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;
50 KKASSERT(lock->refs > 0);
55 nlv = 1 | HAMMER_LOCKF_EXCLUSIVE;
56 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
60 } else if ((lv & HAMMER_LOCKF_EXCLUSIVE) && lock->owner == td) {
62 if (atomic_cmpset_int(&lock->lockval, lv, nlv))
65 if (hammer_debug_locks) {
66 kprintf("hammer_lock_ex: held by %p\n",
69 nlv = lv | HAMMER_LOCKF_WANTED;
70 ++hammer_contention_count;
71 tsleep_interlock(lock, 0);
72 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
73 tsleep(lock, PINTERLOCKED, ident, 0);
74 if (hammer_debug_locks)
75 kprintf("hammer_lock_ex: try again\n");
82 * Try to obtain an exclusive lock
85 hammer_lock_ex_try(struct hammer_lock *lock)
87 thread_t td = curthread;
92 KKASSERT(lock->refs > 0);
97 nlv = 1 | HAMMER_LOCKF_EXCLUSIVE;
98 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
103 } else if ((lv & HAMMER_LOCKF_EXCLUSIVE) && lock->owner == td) {
105 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
118 * Obtain a shared lock
120 * We do not give pending exclusive locks priority over shared locks as
121 * doing so could lead to a deadlock.
124 hammer_lock_sh(struct hammer_lock *lock)
126 thread_t td = curthread;
130 KKASSERT(lock->refs > 0);
134 if ((lv & HAMMER_LOCKF_EXCLUSIVE) == 0) {
136 if (atomic_cmpset_int(&lock->lockval, lv, nlv))
138 } else if (lock->owner == td) {
140 * Disallowed case, drop into kernel debugger for
141 * now. A cont continues w/ an exclusive lock.
144 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
145 if (hammer_debug_critical)
146 Debugger("hammer_lock_sh: holding ex");
150 nlv = lv | HAMMER_LOCKF_WANTED;
151 ++hammer_contention_count;
152 tsleep_interlock(lock, 0);
153 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
154 tsleep(lock, PINTERLOCKED, "hmrlck", 0);
161 hammer_lock_sh_try(struct hammer_lock *lock)
163 thread_t td = curthread;
168 KKASSERT(lock->refs > 0);
172 if ((lv & HAMMER_LOCKF_EXCLUSIVE) == 0) {
174 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
178 } else if (lock->owner == td) {
180 * Disallowed case, drop into kernel debugger for
181 * now. A cont continues w/ an exclusive lock.
184 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
185 if (hammer_debug_critical)
186 Debugger("hammer_lock_sh: holding ex");
199 * Upgrade a shared lock to an exclusively held lock. This function will
200 * return EDEADLK If there is more then one shared holder.
202 * No error occurs and no action is taken if the lock is already exclusively
203 * held by the caller. If the lock is not held at all or held exclusively
204 * by someone else, this function will panic.
207 hammer_lock_upgrade(struct hammer_lock *lock)
209 thread_t td = curthread;
217 if ((lv & ~HAMMER_LOCKF_WANTED) == 1) {
218 nlv = lv | HAMMER_LOCKF_EXCLUSIVE;
219 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
224 } else if (lv & HAMMER_LOCKF_EXCLUSIVE) {
225 if (lock->owner != curthread)
226 panic("hammer_lock_upgrade: illegal state");
229 } else if ((lv & ~HAMMER_LOCKF_WANTED) == 0) {
230 panic("hammer_lock_upgrade: lock is not held");
243 * Downgrade an exclusively held lock to a shared lock.
246 hammer_lock_downgrade(struct hammer_lock *lock)
248 thread_t td __debugvar = curthread;
252 KKASSERT((lock->lockval & ~HAMMER_LOCKF_WANTED) ==
253 (HAMMER_LOCKF_EXCLUSIVE | 1));
254 KKASSERT(lock->owner == td);
257 * NOTE: Must clear owner before releasing exclusivity
263 nlv = lv & ~(HAMMER_LOCKF_EXCLUSIVE | HAMMER_LOCKF_WANTED);
264 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
265 if (lv & HAMMER_LOCKF_WANTED)
273 hammer_unlock(struct hammer_lock *lock)
275 thread_t td __debugvar = curthread;
281 if (lv & HAMMER_LOCKF_EXCLUSIVE)
282 KKASSERT(lock->owner == td);
286 nlv = lv & ~(HAMMER_LOCKF_EXCLUSIVE | HAMMER_LOCKF_WANTED);
289 if (atomic_cmpset_int(&lock->lockval, lv, nlv))
291 } else if (nlv == 1) {
293 if (lv & HAMMER_LOCKF_EXCLUSIVE)
295 if (atomic_cmpset_int(&lock->lockval, lv, nlv)) {
296 if (lv & HAMMER_LOCKF_WANTED)
301 panic("hammer_unlock: lock %p is not held", lock);
307 * The calling thread must be holding a shared or exclusive lock.
308 * Returns < 0 if lock is held shared, and > 0 if held exlusively.
311 hammer_lock_status(struct hammer_lock *lock)
313 u_int lv = lock->lockval;
315 if (lv & HAMMER_LOCKF_EXCLUSIVE)
319 panic("hammer_lock_status: lock must be held: %p", lock);
323 hammer_ref(struct hammer_lock *lock)
325 KKASSERT(lock->refs >= 0);
326 atomic_add_int(&lock->refs, 1);
330 hammer_unref(struct hammer_lock *lock)
332 KKASSERT(lock->refs > 0);
333 atomic_subtract_int(&lock->refs, 1);
337 * The sync_lock must be held when doing any modifying operations on
338 * meta-data. It does not have to be held when modifying non-meta-data buffers
339 * (backend or frontend).
341 * The flusher holds the lock exclusively while all other consumers hold it
342 * shared. All modifying operations made while holding the lock are atomic
343 * in that they will be made part of the same flush group.
345 * Due to the atomicy requirement deadlock recovery code CANNOT release the
346 * sync lock, nor can we give pending exclusive sync locks priority over
347 * a shared sync lock as this could lead to a 3-way deadlock.
350 hammer_sync_lock_ex(hammer_transaction_t trans)
352 ++trans->sync_lock_refs;
353 hammer_lock_ex(&trans->hmp->sync_lock);
357 hammer_sync_lock_sh(hammer_transaction_t trans)
359 ++trans->sync_lock_refs;
360 hammer_lock_sh(&trans->hmp->sync_lock);
364 hammer_sync_lock_sh_try(hammer_transaction_t trans)
368 ++trans->sync_lock_refs;
369 if ((error = hammer_lock_sh_try(&trans->hmp->sync_lock)) != 0)
370 --trans->sync_lock_refs;
375 hammer_sync_unlock(hammer_transaction_t trans)
377 --trans->sync_lock_refs;
378 hammer_unlock(&trans->hmp->sync_lock);
385 hammer_to_unix_xid(uuid_t *uuid)
387 return(*(u_int32_t *)&uuid->node[2]);
391 hammer_guid_to_uuid(uuid_t *uuid, u_int32_t guid)
393 bzero(uuid, sizeof(*uuid));
394 *(u_int32_t *)&uuid->node[2] = guid;
398 hammer_time_to_timespec(u_int64_t xtime, struct timespec *ts)
400 ts->tv_sec = (unsigned long)(xtime / 1000000);
401 ts->tv_nsec = (unsigned int)(xtime % 1000000) * 1000L;
405 hammer_timespec_to_time(struct timespec *ts)
409 xtime = (unsigned)(ts->tv_nsec / 1000) +
410 (unsigned long)ts->tv_sec * 1000000ULL;
416 * Convert a HAMMER filesystem object type to a vnode type
419 hammer_get_vnode_type(u_int8_t obj_type)
422 case HAMMER_OBJTYPE_DIRECTORY:
424 case HAMMER_OBJTYPE_REGFILE:
426 case HAMMER_OBJTYPE_DBFILE:
428 case HAMMER_OBJTYPE_FIFO:
430 case HAMMER_OBJTYPE_SOCKET:
432 case HAMMER_OBJTYPE_CDEV:
434 case HAMMER_OBJTYPE_BDEV:
436 case HAMMER_OBJTYPE_SOFTLINK:
445 hammer_get_dtype(u_int8_t obj_type)
448 case HAMMER_OBJTYPE_DIRECTORY:
450 case HAMMER_OBJTYPE_REGFILE:
452 case HAMMER_OBJTYPE_DBFILE:
454 case HAMMER_OBJTYPE_FIFO:
456 case HAMMER_OBJTYPE_SOCKET:
458 case HAMMER_OBJTYPE_CDEV:
460 case HAMMER_OBJTYPE_BDEV:
462 case HAMMER_OBJTYPE_SOFTLINK:
471 hammer_get_obj_type(enum vtype vtype)
475 return(HAMMER_OBJTYPE_DIRECTORY);
477 return(HAMMER_OBJTYPE_REGFILE);
479 return(HAMMER_OBJTYPE_DBFILE);
481 return(HAMMER_OBJTYPE_FIFO);
483 return(HAMMER_OBJTYPE_SOCKET);
485 return(HAMMER_OBJTYPE_CDEV);
487 return(HAMMER_OBJTYPE_BDEV);
489 return(HAMMER_OBJTYPE_SOFTLINK);
491 return(HAMMER_OBJTYPE_UNKNOWN);
497 * Return flags for hammer_delete_at_cursor()
500 hammer_nohistory(hammer_inode_t ip)
502 if (ip->hmp->hflags & HMNT_NOHISTORY)
503 return(HAMMER_DELETE_DESTROY);
504 if (ip->ino_data.uflags & (SF_NOHISTORY|UF_NOHISTORY))
505 return(HAMMER_DELETE_DESTROY);
510 * ALGORITHM VERSION 1:
511 * Return a namekey hash. The 64 bit namekey hash consists of a 32 bit
512 * crc in the MSB and 0 in the LSB. The caller will use the low 32 bits
513 * to generate a unique key and will scan all entries with the same upper
514 * 32 bits when issuing a lookup.
516 * 0hhhhhhhhhhhhhhh hhhhhhhhhhhhhhhh 0000000000000000 0000000000000000
518 * ALGORITHM VERSION 2:
520 * The 64 bit hash key is generated from the following components. The
521 * first three characters are encoded as 5-bit quantities, the middle
522 * N characters are hashed into a 6 bit quantity, and the last two
523 * characters are encoded as 5-bit quantities. A 32 bit hash of the
524 * entire filename is encoded in the low 32 bits. Bit 0 is set to
525 * 0 to guarantee us a 2^24 bit iteration space.
527 * 0aaaaabbbbbccccc mmmmmmyyyyyzzzzz hhhhhhhhhhhhhhhh hhhhhhhhhhhhhhh0
529 * This gives us a domain sort for the first three characters, the last
530 * two characters, and breaks the middle space into 64 random domains.
531 * The domain sort folds upper case, lower case, digits, and punctuation
532 * spaces together, the idea being the filenames tend to not be a mix
535 * The 64 random domains act as a sub-sort for the middle characters
536 * but may cause a random seek. If the filesystem is being accessed
537 * in sorted order we should tend to get very good linearity for most
538 * filenames and devolve into more random seeks otherwise.
540 * We strip bit 63 in order to provide a positive key, this way a seek
541 * offset of 0 will represent the base of the directory.
543 * This function can never return 0. We use the MSB-0 space to synthesize
544 * artificial directory entries such as "." and "..".
547 hammer_directory_namekey(hammer_inode_t dip, const void *name, int len,
548 u_int32_t *max_iterationsp)
552 const char *aname = name;
554 switch (dip->ino_data.cap_flags & HAMMER_INODE_CAP_DIRHASH_MASK) {
555 case HAMMER_INODE_CAP_DIRHASH_ALG0:
556 key = (int64_t)(crc32(aname, len) & 0x7FFFFFFF) << 32;
558 key |= 0x100000000LL;
559 *max_iterationsp = 0xFFFFFFFFU;
561 case HAMMER_INODE_CAP_DIRHASH_ALG1:
562 key = (u_int32_t)crc32(aname, len) & 0xFFFFFFFEU;
566 crcx = crc32(aname + 3, len - 5);
567 crcx = crcx ^ (crcx >> 6) ^ (crcx >> 12);
568 key |= (int64_t)(crcx & 0x3F) << 42;
574 key |= ((int64_t)(aname[2] & 0x1F) << 48);
577 key |= ((int64_t)(aname[1] & 0x1F) << 53) |
578 ((int64_t)(aname[len-2] & 0x1F) << 37);
581 key |= ((int64_t)(aname[0] & 0x1F) << 58) |
582 ((int64_t)(aname[len-1] & 0x1F) << 32);
587 if ((key & 0xFFFFFFFF00000000LL) == 0)
588 key |= 0x100000000LL;
589 if (hammer_debug_general & 0x0400) {
590 kprintf("namekey2: 0x%016llx %*.*s\n",
591 (long long)key, len, len, aname);
593 *max_iterationsp = 0x00FFFFFF;
595 case HAMMER_INODE_CAP_DIRHASH_ALG2:
596 case HAMMER_INODE_CAP_DIRHASH_ALG3:
598 key = 0; /* compiler warning */
599 *max_iterationsp = 1; /* sanity */
600 panic("hammer_directory_namekey: bad algorithm %p\n", dip);
607 * Convert string after @@ (@@ not included) to TID. Returns 0 on success,
610 * If this function fails *ispfs, *tidp, and *localizationp will not
614 hammer_str_to_tid(const char *str, int *ispfsp,
615 hammer_tid_t *tidp, u_int32_t *localizationp)
618 u_int32_t localization;
624 * Forms allowed for TID: "0x%016llx"
627 tid = strtouq(str, &ptr, 0);
629 if (n == 2 && str[0] == '-' && str[1] == '1') {
631 } else if (n == 18 && str[0] == '0' && (str[1] | 0x20) == 'x') {
638 * Forms allowed for PFS: ":%05d" (i.e. "...:0" would be illegal).
642 localization = strtoul(str + 1, &ptr, 10) << 16;
648 localization = *localizationp;
653 * Any trailing junk invalidates special extension handling.
658 *localizationp = localization;
664 hammer_crc_set_blockmap(hammer_blockmap_t blockmap)
666 blockmap->entry_crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
670 hammer_crc_set_volume(hammer_volume_ondisk_t ondisk)
672 ondisk->vol_crc = crc32(ondisk, HAMMER_VOL_CRCSIZE1) ^
673 crc32(&ondisk->vol_crc + 1, HAMMER_VOL_CRCSIZE2);
677 hammer_crc_test_blockmap(hammer_blockmap_t blockmap)
681 crc = crc32(blockmap, HAMMER_BLOCKMAP_CRCSIZE);
682 return (blockmap->entry_crc == crc);
686 hammer_crc_test_volume(hammer_volume_ondisk_t ondisk)
690 crc = crc32(ondisk, HAMMER_VOL_CRCSIZE1) ^
691 crc32(&ondisk->vol_crc + 1, HAMMER_VOL_CRCSIZE2);
692 return (ondisk->vol_crc == crc);
696 hammer_crc_test_btree(hammer_node_ondisk_t ondisk)
700 crc = crc32(&ondisk->crc + 1, HAMMER_BTREE_CRCSIZE);
701 return (ondisk->crc == crc);
705 * Test or set the leaf->data_crc field. Deal with any special cases given
706 * a generic B-Tree leaf element and its data.
708 * NOTE: Inode-data: the atime and mtime fields are not CRCd, allowing them
709 * to be updated in-place.
712 hammer_crc_test_leaf(void *data, hammer_btree_leaf_elm_t leaf)
716 if (leaf->data_len == 0) {
719 switch(leaf->base.rec_type) {
720 case HAMMER_RECTYPE_INODE:
721 if (leaf->data_len != sizeof(struct hammer_inode_data))
723 crc = crc32(data, HAMMER_INODE_CRCSIZE);
726 crc = crc32(data, leaf->data_len);
730 return (leaf->data_crc == crc);
734 hammer_crc_set_leaf(void *data, hammer_btree_leaf_elm_t leaf)
736 if (leaf->data_len == 0) {
739 switch(leaf->base.rec_type) {
740 case HAMMER_RECTYPE_INODE:
741 KKASSERT(leaf->data_len ==
742 sizeof(struct hammer_inode_data));
743 leaf->data_crc = crc32(data, HAMMER_INODE_CRCSIZE);
746 leaf->data_crc = crc32(data, leaf->data_len);
753 hkprintf(const char *ctl, ...)
757 if (hammer_debug_debug) {
765 * Return the block size at the specified file offset.
768 hammer_blocksize(int64_t file_offset)
770 if (file_offset < HAMMER_XDEMARC)
771 return(HAMMER_BUFSIZE);
773 return(HAMMER_XBUFSIZE);
777 hammer_blockoff(int64_t file_offset)
779 if (file_offset < HAMMER_XDEMARC)
780 return((int)file_offset & HAMMER_BUFMASK);
782 return((int)file_offset & HAMMER_XBUFMASK);
786 * Return the demarkation point between the two offsets where
787 * the block size changes.
790 hammer_blockdemarc(int64_t file_offset1, int64_t file_offset2)
792 if (file_offset1 < HAMMER_XDEMARC) {
793 if (file_offset2 <= HAMMER_XDEMARC)
794 return(file_offset2);
795 return(HAMMER_XDEMARC);
797 panic("hammer_blockdemarc: illegal range %lld %lld\n",
798 (long long)file_offset1, (long long)file_offset2);
802 hammer_fsid_to_udev(uuid_t *uuid)
806 crc = crc32(uuid, sizeof(*uuid));