2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
34 * $DragonFly: src/sys/vfs/hammer/hammer_subs.c,v 1.35 2008/10/15 22:38:37 dillon Exp $
37 * HAMMER structural locking
41 #include <sys/dirent.h>
44 hammer_lock_ex_ident(struct hammer_lock *lock, const char *ident)
46 thread_t td = curthread;
48 KKASSERT(lock->refs > 0);
50 if (lock->locktd != td) {
51 while (lock->locktd != NULL || lock->lockcount) {
54 if (hammer_debug_locks) {
55 kprintf("hammer_lock_ex: held by %p\n",
58 ++hammer_contention_count;
59 tsleep(lock, 0, ident, 0);
60 if (hammer_debug_locks)
61 kprintf("hammer_lock_ex: try again\n");
66 KKASSERT(lock->lockcount >= 0);
72 * Try to obtain an exclusive lock
75 hammer_lock_ex_try(struct hammer_lock *lock)
77 thread_t td = curthread;
79 KKASSERT(lock->refs > 0);
81 if (lock->locktd != td) {
82 if (lock->locktd != NULL || lock->lockcount) {
88 KKASSERT(lock->lockcount >= 0);
95 * Obtain a shared lock
97 * We do not give pending exclusive locks priority over shared locks as
98 * doing so could lead to a deadlock.
101 hammer_lock_sh(struct hammer_lock *lock)
103 KKASSERT(lock->refs > 0);
105 while (lock->locktd != NULL) {
106 if (lock->locktd == curthread) {
107 Debugger("hammer_lock_sh: lock_sh on exclusive");
113 tsleep(lock, 0, "hmrlck", 0);
115 KKASSERT(lock->lockcount <= 0);
121 hammer_lock_sh_try(struct hammer_lock *lock)
123 KKASSERT(lock->refs > 0);
129 KKASSERT(lock->lockcount <= 0);
136 * Upgrade a shared lock to an exclusively held lock. This function will
137 * return EDEADLK If there is more then one shared holder.
139 * No error occurs and no action is taken if the lock is already exclusively
140 * held by the caller. If the lock is not held at all or held exclusively
141 * by someone else, this function will panic.
144 hammer_lock_upgrade(struct hammer_lock *lock)
149 if (lock->lockcount > 0) {
150 if (lock->locktd != curthread)
151 panic("hammer_lock_upgrade: illegal lock state");
153 } else if (lock->lockcount == -1) {
155 lock->locktd = curthread;
157 } else if (lock->lockcount != 0) {
160 panic("hammer_lock_upgrade: lock is not held");
169 * Downgrade an exclusively held lock to a shared lock.
172 hammer_lock_downgrade(struct hammer_lock *lock)
174 KKASSERT(lock->lockcount == 1 && lock->locktd == curthread);
176 lock->lockcount = -1;
183 /* XXX memory barrier */
187 hammer_unlock(struct hammer_lock *lock)
190 KKASSERT(lock->lockcount != 0);
191 if (lock->lockcount < 0) {
192 if (++lock->lockcount == 0 && lock->wanted) {
197 KKASSERT(lock->locktd == curthread);
198 if (--lock->lockcount == 0) {
211 * The calling thread must be holding a shared or exclusive lock.
212 * Returns < 0 if lock is held shared, and > 0 if held exlusively.
215 hammer_lock_status(struct hammer_lock *lock)
217 if (lock->lockcount < 0)
219 if (lock->lockcount > 0)
221 panic("hammer_lock_status: lock must be held: %p", lock);
225 hammer_ref(struct hammer_lock *lock)
227 KKASSERT(lock->refs >= 0);
234 hammer_unref(struct hammer_lock *lock)
236 KKASSERT(lock->refs > 0);
243 * The sync_lock must be held when doing any modifying operations on
244 * meta-data. It does not have to be held when modifying non-meta-data buffers
245 * (backend or frontend).
247 * The flusher holds the lock exclusively while all other consumers hold it
248 * shared. All modifying operations made while holding the lock are atomic
249 * in that they will be made part of the same flush group.
251 * Due to the atomicy requirement deadlock recovery code CANNOT release the
252 * sync lock, nor can we give pending exclusive sync locks priority over
253 * a shared sync lock as this could lead to a 3-way deadlock.
256 hammer_sync_lock_ex(hammer_transaction_t trans)
258 ++trans->sync_lock_refs;
259 hammer_lock_ex(&trans->hmp->sync_lock);
263 hammer_sync_lock_sh(hammer_transaction_t trans)
265 ++trans->sync_lock_refs;
266 hammer_lock_sh(&trans->hmp->sync_lock);
270 hammer_sync_lock_sh_try(hammer_transaction_t trans)
274 ++trans->sync_lock_refs;
275 if ((error = hammer_lock_sh_try(&trans->hmp->sync_lock)) != 0)
276 --trans->sync_lock_refs;
281 hammer_sync_unlock(hammer_transaction_t trans)
283 --trans->sync_lock_refs;
284 hammer_unlock(&trans->hmp->sync_lock);
291 hammer_to_unix_xid(uuid_t *uuid)
293 return(*(u_int32_t *)&uuid->node[2]);
297 hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid)
299 bzero(uuid, sizeof(*uuid));
300 *(u_int32_t *)&uuid->node[2] = guid;
304 hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts)
306 ts->tv_sec = (unsigned long)(xtime / 1000000);
307 ts->tv_nsec = (unsigned int)(xtime % 1000000) * 1000L;
311 hammer_timespec_to_time(struct timespec *ts)
315 xtime = (unsigned)(ts->tv_nsec / 1000) +
316 (unsigned long)ts->tv_sec * 1000000ULL;
322 * Convert a HAMMER filesystem object type to a vnode type
325 hammer_get_vnode_type(u_int8_t obj_type)
328 case HAMMER_OBJTYPE_DIRECTORY:
330 case HAMMER_OBJTYPE_REGFILE:
332 case HAMMER_OBJTYPE_DBFILE:
334 case HAMMER_OBJTYPE_FIFO:
336 case HAMMER_OBJTYPE_SOCKET:
338 case HAMMER_OBJTYPE_CDEV:
340 case HAMMER_OBJTYPE_BDEV:
342 case HAMMER_OBJTYPE_SOFTLINK:
351 hammer_get_dtype(u_int8_t obj_type)
354 case HAMMER_OBJTYPE_DIRECTORY:
356 case HAMMER_OBJTYPE_REGFILE:
358 case HAMMER_OBJTYPE_DBFILE:
360 case HAMMER_OBJTYPE_FIFO:
362 case HAMMER_OBJTYPE_SOCKET:
364 case HAMMER_OBJTYPE_CDEV:
366 case HAMMER_OBJTYPE_BDEV:
368 case HAMMER_OBJTYPE_SOFTLINK:
377 hammer_get_obj_type(enum vtype vtype)
381 return(HAMMER_OBJTYPE_DIRECTORY);
383 return(HAMMER_OBJTYPE_REGFILE);
385 return(HAMMER_OBJTYPE_DBFILE);
387 return(HAMMER_OBJTYPE_FIFO);
389 return(HAMMER_OBJTYPE_SOCKET);
391 return(HAMMER_OBJTYPE_CDEV);
393 return(HAMMER_OBJTYPE_BDEV);
395 return(HAMMER_OBJTYPE_SOFTLINK);
397 return(HAMMER_OBJTYPE_UNKNOWN);
403 * Return flags for hammer_delete_at_cursor()
406 hammer_nohistory(hammer_inode_t ip)
408 if (ip->hmp->hflags & HMNT_NOHISTORY)
409 return(HAMMER_DELETE_DESTROY);
410 if (ip->ino_data.uflags & (SF_NOHISTORY|UF_NOHISTORY))
411 return(HAMMER_DELETE_DESTROY);
416 * ALGORITHM VERSION 1:
417 * Return a namekey hash. The 64 bit namekey hash consists of a 32 bit
418 * crc in the MSB and 0 in the LSB. The caller will use the low 32 bits
419 * to generate a unique key and will scan all entries with the same upper
420 * 32 bits when issuing a lookup.
422 * 0hhhhhhhhhhhhhhh hhhhhhhhhhhhhhhh 0000000000000000 0000000000000000
424 * ALGORITHM VERSION 2:
426 * The 64 bit hash key is generated from the following components. The
427 * first three characters are encoded as 5-bit quantities, the middle
428 * N characters are hashed into a 6 bit quantity, and the last two
429 * characters are encoded as 5-bit quantities. A 32 bit hash of the
430 * entire filename is encoded in the low 32 bits. Bit 0 is set to
431 * 0 to guarantee us a 2^24 bit iteration space.
433 * 0aaaaabbbbbccccc mmmmmmyyyyyzzzzz hhhhhhhhhhhhhhhh hhhhhhhhhhhhhhh0
435 * This gives us a domain sort for the first three characters, the last
436 * two characters, and breaks the middle space into 64 random domains.
437 * The domain sort folds upper case, lower case, digits, and punctuation
438 * spaces together, the idea being the filenames tend to not be a mix
441 * The 64 random domains act as a sub-sort for the middle characters
442 * but may cause a random seek. If the filesystem is being accessed
443 * in sorted order we should tend to get very good linearity for most
444 * filenames and devolve into more random seeks otherwise.
446 * We strip bit 63 in order to provide a positive key, this way a seek
447 * offset of 0 will represent the base of the directory.
449 * This function can never return 0. We use the MSB-0 space to synthesize
450 * artificial directory entries such as "." and "..".
453 hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
454 u_int32_t *max_iterationsp)
458 const char *aname = name;
460 switch (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIRHASH_MASK) {
461 case HAMMER_INODE_CAP_DIRHASH_ALG0:
462 key = (int64_t)(crc32(aname, len) & 0x7FFFFFFF) << 32;
464 key |= 0x100000000LL;
465 *max_iterationsp = 0xFFFFFFFFU;
467 case HAMMER_INODE_CAP_DIRHASH_ALG1:
468 key = (u_int32_t)crc32(aname, len) & 0xFFFFFFFEU;
472 crcx = crc32(aname + 3, len - 5);
473 crcx = crcx ^ (crcx >> 6) ^ (crcx >> 12);
474 key |= (int64_t)(crcx & 0x3F) << 42;
480 key |= ((int64_t)(aname[2] & 0x1F) << 48);
483 key |= ((int64_t)(aname[1] & 0x1F) << 53) |
484 ((int64_t)(aname[len-2] & 0x1F) << 37);
487 key |= ((int64_t)(aname[0] & 0x1F) << 58) |
488 ((int64_t)(aname[len-1] & 0x1F) << 32);
493 if ((key & 0xFFFFFFFF00000000LL) == 0)
494 key |= 0x100000000LL;
495 if (hammer_debug_general & 0x0400) {
496 kprintf("namekey2: 0x%016llx %*.*s\n",
497 (long long)key, len, len, aname);
499 *max_iterationsp = 0x00FFFFFF;
501 case HAMMER_INODE_CAP_DIRHASH_ALG2:
502 case HAMMER_INODE_CAP_DIRHASH_ALG3:
504 key = 0; /* compiler warning */
505 *max_iterationsp = 1; /* sanity */
506 panic("hammer_directory_namekey: bad algorithm %p\n", dip);
513 * Convert string after @@ (@@ not included) to TID. Returns 0 on success,
516 * If this function fails *ispfs, *tidp, and *localizationp will not
520 hammer_str_to_tid(const char *str, int *ispfsp,
521 hammer_tid_t *tidp, u_int32_t *localizationp)
524 u_int32_t localization;
530 * Forms allowed for TID: "0x%016llx"
533 tid = strtouq(str, &ptr, 0);
535 if (n == 2 && str[0] == '-' && str[1] == '1') {
537 } else if (n == 18 && str[0] == '0' && (str[1] | 0x20) == 'x') {
544 * Forms allowed for PFS: ":%05d" (i.e. "...:0" would be illegal).
548 localization = strtoul(str + 1, &ptr, 10) << 16;
554 localization = *localizationp;
559 * Any trailing junk invalidates special extension handling.
564 *localizationp = localization;
570 hammer_crc_set_blockmap(hammer_blockmap_t blockmap)
572 blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
576 hammer_crc_set_volume(hammer_volume_ondisk_t ondisk)
578 ondisk->vol_crc = crc32(ondisk, HAMMER_VOL_CRCSIZE1) ^
579 crc32(&ondisk->vol_crc + 1, HAMMER_VOL_CRCSIZE2);
583 hammer_crc_test_blockmap(hammer_blockmap_t blockmap)
587 crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
588 return (blockmap->entry_crc == crc);
592 hammer_crc_test_volume(hammer_volume_ondisk_t ondisk)
596 crc = crc32(ondisk, HAMMER_VOL_CRCSIZE1) ^
597 crc32(&ondisk->vol_crc + 1, HAMMER_VOL_CRCSIZE2);
598 return (ondisk->vol_crc == crc);
602 hammer_crc_test_btree(hammer_node_ondisk_t ondisk)
606 crc = crc32(&ondisk->crc + 1, HAMMER_BTREE_CRCSIZE);
607 return (ondisk->crc == crc);
611 * Test or set the leaf->data_crc field. Deal with any special cases given
612 * a generic B-Tree leaf element and its data.
614 * NOTE: Inode-data: the atime and mtime fields are not CRCd, allowing them
615 * to be updated in-place.
618 hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf)
622 if (leaf->data_len == 0) {
625 switch(leaf->base.rec_type) {
626 case HAMMER_RECTYPE_INODE:
627 if (leaf->data_len != sizeof(struct hammer_inode_data))
629 crc = crc32(data, HAMMER_INODE_CRCSIZE);
632 crc = crc32(data, leaf->data_len);
636 return (leaf->data_crc == crc);
640 hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf)
642 if (leaf->data_len == 0) {
645 switch(leaf->base.rec_type) {
646 case HAMMER_RECTYPE_INODE:
647 KKASSERT(leaf->data_len ==
648 sizeof(struct hammer_inode_data));
649 leaf->data_crc = crc32(data, HAMMER_INODE_CRCSIZE);
652 leaf->data_crc = crc32(data, leaf->data_len);
659 hkprintf(const char *ctl, ...)
663 if (hammer_debug_debug) {
671 * Return the block size at the specified file offset.
674 hammer_blocksize(int64_t file_offset)
676 if (file_offset < HAMMER_XDEMARC)
677 return(HAMMER_BUFSIZE);
679 return(HAMMER_XBUFSIZE);
683 * Return the demarkation point between the two offsets where
684 * the block size changes.
687 hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2)
689 if (file_offset1 < HAMMER_XDEMARC) {
690 if (file_offset2 <= HAMMER_XDEMARC)
691 return(file_offset2);
692 return(HAMMER_XDEMARC);
694 panic("hammer_blockdemarc: illegal range %lld %lld\n",
695 (long long)file_offset1, (long long)file_offset2);
699 hammer_fsid_to_udev(uuid_t *uuid)
703 crc = crc32(uuid, sizeof(*uuid));