2 * Copyright (c) 2011-2014 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 * Basically everything is represented by a hammer2_chain structure
44 * in-memory and other higher-level structures 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. However, because
48 * chains may represent blocks smaller than the 16KB minimum we impose
49 * on buffer cache buffers, we cannot hold related buffer cache buffers
50 * locked for smaller blocks. In these situations we kmalloc() a copy
53 * When modifications are made to a chain a new filesystem block must be
54 * allocated. Multiple modifications do not necessarily allocate new
55 * blocks. However, when a flush occurs a flush synchronization point
56 * is created and any new modifications made after this point will allocate
57 * a new block even if the chain is already in a modified state.
59 * The in-memory representation may remain cached (for example in order to
60 * placemark clustering locks) even after the related data has been
65 * In order to support concurrent flushes a flush synchronization point
66 * is created represented by a transaction id. Among other things,
67 * operations may move filesystem objects from one part of the topology
68 * to another (for example, if you rename a file or when indirect blocks
69 * are created or destroyed, and a few other things). When this occurs
70 * across a flush synchronization point the flusher needs to be able to
71 * recurse down BOTH the 'before' version of the topology and the 'after'
74 * To facilitate this modifications to chains do what is called a
75 * DELETE-DUPLICATE operation. Chains are not actually moved in-memory.
76 * Instead the chain we wish to move is deleted and a new chain is created
77 * at the target location in the topology. ANY SUBCHAINS PLACED UNDER THE
78 * CHAIN BEING MOVED HAVE TO EXIST IN BOTH PLACES. To make this work
79 * all sub-chains are managed by the hammer2_chain_core structure. This
80 * structure can be multi-homed, meaning that it can have more than one
81 * chain as its parent. When a chain is delete-duplicated the chain's core
82 * becomes shared under both the old and new chain.
86 * When a chain is delete-duplicated the old chain typically becomes stale.
87 * This is detected via the HAMMER2_CHAIN_DUPLICATED flag in chain->flags.
88 * To avoid executing live filesystem operations on stale chains, the inode
89 * locking code will follow stale chains via core->ownerq until it finds
90 * the live chain. The lock prevents ripups by other threads. Lookups
91 * must properly order locking operations to prevent other threads from
92 * racing the lookup operation and will also follow stale chains when
96 #ifndef _VFS_HAMMER2_HAMMER2_H_
97 #define _VFS_HAMMER2_HAMMER2_H_
99 #include <sys/param.h>
100 #include <sys/types.h>
101 #include <sys/kernel.h>
102 #include <sys/conf.h>
103 #include <sys/systm.h>
104 #include <sys/tree.h>
105 #include <sys/malloc.h>
106 #include <sys/mount.h>
107 #include <sys/vnode.h>
108 #include <sys/proc.h>
109 #include <sys/mountctl.h>
110 #include <sys/priv.h>
111 #include <sys/stat.h>
112 #include <sys/thread.h>
113 #include <sys/globaldata.h>
114 #include <sys/lockf.h>
116 #include <sys/queue.h>
117 #include <sys/limits.h>
118 #include <sys/signal2.h>
119 #include <sys/dmsg.h>
120 #include <sys/mutex.h>
121 #include <sys/kern_syscall.h>
123 #include <sys/buf2.h>
124 #include <sys/mutex2.h>
126 #include "hammer2_disk.h"
127 #include "hammer2_mount.h"
128 #include "hammer2_ioctl.h"
129 #include "hammer2_ccms.h"
131 struct hammer2_chain;
132 struct hammer2_cluster;
133 struct hammer2_inode;
134 struct hammer2_mount;
135 struct hammer2_pfsmount;
137 struct hammer2_state;
141 * The chain structure tracks a portion of the media topology from the
142 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
143 * data blocks, and freemap nodes and leafs.
145 * The chain structure can be multi-homed and its topological recursion
146 * (chain->core) can be shared amongst several chains. Chain structures
147 * are topologically stable once placed in the in-memory topology (they
148 * don't move around). Modifications which cross flush synchronization
149 * boundaries, renames, resizing, or any move of the chain to elsewhere
150 * in the topology is accomplished via the DELETE-DUPLICATE mechanism.
152 * Deletions and delete-duplicates:
154 * Any movement of chains within the topology utilize a delete-duplicate
155 * operation instead of a simple rename. That is, the chain must be
156 * deleted from its original location and then duplicated to the new
157 * location. A new chain structure is allocated while the old is
158 * deleted. Deleted chains are removed from the above chain_core's
159 * rbtree but remain linked via the shadow topology for flush
160 * synchronization purposes.
162 * delete_bmap is allocated and a bit set if the chain was originally
163 * loaded via the blockmap.
165 * Flush synchronization:
167 * Flushes must synchronize chains up through the root. To do this
168 * the in-memory topology would normally have to be frozen during the
169 * flush. To avoid freezing the topology and to allow concurrent
170 * foreground / flush activity, any new modifications made while a
171 * flush is in progress retains the original chain in a shadow topology
172 * that is only visible to the flush code. Only one flush can be
173 * running at a time so the shadow hierarchy can be implemented with
174 * just a few link fields in our in-memory data structures.
178 * (1) Fully coherent snapshots can be taken without requiring
179 * a pre-flush, resulting in extremely fast (sub-millisecond)
182 * (2) Multiple synchronization points can be in-flight at the same
183 * time, representing multiple snapshots or flushes.
185 * (3) The algorithms needed to keep track of everything are actually
188 * Special Considerations:
190 * A chain is ref-counted on a per-chain basis, but the chain's lock
191 * is associated with the shared chain_core and is not per-chain.
193 * The power-of-2 nature of the media radix tree ensures that there
194 * will be no overlaps which straddle edges.
196 RB_HEAD(hammer2_chain_tree, hammer2_chain);
197 TAILQ_HEAD(h2_flush_deferral_list, hammer2_chain);
198 TAILQ_HEAD(h2_core_list, hammer2_chain);
200 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
201 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
203 struct hammer2_chain_core {
206 struct h2_core_list ownerq; /* all chains sharing this core */
207 struct hammer2_chain_tree rbtree; /* live chains */
208 struct hammer2_chain_tree dbtree; /* bmapped deletions */
209 struct h2_core_list dbq; /* other deletions */
210 int live_zero; /* blockref array opt */
213 u_int live_count; /* live (not deleted) chains in tree */
214 u_int chain_count; /* live + deleted chains under core */
215 int generation; /* generation number (inserts only) */
218 typedef struct hammer2_chain_core hammer2_chain_core_t;
220 #define HAMMER2_CORE_UNUSED0001 0x0001
221 #define HAMMER2_CORE_COUNTEDBREFS 0x0002
224 * H2 is a copy-on-write filesystem. In order to allow chains to allocate
225 * smaller blocks (down to 64-bytes), but improve performance and make
226 * clustered I/O possible using larger block sizes, the kernel buffer cache
227 * is abstracted via the hammer2_io structure.
229 RB_HEAD(hammer2_io_tree, hammer2_io);
232 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
233 struct spinlock spin;
234 struct hammer2_mount *hmp;
239 void (*callback)(struct hammer2_io *dio,
240 struct hammer2_cluster *cluster,
241 struct hammer2_chain *chain,
242 void *arg1, off_t arg2);
243 struct hammer2_cluster *arg_l; /* INPROG I/O only */
244 struct hammer2_chain *arg_c; /* INPROG I/O only */
245 void *arg_p; /* INPROG I/O only */
246 off_t arg_o; /* INPROG I/O only */
248 int act; /* activity */
251 typedef struct hammer2_io hammer2_io_t;
254 * Primary chain structure keeps track of the topology in-memory.
256 struct hammer2_chain {
257 TAILQ_ENTRY(hammer2_chain) core_entry; /* contemporary chains */
258 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
259 TAILQ_ENTRY(hammer2_chain) db_entry; /* non bmapped deletions */
260 hammer2_blockref_t bref;
261 hammer2_chain_core_t *core;
262 hammer2_chain_core_t *above;
263 struct hammer2_state *state; /* if active cache msg */
264 struct hammer2_mount *hmp;
265 struct hammer2_pfsmount *pmp; /* can be NULL */
267 hammer2_blockref_t dsrc; /* DEBUG */
268 int ninserts; /* DEBUG */
269 int nremoves; /* DEBUG */
270 hammer2_tid_t dsrc_dupfromat; /* DEBUG */
271 uint32_t dsrc_dupfromflags; /* DEBUG */
272 int dsrc_reason; /* DEBUG */
273 int dsrc_ninserts; /* DEBUG */
274 uint32_t dsrc_flags; /* DEBUG */
275 hammer2_tid_t dsrc_modify; /* DEBUG */
276 hammer2_tid_t dsrc_delete; /* DEBUG */
277 hammer2_tid_t dsrc_update_lo; /* DEBUG */
278 struct hammer2_chain *dsrc_original; /* DEBUG */
280 hammer2_tid_t modify_tid; /* flush filter */
281 hammer2_tid_t delete_tid; /* flush filter */
282 hammer2_tid_t update_lo; /* flush propagation */
283 hammer2_tid_t update_hi; /* setsubmod propagation */
284 hammer2_key_t data_count; /* delta's to apply */
285 hammer2_key_t inode_count; /* delta's to apply */
286 hammer2_io_t *dio; /* physical data buffer */
287 u_int bytes; /* physical data size */
291 hammer2_media_data_t *data; /* data pointer shortcut */
292 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush deferral list */
297 typedef struct hammer2_chain hammer2_chain_t;
299 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
300 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
303 * Special notes on flags:
305 * INITIAL - This flag allows a chain to be created and for storage to
306 * be allocated without having to immediately instantiate the
307 * related buffer. The data is assumed to be all-zeros. It
308 * is primarily used for indirect blocks.
310 * MODIFIED- The chain's media data has been modified.
312 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
313 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
314 #define HAMMER2_CHAIN_FLUSH_TEMPORARY 0x00000004
315 #define HAMMER2_CHAIN_FORCECOW 0x00000008 /* force copy-on-wr */
316 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
317 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
318 #define HAMMER2_CHAIN_FLUSH_CREATE 0x00000040 /* needs flush blkadd */
319 #define HAMMER2_CHAIN_FLUSH_DELETE 0x00000080 /* needs flush blkdel */
320 #define HAMMER2_CHAIN_IOFLUSH 0x00000100 /* bawrite on put */
321 #define HAMMER2_CHAIN_DEFERRED 0x00000200 /* on a deferral list */
322 #define HAMMER2_CHAIN_UNLINKED 0x00000400 /* delete on reclaim */
323 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
324 #define HAMMER2_CHAIN_ONDBQ 0x00001000 /* !bmapped deletes */
325 #define HAMMER2_CHAIN_MOUNTED 0x00002000 /* PFS is mounted */
326 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
327 #define HAMMER2_CHAIN_SNAPSHOT 0x00008000 /* snapshot special */
328 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
329 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
330 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* in parent blkmap */
331 #define HAMMER2_CHAIN_ONDBTREE 0x00080000 /* bmapped deletes */
332 #define HAMMER2_CHAIN_DUPLICATED 0x00100000 /* fwd delete-dup */
333 #define HAMMER2_CHAIN_PFSROOT 0x00200000 /* in pfs->cluster */
336 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
338 * NOTE: MATCHIND allows an indirect block / freemap node to be returned
339 * when the passed key range matches the radix. Remember that key_end
340 * is inclusive (e.g. {0x000,0xFFF}, not {0x000,0x1000}).
342 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
343 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
344 #define HAMMER2_LOOKUP_SHARED 0x00000100
345 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
346 #define HAMMER2_LOOKUP_UNUSED0400 0x00000400
347 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
350 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
352 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
353 * blocks in the INITIAL-create state.
355 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
356 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
357 #define HAMMER2_MODIFY_ASSERTNOCOPY 0x00000008 /* assert no del-dup */
358 #define HAMMER2_MODIFY_NOREALLOC 0x00000010
359 #define HAMMER2_MODIFY_INPLACE 0x00000020 /* don't del-dup */
362 * Flags passed to hammer2_chain_lock()
364 #define HAMMER2_RESOLVE_NEVER 1
365 #define HAMMER2_RESOLVE_MAYBE 2
366 #define HAMMER2_RESOLVE_ALWAYS 3
367 #define HAMMER2_RESOLVE_MASK 0x0F
369 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
370 #define HAMMER2_RESOLVE_NOREF 0x20 /* already ref'd on lock */
373 * Flags passed to hammer2_chain_delete()
375 #define HAMMER2_DELETE_UNUSED0001 0x0001
378 * Flags passed to hammer2_chain_delete_duplicate()
380 #define HAMMER2_DELDUP_RECORE 0x0001
383 * Cluster different types of storage together for allocations
385 #define HAMMER2_FREECACHE_INODE 0
386 #define HAMMER2_FREECACHE_INDIR 1
387 #define HAMMER2_FREECACHE_DATA 2
388 #define HAMMER2_FREECACHE_UNUSED3 3
389 #define HAMMER2_FREECACHE_TYPES 4
392 * hammer2_freemap_alloc() block preference
394 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
397 * BMAP read-ahead maximum parameters
399 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
400 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
405 #define HAMMER2_FLUSH_DEPTH_LIMIT 10 /* stack recursion limit */
408 * hammer2_freemap_adjust()
410 #define HAMMER2_FREEMAP_DORECOVER 1
411 #define HAMMER2_FREEMAP_DOMAYFREE 2
412 #define HAMMER2_FREEMAP_DOREALFREE 3
415 * HAMMER2 cluster - A set of chains representing the same entity.
417 * The hammer2_pfsmount structure embeds a hammer2_cluster. All other
418 * hammer2_cluster use cases use temporary allocations.
420 * The cluster API mimics the chain API. Except as used in the pfsmount,
421 * the cluster structure is a temporary 'working copy' of a set of chains
422 * representing targets compatible with the operation. However, for
423 * performance reasons the cluster API does not necessarily issue concurrent
424 * requests to the underlying chain API for all compatible chains all the
425 * time. This may sometimes necessitate revisiting parent cluster nodes
426 * to 'flesh out' (validate more chains).
428 * If an insufficient number of chains remain in a working copy, the operation
429 * may have to be downgraded, retried, or stall until the requisit number
430 * of chains are available.
432 #define HAMMER2_MAXCLUSTER 8
434 struct hammer2_cluster {
435 int status; /* operational status */
436 int refs; /* track for deallocation */
437 struct hammer2_pfsmount *pmp;
440 hammer2_chain_t *focus; /* current focus (or mod) */
441 hammer2_chain_t *array[HAMMER2_MAXCLUSTER];
442 int cache_index[HAMMER2_MAXCLUSTER];
445 typedef struct hammer2_cluster hammer2_cluster_t;
447 #define HAMMER2_CLUSTER_PFS 0x00000001 /* embedded in pfsmount */
448 #define HAMMER2_CLUSTER_INODE 0x00000002 /* embedded in inode */
451 RB_HEAD(hammer2_inode_tree, hammer2_inode);
456 * NOTE: The inode's attribute CST which is also used to lock the inode
457 * is embedded in the chain (chain.cst) and aliased w/ attr_cst.
459 struct hammer2_inode {
460 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
461 ccms_cst_t topo_cst; /* directory topology cst */
462 struct hammer2_pfsmount *pmp; /* PFS mount */
463 struct hammer2_inode *pip; /* parent inode */
465 hammer2_cluster_t cluster;
466 struct lockf advlock;
469 u_int refs; /* +vpref, +flushref */
470 uint8_t comp_heuristic;
475 typedef struct hammer2_inode hammer2_inode_t;
477 #define HAMMER2_INODE_MODIFIED 0x0001
478 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
479 #define HAMMER2_INODE_RENAME_INPROG 0x0004
480 #define HAMMER2_INODE_ONRBTREE 0x0008
481 #define HAMMER2_INODE_RESIZED 0x0010
482 #define HAMMER2_INODE_MTIME 0x0020
484 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
485 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
489 * inode-unlink side-structure
491 struct hammer2_inode_unlink {
492 TAILQ_ENTRY(hammer2_inode_unlink) entry;
495 TAILQ_HEAD(hammer2_unlk_list, hammer2_inode_unlink);
497 typedef struct hammer2_inode_unlink hammer2_inode_unlink_t;
500 * A hammer2 transaction and flush sequencing structure.
502 * This global structure is tied into hammer2_mount and is used
503 * to sequence modifying operations and flushes.
505 * (a) Any modifying operations with sync_tid >= flush_tid will stall until
506 * all modifying operating with sync_tid < flush_tid complete.
508 * The flush related to flush_tid stalls until all modifying operations
509 * with sync_tid < flush_tid complete.
511 * (b) Once unstalled, modifying operations with sync_tid > flush_tid are
512 * allowed to run. All modifications cause modify/duplicate operations
513 * to occur on the related chains. Note that most INDIRECT blocks will
514 * be unaffected because the modifications just overload the RBTREE
515 * structurally instead of actually modifying the indirect blocks.
517 * (c) The actual flush unstalls and RUNS CONCURRENTLY with (b), but only
518 * utilizes the chain structures with sync_tid <= flush_tid. The
519 * flush will modify related indirect blocks and inodes in-place
520 * (rather than duplicate) since the adjustments are compatible with
521 * (b)'s RBTREE overloading
523 * SPECIAL NOTE: Inode modifications have to also propagate along any
524 * modify/duplicate chains. File writes detect the flush
525 * and force out the conflicting buffer cache buffer(s)
526 * before reusing them.
528 * (d) Snapshots can be made instantly but must be flushed and disconnected
529 * from their duplicative source before they can be mounted. This is
530 * because while H2's on-media structure supports forks, its in-memory
531 * structure only supports very simple forking for background flushing
534 * TODO: Flush merging. When fsync() is called on multiple discrete files
535 * concurrently there is no reason to stall the second fsync.
536 * The final flush that reaches to root can cover both fsync()s.
538 * The chains typically terminate as they fly onto the disk. The flush
539 * ultimately reaches the volume header.
541 struct hammer2_trans {
542 TAILQ_ENTRY(hammer2_trans) entry;
543 struct hammer2_pfsmount *pmp; /* might be NULL */
544 struct hammer2_mount *hmp_single; /* if single-targetted */
545 hammer2_tid_t orig_tid;
546 hammer2_tid_t sync_tid; /* effective transaction id */
547 hammer2_tid_t inode_tid;
548 thread_t td; /* pointer */
551 uint8_t inodes_created;
555 typedef struct hammer2_trans hammer2_trans_t;
557 #define HAMMER2_TRANS_ISFLUSH 0x0001 /* formal flush */
558 #define HAMMER2_TRANS_CONCURRENT 0x0002 /* concurrent w/flush */
559 #define HAMMER2_TRANS_BUFCACHE 0x0004 /* from bioq strategy write */
560 #define HAMMER2_TRANS_NEWINODE 0x0008 /* caller allocating inode */
561 #define HAMMER2_TRANS_ISALLOCATING 0x0010 /* in allocator */
562 #define HAMMER2_TRANS_PREFLUSH 0x0020 /* preflush state */
564 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
565 #define HAMMER2_FREEMAP_HEUR_TYPES 8
566 #define HAMMER2_FREEMAP_HEUR (HAMMER2_FREEMAP_HEUR_NRADIX * \
567 HAMMER2_FREEMAP_HEUR_TYPES)
569 #define HAMMER2_CLUSTER_COPY_CHAINS 0x0001 /* copy chains */
570 #define HAMMER2_CLUSTER_COPY_NOREF 0x0002 /* do not ref chains */
573 * Global (per device) mount structure for device (aka vp->v_mount->hmp)
575 TAILQ_HEAD(hammer2_trans_queue, hammer2_trans);
577 struct hammer2_mount {
578 struct vnode *devvp; /* device vnode */
579 int ronly; /* read-only mount */
580 int pmp_count; /* PFS mounts backed by us */
581 TAILQ_ENTRY(hammer2_mount) mntentry; /* hammer2_mntlist */
583 struct malloc_type *mchain;
586 struct spinlock io_spin; /* iotree access */
587 struct hammer2_io_tree iotree;
589 hammer2_chain_t vchain; /* anchor chain (topology) */
590 hammer2_chain_t fchain; /* anchor chain (freemap) */
591 hammer2_inode_t *sroot; /* super-root localized to media */
592 struct lock alloclk; /* lockmgr lock */
593 struct lock voldatalk; /* lockmgr lock */
594 struct hammer2_trans_queue transq; /* all in-progress transactions */
595 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR];
596 int flushcnt; /* #of flush trans on the list */
598 int volhdrno; /* last volhdrno written */
599 hammer2_volume_data_t voldata;
600 hammer2_volume_data_t volsync; /* synchronized voldata */
603 typedef struct hammer2_mount hammer2_mount_t;
606 * HAMMER2 PFS mount point structure (aka vp->v_mount->mnt_data).
607 * This has a 1:1 correspondence to struct mount (note that the
608 * hammer2_mount structure has a N:1 correspondence).
610 * This structure represents a cluster mount and not necessarily a
611 * PFS under a specific device mount (HMP). The distinction is important
612 * because the elements backing a cluster mount can change on the fly.
614 * Usually the first element under the cluster represents the original
615 * user-requested mount that bootstraps the whole mess. In significant
616 * setups the original is usually just a read-only media image (or
617 * representitive file) that simply contains a bootstrap volume header
618 * listing the configuration.
620 struct hammer2_pfsmount {
622 hammer2_cluster_t cluster;
623 hammer2_inode_t *iroot; /* PFS root inode */
624 hammer2_inode_t *ihidden; /* PFS hidden directory */
625 struct lock lock; /* PFS lock for certain ops */
626 hammer2_off_t inode_count; /* copy of inode_count */
627 ccms_domain_t ccms_dom;
628 struct netexport export; /* nfs export */
629 int ronly; /* read-only mount */
630 struct malloc_type *minode;
631 struct malloc_type *mmsg;
633 struct spinlock inum_spin; /* inumber lookup */
634 struct hammer2_inode_tree inum_tree;
636 long inmem_dirty_chains;
637 int count_lwinprog; /* logical write in prog */
638 struct spinlock unlinkq_spin;
639 struct hammer2_unlk_list unlinkq;
640 thread_t wthread_td; /* write thread td */
641 struct bio_queue_head wthread_bioq; /* logical buffer bioq */
642 struct mtx wthread_mtx; /* interlock */
643 int wthread_destroy;/* termination sequencing */
646 typedef struct hammer2_pfsmount hammer2_pfsmount_t;
648 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
649 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
651 #define HAMMER2_LWINPROG_WAITING 0x80000000
652 #define HAMMER2_LWINPROG_MASK 0x7FFFFFFF
656 MALLOC_DECLARE(M_HAMMER2);
658 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
659 #define ITOV(ip) ((ip)->vp)
662 * Currently locked chains retain the locked buffer cache buffer for
663 * indirect blocks, and indirect blocks can be one of two sizes. The
664 * device buffer has to match the case to avoid deadlocking recursive
665 * chains that might otherwise try to access different offsets within
666 * the same device buffer.
670 hammer2_devblkradix(int radix)
672 if (radix <= HAMMER2_LBUFRADIX) {
673 return (HAMMER2_LBUFRADIX);
675 return (HAMMER2_PBUFRADIX);
681 hammer2_devblksize(size_t bytes)
683 if (bytes <= HAMMER2_LBUFSIZE) {
684 return(HAMMER2_LBUFSIZE);
686 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
687 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
688 return (HAMMER2_PBUFSIZE);
695 MPTOPMP(struct mount *mp)
697 return ((hammer2_pfsmount_t *)mp->mnt_data);
700 extern struct vop_ops hammer2_vnode_vops;
701 extern struct vop_ops hammer2_spec_vops;
702 extern struct vop_ops hammer2_fifo_vops;
704 extern int hammer2_debug;
705 extern int hammer2_cluster_enable;
706 extern int hammer2_hardlink_enable;
707 extern int hammer2_flush_pipe;
708 extern int hammer2_synchronous_flush;
709 extern long hammer2_limit_dirty_chains;
710 extern long hammer2_iod_file_read;
711 extern long hammer2_iod_meta_read;
712 extern long hammer2_iod_indr_read;
713 extern long hammer2_iod_fmap_read;
714 extern long hammer2_iod_volu_read;
715 extern long hammer2_iod_file_write;
716 extern long hammer2_iod_meta_write;
717 extern long hammer2_iod_indr_write;
718 extern long hammer2_iod_fmap_write;
719 extern long hammer2_iod_volu_write;
720 extern long hammer2_ioa_file_read;
721 extern long hammer2_ioa_meta_read;
722 extern long hammer2_ioa_indr_read;
723 extern long hammer2_ioa_fmap_read;
724 extern long hammer2_ioa_volu_read;
725 extern long hammer2_ioa_file_write;
726 extern long hammer2_ioa_meta_write;
727 extern long hammer2_ioa_indr_write;
728 extern long hammer2_ioa_fmap_write;
729 extern long hammer2_ioa_volu_write;
731 extern struct objcache *cache_buffer_read;
732 extern struct objcache *cache_buffer_write;
735 extern int write_thread_wakeup;
737 extern mtx_t thread_protect;
742 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
743 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
745 hammer2_cluster_t *hammer2_inode_lock_ex(hammer2_inode_t *ip);
746 hammer2_cluster_t *hammer2_inode_lock_sh(hammer2_inode_t *ip);
747 void hammer2_inode_unlock_ex(hammer2_inode_t *ip, hammer2_cluster_t *chain);
748 void hammer2_inode_unlock_sh(hammer2_inode_t *ip, hammer2_cluster_t *chain);
749 void hammer2_voldata_lock(hammer2_mount_t *hmp);
750 void hammer2_voldata_unlock(hammer2_mount_t *hmp, int modify);
751 ccms_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
752 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip, ccms_state_t ostate);
753 ccms_state_t hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
754 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, ccms_state_t ostate);
756 void hammer2_mount_exlock(hammer2_mount_t *hmp);
757 void hammer2_mount_shlock(hammer2_mount_t *hmp);
758 void hammer2_mount_unlock(hammer2_mount_t *hmp);
760 int hammer2_get_dtype(hammer2_inode_data_t *ipdata);
761 int hammer2_get_vtype(hammer2_inode_data_t *ipdata);
762 u_int8_t hammer2_get_obj_type(enum vtype vtype);
763 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
764 u_int64_t hammer2_timespec_to_time(struct timespec *ts);
765 u_int32_t hammer2_to_unix_xid(uuid_t *uuid);
766 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
768 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
769 int hammer2_getradix(size_t bytes);
771 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
772 hammer2_key_t *lbasep, hammer2_key_t *leofp);
773 int hammer2_calc_physical(hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
774 hammer2_key_t lbase);
775 void hammer2_update_time(uint64_t *timep);
776 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
781 struct vnode *hammer2_igetv(hammer2_inode_t *ip, hammer2_cluster_t *cparent,
783 void hammer2_inode_lock_nlinks(hammer2_inode_t *ip);
784 void hammer2_inode_unlock_nlinks(hammer2_inode_t *ip);
785 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfsmount_t *pmp,
787 hammer2_inode_t *hammer2_inode_get(hammer2_pfsmount_t *pmp,
788 hammer2_inode_t *dip, hammer2_cluster_t *cluster);
789 void hammer2_inode_free(hammer2_inode_t *ip);
790 void hammer2_inode_ref(hammer2_inode_t *ip);
791 void hammer2_inode_drop(hammer2_inode_t *ip);
792 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
793 hammer2_cluster_t *cluster);
794 void hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfsmount_t *pmp);
796 hammer2_inode_t *hammer2_inode_create(hammer2_trans_t *trans,
797 hammer2_inode_t *dip,
798 struct vattr *vap, struct ucred *cred,
799 const uint8_t *name, size_t name_len,
800 hammer2_cluster_t **clusterp, int *errorp);
801 int hammer2_inode_connect(hammer2_trans_t *trans,
802 hammer2_cluster_t **clusterp, int hlink,
803 hammer2_inode_t *dip, hammer2_cluster_t *dcluster,
804 const uint8_t *name, size_t name_len,
806 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
807 hammer2_inode_t *tdip);
808 void hammer2_inode_fsync(hammer2_trans_t *trans, hammer2_inode_t *ip,
809 hammer2_cluster_t *cparent);
810 int hammer2_unlink_file(hammer2_trans_t *trans, hammer2_inode_t *dip,
811 const uint8_t *name, size_t name_len, int isdir,
812 int *hlinkp, struct nchandle *nch);
813 int hammer2_hardlink_consolidate(hammer2_trans_t *trans,
814 hammer2_inode_t *ip, hammer2_cluster_t **clusterp,
815 hammer2_inode_t *cdip, hammer2_cluster_t *cdcluster,
817 int hammer2_hardlink_deconsolidate(hammer2_trans_t *trans, hammer2_inode_t *dip,
818 hammer2_chain_t **chainp, hammer2_chain_t **ochainp);
819 int hammer2_hardlink_find(hammer2_inode_t *dip, hammer2_cluster_t *cluster);
820 void hammer2_inode_install_hidden(hammer2_pfsmount_t *pmp);
825 void hammer2_modify_volume(hammer2_mount_t *hmp);
826 hammer2_chain_t *hammer2_chain_alloc(hammer2_mount_t *hmp,
827 hammer2_pfsmount_t *pmp,
828 hammer2_trans_t *trans,
829 hammer2_blockref_t *bref);
830 void hammer2_chain_core_alloc(hammer2_trans_t *trans, hammer2_chain_t *nchain,
831 hammer2_chain_t *ochain);
832 void hammer2_chain_ref(hammer2_chain_t *chain);
833 void hammer2_chain_drop(hammer2_chain_t *chain);
834 int hammer2_chain_lock(hammer2_chain_t *chain, int how);
835 void hammer2_chain_load_async(hammer2_cluster_t *cluster,
836 void (*func)(hammer2_io_t *dio,
837 hammer2_cluster_t *cluster,
838 hammer2_chain_t *chain,
839 void *arg_p, off_t arg_o),
841 void hammer2_chain_moved(hammer2_chain_t *chain);
842 void hammer2_chain_modify(hammer2_trans_t *trans,
843 hammer2_chain_t **chainp, int flags);
844 void hammer2_chain_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
845 hammer2_chain_t *parent,
846 hammer2_chain_t **chainp,
847 int nradix, int flags);
848 void hammer2_chain_unlock(hammer2_chain_t *chain);
849 void hammer2_chain_wait(hammer2_chain_t *chain);
850 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
851 hammer2_blockref_t *bref);
852 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
853 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
854 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
855 hammer2_key_t *key_nextp,
856 hammer2_key_t key_beg, hammer2_key_t key_end,
857 int *cache_indexp, int flags, int *ddflagp);
858 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
859 hammer2_chain_t *chain,
860 hammer2_key_t *key_nextp,
861 hammer2_key_t key_beg, hammer2_key_t key_end,
862 int *cache_indexp, int flags);
863 hammer2_chain_t *hammer2_chain_scan(hammer2_chain_t *parent,
864 hammer2_chain_t *chain,
865 int *cache_indexp, int flags);
867 int hammer2_chain_create(hammer2_trans_t *trans,
868 hammer2_chain_t **parentp,
869 hammer2_chain_t **chainp,
870 hammer2_key_t key, int keybits,
871 int type, size_t bytes);
872 void hammer2_chain_duplicate(hammer2_trans_t *trans, hammer2_chain_t **parentp,
873 hammer2_chain_t **chainp,
874 hammer2_blockref_t *bref, int snapshot,
875 int duplicate_reason);
876 int hammer2_chain_snapshot(hammer2_trans_t *trans, hammer2_chain_t **chainp,
877 hammer2_ioc_pfs_t *pfs);
878 void hammer2_chain_delete(hammer2_trans_t *trans, hammer2_chain_t *chain,
880 void hammer2_chain_delete_duplicate(hammer2_trans_t *trans,
881 hammer2_chain_t **chainp, int flags);
882 void hammer2_flush(hammer2_trans_t *trans, hammer2_chain_t **chainp);
883 void hammer2_chain_commit(hammer2_trans_t *trans, hammer2_chain_t *chain);
884 void hammer2_chain_setsubmod(hammer2_trans_t *trans, hammer2_chain_t *chain);
885 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
886 hammer2_blockref_t *base, int count);
888 void hammer2_pfs_memory_wait(hammer2_pfsmount_t *pmp);
889 void hammer2_pfs_memory_inc(hammer2_pfsmount_t *pmp);
890 void hammer2_pfs_memory_wakeup(hammer2_pfsmount_t *pmp);
892 int hammer2_base_find(hammer2_chain_t *chain,
893 hammer2_blockref_t *base, int count,
894 int *cache_indexp, hammer2_key_t *key_nextp,
895 hammer2_key_t key_beg, hammer2_key_t key_end,
897 void hammer2_base_delete(hammer2_trans_t *trans, hammer2_chain_t *chain,
898 hammer2_blockref_t *base, int count,
899 int *cache_indexp, hammer2_chain_t *child);
900 void hammer2_base_insert(hammer2_trans_t *trans, hammer2_chain_t *chain,
901 hammer2_blockref_t *base, int count,
902 int *cache_indexp, hammer2_chain_t *child);
903 void hammer2_chain_refactor(hammer2_chain_t **chainp);
908 void hammer2_trans_init(hammer2_trans_t *trans, hammer2_pfsmount_t *pmp,
909 hammer2_mount_t *hmp, int flags);
910 void hammer2_trans_done(hammer2_trans_t *trans);
915 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
916 int fflag, struct ucred *cred);
921 hammer2_io_t *hammer2_io_getblk(hammer2_mount_t *hmp, off_t lbase,
922 int lsize, int *ownerp);
923 void hammer2_io_putblk(hammer2_io_t **diop);
924 void hammer2_io_cleanup(hammer2_mount_t *hmp, struct hammer2_io_tree *tree);
925 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
926 int hammer2_io_new(hammer2_mount_t *hmp, off_t lbase, int lsize,
927 hammer2_io_t **diop);
928 int hammer2_io_newnz(hammer2_mount_t *hmp, off_t lbase, int lsize,
929 hammer2_io_t **diop);
930 int hammer2_io_newq(hammer2_mount_t *hmp, off_t lbase, int lsize,
931 hammer2_io_t **diop);
932 int hammer2_io_bread(hammer2_mount_t *hmp, off_t lbase, int lsize,
933 hammer2_io_t **diop);
934 void hammer2_io_breadcb(hammer2_mount_t *hmp, off_t lbase, int lsize,
935 void (*callback)(hammer2_io_t *dio,
936 hammer2_cluster_t *arg_l,
937 hammer2_chain_t *arg_c,
938 void *arg_p, off_t arg_o),
939 hammer2_cluster_t *arg_l,
940 hammer2_chain_t *arg_c,
941 void *arg_p, off_t arg_o);
942 void hammer2_io_bawrite(hammer2_io_t **diop);
943 void hammer2_io_bdwrite(hammer2_io_t **diop);
944 int hammer2_io_bwrite(hammer2_io_t **diop);
945 int hammer2_io_isdirty(hammer2_io_t *dio);
946 void hammer2_io_setdirty(hammer2_io_t *dio);
947 void hammer2_io_setinval(hammer2_io_t *dio, u_int bytes);
948 void hammer2_io_brelse(hammer2_io_t **diop);
949 void hammer2_io_bqrelse(hammer2_io_t **diop);
954 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
955 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
960 void hammer2_clusterctl_wakeup(kdmsg_iocom_t *iocom);
961 void hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index);
962 void hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp);
963 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
964 void hammer2_bioq_sync(hammer2_pfsmount_t *pmp);
965 int hammer2_vfs_sync(struct mount *mp, int waitflags);
966 void hammer2_lwinprog_ref(hammer2_pfsmount_t *pmp);
967 void hammer2_lwinprog_drop(hammer2_pfsmount_t *pmp);
968 void hammer2_lwinprog_wait(hammer2_pfsmount_t *pmp);
973 int hammer2_freemap_alloc(hammer2_trans_t *trans, hammer2_chain_t *chain,
975 void hammer2_freemap_adjust(hammer2_trans_t *trans, hammer2_mount_t *hmp,
976 hammer2_blockref_t *bref, int how);
981 u_int hammer2_cluster_bytes(hammer2_cluster_t *cluster);
982 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
983 hammer2_media_data_t *hammer2_cluster_data(hammer2_cluster_t *cluster);
984 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
985 int hammer2_cluster_modified(hammer2_cluster_t *cluster);
986 int hammer2_cluster_unlinked(hammer2_cluster_t *cluster);
987 int hammer2_cluster_duplicated(hammer2_cluster_t *cluster);
988 void hammer2_cluster_set_chainflags(hammer2_cluster_t *cluster, uint32_t flags);
989 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
990 void hammer2_cluster_setsubmod(hammer2_trans_t *trans,
991 hammer2_cluster_t *cluster);
992 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfsmount_t *pmp,
993 hammer2_trans_t *trans,
994 hammer2_blockref_t *bref);
995 void hammer2_cluster_core_alloc(hammer2_trans_t *trans,
996 hammer2_cluster_t *ncluster,
997 hammer2_cluster_t *ocluster);
998 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
999 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
1000 void hammer2_cluster_wait(hammer2_cluster_t *cluster);
1001 int hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
1002 void hammer2_cluster_replace(hammer2_cluster_t *dst, hammer2_cluster_t *src);
1003 void hammer2_cluster_replace_locked(hammer2_cluster_t *dst,
1004 hammer2_cluster_t *src);
1005 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster,
1007 void hammer2_cluster_refactor(hammer2_cluster_t *cluster);
1008 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
1009 void hammer2_cluster_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
1010 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1011 int nradix, int flags);
1012 hammer2_inode_data_t *hammer2_cluster_modify_ip(hammer2_trans_t *trans,
1013 hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1015 void hammer2_cluster_modify(hammer2_trans_t *trans, hammer2_cluster_t *cluster,
1017 hammer2_cluster_t *hammer2_cluster_lookup_init(hammer2_cluster_t *cparent,
1019 void hammer2_cluster_lookup_done(hammer2_cluster_t *cparent);
1020 hammer2_cluster_t *hammer2_cluster_lookup(hammer2_cluster_t *cparent,
1021 hammer2_key_t *key_nextp,
1022 hammer2_key_t key_beg, hammer2_key_t key_end,
1023 int flags, int *ddflagp);
1024 hammer2_cluster_t *hammer2_cluster_next(hammer2_cluster_t *cparent,
1025 hammer2_cluster_t *cluster,
1026 hammer2_key_t *key_nextp,
1027 hammer2_key_t key_beg, hammer2_key_t key_end,
1029 hammer2_cluster_t *hammer2_cluster_scan(hammer2_cluster_t *cparent,
1030 hammer2_cluster_t *cluster, int flags);
1031 int hammer2_cluster_create(hammer2_trans_t *trans, hammer2_cluster_t *cparent,
1032 hammer2_cluster_t **clusterp,
1033 hammer2_key_t key, int keybits, int type, size_t bytes);
1034 void hammer2_cluster_duplicate(hammer2_trans_t *trans,
1035 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1036 hammer2_blockref_t *bref,
1037 int snapshot, int duplicate_reason);
1038 void hammer2_cluster_delete_duplicate(hammer2_trans_t *trans,
1039 hammer2_cluster_t *cluster, int flags);
1040 void hammer2_cluster_delete(hammer2_trans_t *trans, hammer2_cluster_t *cluster,
1042 int hammer2_cluster_snapshot(hammer2_trans_t *trans,
1043 hammer2_cluster_t *ocluster, hammer2_ioc_pfs_t *pfs);
1045 #endif /* !_KERNEL */
1046 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */