166325755SMatthew Dillon /* 2b84de5afSMatthew Dillon * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved. 366325755SMatthew Dillon * 466325755SMatthew Dillon * This code is derived from software contributed to The DragonFly Project 566325755SMatthew Dillon * by Matthew Dillon <dillon@backplane.com> 666325755SMatthew Dillon * 766325755SMatthew Dillon * Redistribution and use in source and binary forms, with or without 866325755SMatthew Dillon * modification, are permitted provided that the following conditions 966325755SMatthew Dillon * are met: 1066325755SMatthew Dillon * 1166325755SMatthew Dillon * 1. Redistributions of source code must retain the above copyright 1266325755SMatthew Dillon * notice, this list of conditions and the following disclaimer. 1366325755SMatthew Dillon * 2. Redistributions in binary form must reproduce the above copyright 1466325755SMatthew Dillon * notice, this list of conditions and the following disclaimer in 1566325755SMatthew Dillon * the documentation and/or other materials provided with the 1666325755SMatthew Dillon * distribution. 1766325755SMatthew Dillon * 3. Neither the name of The DragonFly Project nor the names of its 1866325755SMatthew Dillon * contributors may be used to endorse or promote products derived 1966325755SMatthew Dillon * from this software without specific, prior written permission. 2066325755SMatthew Dillon * 2166325755SMatthew Dillon * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 2266325755SMatthew Dillon * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 2366325755SMatthew Dillon * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 2466325755SMatthew Dillon * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 2566325755SMatthew Dillon * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 2666325755SMatthew Dillon * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 2766325755SMatthew Dillon * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 2866325755SMatthew Dillon * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 2966325755SMatthew Dillon * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 3066325755SMatthew Dillon * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 3166325755SMatthew Dillon * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3266325755SMatthew Dillon * SUCH DAMAGE. 3366325755SMatthew Dillon */ 3466325755SMatthew Dillon 3566325755SMatthew Dillon #include "hammer.h" 3666325755SMatthew Dillon 3746137e17STomohiro Kusumi static uint32_t ocp_allocbit(hammer_objid_cache_t ocp, uint32_t n); 380729c8c8SMatthew Dillon 390729c8c8SMatthew Dillon 40b84de5afSMatthew Dillon /* 41b84de5afSMatthew Dillon * Start a standard transaction. 4232fcc103SMatthew Dillon * 4332fcc103SMatthew Dillon * May be called without fs_token 44b84de5afSMatthew Dillon */ 4566325755SMatthew Dillon void 468cd0a023SMatthew Dillon hammer_start_transaction(struct hammer_transaction *trans, 478cd0a023SMatthew Dillon struct hammer_mount *hmp) 4866325755SMatthew Dillon { 49ddfdf542SMatthew Dillon struct timeval tv; 50a89aec1bSMatthew Dillon int error; 5166325755SMatthew Dillon 52b84de5afSMatthew Dillon trans->type = HAMMER_TRANS_STD; 5366325755SMatthew Dillon trans->hmp = hmp; 54a89aec1bSMatthew Dillon trans->rootvol = hammer_get_root_volume(hmp, &error); 55a89aec1bSMatthew Dillon KKASSERT(error == 0); 56b84de5afSMatthew Dillon trans->tid = 0; 572f85fa4dSMatthew Dillon trans->sync_lock_refs = 0; 5821fde338SMatthew Dillon trans->flags = 0; 59ddfdf542SMatthew Dillon 60ddfdf542SMatthew Dillon getmicrotime(&tv); 61dd94f1b1SMatthew Dillon trans->time = (unsigned long)tv.tv_sec * 1000000ULL + tv.tv_usec; 6246137e17STomohiro Kusumi trans->time32 = (uint32_t)tv.tv_sec; 6366325755SMatthew Dillon } 6466325755SMatthew Dillon 65b84de5afSMatthew Dillon /* 66b84de5afSMatthew Dillon * Start a simple read-only transaction. This will not stall. 6732fcc103SMatthew Dillon * 6832fcc103SMatthew Dillon * May be called without fs_token 69b84de5afSMatthew Dillon */ 7066325755SMatthew Dillon void 7136f82b23SMatthew Dillon hammer_simple_transaction(struct hammer_transaction *trans, 7236f82b23SMatthew Dillon struct hammer_mount *hmp) 7336f82b23SMatthew Dillon { 74ddfdf542SMatthew Dillon struct timeval tv; 7536f82b23SMatthew Dillon int error; 7636f82b23SMatthew Dillon 77b84de5afSMatthew Dillon trans->type = HAMMER_TRANS_RO; 7836f82b23SMatthew Dillon trans->hmp = hmp; 7936f82b23SMatthew Dillon trans->rootvol = hammer_get_root_volume(hmp, &error); 8036f82b23SMatthew Dillon KKASSERT(error == 0); 81b84de5afSMatthew Dillon trans->tid = 0; 822f85fa4dSMatthew Dillon trans->sync_lock_refs = 0; 8321fde338SMatthew Dillon trans->flags = 0; 84ddfdf542SMatthew Dillon 85ddfdf542SMatthew Dillon getmicrotime(&tv); 86dd94f1b1SMatthew Dillon trans->time = (unsigned long)tv.tv_sec * 1000000ULL + tv.tv_usec; 8746137e17STomohiro Kusumi trans->time32 = (uint32_t)tv.tv_sec; 8836f82b23SMatthew Dillon } 8936f82b23SMatthew Dillon 90b84de5afSMatthew Dillon /* 91b84de5afSMatthew Dillon * Start a transaction using a particular TID. Used by the sync code. 92b84de5afSMatthew Dillon * This does not stall. 932f85fa4dSMatthew Dillon * 942f85fa4dSMatthew Dillon * This routine may only be called from the flusher thread. We predispose 952f85fa4dSMatthew Dillon * sync_lock_refs, implying serialization against the synchronization stage 962f85fa4dSMatthew Dillon * (which the flusher is responsible for). 97b84de5afSMatthew Dillon */ 9836f82b23SMatthew Dillon void 99b84de5afSMatthew Dillon hammer_start_transaction_fls(struct hammer_transaction *trans, 100b84de5afSMatthew Dillon struct hammer_mount *hmp) 101d113fda1SMatthew Dillon { 102ddfdf542SMatthew Dillon struct timeval tv; 103d113fda1SMatthew Dillon int error; 104d113fda1SMatthew Dillon 1059f5097dcSMatthew Dillon bzero(trans, sizeof(*trans)); 1069f5097dcSMatthew Dillon 107b84de5afSMatthew Dillon trans->type = HAMMER_TRANS_FLS; 108d113fda1SMatthew Dillon trans->hmp = hmp; 109d113fda1SMatthew Dillon trans->rootvol = hammer_get_root_volume(hmp, &error); 110d113fda1SMatthew Dillon KKASSERT(error == 0); 1115de0c0e5SMatthew Dillon trans->tid = hammer_alloc_tid(hmp, 1); 1122f85fa4dSMatthew Dillon trans->sync_lock_refs = 1; 11321fde338SMatthew Dillon trans->flags = 0; 114ddfdf542SMatthew Dillon 115ddfdf542SMatthew Dillon getmicrotime(&tv); 116dd94f1b1SMatthew Dillon trans->time = (unsigned long)tv.tv_sec * 1000000ULL + tv.tv_usec; 11746137e17STomohiro Kusumi trans->time32 = (uint32_t)tv.tv_sec; 118d113fda1SMatthew Dillon } 119d113fda1SMatthew Dillon 12032fcc103SMatthew Dillon /* 12132fcc103SMatthew Dillon * May be called without fs_token 12232fcc103SMatthew Dillon */ 123d113fda1SMatthew Dillon void 124b84de5afSMatthew Dillon hammer_done_transaction(struct hammer_transaction *trans) 12566325755SMatthew Dillon { 126f31f6d84SSascha Wildner int expected_lock_refs __debugvar; 1272f85fa4dSMatthew Dillon 128a89aec1bSMatthew Dillon hammer_rel_volume(trans->rootvol, 0); 129b84de5afSMatthew Dillon trans->rootvol = NULL; 1302f85fa4dSMatthew Dillon expected_lock_refs = (trans->type == HAMMER_TRANS_FLS) ? 1 : 0; 1312f85fa4dSMatthew Dillon KKASSERT(trans->sync_lock_refs == expected_lock_refs); 1322f85fa4dSMatthew Dillon trans->sync_lock_refs = 0; 13382010f9fSMatthew Dillon if (trans->type != HAMMER_TRANS_FLS) { 13432fcc103SMatthew Dillon if (trans->flags & HAMMER_TRANSF_NEWINODE) { 13532fcc103SMatthew Dillon lwkt_gettoken(&trans->hmp->fs_token); 136e98f1b96SMatthew Dillon hammer_inode_waitreclaims(trans); 13732fcc103SMatthew Dillon lwkt_reltoken(&trans->hmp->fs_token); 13832fcc103SMatthew Dillon } 13982010f9fSMatthew Dillon } 14066325755SMatthew Dillon } 14166325755SMatthew Dillon 142d113fda1SMatthew Dillon /* 1435de0c0e5SMatthew Dillon * Allocate (count) TIDs. If running in multi-master mode the returned 1445de0c0e5SMatthew Dillon * base will be aligned to a 16-count plus the master id (0-15). 1455de0c0e5SMatthew Dillon * Multi-master mode allows non-conflicting to run and new objects to be 1465de0c0e5SMatthew Dillon * created on multiple masters in parallel. The transaction id identifies 1475de0c0e5SMatthew Dillon * the original master. The object_id is also subject to this rule in 1485de0c0e5SMatthew Dillon * order to allow objects to be created on multiple masters in parallel. 1495de0c0e5SMatthew Dillon * 1505de0c0e5SMatthew Dillon * Directories may pre-allocate a large number of object ids (100,000). 1515de0c0e5SMatthew Dillon * 1525de0c0e5SMatthew Dillon * NOTE: There is no longer a requirement that successive transaction 1535de0c0e5SMatthew Dillon * ids be 2 apart for separator generation. 15483f2a3aaSMatthew Dillon * 15583f2a3aaSMatthew Dillon * NOTE: When called by pseudo-backends such as ioctls the allocated 15683f2a3aaSMatthew Dillon * TID will be larger then the current flush TID, if a flush is running, 15783f2a3aaSMatthew Dillon * so any mirroring will pick the records up on a later flush. 158c204ebbeSTomohiro Kusumi * 159c204ebbeSTomohiro Kusumi * NOTE: HAMMER1 does not support multi-master clustering as of 2015. 160d113fda1SMatthew Dillon */ 16183f2a3aaSMatthew Dillon hammer_tid_t 1625de0c0e5SMatthew Dillon hammer_alloc_tid(hammer_mount_t hmp, int count) 163a89aec1bSMatthew Dillon { 164a89aec1bSMatthew Dillon hammer_tid_t tid; 165a89aec1bSMatthew Dillon 166732a1697SMatthew Dillon if (hmp->master_id < 0) { 167c82af904SMatthew Dillon tid = hmp->next_tid + 1; 168c82af904SMatthew Dillon hmp->next_tid = tid + count; 169c82af904SMatthew Dillon } else { 170c82af904SMatthew Dillon tid = (hmp->next_tid + HAMMER_MAX_MASTERS) & 171c82af904SMatthew Dillon ~(hammer_tid_t)(HAMMER_MAX_MASTERS - 1); 172c82af904SMatthew Dillon hmp->next_tid = tid + count * HAMMER_MAX_MASTERS; 173732a1697SMatthew Dillon tid |= hmp->master_id; 174c82af904SMatthew Dillon } 175c82af904SMatthew Dillon if (tid >= 0xFFFFFFFFFF000000ULL) 176903fdd05STomohiro Kusumi hpanic("Ran out of TIDs!"); 1775de0c0e5SMatthew Dillon if (hammer_debug_tid) 178*35a5249bSTomohiro Kusumi hdkprintf("%016jx\n", (intmax_t)tid); 179a89aec1bSMatthew Dillon return(tid); 180a89aec1bSMatthew Dillon } 181a89aec1bSMatthew Dillon 1820729c8c8SMatthew Dillon /* 1835a64efa1SMatthew Dillon * Allocate an object id. 1845a64efa1SMatthew Dillon * 1855a64efa1SMatthew Dillon * We use the upper OBJID_CACHE_BITS bits of the namekey to try to match 1865a64efa1SMatthew Dillon * the low bits of the objid we allocate. 1870729c8c8SMatthew Dillon */ 1880729c8c8SMatthew Dillon hammer_tid_t 1895a64efa1SMatthew Dillon hammer_alloc_objid(hammer_mount_t hmp, hammer_inode_t dip, int64_t namekey) 1900729c8c8SMatthew Dillon { 1910729c8c8SMatthew Dillon hammer_objid_cache_t ocp; 1920729c8c8SMatthew Dillon hammer_tid_t tid; 19346137e17STomohiro Kusumi uint32_t n; 1940729c8c8SMatthew Dillon 1950729c8c8SMatthew Dillon while ((ocp = dip->objid_cache) == NULL) { 1965de0c0e5SMatthew Dillon if (hmp->objid_cache_count < OBJID_CACHE_SIZE) { 197bac808feSMatthew Dillon ocp = kmalloc(sizeof(*ocp), hmp->m_misc, 198bac808feSMatthew Dillon M_WAITOK|M_ZERO); 1995a64efa1SMatthew Dillon ocp->base_tid = hammer_alloc_tid(hmp, 2005a64efa1SMatthew Dillon OBJID_CACHE_BULK * 2); 2015a64efa1SMatthew Dillon ocp->base_tid += OBJID_CACHE_BULK_MASK64; 2025a64efa1SMatthew Dillon ocp->base_tid &= ~OBJID_CACHE_BULK_MASK64; 2030729c8c8SMatthew Dillon /* may have blocked, recheck */ 2040729c8c8SMatthew Dillon if (dip->objid_cache == NULL) { 205d1ce1558SMatthew Dillon TAILQ_INSERT_TAIL(&hmp->objid_cache_list, 206d1ce1558SMatthew Dillon ocp, entry); 207d1ce1558SMatthew Dillon ++hmp->objid_cache_count; 2080729c8c8SMatthew Dillon dip->objid_cache = ocp; 2090729c8c8SMatthew Dillon ocp->dip = dip; 210d1ce1558SMatthew Dillon } else { 211d1ce1558SMatthew Dillon kfree(ocp, hmp->m_misc); 2120729c8c8SMatthew Dillon } 2130729c8c8SMatthew Dillon } else { 2145a64efa1SMatthew Dillon /* 2155a64efa1SMatthew Dillon * Steal one from another directory? 2165a64efa1SMatthew Dillon * 2175a64efa1SMatthew Dillon * Throw away ocp's that are more then half full, they 2185a64efa1SMatthew Dillon * aren't worth stealing. 2195a64efa1SMatthew Dillon */ 2205de0c0e5SMatthew Dillon ocp = TAILQ_FIRST(&hmp->objid_cache_list); 2210729c8c8SMatthew Dillon if (ocp->dip) 2220729c8c8SMatthew Dillon ocp->dip->objid_cache = NULL; 2235a64efa1SMatthew Dillon if (ocp->count >= OBJID_CACHE_BULK / 2) { 224d1ce1558SMatthew Dillon TAILQ_REMOVE(&hmp->objid_cache_list, 225d1ce1558SMatthew Dillon ocp, entry); 2265a64efa1SMatthew Dillon --hmp->objid_cache_count; 2275a64efa1SMatthew Dillon kfree(ocp, hmp->m_misc); 2285a64efa1SMatthew Dillon } else { 2290729c8c8SMatthew Dillon dip->objid_cache = ocp; 2300729c8c8SMatthew Dillon ocp->dip = dip; 2310729c8c8SMatthew Dillon } 2320729c8c8SMatthew Dillon } 2335a64efa1SMatthew Dillon } 2345de0c0e5SMatthew Dillon TAILQ_REMOVE(&hmp->objid_cache_list, ocp, entry); 2355de0c0e5SMatthew Dillon 2365de0c0e5SMatthew Dillon /* 237d1ce1558SMatthew Dillon * Allocate inode numbers uniformly. 2385a64efa1SMatthew Dillon */ 239d1ce1558SMatthew Dillon 2405a64efa1SMatthew Dillon n = (namekey >> (63 - OBJID_CACHE_BULK_BITS)) & OBJID_CACHE_BULK_MASK; 2415a64efa1SMatthew Dillon n = ocp_allocbit(ocp, n); 2425a64efa1SMatthew Dillon tid = ocp->base_tid + n; 2435a64efa1SMatthew Dillon 2445a64efa1SMatthew Dillon #if 0 2455a64efa1SMatthew Dillon /* 2465de0c0e5SMatthew Dillon * The TID is incremented by 1 or by 16 depending what mode the 2475de0c0e5SMatthew Dillon * mount is operating in. 2485de0c0e5SMatthew Dillon */ 249732a1697SMatthew Dillon ocp->next_tid += (hmp->master_id < 0) ? 1 : HAMMER_MAX_MASTERS; 2505a64efa1SMatthew Dillon #endif 251d1ce1558SMatthew Dillon if (ocp->count >= OBJID_CACHE_BULK * 3 / 4) { 2520729c8c8SMatthew Dillon dip->objid_cache = NULL; 2535de0c0e5SMatthew Dillon --hmp->objid_cache_count; 2540729c8c8SMatthew Dillon ocp->dip = NULL; 255bac808feSMatthew Dillon kfree(ocp, hmp->m_misc); 2560729c8c8SMatthew Dillon } else { 2575de0c0e5SMatthew Dillon TAILQ_INSERT_TAIL(&hmp->objid_cache_list, ocp, entry); 2580729c8c8SMatthew Dillon } 2590729c8c8SMatthew Dillon return(tid); 2600729c8c8SMatthew Dillon } 2610729c8c8SMatthew Dillon 2625a64efa1SMatthew Dillon /* 2635a64efa1SMatthew Dillon * Allocate a bit starting with bit n. Wrap if necessary. 2645a64efa1SMatthew Dillon * 2655a64efa1SMatthew Dillon * This routine is only ever called if a bit is available somewhere 2665a64efa1SMatthew Dillon * in the bitmap. 2675a64efa1SMatthew Dillon */ 26846137e17STomohiro Kusumi static uint32_t 26946137e17STomohiro Kusumi ocp_allocbit(hammer_objid_cache_t ocp, uint32_t n) 2705a64efa1SMatthew Dillon { 27146137e17STomohiro Kusumi uint32_t n0; 2725a64efa1SMatthew Dillon 2735a64efa1SMatthew Dillon n0 = (n >> 5) & 31; 2745a64efa1SMatthew Dillon n &= 31; 2755a64efa1SMatthew Dillon 2765a64efa1SMatthew Dillon while (ocp->bm1[n0] & (1 << n)) { 2775a64efa1SMatthew Dillon if (ocp->bm0 & (1 << n0)) { 2785a64efa1SMatthew Dillon n0 = (n0 + 1) & 31; 2795a64efa1SMatthew Dillon n = 0; 2805a64efa1SMatthew Dillon } else if (++n == 32) { 2815a64efa1SMatthew Dillon n0 = (n0 + 1) & 31; 2825a64efa1SMatthew Dillon n = 0; 2835a64efa1SMatthew Dillon } 2845a64efa1SMatthew Dillon } 2855a64efa1SMatthew Dillon ++ocp->count; 2865a64efa1SMatthew Dillon ocp->bm1[n0] |= 1 << n; 2875a64efa1SMatthew Dillon if (ocp->bm1[n0] == 0xFFFFFFFFU) 2885a64efa1SMatthew Dillon ocp->bm0 |= 1 << n0; 2895a64efa1SMatthew Dillon return((n0 << 5) + n); 2905a64efa1SMatthew Dillon } 2915a64efa1SMatthew Dillon 2920729c8c8SMatthew Dillon void 2930729c8c8SMatthew Dillon hammer_clear_objid(hammer_inode_t dip) 2940729c8c8SMatthew Dillon { 2950729c8c8SMatthew Dillon hammer_objid_cache_t ocp; 2960729c8c8SMatthew Dillon 2970729c8c8SMatthew Dillon if ((ocp = dip->objid_cache) != NULL) { 2980729c8c8SMatthew Dillon dip->objid_cache = NULL; 2990729c8c8SMatthew Dillon ocp->dip = NULL; 3000729c8c8SMatthew Dillon TAILQ_REMOVE(&dip->hmp->objid_cache_list, ocp, entry); 3010729c8c8SMatthew Dillon TAILQ_INSERT_HEAD(&dip->hmp->objid_cache_list, ocp, entry); 3020729c8c8SMatthew Dillon } 3030729c8c8SMatthew Dillon } 3040729c8c8SMatthew Dillon 3050729c8c8SMatthew Dillon void 3060729c8c8SMatthew Dillon hammer_destroy_objid_cache(hammer_mount_t hmp) 3070729c8c8SMatthew Dillon { 3080729c8c8SMatthew Dillon hammer_objid_cache_t ocp; 3090729c8c8SMatthew Dillon 3100729c8c8SMatthew Dillon while ((ocp = TAILQ_FIRST(&hmp->objid_cache_list)) != NULL) { 3110729c8c8SMatthew Dillon TAILQ_REMOVE(&hmp->objid_cache_list, ocp, entry); 312f437a2abSMatthew Dillon if (ocp->dip) 313f437a2abSMatthew Dillon ocp->dip->objid_cache = NULL; 314bac808feSMatthew Dillon kfree(ocp, hmp->m_misc); 315d1ce1558SMatthew Dillon --hmp->objid_cache_count; 3160729c8c8SMatthew Dillon } 317d1ce1558SMatthew Dillon KKASSERT(hmp->objid_cache_count == 0); 3180729c8c8SMatthew Dillon } 3190729c8c8SMatthew Dillon 320