?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
HAMMER - Rework write pipelining
[dragonfly.git] / sys / vfs / hammer / hammer.h
blob501750cefd3c56abe4993e2d102695cafbb1848a
1 /*
2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * $DragonFly: src/sys/vfs/hammer/hammer.h,v 1.130 2008/11/13 02:18:43 dillon Exp $
35 */
36 /*
37 * This header file contains structures used internally by the HAMMERFS
38 * implementation. See hammer_disk.h for on-disk structures.
39 */
40
41 #include <sys/param.h>
42 #include <sys/types.h>
43 #include <sys/kernel.h>
44 #include <sys/conf.h>
45 #include <sys/systm.h>
46 #include <sys/tree.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/vnode.h>
51 #include <sys/proc.h>
52 #include <sys/priv.h>
53 #include <sys/stat.h>
54 #include <sys/globaldata.h>
55 #include <sys/lockf.h>
56 #include <sys/buf.h>
57 #include <sys/queue.h>
58 #include <sys/ktr.h>
59 #include <sys/globaldata.h>
60
61 #include <sys/buf2.h>
62 #include <sys/signal2.h>
63 #include "hammer_disk.h"
64 #include "hammer_mount.h"
65 #include "hammer_ioctl.h"
66
67 #if defined(_KERNEL) || defined(_KERNEL_STRUCTURES)
68
69 MALLOC_DECLARE(M_HAMMER);
70
71 /*
72 * Kernel trace
73 */
74 #if !defined(KTR_HAMMER)
75 #define KTR_HAMMER KTR_ALL
76 #endif
77 KTR_INFO_MASTER_EXTERN(hammer);
78
79 /*
80 * Misc structures
81 */
82 struct hammer_mount;
83
84 /*
85 * Key structure used for custom RB tree inode lookups. This prototypes
86 * the function hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info).
87 */
88 typedef struct hammer_inode_info {
89 int64_t obj_id; /* (key) object identifier */
90 hammer_tid_t obj_asof; /* (key) snapshot transid or 0 */
91 u_int32_t obj_localization; /* (key) pseudo-fs */
92 union {
93 struct hammer_btree_leaf_elm *leaf;
94 } u;
95 } *hammer_inode_info_t;
96
97 typedef enum hammer_transaction_type {
98 HAMMER_TRANS_RO,
99 HAMMER_TRANS_STD,
100 HAMMER_TRANS_FLS
101 } hammer_transaction_type_t;
102
103 /*
104 * HAMMER Transaction tracking
105 */
106 struct hammer_transaction {
107 hammer_transaction_type_t type;
108 struct hammer_mount *hmp;
109 hammer_tid_t tid;
110 u_int64_t time;
111 u_int32_t time32;
112 int sync_lock_refs;
113 int flags;
114 struct hammer_volume *rootvol;
115 };
116
117 typedef struct hammer_transaction *hammer_transaction_t;
118
119 #define HAMMER_TRANSF_NEWINODE 0x0001
120 #define HAMMER_TRANSF_DIDIO 0x0002
121 #define HAMMER_TRANSF_CRCDOM 0x0004 /* EDOM on CRC error, less critical */
122
123 /*
124 * HAMMER locks
125 */
126 struct hammer_lock {
127 int refs; /* active references delay writes */
128 volatile u_int lockval; /* lock count and control bits */
129 struct thread *owner; /* owner if exclusively held */
130 };
131
132 #define HAMMER_LOCKF_EXCLUSIVE 0x40000000
133 #define HAMMER_LOCKF_WANTED 0x80000000
134
135 static __inline int
136 hammer_notlocked(struct hammer_lock *lock)
137 {
138 return(lock->lockval == 0);
139 }
140
141 static __inline int
142 hammer_islocked(struct hammer_lock *lock)
143 {
144 return(lock->lockval != 0);
145 }
146
147 static __inline int
148 hammer_isactive(struct hammer_lock *lock)
149 {
150 return(lock->refs != 0);
151 }
152
153 static __inline int
154 hammer_islastref(struct hammer_lock *lock)
155 {
156 return(lock->refs == 1);
157 }
158
159 /*
160 * Return if we specifically own the lock exclusively.
161 */
162 static __inline int
163 hammer_lock_excl_owned(struct hammer_lock *lock, thread_t td)
164 {
165 if ((lock->lockval & HAMMER_LOCKF_EXCLUSIVE) &&
166 lock->owner == td) {
167 return(1);
168 }
169 return(0);
170 }
171
172 /*
173 * Flush state, used by various structures
174 */
175 typedef enum hammer_inode_state {
176 HAMMER_FST_IDLE,
177 HAMMER_FST_SETUP,
178 HAMMER_FST_FLUSH
179 } hammer_inode_state_t;
180
181 TAILQ_HEAD(hammer_record_list, hammer_record);
182
183 /*
184 * Pseudo-filesystem extended data tracking
185 */
186 struct hammer_pfs_rb_tree;
187 struct hammer_pseudofs_inmem;
188 RB_HEAD(hammer_pfs_rb_tree, hammer_pseudofs_inmem);
189 RB_PROTOTYPE2(hammer_pfs_rb_tree, hammer_pseudofs_inmem, rb_node,
190 hammer_pfs_rb_compare, u_int32_t);
191
192 struct hammer_pseudofs_inmem {
193 RB_ENTRY(hammer_pseudofs_inmem) rb_node;
194 struct hammer_lock lock;
195 u_int32_t localization;
196 hammer_tid_t create_tid;
197 int flags;
198 udev_t fsid_udev;
199 struct hammer_pseudofs_data pfsd;
200 };
201
202 typedef struct hammer_pseudofs_inmem *hammer_pseudofs_inmem_t;
203
204 #define HAMMER_PFSM_DELETED 0x0001
205
206 /*
207 * Cache object ids. A fixed number of objid cache structures are
208 * created to reserve object id's for newly created files in multiples
209 * of 100,000, localized to a particular directory, and recycled as
210 * needed. This allows parallel create operations in different
211 * directories to retain fairly localized object ids which in turn
212 * improves reblocking performance and layout.
213 */
214 #define OBJID_CACHE_SIZE 1024
215 #define OBJID_CACHE_BULK_BITS 10 /* 10 bits (1024) */
216 #define OBJID_CACHE_BULK (32 * 32) /* two level (1024) */
217 #define OBJID_CACHE_BULK_MASK (OBJID_CACHE_BULK - 1)
218 #define OBJID_CACHE_BULK_MASK64 ((u_int64_t)(OBJID_CACHE_BULK - 1))
219
220 typedef struct hammer_objid_cache {
221 TAILQ_ENTRY(hammer_objid_cache) entry;
222 struct hammer_inode *dip;
223 hammer_tid_t base_tid;
224 int count;
225 u_int32_t bm0;
226 u_int32_t bm1[32];
227 } *hammer_objid_cache_t;
228
229 /*
230 * Associate an inode with a B-Tree node to cache search start positions
231 */
232 typedef struct hammer_node_cache {
233 TAILQ_ENTRY(hammer_node_cache) entry;
234 struct hammer_node *node;
235 struct hammer_inode *ip;
236 } *hammer_node_cache_t;
237
238 TAILQ_HEAD(hammer_node_cache_list, hammer_node_cache);
239
240 /*
241 * Structure used to organize flush groups. Flush groups must be
242 * organized into chunks in order to avoid blowing out the UNDO FIFO.
243 * Without this a 'sync' could end up flushing 50,000 inodes in a single
244 * transaction.
245 */
246 struct hammer_flush_group {
247 TAILQ_ENTRY(hammer_flush_group) flush_entry;
248 TAILQ_HEAD(, hammer_inode) flush_list;
249 int unused01; /* inode load */
250 int total_count; /* record load */
251 int running; /* group is running */
252 int closed;
253 int refs;
254 };
255
256 typedef struct hammer_flush_group *hammer_flush_group_t;
257
258 TAILQ_HEAD(hammer_flush_group_list, hammer_flush_group);
259
260 /*
261 * Structure used to represent an inode in-memory.
262 *
263 * The record and data associated with an inode may be out of sync with
264 * the disk (xDIRTY flags), or not even on the disk at all (ONDISK flag
265 * clear).
266 *
267 * An inode may also hold a cache of unsynchronized records, used for
268 * database and directories only. Unsynchronized regular file data is
269 * stored in the buffer cache.
270 *
271 * NOTE: A file which is created and destroyed within the initial
272 * synchronization period can wind up not doing any disk I/O at all.
273 *
274 * Finally, an inode may cache numerous disk-referencing B-Tree cursors.
275 */
276 struct hammer_ino_rb_tree;
277 struct hammer_inode;
278 RB_HEAD(hammer_ino_rb_tree, hammer_inode);
279 RB_PROTOTYPEX(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
280 hammer_ino_rb_compare, hammer_inode_info_t);
281
282 struct hammer_rec_rb_tree;
283 struct hammer_record;
284 RB_HEAD(hammer_rec_rb_tree, hammer_record);
285 RB_PROTOTYPEX(hammer_rec_rb_tree, INFO, hammer_record, rb_node,
286 hammer_rec_rb_compare, hammer_btree_leaf_elm_t);
287
288 TAILQ_HEAD(hammer_node_list, hammer_node);
289
290 struct hammer_inode {
291 RB_ENTRY(hammer_inode) rb_node;
292 hammer_inode_state_t flush_state;
293 hammer_flush_group_t flush_group;
294 TAILQ_ENTRY(hammer_inode) flush_entry;
295 struct hammer_record_list target_list; /* target of dependant recs */
296 int64_t obj_id; /* (key) object identifier */
297 hammer_tid_t obj_asof; /* (key) snapshot or 0 */
298 u_int32_t obj_localization; /* (key) pseudo-fs */
299 struct hammer_mount *hmp;
300 hammer_objid_cache_t objid_cache;
301 int flags;
302 int error; /* flush error */
303 int cursor_ip_refs; /* sanity */
304 int rsv_recs;
305 struct vnode *vp;
306 hammer_pseudofs_inmem_t pfsm;
307 struct lockf advlock;
308 struct hammer_lock lock; /* sync copy interlock */
309 off_t trunc_off;
310 struct hammer_btree_leaf_elm ino_leaf; /* in-memory cache */
311 struct hammer_inode_data ino_data; /* in-memory cache */
312 struct hammer_rec_rb_tree rec_tree; /* in-memory cache */
313 int rec_generation;
314 struct hammer_node_cache cache[4]; /* search initiate cache */
315
316 /*
317 * When a demark is created to synchronize an inode to
318 * disk, certain fields are copied so the front-end VOPs
319 * can continue to run in parallel with the synchronization
320 * occuring in the background.
321 */
322 int sync_flags; /* to-sync flags cache */
323 off_t sync_trunc_off; /* to-sync truncation */
324 off_t save_trunc_off; /* write optimization */
325 struct hammer_btree_leaf_elm sync_ino_leaf; /* to-sync cache */
326 struct hammer_inode_data sync_ino_data; /* to-sync cache */
327 };
328
329 typedef struct hammer_inode *hammer_inode_t;
330
331 #define VTOI(vp) ((struct hammer_inode *)(vp)->v_data)
332
333 #define HAMMER_INODE_DDIRTY 0x0001 /* in-memory ino_data is dirty */
334 /* (not including atime/mtime) */
335 #define HAMMER_INODE_RSV_INODES 0x0002 /* hmp->rsv_inodes bumped */
336 #define HAMMER_INODE_CONN_DOWN 0x0004 /* include in downward recursion */
337 #define HAMMER_INODE_XDIRTY 0x0008 /* in-memory records */
338 #define HAMMER_INODE_ONDISK 0x0010 /* inode is on-disk (else not yet) */
339 #define HAMMER_INODE_FLUSH 0x0020 /* flush on last ref */
340 #define HAMMER_INODE_DELETED 0x0080 /* inode delete (backend) */
341 #define HAMMER_INODE_DELONDISK 0x0100 /* delete synchronized to disk */
342 #define HAMMER_INODE_RO 0x0200 /* read-only (because of as-of) */
343 #define HAMMER_INODE_VHELD 0x0400 /* vnode held on sync */
344 #define HAMMER_INODE_DONDISK 0x0800 /* data records may be on disk */
345 #define HAMMER_INODE_BUFS 0x1000 /* dirty high level bps present */
346 #define HAMMER_INODE_REFLUSH 0x2000 /* flush on dependancy / reflush */
347 #define HAMMER_INODE_RECLAIM 0x4000 /* trying to reclaim */
348 #define HAMMER_INODE_FLUSHW 0x8000 /* Someone waiting for flush */
349
350 #define HAMMER_INODE_TRUNCATED 0x00010000
351 #define HAMMER_INODE_DELETING 0x00020000 /* inode delete request (frontend)*/
352 #define HAMMER_INODE_RESIGNAL 0x00040000 /* re-signal on re-flush */
353 #define HAMMER_INODE_ATIME 0x00100000 /* in-memory atime modified */
354 #define HAMMER_INODE_MTIME 0x00200000 /* in-memory mtime modified */
355 #define HAMMER_INODE_WOULDBLOCK 0x00400000 /* re-issue to new flush group */
356 #define HAMMER_INODE_DUMMY 0x00800000 /* dummy inode covering bad file */
357
358 #define HAMMER_INODE_MODMASK (HAMMER_INODE_DDIRTY| \
359 HAMMER_INODE_XDIRTY|HAMMER_INODE_BUFS| \
360 HAMMER_INODE_ATIME|HAMMER_INODE_MTIME| \
361 HAMMER_INODE_TRUNCATED|HAMMER_INODE_DELETING)
362
363 #define HAMMER_INODE_MODMASK_NOXDIRTY \
364 (HAMMER_INODE_MODMASK & ~HAMMER_INODE_XDIRTY)
365
366 #define HAMMER_FLUSH_GROUP_SIZE 64
367
368 #define HAMMER_FLUSH_SIGNAL 0x0001
369 #define HAMMER_FLUSH_RECURSION 0x0002
370
371 /*
372 * Used by the inode reclaim code to pipeline reclaims and avoid
373 * blowing out kernel memory or letting the flusher get too far
374 * behind. The reclaim wakes up when count reaches 0 or the
375 * timer expires.
376 */
377 struct hammer_reclaim {
378 TAILQ_ENTRY(hammer_reclaim) entry;
379 int count;
380 };
381
382 #define HAMMER_RECLAIM_FLUSH 2000
383 #define HAMMER_RECLAIM_WAIT 4000
384
385 /*
386 * Structure used to represent an unsynchronized record in-memory. These
387 * records typically represent directory entries. Only non-historical
388 * records are kept in-memory.
389 *
390 * Records are organized as a per-inode RB-Tree. If the inode is not
391 * on disk then neither are any records and the in-memory record tree
392 * represents the entire contents of the inode. If the inode is on disk
393 * then the on-disk B-Tree is scanned in parallel with the in-memory
394 * RB-Tree to synthesize the current state of the file.
395 *
396 * Records are also used to enforce the ordering of directory create/delete
397 * operations. A new inode will not be flushed to disk unless its related
398 * directory entry is also being flushed at the same time. A directory entry
399 * will not be removed unless its related inode is also being removed at the
400 * same time.
401 */
402 typedef enum hammer_record_type {
403 HAMMER_MEM_RECORD_GENERAL, /* misc record */
404 HAMMER_MEM_RECORD_INODE, /* inode record */
405 HAMMER_MEM_RECORD_ADD, /* positive memory cache record */
406 HAMMER_MEM_RECORD_DEL, /* negative delete-on-disk record */
407 HAMMER_MEM_RECORD_DATA /* bulk-data record w/on-disk ref */
408 } hammer_record_type_t;
409
410 struct hammer_record {
411 RB_ENTRY(hammer_record) rb_node;
412 TAILQ_ENTRY(hammer_record) target_entry;
413 hammer_inode_state_t flush_state;
414 hammer_flush_group_t flush_group;
415 hammer_record_type_t type;
416 struct hammer_lock lock;
417 struct hammer_reserve *resv;
418 struct hammer_inode *ip;
419 struct hammer_inode *target_ip;
420 struct hammer_btree_leaf_elm leaf;
421 union hammer_data_ondisk *data;
422 int flags;
423 hammer_off_t zone2_offset; /* direct-write only */
424 };
425
426 typedef struct hammer_record *hammer_record_t;
427
428 /*
429 * Record flags. Note that FE can only be set by the frontend if the
430 * record has not been interlocked by the backend w/ BE.
431 */
432 #define HAMMER_RECF_ALLOCDATA 0x0001
433 #define HAMMER_RECF_ONRBTREE 0x0002
434 #define HAMMER_RECF_DELETED_FE 0x0004 /* deleted (frontend) */
435 #define HAMMER_RECF_DELETED_BE 0x0008 /* deleted (backend) */
436 #define HAMMER_RECF_COMMITTED 0x0010 /* committed to the B-Tree */
437 #define HAMMER_RECF_INTERLOCK_BE 0x0020 /* backend interlock */
438 #define HAMMER_RECF_WANTED 0x0040 /* wanted by the frontend */
439 #define HAMMER_RECF_CONVERT_DELETE 0x0100 /* special case */
440 #define HAMMER_RECF_DIRECT_IO 0x0200 /* related direct I/O running*/
441 #define HAMMER_RECF_DIRECT_WAIT 0x0400 /* related direct I/O running*/
442 #define HAMMER_RECF_DIRECT_INVAL 0x0800 /* buffer alias invalidation */
443
444 /*
445 * hammer_delete_at_cursor() flags
446 */
447 #define HAMMER_DELETE_ADJUST 0x0001
448 #define HAMMER_DELETE_DESTROY 0x0002
449
450 /*
451 * In-memory structures representing on-disk structures.
452 */
453 struct hammer_volume;
454 struct hammer_buffer;
455 struct hammer_node;
456 struct hammer_undo;
457 struct hammer_reserve;
458
459 RB_HEAD(hammer_vol_rb_tree, hammer_volume);
460 RB_HEAD(hammer_buf_rb_tree, hammer_buffer);
461 RB_HEAD(hammer_nod_rb_tree, hammer_node);
462 RB_HEAD(hammer_und_rb_tree, hammer_undo);
463 RB_HEAD(hammer_res_rb_tree, hammer_reserve);
464
465 RB_PROTOTYPE2(hammer_vol_rb_tree, hammer_volume, rb_node,
466 hammer_vol_rb_compare, int32_t);
467 RB_PROTOTYPE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
468 hammer_buf_rb_compare, hammer_off_t);
469 RB_PROTOTYPE2(hammer_nod_rb_tree, hammer_node, rb_node,
470 hammer_nod_rb_compare, hammer_off_t);
471 RB_PROTOTYPE2(hammer_und_rb_tree, hammer_undo, rb_node,
472 hammer_und_rb_compare, hammer_off_t);
473 RB_PROTOTYPE2(hammer_res_rb_tree, hammer_reserve, rb_node,
474 hammer_res_rb_compare, hammer_off_t);
475
476 /*
477 * IO management - embedded at the head of various in-memory structures
478 *
479 * VOLUME - hammer_volume containing meta-data
480 * META_BUFFER - hammer_buffer containing meta-data
481 * DATA_BUFFER - hammer_buffer containing pure-data
482 *
483 * Dirty volume headers and dirty meta-data buffers are locked until the
484 * flusher can sequence them out. Dirty pure-data buffers can be written.
485 * Clean buffers can be passively released.
486 */
487 typedef enum hammer_io_type {
488 HAMMER_STRUCTURE_VOLUME,
489 HAMMER_STRUCTURE_META_BUFFER,
490 HAMMER_STRUCTURE_UNDO_BUFFER,
491 HAMMER_STRUCTURE_DATA_BUFFER
492 } hammer_io_type_t;
493
494 union hammer_io_structure;
495 struct hammer_io;
496
497 struct worklist {
498 LIST_ENTRY(worklist) node;
499 };
500
501 TAILQ_HEAD(hammer_io_list, hammer_io);
502 typedef struct hammer_io_list *hammer_io_list_t;
503
504 struct hammer_io {
505 struct worklist worklist;
506 struct hammer_lock lock;
507 enum hammer_io_type type;
508 struct hammer_mount *hmp;
509 struct hammer_volume *volume;
510 TAILQ_ENTRY(hammer_io) mod_entry; /* list entry if modified */
511 hammer_io_list_t mod_list;
512 struct buf *bp;
513 int64_t offset; /* zone-2 offset */
514 int bytes; /* buffer cache buffer size */
515 int loading; /* loading/unloading interlock */
516 int modify_refs;
517
518 u_int modified : 1; /* bp's data was modified */
519 u_int released : 1; /* bp released (w/ B_LOCKED set) */
520 u_int running : 1; /* bp write IO in progress */
521 u_int waiting : 1; /* someone is waiting on us */
522 u_int validated : 1; /* ondisk has been validated */
523 u_int waitdep : 1; /* flush waits for dependancies */
524 u_int recovered : 1; /* has recovery ref */
525 u_int waitmod : 1; /* waiting for modify_refs */
526 u_int reclaim : 1; /* reclaim requested */
527 u_int gencrc : 1; /* crc needs to be generated */
528 u_int ioerror : 1; /* abort on io-error */
529 };
530
531 typedef struct hammer_io *hammer_io_t;
532
533 #define HAMMER_CLUSTER_SIZE (64 * 1024)
534 #if HAMMER_CLUSTER_SIZE > MAXBSIZE
535 #undef HAMMER_CLUSTER_SIZE
536 #define HAMMER_CLUSTER_SIZE MAXBSIZE
537 #endif
538 #define HAMMER_CLUSTER_BUFS (HAMMER_CLUSTER_SIZE / HAMMER_BUFSIZE)
539
540 /*
541 * In-memory volume representing on-disk buffer
542 */
543 struct hammer_volume {
544 struct hammer_io io;
545 RB_ENTRY(hammer_volume) rb_node;
546 struct hammer_volume_ondisk *ondisk;
547 int32_t vol_no;
548 int64_t nblocks; /* note: special calculation for statfs */
549 int64_t buffer_base; /* base offset of buffer 0 */
550 hammer_off_t maxbuf_off; /* Maximum buffer offset (zone-2) */
551 hammer_off_t maxraw_off; /* Maximum raw offset for device */
552 char *vol_name;
553 struct vnode *devvp;
554 int vol_flags;
555 };
556
557 typedef struct hammer_volume *hammer_volume_t;
558
559 /*
560 * In-memory buffer (other then volume, super-cluster, or cluster),
561 * representing an on-disk buffer.
562 */
563 struct hammer_buffer {
564 struct hammer_io io;
565 RB_ENTRY(hammer_buffer) rb_node;
566 void *ondisk;
567 hammer_off_t zoneX_offset;
568 hammer_off_t zone2_offset;
569 struct hammer_reserve *resv;
570 struct hammer_node_list clist;
571 };
572
573 typedef struct hammer_buffer *hammer_buffer_t;
574
575 /*
576 * In-memory B-Tree node, representing an on-disk B-Tree node.
577 *
578 * This is a hang-on structure which is backed by a hammer_buffer,
579 * indexed by a hammer_cluster, and used for fine-grained locking of
580 * B-Tree nodes in order to properly control lock ordering. A hammer_buffer
581 * can contain multiple nodes representing wildly disassociated portions
582 * of the B-Tree so locking cannot be done on a buffer-by-buffer basis.
583 *
584 * This structure uses a cluster-relative index to reduce the number
585 * of layers required to access it, and also because all on-disk B-Tree
586 * references are cluster-relative offsets.
587 */
588 struct hammer_node {
589 struct hammer_lock lock; /* node-by-node lock */
590 TAILQ_ENTRY(hammer_node) entry; /* per-buffer linkage */
591 RB_ENTRY(hammer_node) rb_node; /* per-cluster linkage */
592 hammer_off_t node_offset; /* full offset spec */
593 struct hammer_mount *hmp;
594 struct hammer_buffer *buffer; /* backing buffer */
595 hammer_node_ondisk_t ondisk; /* ptr to on-disk structure */
596 TAILQ_HEAD(, hammer_cursor) cursor_list; /* deadlock recovery */
597 struct hammer_node_cache_list cache_list; /* passive caches */
598 int flags;
599 int loading; /* load interlock */
600 };
601
602 #define HAMMER_NODE_DELETED 0x0001
603 #define HAMMER_NODE_FLUSH 0x0002
604 #define HAMMER_NODE_CRCGOOD 0x0004
605 #define HAMMER_NODE_NEEDSCRC 0x0008
606 #define HAMMER_NODE_NEEDSMIRROR 0x0010
607 #define HAMMER_NODE_CRCBAD 0x0020
608
609 #define HAMMER_NODE_CRCANY (HAMMER_NODE_CRCGOOD | HAMMER_NODE_CRCBAD)
610
611 typedef struct hammer_node *hammer_node_t;
612
613 /*
614 * List of locked nodes. This structure is used to lock potentially large
615 * numbers of nodes as an aid for complex B-Tree operations.
616 */
617 struct hammer_node_lock;
618 TAILQ_HEAD(hammer_node_lock_list, hammer_node_lock);
619
620 struct hammer_node_lock {
621 TAILQ_ENTRY(hammer_node_lock) entry;
622 struct hammer_node_lock_list list;
623 struct hammer_node_lock *parent;
624 hammer_node_t node;
625 hammer_node_ondisk_t copy; /* copy of on-disk data */
626 int index; /* index of this node in parent */
627 int count; /* count children */
628 int flags;
629 };
630
631 typedef struct hammer_node_lock *hammer_node_lock_t;
632
633 #define HAMMER_NODE_LOCK_UPDATED 0x0001
634
635 /*
636 * Common I/O management structure - embedded in in-memory structures
637 * which are backed by filesystem buffers.
638 */
639 union hammer_io_structure {
640 struct hammer_io io;
641 struct hammer_volume volume;
642 struct hammer_buffer buffer;
643 };
644
645 typedef union hammer_io_structure *hammer_io_structure_t;
646
647 /*
648 * The reserve structure prevents the blockmap from allocating
649 * out of a reserved bigblock. Such reservations are used by
650 * the direct-write mechanism.
651 *
652 * The structure is also used to hold off on reallocations of
653 * big blocks from the freemap until flush dependancies have
654 * been dealt with.
655 */
656 struct hammer_reserve {
657 RB_ENTRY(hammer_reserve) rb_node;
658 TAILQ_ENTRY(hammer_reserve) delay_entry;
659 int flush_group;
660 int flags;
661 int refs;
662 int zone;
663 int append_off;
664 hammer_off_t zone_offset;
665 };
666
667 typedef struct hammer_reserve *hammer_reserve_t;
668
669 #define HAMMER_RESF_ONDELAY 0x0001
670 #define HAMMER_RESF_LAYER2FREE 0x0002
671
672 #include "hammer_cursor.h"
673
674 /*
675 * The undo structure tracks recent undos to avoid laying down duplicate
676 * undos within a flush group, saving us a significant amount of overhead.
677 *
678 * This is strictly a heuristic.
679 */
680 #define HAMMER_MAX_UNDOS 1024
681 #define HAMMER_MAX_FLUSHERS 4
682
683 struct hammer_undo {
684 RB_ENTRY(hammer_undo) rb_node;
685 TAILQ_ENTRY(hammer_undo) lru_entry;
686 hammer_off_t offset;
687 int bytes;
688 };
689
690 typedef struct hammer_undo *hammer_undo_t;
691
692 struct hammer_flusher_info;
693 TAILQ_HEAD(hammer_flusher_info_list, hammer_flusher_info);
694
695 struct hammer_flusher {
696 int signal; /* flusher thread sequencer */
697 int act; /* currently active flush group */
698 int done; /* set to act when complete */
699 int next; /* next flush group */
700 int group_lock; /* lock sequencing of the next flush */
701 int exiting; /* request master exit */
702 thread_t td; /* master flusher thread */
703 hammer_tid_t tid; /* last flushed transaction id */
704 int finalize_want; /* serialize finalization */
705 struct hammer_lock finalize_lock; /* serialize finalization */
706 struct hammer_transaction trans; /* shared transaction */
707 struct hammer_flusher_info_list run_list;
708 struct hammer_flusher_info_list ready_list;
709 };
710
711 /*
712 * Internal hammer mount data structure
713 */
714 struct hammer_mount {
715 struct mount *mp;
716 /*struct vnode *rootvp;*/
717 struct hammer_ino_rb_tree rb_inos_root;
718 struct hammer_vol_rb_tree rb_vols_root;
719 struct hammer_nod_rb_tree rb_nods_root;
720 struct hammer_und_rb_tree rb_undo_root;
721 struct hammer_res_rb_tree rb_resv_root;
722 struct hammer_buf_rb_tree rb_bufs_root;
723 struct hammer_pfs_rb_tree rb_pfsm_root;
724 struct hammer_volume *rootvol;
725 struct hammer_base_elm root_btree_beg;
726 struct hammer_base_elm root_btree_end;
727
728 struct malloc_type *m_misc;
729 struct malloc_type *m_inodes;
730
731 int flags; /* HAMMER_MOUNT_xxx flags */
732 int hflags;
733 int ronly;
734 int nvolumes;
735 int volume_iterator;
736 int master_id; /* -1 or 0-15 - clustering and mirroring */
737 int version; /* hammer filesystem version to use */
738 int rsv_inodes; /* reserved space due to dirty inodes */
739 int64_t rsv_databytes; /* reserved space due to record data */
740 int rsv_recs; /* reserved space due to dirty records */
741 int rsv_fromdelay; /* bigblocks reserved due to flush delay */
742 int undo_rec_limit; /* based on size of undo area */
743 int last_newrecords;
744 int count_newrecords;
745
746 int inode_reclaims; /* inodes pending reclaim by flusher */
747 int count_inodes; /* total number of inodes */
748 int count_iqueued; /* inodes queued to flusher */
749
750 struct hammer_flusher flusher;
751
752 u_int check_interrupt;
753 u_int check_yield;
754 uuid_t fsid;
755 struct hammer_io_list volu_list; /* dirty undo buffers */
756 struct hammer_io_list undo_list; /* dirty undo buffers */
757 struct hammer_io_list data_list; /* dirty data buffers */
758 struct hammer_io_list alt_data_list; /* dirty data buffers */
759 struct hammer_io_list meta_list; /* dirty meta bufs */
760 struct hammer_io_list lose_list; /* loose buffers */
761 int locked_dirty_space; /* meta/volu count */
762 int io_running_space;
763 int objid_cache_count;
764 int error; /* critical I/O error */
765 struct krate krate; /* rate limited kprintf */
766 hammer_tid_t asof; /* snapshot mount */
767 hammer_tid_t next_tid;
768 hammer_tid_t flush_tid1; /* flusher tid sequencing */
769 hammer_tid_t flush_tid2; /* flusher tid sequencing */
770 int64_t copy_stat_freebigblocks; /* number of free bigblocks */
771
772 struct netexport export;
773 struct hammer_lock sync_lock;
774 struct hammer_lock free_lock;
775 struct hammer_lock undo_lock;
776 struct hammer_lock blkmap_lock;
777 struct hammer_blockmap blockmap[HAMMER_MAX_ZONES];
778 struct hammer_undo undos[HAMMER_MAX_UNDOS];
779 int undo_alloc;
780 TAILQ_HEAD(, hammer_undo) undo_lru_list;
781 TAILQ_HEAD(, hammer_reserve) delay_list;
782 struct hammer_flush_group_list flush_group_list;
783 hammer_flush_group_t next_flush_group;
784 TAILQ_HEAD(, hammer_objid_cache) objid_cache_list;
785 TAILQ_HEAD(, hammer_reclaim) reclaim_list;
786 };
787
788 typedef struct hammer_mount *hammer_mount_t;
789
790 #define HAMMER_MOUNT_CRITICAL_ERROR 0x0001
791 #define HAMMER_MOUNT_FLUSH_RECOVERY 0x0002
792
793 struct hammer_sync_info {
794 int error;
795 int waitfor;
796 };
797
798 #endif
799
800 /*
801 * checkspace slop (8MB chunks), higher numbers are more conservative.
802 */
803 #define HAMMER_CHKSPC_REBLOCK 25
804 #define HAMMER_CHKSPC_MIRROR 20
805 #define HAMMER_CHKSPC_WRITE 20
806 #define HAMMER_CHKSPC_CREATE 20
807 #define HAMMER_CHKSPC_REMOVE 10
808 #define HAMMER_CHKSPC_EMERGENCY 0
809
810 #if defined(_KERNEL)
811
812 extern struct vop_ops hammer_vnode_vops;
813 extern struct vop_ops hammer_spec_vops;
814 extern struct vop_ops hammer_fifo_vops;
815 extern struct bio_ops hammer_bioops;
816
817 extern int hammer_debug_io;
818 extern int hammer_debug_general;
819 extern int hammer_debug_debug;
820 extern int hammer_debug_inode;
821 extern int hammer_debug_locks;
822 extern int hammer_debug_btree;
823 extern int hammer_debug_tid;
824 extern int hammer_debug_recover;
825 extern int hammer_debug_recover_faults;
826 extern int hammer_cluster_enable;
827 extern int hammer_count_fsyncs;
828 extern int hammer_count_inodes;
829 extern int hammer_count_iqueued;
830 extern int hammer_count_reclaiming;
831 extern int hammer_count_records;
832 extern int hammer_count_record_datas;
833 extern int hammer_count_volumes;
834 extern int hammer_count_buffers;
835 extern int hammer_count_nodes;
836 extern int64_t hammer_count_extra_space_used;
837 extern int64_t hammer_stats_btree_lookups;
838 extern int64_t hammer_stats_btree_searches;
839 extern int64_t hammer_stats_btree_inserts;
840 extern int64_t hammer_stats_btree_deletes;
841 extern int64_t hammer_stats_btree_elements;
842 extern int64_t hammer_stats_btree_splits;
843 extern int64_t hammer_stats_btree_iterations;
844 extern int64_t hammer_stats_btree_root_iterations;
845 extern int64_t hammer_stats_record_iterations;
846 extern int64_t hammer_stats_file_read;
847 extern int64_t hammer_stats_file_write;
848 extern int64_t hammer_stats_file_iopsr;
849 extern int64_t hammer_stats_file_iopsw;
850 extern int64_t hammer_stats_disk_read;
851 extern int64_t hammer_stats_disk_write;
852 extern int64_t hammer_stats_inode_flushes;
853 extern int64_t hammer_stats_commits;
854 extern int hammer_count_dirtybufspace;
855 extern int hammer_count_refedbufs;
856 extern int hammer_count_reservations;
857 extern int hammer_count_io_running_read;
858 extern int hammer_count_io_running_write;
859 extern int hammer_count_io_locked;
860 extern int hammer_limit_dirtybufspace;
861 extern int hammer_limit_recs;
862 extern int hammer_limit_inode_recs;
863 extern int hammer_bio_count;
864 extern int hammer_verify_zone;
865 extern int hammer_verify_data;
866 extern int hammer_write_mode;
867 extern int hammer_yield_check;
868 extern int hammer_autoflush;
869 extern int64_t hammer_contention_count;
870
871 void hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
872 int error, const char *msg);
873 int hammer_vop_inactive(struct vop_inactive_args *);
874 int hammer_vop_reclaim(struct vop_reclaim_args *);
875 int hammer_get_vnode(struct hammer_inode *ip, struct vnode **vpp);
876 struct hammer_inode *hammer_get_inode(hammer_transaction_t trans,
877 hammer_inode_t dip, int64_t obj_id,
878 hammer_tid_t asof, u_int32_t localization,
879 int flags, int *errorp);
880 struct hammer_inode *hammer_get_dummy_inode(hammer_transaction_t trans,
881 hammer_inode_t dip, int64_t obj_id,
882 hammer_tid_t asof, u_int32_t localization,
883 int flags, int *errorp);
884 struct hammer_inode *hammer_find_inode(hammer_transaction_t trans,
885 int64_t obj_id, hammer_tid_t asof,
886 u_int32_t localization);
887 void hammer_scan_inode_snapshots(hammer_mount_t hmp,
888 hammer_inode_info_t iinfo,
889 int (*callback)(hammer_inode_t ip, void *data),
890 void *data);
891 void hammer_put_inode(struct hammer_inode *ip);
892 void hammer_put_inode_ref(struct hammer_inode *ip);
893 void hammer_inode_waitreclaims(hammer_mount_t hmp);
894 void hammer_inode_waithard(hammer_mount_t hmp);
895
896 int hammer_unload_volume(hammer_volume_t volume, void *data __unused);
897 int hammer_adjust_volume_mode(hammer_volume_t volume, void *data __unused);
898
899 int hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused);
900 int hammer_install_volume(hammer_mount_t hmp, const char *volname,
901 struct vnode *devvp);
902 int hammer_mountcheck_volumes(hammer_mount_t hmp);
903
904 int hammer_mem_add(hammer_record_t record);
905 int hammer_ip_lookup(hammer_cursor_t cursor);
906 int hammer_ip_first(hammer_cursor_t cursor);
907 int hammer_ip_next(hammer_cursor_t cursor);
908 int hammer_ip_resolve_data(hammer_cursor_t cursor);
909 int hammer_ip_delete_record(hammer_cursor_t cursor, hammer_inode_t ip,
910 hammer_tid_t tid);
911 int hammer_delete_at_cursor(hammer_cursor_t cursor, int delete_flags,
912 hammer_tid_t delete_tid, u_int32_t delete_ts,
913 int track, int64_t *stat_bytes);
914 int hammer_ip_check_directory_empty(hammer_transaction_t trans,
915 hammer_inode_t ip);
916 int hammer_sync_hmp(hammer_mount_t hmp, int waitfor);
917 int hammer_queue_inodes_flusher(hammer_mount_t hmp, int waitfor);
918
919 hammer_record_t
920 hammer_alloc_mem_record(hammer_inode_t ip, int data_len);
921 void hammer_flush_record_done(hammer_record_t record, int error);
922 void hammer_wait_mem_record_ident(hammer_record_t record, const char *ident);
923 void hammer_rel_mem_record(hammer_record_t record);
924
925 int hammer_cursor_up(hammer_cursor_t cursor);
926 int hammer_cursor_up_locked(hammer_cursor_t cursor);
927 int hammer_cursor_down(hammer_cursor_t cursor);
928 int hammer_cursor_upgrade(hammer_cursor_t cursor);
929 int hammer_cursor_upgrade_node(hammer_cursor_t cursor);
930 void hammer_cursor_downgrade(hammer_cursor_t cursor);
931 int hammer_cursor_seek(hammer_cursor_t cursor, hammer_node_t node,
932 int index);
933 void hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident);
934 int hammer_lock_ex_try(struct hammer_lock *lock);
935 void hammer_lock_sh(struct hammer_lock *lock);
936 int hammer_lock_sh_try(struct hammer_lock *lock);
937 int hammer_lock_upgrade(struct hammer_lock *lock);
938 void hammer_lock_downgrade(struct hammer_lock *lock);
939 int hammer_lock_status(struct hammer_lock *lock);
940 void hammer_unlock(struct hammer_lock *lock);
941 void hammer_ref(struct hammer_lock *lock);
942 void hammer_unref(struct hammer_lock *lock);
943
944 void hammer_sync_lock_ex(hammer_transaction_t trans);
945 void hammer_sync_lock_sh(hammer_transaction_t trans);
946 int hammer_sync_lock_sh_try(hammer_transaction_t trans);
947 void hammer_sync_unlock(hammer_transaction_t trans);
948
949 u_int32_t hammer_to_unix_xid(uuid_t *uuid);
950 void hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid);
951 void hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts);
952 u_int64_t hammer_timespec_to_time(struct timespec *ts);
953 int hammer_str_to_tid(const char *str, int *ispfsp,
954 hammer_tid_t *tidp, u_int32_t *localizationp);
955 int hammer_is_atatext(const char *name, int len);
956 hammer_tid_t hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip,
957 int64_t namekey);
958 void hammer_clear_objid(hammer_inode_t dip);
959 void hammer_destroy_objid_cache(hammer_mount_t hmp);
960
961 int hammer_enter_undo_history(hammer_mount_t hmp, hammer_off_t offset,
962 int bytes);
963 void hammer_clear_undo_history(hammer_mount_t hmp);
964 enum vtype hammer_get_vnode_type(u_int8_t obj_type);
965 int hammer_get_dtype(u_int8_t obj_type);
966 u_int8_t hammer_get_obj_type(enum vtype vtype);
967 int64_t hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
968 u_int32_t *max_iterationsp);
969 int hammer_nohistory(hammer_inode_t ip);
970
971 int hammer_init_cursor(hammer_transaction_t trans, hammer_cursor_t cursor,
972 hammer_node_cache_t cache, hammer_inode_t ip);
973 void hammer_normalize_cursor(hammer_cursor_t cursor);
974 void hammer_done_cursor(hammer_cursor_t cursor);
975 int hammer_recover_cursor(hammer_cursor_t cursor);
976 void hammer_unlock_cursor(hammer_cursor_t cursor);
977 int hammer_lock_cursor(hammer_cursor_t cursor);
978 hammer_cursor_t hammer_push_cursor(hammer_cursor_t ocursor);
979 void hammer_pop_cursor(hammer_cursor_t ocursor, hammer_cursor_t ncursor);
980
981 void hammer_cursor_replaced_node(hammer_node_t onode, hammer_node_t nnode);
982 void hammer_cursor_removed_node(hammer_node_t onode, hammer_node_t parent,
983 int index);
984 void hammer_cursor_split_node(hammer_node_t onode, hammer_node_t nnode,
985 int index);
986 void hammer_cursor_moved_element(hammer_node_t onode, hammer_node_t nnode,
987 int oindex, int nindex);
988 void hammer_cursor_parent_changed(hammer_node_t node, hammer_node_t oparent,
989 hammer_node_t nparent, int nindex);
990 void hammer_cursor_inserted_element(hammer_node_t node, int index);
991 void hammer_cursor_deleted_element(hammer_node_t node, int index);
992
993 int hammer_btree_lookup(hammer_cursor_t cursor);
994 int hammer_btree_first(hammer_cursor_t cursor);
995 int hammer_btree_last(hammer_cursor_t cursor);
996 int hammer_btree_extract(hammer_cursor_t cursor, int flags);
997 int hammer_btree_iterate(hammer_cursor_t cursor);
998 int hammer_btree_iterate_reverse(hammer_cursor_t cursor);
999 int hammer_btree_insert(hammer_cursor_t cursor,
1000 hammer_btree_leaf_elm_t elm, int *doprop);
1001 int hammer_btree_delete(hammer_cursor_t cursor);
1002 void hammer_btree_do_propagation(hammer_cursor_t cursor,
1003 hammer_pseudofs_inmem_t pfsm,
1004 hammer_btree_leaf_elm_t leaf);
1005 int hammer_btree_cmp(hammer_base_elm_t key1, hammer_base_elm_t key2);
1006 int hammer_btree_chkts(hammer_tid_t ts, hammer_base_elm_t key);
1007 int hammer_btree_correct_rhb(hammer_cursor_t cursor, hammer_tid_t tid);
1008 int hammer_btree_correct_lhb(hammer_cursor_t cursor, hammer_tid_t tid);
1009
1010 int btree_set_parent(hammer_transaction_t trans, hammer_node_t node,
1011 hammer_btree_elm_t elm);
1012 void hammer_node_lock_init(hammer_node_lock_t parent, hammer_node_t node);
1013 int hammer_btree_lock_children(hammer_cursor_t cursor, int depth,
1014 hammer_node_lock_t parent);
1015 void hammer_btree_lock_copy(hammer_cursor_t cursor,
1016 hammer_node_lock_t parent);
1017 int hammer_btree_sync_copy(hammer_cursor_t cursor,
1018 hammer_node_lock_t parent);
1019 void hammer_btree_unlock_children(hammer_cursor_t cursor,
1020 hammer_node_lock_t parent);
1021 int hammer_btree_search_node(hammer_base_elm_t elm, hammer_node_ondisk_t node);
1022 hammer_node_t hammer_btree_get_parent(hammer_transaction_t trans,
1023 hammer_node_t node, int *parent_indexp,
1024 int *errorp, int try_exclusive);
1025
1026 void hammer_print_btree_node(hammer_node_ondisk_t ondisk);
1027 void hammer_print_btree_elm(hammer_btree_elm_t elm, u_int8_t type, int i);
1028
1029 void *hammer_bread(struct hammer_mount *hmp, hammer_off_t off,
1030 int *errorp, struct hammer_buffer **bufferp);
1031 void *hammer_bnew(struct hammer_mount *hmp, hammer_off_t off,
1032 int *errorp, struct hammer_buffer **bufferp);
1033 void *hammer_bread_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1034 int *errorp, struct hammer_buffer **bufferp);
1035 void *hammer_bnew_ext(struct hammer_mount *hmp, hammer_off_t off, int bytes,
1036 int *errorp, struct hammer_buffer **bufferp);
1037
1038 hammer_volume_t hammer_get_root_volume(hammer_mount_t hmp, int *errorp);
1039
1040 hammer_volume_t hammer_get_volume(hammer_mount_t hmp,
1041 int32_t vol_no, int *errorp);
1042 hammer_buffer_t hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
1043 int bytes, int isnew, int *errorp);
1044 void hammer_sync_buffers(hammer_mount_t hmp,
1045 hammer_off_t base_offset, int bytes);
1046 int hammer_del_buffers(hammer_mount_t hmp,
1047 hammer_off_t base_offset,
1048 hammer_off_t zone2_offset, int bytes,
1049 int report_conflicts);
1050
1051 int hammer_ref_volume(hammer_volume_t volume);
1052 int hammer_ref_buffer(hammer_buffer_t buffer);
1053 void hammer_flush_buffer_nodes(hammer_buffer_t buffer);
1054
1055 void hammer_rel_volume(hammer_volume_t volume, int flush);
1056 void hammer_rel_buffer(hammer_buffer_t buffer, int flush);
1057
1058 int hammer_vfs_export(struct mount *mp, int op,
1059 const struct export_args *export);
1060 hammer_node_t hammer_get_node(hammer_transaction_t trans,
1061 hammer_off_t node_offset, int isnew, int *errorp);
1062 void hammer_ref_node(hammer_node_t node);
1063 hammer_node_t hammer_ref_node_safe(hammer_transaction_t trans,
1064 hammer_node_cache_t cache, int *errorp);
1065 void hammer_rel_node(hammer_node_t node);
1066 void hammer_delete_node(hammer_transaction_t trans,
1067 hammer_node_t node);
1068 void hammer_cache_node(hammer_node_cache_t cache,
1069 hammer_node_t node);
1070 void hammer_uncache_node(hammer_node_cache_t cache);
1071 void hammer_flush_node(hammer_node_t node);
1072
1073 void hammer_dup_buffer(struct hammer_buffer **bufferp,
1074 struct hammer_buffer *buffer);
1075 hammer_node_t hammer_alloc_btree(hammer_transaction_t trans,
1076 hammer_off_t hint, int *errorp);
1077 void *hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1078 u_int16_t rec_type, hammer_off_t *data_offsetp,
1079 struct hammer_buffer **data_bufferp,
1080 hammer_off_t hint, int *errorp);
1081
1082 int hammer_generate_undo(hammer_transaction_t trans, hammer_io_t io,
1083 hammer_off_t zone1_offset, void *base, int len);
1084
1085 void hammer_put_volume(struct hammer_volume *volume, int flush);
1086 void hammer_put_buffer(struct hammer_buffer *buffer, int flush);
1087
1088 hammer_off_t hammer_freemap_alloc(hammer_transaction_t trans,
1089 hammer_off_t owner, int *errorp);
1090 void hammer_freemap_free(hammer_transaction_t trans, hammer_off_t phys_offset,
1091 hammer_off_t owner, int *errorp);
1092 int _hammer_checkspace(hammer_mount_t hmp, int slop, int64_t *resp);
1093 hammer_off_t hammer_blockmap_alloc(hammer_transaction_t trans, int zone,
1094 int bytes, hammer_off_t hint, int *errorp);
1095 hammer_reserve_t hammer_blockmap_reserve(hammer_mount_t hmp, int zone,
1096 int bytes, hammer_off_t *zone_offp, int *errorp);
1097 void hammer_blockmap_reserve_complete(hammer_mount_t hmp,
1098 hammer_reserve_t resv);
1099 void hammer_reserve_clrdelay(hammer_mount_t hmp, hammer_reserve_t resv);
1100 void hammer_blockmap_free(hammer_transaction_t trans,
1101 hammer_off_t bmap_off, int bytes);
1102 int hammer_blockmap_finalize(hammer_transaction_t trans,
1103 hammer_reserve_t resv,
1104 hammer_off_t bmap_off, int bytes);
1105 int hammer_blockmap_getfree(hammer_mount_t hmp, hammer_off_t bmap_off,
1106 int *curp, int *errorp);
1107 hammer_off_t hammer_blockmap_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1108 int *errorp);
1109 hammer_off_t hammer_undo_lookup(hammer_mount_t hmp, hammer_off_t bmap_off,
1110 int *errorp);
1111 int64_t hammer_undo_used(hammer_transaction_t trans);
1112 int64_t hammer_undo_space(hammer_transaction_t trans);
1113 int64_t hammer_undo_max(hammer_mount_t hmp);
1114
1115 void hammer_start_transaction(struct hammer_transaction *trans,
1116 struct hammer_mount *hmp);
1117 void hammer_simple_transaction(struct hammer_transaction *trans,
1118 struct hammer_mount *hmp);
1119 void hammer_start_transaction_fls(struct hammer_transaction *trans,
1120 struct hammer_mount *hmp);
1121 void hammer_done_transaction(struct hammer_transaction *trans);
1122
1123 void hammer_modify_inode(hammer_inode_t ip, int flags);
1124 void hammer_flush_inode(hammer_inode_t ip, int flags);
1125 void hammer_flush_inode_done(hammer_inode_t ip, int error);
1126 void hammer_wait_inode(hammer_inode_t ip);
1127
1128 int hammer_create_inode(struct hammer_transaction *trans, struct vattr *vap,
1129 struct ucred *cred, struct hammer_inode *dip,
1130 const char *name, int namelen,
1131 hammer_pseudofs_inmem_t pfsm,
1132 struct hammer_inode **ipp);
1133 void hammer_rel_inode(hammer_inode_t ip, int flush);
1134 int hammer_reload_inode(hammer_inode_t ip, void *arg __unused);
1135 int hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2);
1136 int hammer_destroy_inode_callback(hammer_inode_t ip, void *data __unused);
1137
1138 int hammer_sync_inode(hammer_transaction_t trans, hammer_inode_t ip);
1139 void hammer_test_inode(hammer_inode_t dip);
1140 void hammer_inode_unloadable_check(hammer_inode_t ip, int getvp);
1141
1142 int hammer_ip_add_directory(struct hammer_transaction *trans,
1143 hammer_inode_t dip, const char *name, int bytes,
1144 hammer_inode_t nip);
1145 int hammer_ip_del_directory(struct hammer_transaction *trans,
1146 hammer_cursor_t cursor, hammer_inode_t dip,
1147 hammer_inode_t ip);
1148 hammer_record_t hammer_ip_add_bulk(hammer_inode_t ip, off_t file_offset,
1149 void *data, int bytes, int *errorp);
1150 int hammer_ip_frontend_trunc(struct hammer_inode *ip, off_t file_size);
1151 int hammer_ip_add_record(struct hammer_transaction *trans,
1152 hammer_record_t record);
1153 int hammer_ip_delete_range(hammer_cursor_t cursor, hammer_inode_t ip,
1154 int64_t ran_beg, int64_t ran_end, int truncating);
1155 int hammer_ip_delete_clean(hammer_cursor_t cursor, hammer_inode_t ip,
1156 int *countp);
1157 int hammer_ip_sync_data(hammer_cursor_t cursor, hammer_inode_t ip,
1158 int64_t offset, void *data, int bytes);
1159 int hammer_ip_sync_record(hammer_transaction_t trans, hammer_record_t rec);
1160 int hammer_ip_sync_record_cursor(hammer_cursor_t cursor, hammer_record_t rec);
1161 hammer_pseudofs_inmem_t hammer_load_pseudofs(hammer_transaction_t trans,
1162 u_int32_t localization, int *errorp);
1163 int hammer_mkroot_pseudofs(hammer_transaction_t trans, struct ucred *cred,
1164 hammer_pseudofs_inmem_t pfsm);
1165 int hammer_save_pseudofs(hammer_transaction_t trans,
1166 hammer_pseudofs_inmem_t pfsm);
1167 int hammer_unload_pseudofs(hammer_transaction_t trans, u_int32_t localization);
1168 void hammer_rel_pseudofs(hammer_mount_t hmp, hammer_pseudofs_inmem_t pfsm);
1169 int hammer_ioctl(hammer_inode_t ip, u_long com, caddr_t data, int fflag,
1170 struct ucred *cred);
1171
1172 void hammer_io_init(hammer_io_t io, hammer_volume_t volume,
1173 enum hammer_io_type type);
1174 int hammer_io_read(struct vnode *devvp, struct hammer_io *io,
1175 hammer_off_t limit);
1176 int hammer_io_new(struct vnode *devvp, struct hammer_io *io);
1177 int hammer_io_inval(hammer_volume_t volume, hammer_off_t zone2_offset);
1178 struct buf *hammer_io_release(struct hammer_io *io, int flush);
1179 void hammer_io_flush(struct hammer_io *io);
1180 void hammer_io_wait(struct hammer_io *io);
1181 void hammer_io_waitdep(struct hammer_io *io);
1182 void hammer_io_wait_all(hammer_mount_t hmp, const char *ident);
1183 int hammer_io_direct_read(hammer_mount_t hmp, struct bio *bio,
1184 hammer_btree_leaf_elm_t leaf);
1185 int hammer_io_direct_write(hammer_mount_t hmp, hammer_record_t record,
1186 struct bio *bio);
1187 void hammer_io_direct_wait(hammer_record_t record);
1188 void hammer_io_direct_uncache(hammer_mount_t hmp, hammer_btree_leaf_elm_t leaf);
1189 void hammer_io_write_interlock(hammer_io_t io);
1190 void hammer_io_done_interlock(hammer_io_t io);
1191 void hammer_io_clear_modify(struct hammer_io *io, int inval);
1192 void hammer_io_clear_modlist(struct hammer_io *io);
1193 void hammer_io_flush_sync(hammer_mount_t hmp);
1194
1195 void hammer_modify_volume(hammer_transaction_t trans, hammer_volume_t volume,
1196 void *base, int len);
1197 void hammer_modify_buffer(hammer_transaction_t trans, hammer_buffer_t buffer,
1198 void *base, int len);
1199 void hammer_modify_volume_done(hammer_volume_t volume);
1200 void hammer_modify_buffer_done(hammer_buffer_t buffer);
1201
1202 int hammer_ioc_reblock(hammer_transaction_t trans, hammer_inode_t ip,
1203 struct hammer_ioc_reblock *reblock);
1204 int hammer_ioc_rebalance(hammer_transaction_t trans, hammer_inode_t ip,
1205 struct hammer_ioc_rebalance *rebal);
1206 int hammer_ioc_prune(hammer_transaction_t trans, hammer_inode_t ip,
1207 struct hammer_ioc_prune *prune);
1208 int hammer_ioc_mirror_read(hammer_transaction_t trans, hammer_inode_t ip,
1209 struct hammer_ioc_mirror_rw *mirror);
1210 int hammer_ioc_mirror_write(hammer_transaction_t trans, hammer_inode_t ip,
1211 struct hammer_ioc_mirror_rw *mirror);
1212 int hammer_ioc_set_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1213 struct ucred *cred, struct hammer_ioc_pseudofs_rw *pfs);
1214 int hammer_ioc_get_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1215 struct hammer_ioc_pseudofs_rw *pfs);
1216 int hammer_ioc_destroy_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1217 struct hammer_ioc_pseudofs_rw *pfs);
1218 int hammer_ioc_downgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1219 struct hammer_ioc_pseudofs_rw *pfs);
1220 int hammer_ioc_upgrade_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1221 struct hammer_ioc_pseudofs_rw *pfs);
1222 int hammer_ioc_wait_pseudofs(hammer_transaction_t trans, hammer_inode_t ip,
1223 struct hammer_ioc_pseudofs_rw *pfs);
1224 int hammer_ioc_expand(hammer_transaction_t trans, hammer_inode_t ip,
1225 struct hammer_ioc_expand *expand);
1226
1227 int hammer_signal_check(hammer_mount_t hmp);
1228
1229 void hammer_flusher_create(hammer_mount_t hmp);
1230 void hammer_flusher_destroy(hammer_mount_t hmp);
1231 void hammer_flusher_sync(hammer_mount_t hmp);
1232 int hammer_flusher_async(hammer_mount_t hmp, hammer_flush_group_t flg);
1233 int hammer_flusher_async_one(hammer_mount_t hmp);
1234 void hammer_flusher_wait(hammer_mount_t hmp, int seq);
1235 void hammer_flusher_wait_next(hammer_mount_t hmp);
1236 int hammer_flusher_meta_limit(hammer_mount_t hmp);
1237 int hammer_flusher_meta_halflimit(hammer_mount_t hmp);
1238 int hammer_flusher_undo_exhausted(hammer_transaction_t trans, int quarter);
1239 void hammer_flusher_clean_loose_ios(hammer_mount_t hmp);
1240 void hammer_flusher_finalize(hammer_transaction_t trans, int final);
1241 int hammer_flusher_haswork(hammer_mount_t hmp);
1242
1243
1244 int hammer_recover(hammer_mount_t hmp, hammer_volume_t rootvol);
1245 void hammer_recover_flush_buffers(hammer_mount_t hmp,
1246 hammer_volume_t root_volume, int final);
1247
1248 void hammer_crc_set_blockmap(hammer_blockmap_t blockmap);
1249 void hammer_crc_set_volume(hammer_volume_ondisk_t ondisk);
1250 void hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1251
1252 int hammer_crc_test_blockmap(hammer_blockmap_t blockmap);
1253 int hammer_crc_test_volume(hammer_volume_ondisk_t ondisk);
1254 int hammer_crc_test_btree(hammer_node_ondisk_t ondisk);
1255 int hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf);
1256 void hkprintf(const char *ctl, ...);
1257 udev_t hammer_fsid_to_udev(uuid_t *uuid);
1258
1259
1260 int hammer_blocksize(int64_t file_offset);
1261 int64_t hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2);
1262
1263 /*
1264 * Shortcut for _hammer_checkspace(), used all over the code.
1265 */
1266 static __inline int
1267 hammer_checkspace(hammer_mount_t hmp, int slop)
1268 {
1269 return(_hammer_checkspace(hmp, slop, NULL));
1270 }
1271
1272 #endif
1273
1274 static __inline void
1275 hammer_wait_mem_record(hammer_record_t record)
1276 {
1277 hammer_wait_mem_record_ident(record, "hmmwai");
1278 }
1279
1280 static __inline void
1281 hammer_lock_ex(struct hammer_lock *lock)
1282 {
1283 hammer_lock_ex_ident(lock, "hmrlck");
1284 }
1285
1286 /*
1287 * Indicate that a B-Tree node is being modified.
1288 */
1289 static __inline void
1290 hammer_modify_node_noundo(hammer_transaction_t trans, hammer_node_t node)
1291 {
1292 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1293 hammer_modify_buffer(trans, node->buffer, NULL, 0);
1294 }
1295
1296 static __inline void
1297 hammer_modify_node_all(hammer_transaction_t trans, struct hammer_node *node)
1298 {
1299 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1300 hammer_modify_buffer(trans, node->buffer,
1301 node->ondisk, sizeof(*node->ondisk));
1302 }
1303
1304 static __inline void
1305 hammer_modify_node(hammer_transaction_t trans, hammer_node_t node,
1306 void *base, int len)
1307 {
1308 hammer_crc_t *crcptr;
1309
1310 KKASSERT((char *)base >= (char *)node->ondisk &&
1311 (char *)base + len <=
1312 (char *)node->ondisk + sizeof(*node->ondisk));
1313 KKASSERT((node->flags & HAMMER_NODE_CRCBAD) == 0);
1314 hammer_modify_buffer(trans, node->buffer, base, len);
1315 crcptr = &node->ondisk->crc;
1316 hammer_modify_buffer(trans, node->buffer, crcptr, sizeof(hammer_crc_t));
1317 --node->buffer->io.modify_refs; /* only want one ref */
1318 }
1319
1320 /*
1321 * Indicate that the specified modifications have been completed.
1322 *
1323 * Do not try to generate the crc here, it's very expensive to do and a
1324 * sequence of insertions or deletions can result in many calls to this
1325 * function on the same node.
1326 */
1327 static __inline void
1328 hammer_modify_node_done(hammer_node_t node)
1329 {
1330 node->flags |= HAMMER_NODE_CRCGOOD;
1331 if ((node->flags & HAMMER_NODE_NEEDSCRC) == 0) {
1332 node->flags |= HAMMER_NODE_NEEDSCRC;
1333 node->buffer->io.gencrc = 1;
1334 hammer_ref_node(node);
1335 }
1336 hammer_modify_buffer_done(node->buffer);
1337 }
1338
1339 #define hammer_modify_volume_field(trans, vol, field) \
1340 hammer_modify_volume(trans, vol, &(vol)->ondisk->field, \
1341 sizeof((vol)->ondisk->field))
1342
1343 #define hammer_modify_node_field(trans, node, field) \
1344 hammer_modify_node(trans, node, &(node)->ondisk->field, \
1345 sizeof((node)->ondisk->field))
1346
1347 /*
1348 * The HAMMER_INODE_CAP_DIR_LOCAL_INO capability is set on newly
1349 * created directories for HAMMER version 2 or greater and causes
1350 * directory entries to be placed the inode localization zone in
1351 * the B-Tree instead of the misc zone.
1352 *
1353 * This greatly improves localization between directory entries and
1354 * inodes
1355 */
1356 static __inline u_int32_t
1357 hammer_dir_localization(hammer_inode_t dip)
1358 {
1359 if (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO)
1360 return(HAMMER_LOCALIZE_INODE);
1361 else
1362 return(HAMMER_LOCALIZE_MISC);
1363 }
DragonFly BSD