostate);
1524 int hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
1525 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int);
1527 void hammer2_dev_exlock(hammer2_dev_t *hmp);
1528 void hammer2_dev_shlock(hammer2_dev_t *hmp);
1529 void hammer2_dev_unlock(hammer2_dev_t *hmp);
1531 int hammer2_get_dtype(uint8_t type);
1532 int hammer2_get_vtype(uint8_t type);
1533 uint8_t hammer2_get_obj_type(enum vtype vtype);
1534 void hammer2_time_to_timespec(uint64_t xtime, struct timespec *ts);
1535 uint64_t hammer2_timespec_to_time(const struct timespec *ts);
1536 uint32_t hammer2_to_unix_xid(const uuid_t *uuid);
1537 void hammer2_guid_to_uuid(uuid_t *uuid, uint32_t guid);
1538 void hammer2_trans_manage_init(hammer2_pfs_t *pmp);
1540 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
1541 int hammer2_getradix(size_t bytes);
1543 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
1544 hammer2_key_t *lbasep, hammer2_key_t *leofp);
1545 int hammer2_calc_physical(hammer2_inode_t *ip, hammer2_key_t lbase);
1546 void hammer2_update_time(uint64_t *timep);
1547 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
1552 struct vnode *hammer2_igetv(hammer2_inode_t *ip, int *errorp);
1553 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfs_t *pmp,
1554 hammer2_tid_t inum);
1555 hammer2_inode_t *hammer2_inode_get(hammer2_pfs_t *pmp,
1556 hammer2_xop_head_t *xop, hammer2_tid_t inum, int idx);
1557 void hammer2_inode_free(hammer2_inode_t *ip);
1558 void hammer2_inode_ref(hammer2_inode_t *ip);
1559 void hammer2_inode_drop(hammer2_inode_t *ip);
1560 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
1561 hammer2_cluster_t *cluster);
1562 void hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1564 void hammer2_inode_modify(hammer2_inode_t *ip);
1565 void hammer2_inode_run_sideq(hammer2_pfs_t *pmp, int doall);
1567 hammer2_inode_t *hammer2_inode_create_normal(hammer2_inode_t *pip,
1568 struct vattr *vap, struct ucred *cred,
1569 hammer2_key_t inum, int *errorp);
1570 hammer2_inode_t *hammer2_inode_create_pfs(hammer2_pfs_t *spmp,
1571 const uint8_t *name, size_t name_len,
1573 int hammer2_inode_chain_ins(hammer2_inode_t *ip);
1574 int hammer2_inode_chain_des(hammer2_inode_t *ip);
1575 int hammer2_inode_chain_sync(hammer2_inode_t *ip);
1576 int hammer2_inode_chain_flush(hammer2_inode_t *ip, int flags);
1577 int hammer2_inode_unlink_finisher(hammer2_inode_t *ip, int isopen);
1578 int hammer2_dirent_create(hammer2_inode_t *dip, const char *name,
1579 size_t name_len, hammer2_key_t inum, uint8_t type);
1584 void hammer2_voldata_lock(hammer2_dev_t *hmp);
1585 void hammer2_voldata_unlock(hammer2_dev_t *hmp);
1586 void hammer2_voldata_modify(hammer2_dev_t *hmp);
1587 hammer2_chain_t *hammer2_chain_alloc(hammer2_dev_t *hmp,
1589 hammer2_blockref_t *bref);
1590 void hammer2_chain_core_init(hammer2_chain_t *chain);
1591 void hammer2_chain_ref(hammer2_chain_t *chain);
1592 void hammer2_chain_ref_hold(hammer2_chain_t *chain);
1593 void hammer2_chain_drop(hammer2_chain_t *chain);
1594 void hammer2_chain_drop_unhold(hammer2_chain_t *chain);
1595 void hammer2_chain_unhold(hammer2_chain_t *chain);
1596 void hammer2_chain_rehold(hammer2_chain_t *chain);
1597 int hammer2_chain_lock(hammer2_chain_t *chain, int how);
1598 void hammer2_chain_lock_unhold(hammer2_chain_t *chain, int how);
1599 void hammer2_chain_load_data(hammer2_chain_t *chain);
1600 const hammer2_media_data_t *hammer2_chain_rdata(hammer2_chain_t *chain);
1601 hammer2_media_data_t *hammer2_chain_wdata(hammer2_chain_t *chain);
1603 int hammer2_chain_inode_find(hammer2_pfs_t *pmp, hammer2_key_t inum,
1604 int clindex, int flags,
1605 hammer2_chain_t **parentp,
1606 hammer2_chain_t **chainp);
1607 int hammer2_chain_modify(hammer2_chain_t *chain, hammer2_tid_t mtid,
1608 hammer2_off_t dedup_off, int flags);
1609 int hammer2_chain_modify_ip(hammer2_inode_t *ip, hammer2_chain_t *chain,
1610 hammer2_tid_t mtid, int flags);
1611 int hammer2_chain_resize(hammer2_chain_t *chain,
1612 hammer2_tid_t mtid, hammer2_off_t dedup_off,
1613 int nradix, int flags);
1614 void hammer2_chain_unlock(hammer2_chain_t *chain);
1615 void hammer2_chain_unlock_hold(hammer2_chain_t *chain);
1616 void hammer2_chain_wait(hammer2_chain_t *chain);
1617 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
1618 hammer2_blockref_t *bref, int how);
1619 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
1620 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
1621 hammer2_chain_t *hammer2_chain_getparent(hammer2_chain_t *chain, int flags);
1622 hammer2_chain_t *hammer2_chain_repparent(hammer2_chain_t **chainp, int flags);
1623 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
1624 hammer2_key_t *key_nextp,
1625 hammer2_key_t key_beg, hammer2_key_t key_end,
1626 int *errorp, int flags);
1627 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
1628 hammer2_chain_t *chain,
1629 hammer2_key_t *key_nextp,
1630 hammer2_key_t key_beg, hammer2_key_t key_end,
1631 int *errorp, int flags);
1632 int hammer2_chain_scan(hammer2_chain_t *parent,
1633 hammer2_chain_t **chainp,
1634 hammer2_blockref_t *bref,
1635 int *firstp, int flags);
1637 int hammer2_chain_create(hammer2_chain_t **parentp, hammer2_chain_t **chainp,
1638 hammer2_dev_t *hmp, hammer2_pfs_t *pmp,
1639 int methods, hammer2_key_t key, int keybits,
1640 int type, size_t bytes, hammer2_tid_t mtid,
1641 hammer2_off_t dedup_off, int flags);
1642 void hammer2_chain_rename(hammer2_chain_t **parentp,
1643 hammer2_chain_t *chain,
1644 hammer2_tid_t mtid, int flags);
1645 int hammer2_chain_delete(hammer2_chain_t *parent, hammer2_chain_t *chain,
1646 hammer2_tid_t mtid, int flags);
1647 int hammer2_chain_indirect_maintenance(hammer2_chain_t *parent,
1648 hammer2_chain_t *chain);
1649 void hammer2_chain_setflush(hammer2_chain_t *chain);
1650 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
1651 hammer2_blockref_t *base, int count);
1652 hammer2_chain_t *hammer2_chain_bulksnap(hammer2_dev_t *hmp);
1653 void hammer2_chain_bulkdrop(hammer2_chain_t *copy);
1655 void hammer2_chain_setcheck(hammer2_chain_t *chain, void *bdata);
1656 int hammer2_chain_testcheck(hammer2_chain_t *chain, void *bdata);
1657 int hammer2_chain_dirent_test(hammer2_chain_t *chain, const char *name,
1660 void hammer2_pfs_memory_wait(hammer2_pfs_t *pmp);
1661 void hammer2_pfs_memory_inc(hammer2_pfs_t *pmp);
1662 void hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp);
1664 void hammer2_base_delete(hammer2_chain_t *parent,
1665 hammer2_blockref_t *base, int count,
1666 hammer2_chain_t *chain,
1667 hammer2_blockref_t *obref);
1668 void hammer2_base_insert(hammer2_chain_t *parent,
1669 hammer2_blockref_t *base, int count,
1670 hammer2_chain_t *chain,
1671 hammer2_blockref_t *elm);
1676 int hammer2_flush(hammer2_chain_t *chain, int istop);
1677 void hammer2_delayed_flush(hammer2_chain_t *chain);
1682 void hammer2_trans_init(hammer2_pfs_t *pmp, uint32_t flags);
1683 void hammer2_trans_setflags(hammer2_pfs_t *pmp, uint32_t flags);
1684 void hammer2_trans_clearflags(hammer2_pfs_t *pmp, uint32_t flags);
1685 hammer2_tid_t hammer2_trans_sub(hammer2_pfs_t *pmp);
1686 void hammer2_trans_done(hammer2_pfs_t *pmp, uint32_t flags);
1687 hammer2_tid_t hammer2_trans_newinum(hammer2_pfs_t *pmp);
1688 void hammer2_trans_assert_strategy(hammer2_pfs_t *pmp);
1689 void hammer2_dedup_record(hammer2_chain_t *chain, hammer2_io_t *dio,
1695 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
1696 int fflag, struct ucred *cred);
1701 void hammer2_io_inval(hammer2_io_t *dio, hammer2_off_t data_off, u_int bytes);
1702 void hammer2_io_cleanup(hammer2_dev_t *hmp, struct hammer2_io_tree *tree);
1703 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
1704 void hammer2_io_bkvasync(hammer2_io_t *dio);
1705 void hammer2_io_dedup_set(hammer2_dev_t *hmp, hammer2_blockref_t *bref);
1706 void hammer2_io_dedup_delete(hammer2_dev_t *hmp, uint8_t btype,
1707 hammer2_off_t data_off, u_int bytes);
1708 void hammer2_io_dedup_assert(hammer2_dev_t *hmp, hammer2_off_t data_off,
1710 void hammer2_io_callback(struct bio *bio);
1711 int hammer2_io_new(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1712 hammer2_io_t **diop);
1713 int hammer2_io_newnz(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1714 hammer2_io_t **diop);
1715 int _hammer2_io_bread(hammer2_dev_t *hmp, int btype, off_t lbase, int lsize,
1716 hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1717 void hammer2_io_setdirty(hammer2_io_t *dio);
1719 hammer2_io_t *_hammer2_io_getblk(hammer2_dev_t *hmp, int btype, off_t lbase,
1720 int lsize, int op HAMMER2_IO_DEBUG_ARGS);
1721 hammer2_io_t *_hammer2_io_getquick(hammer2_dev_t *hmp, off_t lbase,
1722 int lsize HAMMER2_IO_DEBUG_ARGS);
1723 void _hammer2_io_putblk(hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1724 int _hammer2_io_bwrite(hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1725 void _hammer2_io_bawrite(hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1726 void _hammer2_io_bdwrite(hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1727 void _hammer2_io_brelse(hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1728 void _hammer2_io_bqrelse(hammer2_io_t **diop HAMMER2_IO_DEBUG_ARGS);
1729 void _hammer2_io_ref(hammer2_io_t *dio HAMMER2_IO_DEBUG_ARGS);
1731 #ifndef HAMMER2_IO_DEBUG
1733 #define hammer2_io_getblk(hmp, btype, lbase, lsize, op) \
1734 _hammer2_io_getblk((hmp), (btype), (lbase), (lsize), (op))
1735 #define hammer2_io_getquick(hmp, lbase, lsize) \
1736 _hammer2_io_getquick((hmp), (lbase), (lsize))
1737 #define hammer2_io_putblk(diop) \
1738 _hammer2_io_putblk(diop)
1739 #define hammer2_io_bwrite(diop) \
1740 _hammer2_io_bwrite((diop))
1741 #define hammer2_io_bawrite(diop) \
1742 _hammer2_io_bawrite((diop))
1743 #define hammer2_io_bdwrite(diop) \
1744 _hammer2_io_bdwrite((diop))
1745 #define hammer2_io_brelse(diop) \
1746 _hammer2_io_brelse((diop))
1747 #define hammer2_io_bqrelse(diop) \
1748 _hammer2_io_bqrelse((diop))
1749 #define hammer2_io_ref(dio) \
1750 _hammer2_io_ref((dio))
1752 #define hammer2_io_bread(hmp, btype, lbase, lsize, diop) \
1753 _hammer2_io_bread((hmp), (btype), (lbase), (lsize), (diop))
1757 #define hammer2_io_getblk(hmp, btype, lbase, lsize, op) \
1758 _hammer2_io_getblk((hmp), (btype), (lbase), (lsize), (op), \
1761 #define hammer2_io_getquick(hmp, lbase, lsize) \
1762 _hammer2_io_getquick((hmp), (lbase), (lsize), __FILE__, __LINE__)
1764 #define hammer2_io_putblk(diop) \
1765 _hammer2_io_putblk(diop, __FILE__, __LINE__)
1767 #define hammer2_io_bwrite(diop) \
1768 _hammer2_io_bwrite((diop), __FILE__, __LINE__)
1769 #define hammer2_io_bawrite(diop) \
1770 _hammer2_io_bawrite((diop), __FILE__, __LINE__)
1771 #define hammer2_io_bdwrite(diop) \
1772 _hammer2_io_bdwrite((diop), __FILE__, __LINE__)
1773 #define hammer2_io_brelse(diop) \
1774 _hammer2_io_brelse((diop), __FILE__, __LINE__)
1775 #define hammer2_io_bqrelse(diop) \
1776 _hammer2_io_bqrelse((diop), __FILE__, __LINE__)
1777 #define hammer2_io_ref(dio) \
1778 _hammer2_io_ref((dio), __FILE__, __LINE__)
1780 #define hammer2_io_bread(hmp, btype, lbase, lsize, diop) \
1781 _hammer2_io_bread((hmp), (btype), (lbase), (lsize), (diop), \
1789 void hammer2_thr_signal(hammer2_thread_t *thr, uint32_t flags);
1790 void hammer2_thr_signal2(hammer2_thread_t *thr,
1791 uint32_t pflags, uint32_t nflags);
1792 void hammer2_thr_wait(hammer2_thread_t *thr, uint32_t flags);
1793 void hammer2_thr_wait_neg(hammer2_thread_t *thr, uint32_t flags);
1794 int hammer2_thr_wait_any(hammer2_thread_t *thr, uint32_t flags, int timo);
1795 void hammer2_thr_create(hammer2_thread_t *thr,
1796 hammer2_pfs_t *pmp, hammer2_dev_t *hmp,
1797 const char *id, int clindex, int repidx,
1798 void (*func)(void *arg));
1799 void hammer2_thr_delete(hammer2_thread_t *thr);
1800 void hammer2_thr_remaster(hammer2_thread_t *thr);
1801 void hammer2_thr_freeze_async(hammer2_thread_t *thr);
1802 void hammer2_thr_freeze(hammer2_thread_t *thr);
1803 void hammer2_thr_unfreeze(hammer2_thread_t *thr);
1804 int hammer2_thr_break(hammer2_thread_t *thr);
1805 void hammer2_primary_xops_thread(void *arg);
1808 * hammer2_thread.c (XOP API)
1810 void hammer2_xop_group_init(hammer2_pfs_t *pmp, hammer2_xop_group_t *xgrp);
1811 void *hammer2_xop_alloc(hammer2_inode_t *ip, int flags);
1812 void hammer2_xop_setname(hammer2_xop_head_t *xop,
1813 const char *name, size_t name_len);
1814 void hammer2_xop_setname2(hammer2_xop_head_t *xop,
1815 const char *name, size_t name_len);
1816 size_t hammer2_xop_setname_inum(hammer2_xop_head_t *xop, hammer2_key_t inum);
1817 void hammer2_xop_setip2(hammer2_xop_head_t *xop, hammer2_inode_t *ip2);
1818 void hammer2_xop_setip3(hammer2_xop_head_t *xop, hammer2_inode_t *ip3);
1819 void hammer2_xop_reinit(hammer2_xop_head_t *xop);
1820 void hammer2_xop_helper_create(hammer2_pfs_t *pmp);
1821 void hammer2_xop_helper_cleanup(hammer2_pfs_t *pmp);
1822 void hammer2_xop_start(hammer2_xop_head_t *xop, hammer2_xop_desc_t *desc);
1823 void hammer2_xop_start_except(hammer2_xop_head_t *xop, hammer2_xop_desc_t *desc,
1825 int hammer2_xop_collect(hammer2_xop_head_t *xop, int flags);
1826 void hammer2_xop_retire(hammer2_xop_head_t *xop, uint64_t mask);
1827 int hammer2_xop_active(hammer2_xop_head_t *xop);
1828 int hammer2_xop_feed(hammer2_xop_head_t *xop, hammer2_chain_t *chain,
1829 int clindex, int error);
1834 void hammer2_primary_sync_thread(void *arg);
1837 * XOP backends in hammer2_xops.c, primarily for VNOPS. Other XOP backends
1838 * may be integrated into other source files.
1840 void hammer2_xop_ipcluster(hammer2_xop_t *xop, void *scratch, int clindex);
1841 void hammer2_xop_readdir(hammer2_xop_t *xop, void *scratch, int clindex);
1842 void hammer2_xop_nresolve(hammer2_xop_t *xop, void *scratch, int clindex);
1843 void hammer2_xop_unlink(hammer2_xop_t *xop, void *scratch, int clindex);
1844 void hammer2_xop_nrename(hammer2_xop_t *xop, void *scratch, int clindex);
1845 void hammer2_xop_scanlhc(hammer2_xop_t *xop, void *scratch, int clindex);
1846 void hammer2_xop_scanall(hammer2_xop_t *xop, void *scratch, int clindex);
1847 void hammer2_xop_lookup(hammer2_xop_t *xop, void *scratch, int clindex);
1848 void hammer2_xop_delete(hammer2_xop_t *xop, void *scratch, int clindex);
1849 void hammer2_xop_inode_mkdirent(hammer2_xop_t *xop, void *scratch, int clindex);
1850 void hammer2_xop_inode_create(hammer2_xop_t *xop, void *scratch, int clindex);
1851 void hammer2_xop_inode_create_det(hammer2_xop_t *xop,
1852 void *scratch, int clindex);
1853 void hammer2_xop_inode_create_ins(hammer2_xop_t *xop,
1854 void *scratch, int clindex);
1855 void hammer2_xop_inode_destroy(hammer2_xop_t *xop, void *scratch, int clindex);
1856 void hammer2_xop_inode_chain_sync(hammer2_xop_t *xop, void *scratch,
1858 void hammer2_xop_inode_unlinkall(hammer2_xop_t *xop, void *scratch,
1860 void hammer2_xop_inode_connect(hammer2_xop_t *xop, void *scratch, int clindex);
1861 void hammer2_xop_inode_flush(hammer2_xop_t *xop, void *scratch, int clindex);
1862 void hammer2_xop_strategy_read(hammer2_xop_t *xop, void *scratch, int clindex);
1863 void hammer2_xop_strategy_write(hammer2_xop_t *xop, void *scratch, int clindex);
1865 void hammer2_dmsg_ipcluster(hammer2_xop_t *xop, void *scratch, int clindex);
1866 void hammer2_dmsg_readdir(hammer2_xop_t *xop, void *scratch, int clindex);
1867 void hammer2_dmsg_nresolve(hammer2_xop_t *xop, void *scratch, int clindex);
1868 void hammer2_dmsg_unlink(hammer2_xop_t *xop, void *scratch, int clindex);
1869 void hammer2_dmsg_nrename(hammer2_xop_t *xop, void *scratch, int clindex);
1870 void hammer2_dmsg_scanlhc(hammer2_xop_t *xop, void *scratch, int clindex);
1871 void hammer2_dmsg_scanall(hammer2_xop_t *xop, void *scratch, int clindex);
1872 void hammer2_dmsg_lookup(hammer2_xop_t *xop, void *scratch, int clindex);
1873 void hammer2_dmsg_inode_mkdirent(hammer2_xop_t *xop, void *scratch,
1875 void hammer2_dmsg_inode_create(hammer2_xop_t *xop, void *scratch, int clindex);
1876 void hammer2_dmsg_inode_destroy(hammer2_xop_t *xop, void *scratch, int clindex);
1877 void hammer2_dmsg_inode_chain_sync(hammer2_xop_t *xop, void *scratch,
1879 void hammer2_dmsg_inode_unlinkall(hammer2_xop_t *xop, void *scratch,
1881 void hammer2_dmsg_inode_connect(hammer2_xop_t *xop, void *scratch, int clindex);
1882 void hammer2_dmsg_inode_flush(hammer2_xop_t *xop, void *scratch, int clindex);
1883 void hammer2_dmsg_strategy_read(hammer2_xop_t *xop, void *scratch, int clindex);
1884 void hammer2_dmsg_strategy_write(hammer2_xop_t *xop, void *scratch,
1887 void hammer2_rmsg_ipcluster(hammer2_xop_t *xop, void *scratch, int clindex);
1888 void hammer2_rmsg_readdir(hammer2_xop_t *xop, void *scratch, int clindex);
1889 void hammer2_rmsg_nresolve(hammer2_xop_t *xop, void *scratch, int clindex);
1890 void hammer2_rmsg_unlink(hammer2_xop_t *xop, void *scratch, int clindex);
1891 void hammer2_rmsg_nrename(hammer2_xop_t *xop, void *scratch, int clindex);
1892 void hammer2_rmsg_scanlhc(hammer2_xop_t *xop, void *scratch, int clindex);
1893 void hammer2_rmsg_scanall(hammer2_xop_t *xop, void *scratch, int clindex);
1894 void hammer2_rmsg_lookup(hammer2_xop_t *xop, void *scratch, int clindex);
1895 void hammer2_rmsg_inode_mkdirent(hammer2_xop_t *xop, void *scratch,
1897 void hammer2_rmsg_inode_create(hammer2_xop_t *xop, void *scratch, int clindex);
1898 void hammer2_rmsg_inode_destroy(hammer2_xop_t *xop, void *scratch, int clindex);
1899 void hammer2_rmsg_inode_chain_sync(hammer2_xop_t *xop, void *scratch,
1901 void hammer2_rmsg_inode_unlinkall(hammer2_xop_t *xop, void *scratch,
1903 void hammer2_rmsg_inode_connect(hammer2_xop_t *xop, void *scratch, int clindex);
1904 void hammer2_rmsg_inode_flush(hammer2_xop_t *xop, void *scratch, int clindex);
1905 void hammer2_rmsg_strategy_read(hammer2_xop_t *xop, void *scratch, int clindex);
1906 void hammer2_rmsg_strategy_write(hammer2_xop_t *xop, void *scratch,
1909 extern hammer2_xop_desc_t hammer2_ipcluster_desc;
1910 extern hammer2_xop_desc_t hammer2_readdir_desc;
1911 extern hammer2_xop_desc_t hammer2_nresolve_desc;
1912 extern hammer2_xop_desc_t hammer2_unlink_desc;
1913 extern hammer2_xop_desc_t hammer2_nrename_desc;
1914 extern hammer2_xop_desc_t hammer2_scanlhc_desc;
1915 extern hammer2_xop_desc_t hammer2_scanall_desc;
1916 extern hammer2_xop_desc_t hammer2_lookup_desc;
1917 extern hammer2_xop_desc_t hammer2_delete_desc;
1918 extern hammer2_xop_desc_t hammer2_inode_mkdirent_desc;
1919 extern hammer2_xop_desc_t hammer2_inode_create_desc;
1920 extern hammer2_xop_desc_t hammer2_inode_create_det_desc;
1921 extern hammer2_xop_desc_t hammer2_inode_create_ins_desc;
1922 extern hammer2_xop_desc_t hammer2_inode_destroy_desc;
1923 extern hammer2_xop_desc_t hammer2_inode_chain_sync_desc;
1924 extern hammer2_xop_desc_t hammer2_inode_unlinkall_desc;
1925 extern hammer2_xop_desc_t hammer2_inode_connect_desc;
1926 extern hammer2_xop_desc_t hammer2_inode_flush_desc;
1927 extern hammer2_xop_desc_t hammer2_strategy_read_desc;
1928 extern hammer2_xop_desc_t hammer2_strategy_write_desc;
1933 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
1934 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
1939 void hammer2_volconf_update(hammer2_dev_t *hmp, int index);
1940 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx,
1942 int hammer2_vfs_sync(struct mount *mp, int waitflags);
1943 int hammer2_vfs_sync_pmp(hammer2_pfs_t *pmp, int waitfor);
1944 int hammer2_vfs_enospace(hammer2_inode_t *ip, off_t bytes, struct ucred *cred);
1946 hammer2_pfs_t *hammer2_pfsalloc(hammer2_chain_t *chain,
1947 const hammer2_inode_data_t *ripdata,
1948 hammer2_tid_t modify_tid,
1949 hammer2_dev_t *force_local);
1950 void hammer2_pfsdealloc(hammer2_pfs_t *pmp, int clindex, int destroying);
1951 int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1952 ino_t ino, struct vnode **vpp);
1954 void hammer2_lwinprog_ref(hammer2_pfs_t *pmp);
1955 void hammer2_lwinprog_drop(hammer2_pfs_t *pmp);
1956 void hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int pipe);
1961 int hammer2_freemap_alloc(hammer2_chain_t *chain, size_t bytes);
1962 void hammer2_freemap_adjust(hammer2_dev_t *hmp,
1963 hammer2_blockref_t *bref, int how);
1968 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
1969 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
1970 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfs_t *pmp,
1971 hammer2_blockref_t *bref);
1972 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
1973 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1974 void hammer2_cluster_unhold(hammer2_cluster_t *cluster);
1975 void hammer2_cluster_rehold(hammer2_cluster_t *cluster);
1976 void hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1977 int hammer2_cluster_check(hammer2_cluster_t *cluster, hammer2_key_t lokey,
1979 void hammer2_cluster_resolve(hammer2_cluster_t *cluster);
1980 void hammer2_cluster_forcegood(hammer2_cluster_t *cluster);
1981 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1983 void hammer2_bulkfree_init(hammer2_dev_t *hmp);
1984 void hammer2_bulkfree_uninit(hammer2_dev_t *hmp);
1985 int hammer2_bulkfree_pass(hammer2_dev_t *hmp, hammer2_chain_t *vchain,
1986 struct hammer2_ioc_bulkfree *bfi);
1987 void hammer2_dummy_xop_from_chain(hammer2_xop_head_t *xop,
1988 hammer2_chain_t *chain);
1993 void hammer2_iocom_init(hammer2_dev_t *hmp);
1994 void hammer2_iocom_uninit(hammer2_dev_t *hmp);
1995 void hammer2_cluster_reconnect(hammer2_dev_t *hmp, struct file *fp);
1998 * hammer2_strategy.c
2000 int hammer2_vop_strategy(struct vop_strategy_args *ap);
2001 int hammer2_vop_bmap(struct vop_bmap_args *ap);
2002 void hammer2_write_thread(void *arg);
2003 void hammer2_bioq_sync(hammer2_pfs_t *pmp);
2004 void hammer2_dedup_clear(hammer2_dev_t *hmp);
2007 * More complex inlines
2010 #define hammer2_xop_gdata(xop) _hammer2_xop_gdata((xop), __FILE__, __LINE__)
2013 const hammer2_media_data_t *
2014 _hammer2_xop_gdata(hammer2_xop_head_t *xop, const char *file, int line)
2016 hammer2_chain_t *focus;
2019 focus = xop->cluster.focus;
2021 lockmgr(&focus->diolk, LK_SHARED);
2022 if ((xop->focus_dio = focus->dio) != NULL) {
2023 _hammer2_io_ref(xop->focus_dio HAMMER2_IO_DEBUG_CALL);
2024 hammer2_io_bkvasync(xop->focus_dio);
2027 lockmgr(&focus->diolk, LK_RELEASE);
2035 #define hammer2_xop_pdata(xop) _hammer2_xop_pdata((xop), __FILE__, __LINE__)
2039 _hammer2_xop_pdata(hammer2_xop_head_t *xop, const char *file, int line)
2042 _hammer2_io_putblk(&xop->focus_dio HAMMER2_IO_DEBUG_CALL);
2045 #endif /* !_KERNEL */
2046 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */