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 * Almost everything is represented by a hammer2_chain structure in-memory.
44 * Other higher-level structures typically map to chains.
46 * A great deal of data is accessed simply via its buffer cache buffer,
47 * which is mapped for the duration of the chain's lock. Hammer2 must
48 * implement its own buffer cache layer on top of the system layer to
49 * allow for different threads to lock different sub-block-sized buffers.
51 * When modifications are made to a chain a new filesystem block must be
52 * allocated. Multiple modifications do not typically allocate new blocks
53 * until the current block has been flushed. Flushes do not block the
54 * front-end unless the front-end operation crosses the current inode being
57 * The in-memory representation may remain cached (for example in order to
58 * placemark clustering locks) even after the related data has been
62 #ifndef _VFS_HAMMER2_HAMMER2_H_
63 #define _VFS_HAMMER2_HAMMER2_H_
65 #include <sys/param.h>
66 #include <sys/types.h>
67 #include <sys/kernel.h>
69 #include <sys/systm.h>
71 #include <sys/malloc.h>
72 #include <sys/mount.h>
73 #include <sys/vnode.h>
75 #include <sys/mountctl.h>
78 #include <sys/thread.h>
79 #include <sys/globaldata.h>
80 #include <sys/lockf.h>
82 #include <sys/queue.h>
83 #include <sys/limits.h>
84 #include <sys/signal2.h>
86 #include <sys/mutex.h>
87 #include <sys/kern_syscall.h>
90 #include <sys/mutex2.h>
92 #include "hammer2_disk.h"
93 #include "hammer2_mount.h"
94 #include "hammer2_ioctl.h"
95 #include "hammer2_ccms.h"
98 struct hammer2_cluster;
100 struct hammer2_mount;
101 struct hammer2_pfsmount;
103 struct hammer2_state;
107 * The xid tracks internal transactional updates.
109 * XXX fix-me, really needs to be 64-bits
111 typedef uint32_t hammer2_xid_t;
113 #define HAMMER2_XID_MIN 0x00000000U
114 #define HAMMER2_XID_MAX 0x7FFFFFFFU
117 * The chain structure tracks a portion of the media topology from the
118 * root (volume) down. Chains represent volumes, inodes, indirect blocks,
119 * data blocks, and freemap nodes and leafs.
121 * The chain structure utilizes a simple singly-homed topology and the
122 * chain's in-memory topology will move around as the chains do, due mainly
123 * to renames and indirect block creation.
125 * Block Table Updates
127 * Block table updates for insertions and updates are delayed until the
128 * flush. This allows us to avoid having to modify the parent chain
129 * all the way to the root.
131 * Block table deletions are performed immediately (modifying the parent
132 * in the process) because the flush code uses the chain structure to
133 * track delayed updates and the chain will be (likely) gone or moved to
134 * another location in the topology after a deletion.
136 * A prior iteration of the code tried to keep the relationship intact
137 * on deletes by doing a delete-duplicate operation on the chain, but
138 * it added way too much complexity to the codebase.
140 * Flush Synchronization
142 * The flush code must flush modified chains bottom-up. Because chain
143 * structures can shift around and are NOT topologically stable,
144 * modified chains are independently indexed for the flush. As the flush
145 * runs it modifies (or further modifies) and updates the parents,
146 * propagating the flush all the way to the volume root.
148 * Modifying front-end operations can occur during a flush but will block
149 * in two cases: (1) when the front-end tries to operate on the inode
150 * currently in the midst of being flushed and (2) if the front-end
151 * crosses an inode currently being flushed (such as during a rename).
152 * So, for example, if you rename directory "x" to "a/b/c/d/e/f/g/x" and
153 * the flusher is currently working on "a/b/c", the rename will block
154 * temporarily in order to ensure that "x" exists in one place or the
157 * Meta-data statistics are updated by the flusher. The front-end will
158 * make estimates but meta-data must be fully synchronized only during a
159 * flush in order to ensure that it remains correct across a crash.
161 * Multiple flush synchronizations can theoretically be in-flight at the
162 * same time but the implementation is not coded to handle the case and
163 * currently serializes them.
167 * Snapshots currently require the subdirectory tree being snapshotted
168 * to be flushed. The snapshot then creates a new super-root inode which
169 * copies the flushed blockdata of the directory or file that was
174 * - Note that the radix tree runs in powers of 2 only so sub-trees
175 * cannot straddle edges.
177 RB_HEAD(hammer2_chain_tree, hammer2_chain);
178 TAILQ_HEAD(h2_flush_list, hammer2_chain);
179 TAILQ_HEAD(h2_core_list, hammer2_chain);
181 #define CHAIN_CORE_DELETE_BMAP_ENTRIES \
182 (HAMMER2_PBUFSIZE / sizeof(hammer2_blockref_t) / sizeof(uint32_t))
184 struct hammer2_chain_core {
186 struct hammer2_chain_tree rbtree; /* sub-chains */
187 int live_zero; /* blockref array opt */
189 u_int live_count; /* live (not deleted) chains in tree */
190 u_int chain_count; /* live + deleted chains under core */
191 int generation; /* generation number (inserts only) */
194 typedef struct hammer2_chain_core hammer2_chain_core_t;
196 #define HAMMER2_CORE_UNUSED0001 0x0001
197 #define HAMMER2_CORE_COUNTEDBREFS 0x0002
200 * H2 is a copy-on-write filesystem. In order to allow chains to allocate
201 * smaller blocks (down to 64-bytes), but improve performance and make
202 * clustered I/O possible using larger block sizes, the kernel buffer cache
203 * is abstracted via the hammer2_io structure.
205 RB_HEAD(hammer2_io_tree, hammer2_io);
208 RB_ENTRY(hammer2_io) rbnode; /* indexed by device offset */
209 struct spinlock spin;
210 struct hammer2_mount *hmp;
215 void (*callback)(struct hammer2_io *dio,
216 struct hammer2_cluster *cluster,
217 struct hammer2_chain *chain,
218 void *arg1, off_t arg2);
219 struct hammer2_cluster *arg_l; /* INPROG I/O only */
220 struct hammer2_chain *arg_c; /* INPROG I/O only */
221 void *arg_p; /* INPROG I/O only */
222 off_t arg_o; /* INPROG I/O only */
224 int act; /* activity */
227 typedef struct hammer2_io hammer2_io_t;
230 * Primary chain structure keeps track of the topology in-memory.
232 struct hammer2_chain {
233 hammer2_chain_core_t core;
234 RB_ENTRY(hammer2_chain) rbnode; /* live chain(s) */
235 hammer2_blockref_t bref;
236 struct hammer2_chain *parent;
237 struct hammer2_state *state; /* if active cache msg */
238 struct hammer2_mount *hmp;
239 struct hammer2_pfsmount *pmp; /* (pfs-cluster pmp or spmp) */
241 hammer2_xid_t flush_xid; /* flush sequencing */
242 hammer2_key_t data_count; /* delta's to apply */
243 hammer2_key_t inode_count; /* delta's to apply */
244 hammer2_key_t data_count_up; /* delta's to apply */
245 hammer2_key_t inode_count_up; /* delta's to apply */
246 hammer2_io_t *dio; /* physical data buffer */
247 u_int bytes; /* physical data size */
251 hammer2_media_data_t *data; /* data pointer shortcut */
252 TAILQ_ENTRY(hammer2_chain) flush_node; /* flush list */
255 typedef struct hammer2_chain hammer2_chain_t;
257 int hammer2_chain_cmp(hammer2_chain_t *chain1, hammer2_chain_t *chain2);
258 RB_PROTOTYPE(hammer2_chain_tree, hammer2_chain, rbnode, hammer2_chain_cmp);
261 * Special notes on flags:
263 * INITIAL - This flag allows a chain to be created and for storage to
264 * be allocated without having to immediately instantiate the
265 * related buffer. The data is assumed to be all-zeros. It
266 * is primarily used for indirect blocks.
268 * MODIFIED- The chain's media data has been modified.
269 * UPDATE - Chain might not be modified but parent blocktable needs update
271 * BMAPPED - Indicates that the chain is present in the parent blockmap.
272 * BMAPUPD - Indicates that the chain is present but needs to be updated
273 * in the parent blockmap.
275 #define HAMMER2_CHAIN_MODIFIED 0x00000001 /* dirty chain data */
276 #define HAMMER2_CHAIN_ALLOCATED 0x00000002 /* kmalloc'd chain */
277 #define HAMMER2_CHAIN_DESTROY 0x00000004
278 #define HAMMER2_CHAIN_FORCECOW 0x00000008 /* force copy-on-wr */
279 #define HAMMER2_CHAIN_DELETED 0x00000010 /* deleted chain */
280 #define HAMMER2_CHAIN_INITIAL 0x00000020 /* initial create */
281 #define HAMMER2_CHAIN_UPDATE 0x00000040 /* need parent update */
282 #define HAMMER2_CHAIN_DEFERRED 0x00000080 /* flush depth defer */
283 #define HAMMER2_CHAIN_IOFLUSH 0x00000100 /* bawrite on put */
284 #define HAMMER2_CHAIN_ONFLUSH 0x00000200 /* on a flush list */
285 #define HAMMER2_CHAIN_UNUSED00000400 0x00000400
286 #define HAMMER2_CHAIN_VOLUMESYNC 0x00000800 /* needs volume sync */
287 #define HAMMER2_CHAIN_UNUSED00001000 0x00001000
288 #define HAMMER2_CHAIN_MOUNTED 0x00002000 /* PFS is mounted */
289 #define HAMMER2_CHAIN_ONRBTREE 0x00004000 /* on parent RB tree */
290 #define HAMMER2_CHAIN_SNAPSHOT 0x00008000 /* snapshot special */
291 #define HAMMER2_CHAIN_EMBEDDED 0x00010000 /* embedded data */
292 #define HAMMER2_CHAIN_RELEASE 0x00020000 /* don't keep around */
293 #define HAMMER2_CHAIN_BMAPPED 0x00040000 /* present in blkmap */
294 #define HAMMER2_CHAIN_BMAPUPD 0x00080000 /* +needs updating */
295 #define HAMMER2_CHAIN_UNUSED00100000 0x00100000
296 #define HAMMER2_CHAIN_PFSROOT 0x00200000 /* in pfs->cluster */
297 #define HAMMER2_CHAIN_PFSBOUNDARY 0x00400000 /* super->pfs inode */
299 #define HAMMER2_CHAIN_FLUSH_MASK (HAMMER2_CHAIN_MODIFIED | \
300 HAMMER2_CHAIN_UPDATE | \
301 HAMMER2_CHAIN_ONFLUSH)
304 * Flags passed to hammer2_chain_lookup() and hammer2_chain_next()
306 * NOTE: MATCHIND allows an indirect block / freemap node to be returned
307 * when the passed key range matches the radix. Remember that key_end
308 * is inclusive (e.g. {0x000,0xFFF}, not {0x000,0x1000}).
310 #define HAMMER2_LOOKUP_NOLOCK 0x00000001 /* ref only */
311 #define HAMMER2_LOOKUP_NODATA 0x00000002 /* data left NULL */
312 #define HAMMER2_LOOKUP_SHARED 0x00000100
313 #define HAMMER2_LOOKUP_MATCHIND 0x00000200 /* return all chains */
314 #define HAMMER2_LOOKUP_UNUSED0400 0x00000400
315 #define HAMMER2_LOOKUP_ALWAYS 0x00000800 /* resolve data */
318 * Flags passed to hammer2_chain_modify() and hammer2_chain_resize()
320 * NOTE: OPTDATA allows us to avoid instantiating buffers for INDIRECT
321 * blocks in the INITIAL-create state.
323 #define HAMMER2_MODIFY_OPTDATA 0x00000002 /* data can be NULL */
324 #define HAMMER2_MODIFY_NO_MODIFY_TID 0x00000004
325 #define HAMMER2_MODIFY_UNUSED0008 0x00000008
326 #define HAMMER2_MODIFY_NOREALLOC 0x00000010
329 * Flags passed to hammer2_chain_lock()
331 #define HAMMER2_RESOLVE_NEVER 1
332 #define HAMMER2_RESOLVE_MAYBE 2
333 #define HAMMER2_RESOLVE_ALWAYS 3
334 #define HAMMER2_RESOLVE_MASK 0x0F
336 #define HAMMER2_RESOLVE_SHARED 0x10 /* request shared lock */
337 #define HAMMER2_RESOLVE_NOREF 0x20 /* already ref'd on lock */
340 * Flags passed to hammer2_chain_delete()
342 #define HAMMER2_DELETE_PERMANENT 0x0001
343 #define HAMMER2_DELETE_NOSTATS 0x0002
345 #define HAMMER2_INSERT_NOSTATS 0x0002
348 * Flags passed to hammer2_chain_delete_duplicate()
350 #define HAMMER2_DELDUP_RECORE 0x0001
353 * Cluster different types of storage together for allocations
355 #define HAMMER2_FREECACHE_INODE 0
356 #define HAMMER2_FREECACHE_INDIR 1
357 #define HAMMER2_FREECACHE_DATA 2
358 #define HAMMER2_FREECACHE_UNUSED3 3
359 #define HAMMER2_FREECACHE_TYPES 4
362 * hammer2_freemap_alloc() block preference
364 #define HAMMER2_OFF_NOPREF ((hammer2_off_t)-1)
367 * BMAP read-ahead maximum parameters
369 #define HAMMER2_BMAP_COUNT 16 /* max bmap read-ahead */
370 #define HAMMER2_BMAP_BYTES (HAMMER2_PBUFSIZE * HAMMER2_BMAP_COUNT)
375 #define HAMMER2_FLUSH_DEPTH_LIMIT 10 /* stack recursion limit */
378 * hammer2_freemap_adjust()
380 #define HAMMER2_FREEMAP_DORECOVER 1
381 #define HAMMER2_FREEMAP_DOMAYFREE 2
382 #define HAMMER2_FREEMAP_DOREALFREE 3
385 * HAMMER2 cluster - A set of chains representing the same entity.
387 * The hammer2_pfsmount structure embeds a hammer2_cluster. All other
388 * hammer2_cluster use cases use temporary allocations.
390 * The cluster API mimics the chain API. Except as used in the pfsmount,
391 * the cluster structure is a temporary 'working copy' of a set of chains
392 * representing targets compatible with the operation. However, for
393 * performance reasons the cluster API does not necessarily issue concurrent
394 * requests to the underlying chain API for all compatible chains all the
395 * time. This may sometimes necessitate revisiting parent cluster nodes
396 * to 'flesh out' (validate more chains).
398 * If an insufficient number of chains remain in a working copy, the operation
399 * may have to be downgraded, retried, or stall until the requisit number
400 * of chains are available.
402 #define HAMMER2_MAXCLUSTER 8
404 struct hammer2_cluster {
405 int status; /* operational status */
406 int refs; /* track for deallocation */
407 struct hammer2_pfsmount *pmp;
410 hammer2_chain_t *focus; /* current focus (or mod) */
411 hammer2_chain_t *array[HAMMER2_MAXCLUSTER];
412 char missed[HAMMER2_MAXCLUSTER];
413 int cache_index[HAMMER2_MAXCLUSTER];
416 typedef struct hammer2_cluster hammer2_cluster_t;
418 #define HAMMER2_CLUSTER_INODE 0x00000001 /* embedded in inode */
419 #define HAMMER2_CLUSTER_NOSYNC 0x00000002 /* not in sync (cumulative) */
422 RB_HEAD(hammer2_inode_tree, hammer2_inode);
427 * NOTE: The inode's attribute CST which is also used to lock the inode
428 * is embedded in the chain (chain.cst) and aliased w/ attr_cst.
430 struct hammer2_inode {
431 RB_ENTRY(hammer2_inode) rbnode; /* inumber lookup (HL) */
432 ccms_cst_t topo_cst; /* directory topology cst */
433 struct hammer2_pfsmount *pmp; /* PFS mount */
434 struct hammer2_inode *pip; /* parent inode */
436 hammer2_cluster_t cluster;
437 struct lockf advlock;
440 u_int refs; /* +vpref, +flushref */
441 uint8_t comp_heuristic;
446 typedef struct hammer2_inode hammer2_inode_t;
448 #define HAMMER2_INODE_MODIFIED 0x0001
449 #define HAMMER2_INODE_SROOT 0x0002 /* kmalloc special case */
450 #define HAMMER2_INODE_RENAME_INPROG 0x0004
451 #define HAMMER2_INODE_ONRBTREE 0x0008
452 #define HAMMER2_INODE_RESIZED 0x0010
453 #define HAMMER2_INODE_MTIME 0x0020
454 #define HAMMER2_INODE_UNLINKED 0x0040
456 int hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2);
457 RB_PROTOTYPE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
461 * inode-unlink side-structure
463 struct hammer2_inode_unlink {
464 TAILQ_ENTRY(hammer2_inode_unlink) entry;
467 TAILQ_HEAD(h2_unlk_list, hammer2_inode_unlink);
469 typedef struct hammer2_inode_unlink hammer2_inode_unlink_t;
472 * A hammer2 transaction and flush sequencing structure.
474 * This global structure is tied into hammer2_mount and is used
475 * to sequence modifying operations and flushes.
477 * (a) Any modifying operations with sync_tid >= flush_tid will stall until
478 * all modifying operating with sync_tid < flush_tid complete.
480 * The flush related to flush_tid stalls until all modifying operations
481 * with sync_tid < flush_tid complete.
483 * (b) Once unstalled, modifying operations with sync_tid > flush_tid are
484 * allowed to run. All modifications cause modify/duplicate operations
485 * to occur on the related chains. Note that most INDIRECT blocks will
486 * be unaffected because the modifications just overload the RBTREE
487 * structurally instead of actually modifying the indirect blocks.
489 * (c) The actual flush unstalls and RUNS CONCURRENTLY with (b), but only
490 * utilizes the chain structures with sync_tid <= flush_tid. The
491 * flush will modify related indirect blocks and inodes in-place
492 * (rather than duplicate) since the adjustments are compatible with
493 * (b)'s RBTREE overloading
495 * SPECIAL NOTE: Inode modifications have to also propagate along any
496 * modify/duplicate chains. File writes detect the flush
497 * and force out the conflicting buffer cache buffer(s)
498 * before reusing them.
500 * (d) Snapshots can be made instantly but must be flushed and disconnected
501 * from their duplicative source before they can be mounted. This is
502 * because while H2's on-media structure supports forks, its in-memory
503 * structure only supports very simple forking for background flushing
506 * TODO: Flush merging. When fsync() is called on multiple discrete files
507 * concurrently there is no reason to stall the second fsync.
508 * The final flush that reaches to root can cover both fsync()s.
510 * The chains typically terminate as they fly onto the disk. The flush
511 * ultimately reaches the volume header.
513 struct hammer2_trans {
514 TAILQ_ENTRY(hammer2_trans) entry;
515 struct hammer2_pfsmount *pmp;
516 hammer2_xid_t sync_xid;
517 hammer2_tid_t inode_tid; /* inode number assignment */
518 thread_t td; /* pointer */
521 uint8_t inodes_created;
525 typedef struct hammer2_trans hammer2_trans_t;
527 #define HAMMER2_TRANS_ISFLUSH 0x0001 /* formal flush */
528 #define HAMMER2_TRANS_CONCURRENT 0x0002 /* concurrent w/flush */
529 #define HAMMER2_TRANS_BUFCACHE 0x0004 /* from bioq strategy write */
530 #define HAMMER2_TRANS_NEWINODE 0x0008 /* caller allocating inode */
531 #define HAMMER2_TRANS_UNUSED0010 0x0010
532 #define HAMMER2_TRANS_PREFLUSH 0x0020 /* preflush state */
534 #define HAMMER2_FREEMAP_HEUR_NRADIX 4 /* pwr 2 PBUFRADIX-MINIORADIX */
535 #define HAMMER2_FREEMAP_HEUR_TYPES 8
536 #define HAMMER2_FREEMAP_HEUR (HAMMER2_FREEMAP_HEUR_NRADIX * \
537 HAMMER2_FREEMAP_HEUR_TYPES)
539 #define HAMMER2_CLUSTER_COPY_NOCHAINS 0x0001 /* do not copy or ref chains */
540 #define HAMMER2_CLUSTER_COPY_NOREF 0x0002 /* do not ref chains or cl */
543 * Transaction Rendezvous
545 TAILQ_HEAD(hammer2_trans_queue, hammer2_trans);
547 struct hammer2_trans_manage {
548 hammer2_xid_t flush_xid; /* last flush transaction */
549 hammer2_xid_t alloc_xid;
550 struct lock translk; /* lockmgr lock */
551 struct hammer2_trans_queue transq; /* modifying transactions */
552 int flushcnt; /* track flush trans */
555 typedef struct hammer2_trans_manage hammer2_trans_manage_t;
558 * Global (per device) mount structure for device (aka vp->v_mount->hmp)
560 struct hammer2_mount {
561 struct vnode *devvp; /* device vnode */
562 int ronly; /* read-only mount */
563 int pmp_count; /* PFS mounts backed by us */
564 TAILQ_ENTRY(hammer2_mount) mntentry; /* hammer2_mntlist */
566 struct malloc_type *mchain;
569 kdmsg_iocom_t iocom; /* volume-level dmsg interface */
570 struct spinlock io_spin; /* iotree access */
571 struct hammer2_io_tree iotree;
573 hammer2_chain_t vchain; /* anchor chain (topology) */
574 hammer2_chain_t fchain; /* anchor chain (freemap) */
575 struct spinlock list_spin;
576 struct h2_flush_list flushq; /* flush seeds */
577 struct hammer2_pfsmount *spmp; /* super-root pmp for transactions */
578 struct lock vollk; /* lockmgr lock */
579 hammer2_off_t heur_freemap[HAMMER2_FREEMAP_HEUR];
580 int volhdrno; /* last volhdrno written */
581 hammer2_volume_data_t voldata;
582 hammer2_volume_data_t volsync; /* synchronized voldata */
585 typedef struct hammer2_mount hammer2_mount_t;
588 * HAMMER2 PFS mount point structure (aka vp->v_mount->mnt_data).
589 * This has a 1:1 correspondence to struct mount (note that the
590 * hammer2_mount structure has a N:1 correspondence).
592 * This structure represents a cluster mount and not necessarily a
593 * PFS under a specific device mount (HMP). The distinction is important
594 * because the elements backing a cluster mount can change on the fly.
596 * Usually the first element under the cluster represents the original
597 * user-requested mount that bootstraps the whole mess. In significant
598 * setups the original is usually just a read-only media image (or
599 * representitive file) that simply contains a bootstrap volume header
600 * listing the configuration.
602 struct hammer2_pfsmount {
604 TAILQ_ENTRY(hammer2_pfsmount) mntentry; /* hammer2_pfslist */
607 hammer2_mount_t *spmp_hmp; /* (spmp only) */
608 hammer2_inode_t *iroot; /* PFS root inode */
609 hammer2_inode_t *ihidden; /* PFS hidden directory */
610 struct lock lock; /* PFS lock for certain ops */
611 hammer2_off_t inode_count; /* copy of inode_count */
612 ccms_domain_t ccms_dom;
613 struct netexport export; /* nfs export */
614 int ronly; /* read-only mount */
615 struct malloc_type *minode;
616 struct malloc_type *mmsg;
617 struct spinlock inum_spin; /* inumber lookup */
618 struct hammer2_inode_tree inum_tree; /* (not applicable to spmp) */
619 hammer2_tid_t alloc_tid;
620 hammer2_tid_t flush_tid;
621 hammer2_tid_t inode_tid;
623 uint32_t inmem_dirty_chains;
624 int count_lwinprog; /* logical write in prog */
625 struct spinlock list_spin;
626 struct h2_unlk_list unlinkq; /* last-close unlink */
627 thread_t wthread_td; /* write thread td */
628 struct bio_queue_head wthread_bioq; /* logical buffer bioq */
629 struct mtx wthread_mtx; /* interlock */
630 int wthread_destroy;/* termination sequencing */
633 typedef struct hammer2_pfsmount hammer2_pfsmount_t;
635 #define HAMMER2_DIRTYCHAIN_WAITING 0x80000000
636 #define HAMMER2_DIRTYCHAIN_MASK 0x7FFFFFFF
638 #define HAMMER2_LWINPROG_WAITING 0x80000000
639 #define HAMMER2_LWINPROG_MASK 0x7FFFFFFF
643 MALLOC_DECLARE(M_HAMMER2);
645 #define VTOI(vp) ((hammer2_inode_t *)(vp)->v_data)
646 #define ITOV(ip) ((ip)->vp)
649 * Currently locked chains retain the locked buffer cache buffer for
650 * indirect blocks, and indirect blocks can be one of two sizes. The
651 * device buffer has to match the case to avoid deadlocking recursive
652 * chains that might otherwise try to access different offsets within
653 * the same device buffer.
657 hammer2_devblkradix(int radix)
659 if (radix <= HAMMER2_LBUFRADIX) {
660 return (HAMMER2_LBUFRADIX);
662 return (HAMMER2_PBUFRADIX);
668 hammer2_devblksize(size_t bytes)
670 if (bytes <= HAMMER2_LBUFSIZE) {
671 return(HAMMER2_LBUFSIZE);
673 KKASSERT(bytes <= HAMMER2_PBUFSIZE &&
674 (bytes ^ (bytes - 1)) == ((bytes << 1) - 1));
675 return (HAMMER2_PBUFSIZE);
682 MPTOPMP(struct mount *mp)
684 return ((hammer2_pfsmount_t *)mp->mnt_data);
687 #define LOCKSTART int __nlocks = curthread->td_locks
688 #define LOCKENTER (++curthread->td_locks)
689 #define LOCKEXIT (--curthread->td_locks)
690 #define LOCKSTOP KKASSERT(curthread->td_locks == __nlocks)
692 extern struct vop_ops hammer2_vnode_vops;
693 extern struct vop_ops hammer2_spec_vops;
694 extern struct vop_ops hammer2_fifo_vops;
696 extern int hammer2_debug;
697 extern int hammer2_cluster_enable;
698 extern int hammer2_hardlink_enable;
699 extern int hammer2_flush_pipe;
700 extern int hammer2_synchronous_flush;
701 extern int hammer2_dio_count;
702 extern long hammer2_limit_dirty_chains;
703 extern long hammer2_iod_file_read;
704 extern long hammer2_iod_meta_read;
705 extern long hammer2_iod_indr_read;
706 extern long hammer2_iod_fmap_read;
707 extern long hammer2_iod_volu_read;
708 extern long hammer2_iod_file_write;
709 extern long hammer2_iod_meta_write;
710 extern long hammer2_iod_indr_write;
711 extern long hammer2_iod_fmap_write;
712 extern long hammer2_iod_volu_write;
713 extern long hammer2_ioa_file_read;
714 extern long hammer2_ioa_meta_read;
715 extern long hammer2_ioa_indr_read;
716 extern long hammer2_ioa_fmap_read;
717 extern long hammer2_ioa_volu_read;
718 extern long hammer2_ioa_file_write;
719 extern long hammer2_ioa_meta_write;
720 extern long hammer2_ioa_indr_write;
721 extern long hammer2_ioa_fmap_write;
722 extern long hammer2_ioa_volu_write;
724 extern struct objcache *cache_buffer_read;
725 extern struct objcache *cache_buffer_write;
728 extern int write_thread_wakeup;
730 extern mtx_t thread_protect;
735 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size))
736 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc))
738 hammer2_cluster_t *hammer2_inode_lock_ex(hammer2_inode_t *ip);
739 hammer2_cluster_t *hammer2_inode_lock_sh(hammer2_inode_t *ip);
740 void hammer2_inode_unlock_ex(hammer2_inode_t *ip, hammer2_cluster_t *chain);
741 void hammer2_inode_unlock_sh(hammer2_inode_t *ip, hammer2_cluster_t *chain);
742 ccms_state_t hammer2_inode_lock_temp_release(hammer2_inode_t *ip);
743 void hammer2_inode_lock_temp_restore(hammer2_inode_t *ip, ccms_state_t ostate);
744 ccms_state_t hammer2_inode_lock_upgrade(hammer2_inode_t *ip);
745 void hammer2_inode_lock_downgrade(hammer2_inode_t *ip, ccms_state_t ostate);
747 void hammer2_mount_exlock(hammer2_mount_t *hmp);
748 void hammer2_mount_shlock(hammer2_mount_t *hmp);
749 void hammer2_mount_unlock(hammer2_mount_t *hmp);
751 int hammer2_get_dtype(const hammer2_inode_data_t *ipdata);
752 int hammer2_get_vtype(const hammer2_inode_data_t *ipdata);
753 u_int8_t hammer2_get_obj_type(enum vtype vtype);
754 void hammer2_time_to_timespec(u_int64_t xtime, struct timespec *ts);
755 u_int64_t hammer2_timespec_to_time(const struct timespec *ts);
756 u_int32_t hammer2_to_unix_xid(const uuid_t *uuid);
757 void hammer2_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
758 hammer2_xid_t hammer2_trans_newxid(hammer2_pfsmount_t *pmp);
759 void hammer2_trans_manage_init(void);
761 hammer2_key_t hammer2_dirhash(const unsigned char *name, size_t len);
762 int hammer2_getradix(size_t bytes);
764 int hammer2_calc_logical(hammer2_inode_t *ip, hammer2_off_t uoff,
765 hammer2_key_t *lbasep, hammer2_key_t *leofp);
766 int hammer2_calc_physical(hammer2_inode_t *ip,
767 const hammer2_inode_data_t *ipdata,
768 hammer2_key_t lbase);
769 void hammer2_update_time(uint64_t *timep);
770 void hammer2_adjreadcounter(hammer2_blockref_t *bref, size_t bytes);
775 struct vnode *hammer2_igetv(hammer2_inode_t *ip, hammer2_cluster_t *cparent,
777 void hammer2_inode_lock_nlinks(hammer2_inode_t *ip);
778 void hammer2_inode_unlock_nlinks(hammer2_inode_t *ip);
779 hammer2_inode_t *hammer2_inode_lookup(hammer2_pfsmount_t *pmp,
781 hammer2_inode_t *hammer2_inode_get(hammer2_pfsmount_t *pmp,
782 hammer2_inode_t *dip, hammer2_cluster_t *cluster);
783 void hammer2_inode_free(hammer2_inode_t *ip);
784 void hammer2_inode_ref(hammer2_inode_t *ip);
785 void hammer2_inode_drop(hammer2_inode_t *ip);
786 void hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
787 hammer2_cluster_t *cluster);
788 void hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfsmount_t *pmp);
790 hammer2_inode_t *hammer2_inode_create(hammer2_trans_t *trans,
791 hammer2_inode_t *dip,
792 struct vattr *vap, struct ucred *cred,
793 const uint8_t *name, size_t name_len,
794 hammer2_cluster_t **clusterp, int *errorp);
795 int hammer2_inode_connect(hammer2_trans_t *trans,
796 hammer2_cluster_t **clusterp, int hlink,
797 hammer2_inode_t *dip, hammer2_cluster_t *dcluster,
798 const uint8_t *name, size_t name_len,
800 hammer2_inode_t *hammer2_inode_common_parent(hammer2_inode_t *fdip,
801 hammer2_inode_t *tdip);
802 void hammer2_inode_fsync(hammer2_trans_t *trans, hammer2_inode_t *ip,
803 hammer2_cluster_t *cparent);
804 int hammer2_unlink_file(hammer2_trans_t *trans, hammer2_inode_t *dip,
805 const uint8_t *name, size_t name_len, int isdir,
806 int *hlinkp, struct nchandle *nch, int nlinks);
807 int hammer2_hardlink_consolidate(hammer2_trans_t *trans,
808 hammer2_inode_t *ip, hammer2_cluster_t **clusterp,
809 hammer2_inode_t *cdip, hammer2_cluster_t *cdcluster,
811 int hammer2_hardlink_deconsolidate(hammer2_trans_t *trans, hammer2_inode_t *dip,
812 hammer2_chain_t **chainp, hammer2_chain_t **ochainp);
813 int hammer2_hardlink_find(hammer2_inode_t *dip, hammer2_cluster_t **cparentp,
814 hammer2_cluster_t *cluster);
815 int hammer2_parent_find(hammer2_cluster_t **cparentp,
816 hammer2_cluster_t *cluster);
817 void hammer2_inode_install_hidden(hammer2_pfsmount_t *pmp);
822 void hammer2_voldata_lock(hammer2_mount_t *hmp);
823 void hammer2_voldata_unlock(hammer2_mount_t *hmp);
824 void hammer2_voldata_modify(hammer2_mount_t *hmp);
825 hammer2_chain_t *hammer2_chain_alloc(hammer2_mount_t *hmp,
826 hammer2_pfsmount_t *pmp,
827 hammer2_trans_t *trans,
828 hammer2_blockref_t *bref);
829 void hammer2_chain_core_alloc(hammer2_trans_t *trans, hammer2_chain_t *chain);
830 void hammer2_chain_ref(hammer2_chain_t *chain);
831 void hammer2_chain_drop(hammer2_chain_t *chain);
832 int hammer2_chain_lock(hammer2_chain_t *chain, int how);
833 void hammer2_chain_load_async(hammer2_cluster_t *cluster,
834 void (*func)(hammer2_io_t *dio,
835 hammer2_cluster_t *cluster,
836 hammer2_chain_t *chain,
837 void *arg_p, off_t arg_o),
839 void hammer2_chain_moved(hammer2_chain_t *chain);
840 void hammer2_chain_modify(hammer2_trans_t *trans,
841 hammer2_chain_t *chain, int flags);
842 void hammer2_chain_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
843 hammer2_chain_t *parent,
844 hammer2_chain_t *chain,
845 int nradix, int flags);
846 void hammer2_chain_unlock(hammer2_chain_t *chain);
847 void hammer2_chain_wait(hammer2_chain_t *chain);
848 hammer2_chain_t *hammer2_chain_get(hammer2_chain_t *parent, int generation,
849 hammer2_blockref_t *bref);
850 hammer2_chain_t *hammer2_chain_lookup_init(hammer2_chain_t *parent, int flags);
851 void hammer2_chain_lookup_done(hammer2_chain_t *parent);
852 hammer2_chain_t *hammer2_chain_lookup(hammer2_chain_t **parentp,
853 hammer2_key_t *key_nextp,
854 hammer2_key_t key_beg, hammer2_key_t key_end,
855 int *cache_indexp, int flags, int *ddflagp);
856 hammer2_chain_t *hammer2_chain_next(hammer2_chain_t **parentp,
857 hammer2_chain_t *chain,
858 hammer2_key_t *key_nextp,
859 hammer2_key_t key_beg, hammer2_key_t key_end,
860 int *cache_indexp, int flags);
861 hammer2_chain_t *hammer2_chain_scan(hammer2_chain_t *parent,
862 hammer2_chain_t *chain,
863 int *cache_indexp, int flags);
865 int hammer2_chain_create(hammer2_trans_t *trans, hammer2_chain_t **parentp,
866 hammer2_chain_t **chainp,
867 hammer2_pfsmount_t *pmp,
868 hammer2_key_t key, int keybits,
869 int type, size_t bytes, int flags);
870 void hammer2_chain_rename(hammer2_trans_t *trans, hammer2_blockref_t *bref,
871 hammer2_chain_t **parentp,
872 hammer2_chain_t *chain, int flags);
873 int hammer2_chain_snapshot(hammer2_trans_t *trans, hammer2_chain_t **chainp,
874 hammer2_ioc_pfs_t *pfs);
875 void hammer2_chain_delete(hammer2_trans_t *trans, hammer2_chain_t *parent,
876 hammer2_chain_t *chain, int flags);
877 void hammer2_chain_delete_duplicate(hammer2_trans_t *trans,
878 hammer2_chain_t **chainp, int flags);
879 void hammer2_flush(hammer2_trans_t *trans, hammer2_chain_t *chain);
880 void hammer2_chain_commit(hammer2_trans_t *trans, hammer2_chain_t *chain);
881 void hammer2_chain_setflush(hammer2_trans_t *trans, hammer2_chain_t *chain);
882 void hammer2_chain_countbrefs(hammer2_chain_t *chain,
883 hammer2_blockref_t *base, int count);
885 void hammer2_pfs_memory_wait(hammer2_pfsmount_t *pmp);
886 void hammer2_pfs_memory_inc(hammer2_pfsmount_t *pmp);
887 void hammer2_pfs_memory_wakeup(hammer2_pfsmount_t *pmp);
889 void hammer2_base_delete(hammer2_trans_t *trans, hammer2_chain_t *chain,
890 hammer2_blockref_t *base, int count,
891 int *cache_indexp, hammer2_chain_t *child);
892 void hammer2_base_insert(hammer2_trans_t *trans, hammer2_chain_t *chain,
893 hammer2_blockref_t *base, int count,
894 int *cache_indexp, hammer2_chain_t *child);
899 void hammer2_trans_init(hammer2_trans_t *trans, hammer2_pfsmount_t *pmp,
901 void hammer2_trans_spmp(hammer2_trans_t *trans, hammer2_pfsmount_t *pmp);
902 void hammer2_trans_done(hammer2_trans_t *trans);
907 int hammer2_ioctl(hammer2_inode_t *ip, u_long com, void *data,
908 int fflag, struct ucred *cred);
913 hammer2_io_t *hammer2_io_getblk(hammer2_mount_t *hmp, off_t lbase,
914 int lsize, int *ownerp);
915 void hammer2_io_putblk(hammer2_io_t **diop);
916 void hammer2_io_cleanup(hammer2_mount_t *hmp, struct hammer2_io_tree *tree);
917 char *hammer2_io_data(hammer2_io_t *dio, off_t lbase);
918 int hammer2_io_new(hammer2_mount_t *hmp, off_t lbase, int lsize,
919 hammer2_io_t **diop);
920 int hammer2_io_newnz(hammer2_mount_t *hmp, off_t lbase, int lsize,
921 hammer2_io_t **diop);
922 int hammer2_io_newq(hammer2_mount_t *hmp, off_t lbase, int lsize,
923 hammer2_io_t **diop);
924 int hammer2_io_bread(hammer2_mount_t *hmp, off_t lbase, int lsize,
925 hammer2_io_t **diop);
926 void hammer2_io_breadcb(hammer2_mount_t *hmp, off_t lbase, int lsize,
927 void (*callback)(hammer2_io_t *dio,
928 hammer2_cluster_t *arg_l,
929 hammer2_chain_t *arg_c,
930 void *arg_p, off_t arg_o),
931 hammer2_cluster_t *arg_l,
932 hammer2_chain_t *arg_c,
933 void *arg_p, off_t arg_o);
934 void hammer2_io_bawrite(hammer2_io_t **diop);
935 void hammer2_io_bdwrite(hammer2_io_t **diop);
936 int hammer2_io_bwrite(hammer2_io_t **diop);
937 int hammer2_io_isdirty(hammer2_io_t *dio);
938 void hammer2_io_setdirty(hammer2_io_t *dio);
939 void hammer2_io_setinval(hammer2_io_t *dio, u_int bytes);
940 void hammer2_io_brelse(hammer2_io_t **diop);
941 void hammer2_io_bqrelse(hammer2_io_t **diop);
946 int hammer2_msg_dbg_rcvmsg(kdmsg_msg_t *msg);
947 int hammer2_msg_adhoc_input(kdmsg_msg_t *msg);
952 void hammer2_clusterctl_wakeup(kdmsg_iocom_t *iocom);
953 void hammer2_volconf_update(hammer2_mount_t *hmp, int index);
954 void hammer2_cluster_reconnect(hammer2_mount_t *hmp, struct file *fp);
955 void hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx);
956 void hammer2_bioq_sync(hammer2_pfsmount_t *pmp);
957 int hammer2_vfs_sync(struct mount *mp, int waitflags);
958 void hammer2_lwinprog_ref(hammer2_pfsmount_t *pmp);
959 void hammer2_lwinprog_drop(hammer2_pfsmount_t *pmp);
960 void hammer2_lwinprog_wait(hammer2_pfsmount_t *pmp);
965 int hammer2_freemap_alloc(hammer2_trans_t *trans, hammer2_chain_t *chain,
967 void hammer2_freemap_adjust(hammer2_trans_t *trans, hammer2_mount_t *hmp,
968 hammer2_blockref_t *bref, int how);
973 int hammer2_cluster_need_resize(hammer2_cluster_t *cluster, int bytes);
974 uint8_t hammer2_cluster_type(hammer2_cluster_t *cluster);
975 const hammer2_media_data_t *hammer2_cluster_data(hammer2_cluster_t *cluster);
976 hammer2_media_data_t *hammer2_cluster_wdata(hammer2_cluster_t *cluster);
977 hammer2_cluster_t *hammer2_cluster_from_chain(hammer2_chain_t *chain);
978 int hammer2_cluster_modified(hammer2_cluster_t *cluster);
979 int hammer2_cluster_duplicated(hammer2_cluster_t *cluster);
980 void hammer2_cluster_set_chainflags(hammer2_cluster_t *cluster, uint32_t flags);
981 void hammer2_cluster_bref(hammer2_cluster_t *cluster, hammer2_blockref_t *bref);
982 void hammer2_cluster_setflush(hammer2_trans_t *trans,
983 hammer2_cluster_t *cluster);
984 hammer2_cluster_t *hammer2_cluster_alloc(hammer2_pfsmount_t *pmp,
985 hammer2_trans_t *trans,
986 hammer2_blockref_t *bref);
987 void hammer2_cluster_ref(hammer2_cluster_t *cluster);
988 void hammer2_cluster_drop(hammer2_cluster_t *cluster);
989 void hammer2_cluster_wait(hammer2_cluster_t *cluster);
990 int hammer2_cluster_lock(hammer2_cluster_t *cluster, int how);
991 void hammer2_cluster_replace(hammer2_cluster_t *dst, hammer2_cluster_t *src);
992 void hammer2_cluster_replace_locked(hammer2_cluster_t *dst,
993 hammer2_cluster_t *src);
994 hammer2_cluster_t *hammer2_cluster_copy(hammer2_cluster_t *ocluster,
996 void hammer2_cluster_unlock(hammer2_cluster_t *cluster);
997 void hammer2_cluster_resize(hammer2_trans_t *trans, hammer2_inode_t *ip,
998 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
999 int nradix, int flags);
1000 hammer2_inode_data_t *hammer2_cluster_modify_ip(hammer2_trans_t *trans,
1001 hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1003 void hammer2_cluster_modify(hammer2_trans_t *trans, hammer2_cluster_t *cluster,
1005 void hammer2_cluster_modsync(hammer2_cluster_t *cluster);
1006 hammer2_cluster_t *hammer2_cluster_lookup_init(hammer2_cluster_t *cparent,
1008 void hammer2_cluster_lookup_done(hammer2_cluster_t *cparent);
1009 hammer2_cluster_t *hammer2_cluster_lookup(hammer2_cluster_t *cparent,
1010 hammer2_key_t *key_nextp,
1011 hammer2_key_t key_beg, hammer2_key_t key_end,
1012 int flags, int *ddflagp);
1013 hammer2_cluster_t *hammer2_cluster_next(hammer2_cluster_t *cparent,
1014 hammer2_cluster_t *cluster,
1015 hammer2_key_t *key_nextp,
1016 hammer2_key_t key_beg, hammer2_key_t key_end,
1018 hammer2_cluster_t *hammer2_cluster_scan(hammer2_cluster_t *cparent,
1019 hammer2_cluster_t *cluster, int flags);
1020 int hammer2_cluster_create(hammer2_trans_t *trans, hammer2_cluster_t *cparent,
1021 hammer2_cluster_t **clusterp,
1022 hammer2_key_t key, int keybits,
1023 int type, size_t bytes, int flags);
1024 void hammer2_cluster_rename(hammer2_trans_t *trans, hammer2_blockref_t *bref,
1025 hammer2_cluster_t *cparent, hammer2_cluster_t *cluster,
1027 void hammer2_cluster_delete(hammer2_trans_t *trans, hammer2_cluster_t *pcluster,
1028 hammer2_cluster_t *cluster, int flags);
1029 int hammer2_cluster_snapshot(hammer2_trans_t *trans,
1030 hammer2_cluster_t *ocluster, hammer2_ioc_pfs_t *pfs);
1031 hammer2_cluster_t *hammer2_cluster_parent(hammer2_cluster_t *cluster);
1034 #endif /* !_KERNEL */
1035 #endif /* !_VFS_HAMMER2_HAMMER2_H_ */