core/BTreeNodeReserve.c

   1 /*
   2  * Copyright (c) 2004-2015 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 #include "BTreesPrivate.h"
  29 #include "sys/malloc.h"
  30 #include <kern/locks.h>
  31
  32
  33 /*
  34  * B-tree Node Reserve
  35  *
  36  * BTReserveSpace
  37  * BTReleaseReserve
  38  * BTUpdateReserve
  39  *
  40  * Each kernel thread can have it's own reserve of b-tree
  41  * nodes. This reserve info is kept in a hash table.
  42  *
  43  * Don't forget to call BTReleaseReserve when you're finished
  44  * or you will leave stale node reserves in the hash.
  45  */
  46
  47
  48 /*
  49  * BE CAREFUL WHEN INCREASING THE SIZE OF THIS STRUCT!
  50  *
  51  * It must remain equal in size to the opaque cat_cookie_t
  52  * struct (in hfs_catalog.h).
  53  */
  54 struct nreserve {
  55         LIST_ENTRY(nreserve) nr_hash;  /* hash chain */
  56         int  nr_nodecnt;               /* count of nodes held in reserve */
  57         int  nr_newnodes;              /* nodes that were allocated */
  58         struct  vnode *nr_btvp;        /* b-tree file vnode */
  59         void  *nr_tag;                 /* unique tag (per thread) */
  60 };
  61
  62 #define NR_GET_TAG()    (current_thread())
  63
  64 #define NR_CACHE 17
  65
  66 #define NR_HASH(btvp, tag) \
  67         (&nr_hashtbl[((((intptr_t)(btvp)) >> 8) ^ ((intptr_t)(tag) >> 4)) & nr_hashmask])
  68
  69 LIST_HEAD(nodereserve, nreserve) *nr_hashtbl;
  70
  71 u_long nr_hashmask;
  72
  73 lck_grp_t * nr_lck_grp;
  74 lck_grp_attr_t * nr_lck_grp_attr;
  75 lck_attr_t * nr_lck_attr;
  76
  77 lck_mtx_t  nr_mutex;
  78
  79 /* Internal Node Reserve Hash Routines (private) */
  80 static void nr_insert (struct vnode *, struct nreserve *nrp, int);
  81 static void nr_delete (struct vnode *, struct nreserve *nrp, int *);
  82 static void nr_update (struct vnode *, int);
  83
  84
  85 /*
  86  * BTReserveSetup - initialize the node reserve hash table
  87  */
  88 void BTReserveSetup(void)
  89 {
  90         if (sizeof(struct nreserve) != sizeof(cat_cookie_t))
  91                 panic("hfs: BTReserveSetup: nreserve size != opaque struct size");
  92
  93         nr_hashtbl = hashinit(NR_CACHE, M_TEMP, &nr_hashmask);
  94
  95         nr_lck_grp_attr= lck_grp_attr_alloc_init();
  96         nr_lck_grp  = lck_grp_alloc_init("btree_node_reserve", nr_lck_grp_attr);
  97
  98         nr_lck_attr = lck_attr_alloc_init();
  99
 100         lck_mtx_init(&nr_mutex, nr_lck_grp, nr_lck_attr);
 101 }
 102
 103
 104 /*
 105  * BTReserveSpace - obtain a node reserve (for current thread)
 106  *
 107  * Used by the Catalog Layer (hfs_catalog.c) to reserve space.
 108  *
 109  * When data is NULL, we only insure that there's enough space
 110  * but it is not reserved (assumes you keep the b-tree lock).
 111  */
 112 int
 113 BTReserveSpace(FCB *file, int operations, void* data)
 114 {
 115         BTreeControlBlock *btree;
 116         int rsrvNodes, availNodes, totalNodes;
 117         int height;
 118         int inserts, deletes;
 119         u_int32_t clumpsize;
 120         int err = 0;
 121
 122         btree = (BTreeControlBlockPtr)file->fcbBTCBPtr;
 123         clumpsize = file->ff_clumpsize;
 124
 125         REQUIRE_FILE_LOCK(btree->fileRefNum, true);
 126
 127         /*
 128          * The node reserve is based on the number of b-tree
 129          * operations (insert/deletes) and the height of the
 130          * tree.
 131          */
 132         height = btree->treeDepth;
 133         if (height < 2)
 134                 height = 2;  /* prevent underflow in rsrvNodes calculation */
 135         inserts = operations & 0xffff;
 136         deletes = operations >> 16;
 137
 138         /*
 139          * Allow for at least one root split.
 140          *
 141          * Each delete operation can propogate a big key up the
 142          * index. This can cause a split at each level up.
 143          *
 144          * Each insert operation can cause a local split and a
 145          * split at each level up.
 146          */
 147         rsrvNodes = 1 + (deletes * (height - 2)) + (inserts * (height - 1));
 148
 149         availNodes = btree->freeNodes - btree->reservedNodes;
 150
 151         if (rsrvNodes > availNodes) {
 152                 u_int32_t reqblks, freeblks, rsrvblks;
 153                 uint32_t bt_rsrv;
 154                 struct hfsmount *hfsmp;
 155
 156                 /*
 157                  * For UNIX conformance, we try and reserve the MIN of either 5% of
 158                  * total file blocks or 10MB worth of blocks, for growing existing
 159                  * files.  On non-HFS filesystems, creating a new directory entry may
 160                  * not cause additional disk space to be allocated, but on HFS, creating
 161                  * a new entry could cause the b-tree to grow.  As a result, we take
 162                  * some precautions here to prevent that on configurations that try to
 163                  * satisfy conformance.
 164                  */
 165                 hfsmp = VTOVCB(btree->fileRefNum);
 166                 rsrvblks = ((u_int64_t)hfsmp->allocLimit * 5) / 100;
 167                 if (hfsmp->blockSize > HFS_BT_MAXRESERVE) {
 168                         bt_rsrv = 1;
 169                 }
 170                 else {
 171                         bt_rsrv = (HFS_BT_MAXRESERVE / hfsmp->blockSize);
 172                 }
 173                 rsrvblks = MIN(rsrvblks, bt_rsrv);
 174
 175                 freeblks = hfs_freeblks(hfsmp, 0);
 176                 if (freeblks <= rsrvblks) {
 177                         /* When running low, disallow adding new items. */
 178                         if ((inserts > 0) && (deletes == 0)) {
 179                                 return (ENOSPC);
 180                         }
 181                         freeblks = 0;
 182                 } else {
 183                         freeblks -= rsrvblks;
 184                 }
 185                 reqblks = clumpsize / hfsmp->blockSize;
 186
 187                 if (reqblks > freeblks) {
 188                         reqblks = ((rsrvNodes - availNodes) * btree->nodeSize) / hfsmp->blockSize;
 189                         /* When running low, disallow adding new items. */
 190                         if ((reqblks > freeblks) && (inserts > 0) && (deletes == 0)) {
 191                                 return (ENOSPC);
 192                         }
 193                         file->ff_clumpsize = freeblks * hfsmp->blockSize;
 194                 }
 195                 totalNodes = rsrvNodes + btree->totalNodes - availNodes;
 196
 197                 /* See if we also need a map node */
 198                 if (totalNodes > (int)CalcMapBits(btree)) {
 199                         ++totalNodes;
 200                 }
 201                 if ((err = ExtendBTree(btree, totalNodes))) {
 202                         goto out;
 203                 }
 204         }
 205         /* Save this reserve if this is a persistent request. */
 206         if (data) {
 207                 btree->reservedNodes += rsrvNodes;
 208                 nr_insert(btree->fileRefNum, (struct nreserve *)data, rsrvNodes);
 209         }
 210 out:
 211         /* Put clump size back if it was changed. */
 212         if (file->ff_clumpsize != clumpsize)
 213                 file->ff_clumpsize = clumpsize;
 214
 215         return (err);
 216 }
 217
 218
 219 /*
 220  * BTReleaseReserve - release the node reserve held by current thread
 221  *
 222  * Used by the Catalog Layer (hfs_catalog.c) to relinquish reserved space.
 223  */
 224 int
 225 BTReleaseReserve(FCB *file, void* data)
 226 {
 227         BTreeControlBlock *btree;
 228         int nodecnt;
 229
 230         btree = (BTreeControlBlockPtr)file->fcbBTCBPtr;
 231
 232         REQUIRE_FILE_LOCK(btree->fileRefNum, true);
 233
 234         nr_delete(btree->fileRefNum, (struct nreserve *)data, &nodecnt);
 235
 236         if (nodecnt)
 237                 btree->reservedNodes -= nodecnt;
 238
 239         return (0);
 240 }
 241
 242 /*
 243  * BTUpdateReserve - update a node reserve for allocations that occurred.
 244  */
 245 void
 246 BTUpdateReserve(BTreeControlBlockPtr btreePtr, int nodes)
 247 {
 248         nr_update(btreePtr->fileRefNum, nodes);
 249 }
 250
 251
 252 /*----------------------------------------------------------------------------*/
 253 /* Node Reserve Hash Functions (private) */
 254
 255
 256 int nrinserts = 0;
 257 int nrdeletes = 0;
 258
 259 /*
 260  * Insert a new node reserve.
 261  */
 262 static void
 263 nr_insert(struct vnode * btvp, struct nreserve *nrp, int nodecnt)
 264 {
 265         struct nodereserve *nrhead;
 266         struct nreserve *tmp_nrp;
 267         void * tag = NR_GET_TAG();
 268
 269         /*
 270          * Check the cache - there may already be a reserve
 271          */
 272         lck_mtx_lock(&nr_mutex);
 273         nrhead = NR_HASH(btvp, tag);
 274         for (tmp_nrp = nrhead->lh_first; tmp_nrp;
 275              tmp_nrp = tmp_nrp->nr_hash.le_next) {
 276                 if ((tmp_nrp->nr_tag == tag) && (tmp_nrp->nr_btvp == btvp)) {
 277                         nrp->nr_tag = 0;
 278                         tmp_nrp->nr_nodecnt += nodecnt;
 279                         lck_mtx_unlock(&nr_mutex);
 280                         return;
 281                 }
 282         }
 283
 284         nrp->nr_nodecnt = nodecnt;
 285         nrp->nr_newnodes = 0;
 286         nrp->nr_btvp = btvp;
 287         nrp->nr_tag = tag;
 288         LIST_INSERT_HEAD(nrhead, nrp, nr_hash);
 289         ++nrinserts;
 290         lck_mtx_unlock(&nr_mutex);
 291 }
 292
 293 /*
 294  * Delete a node reserve.
 295  */
 296 static void
 297 nr_delete(struct vnode * btvp, struct nreserve *nrp, int *nodecnt)
 298 {
 299         void * tag = NR_GET_TAG();
 300
 301         lck_mtx_lock(&nr_mutex);
 302         if (nrp->nr_tag) {
 303                 if ((nrp->nr_tag != tag) || (nrp->nr_btvp != btvp))
 304                         panic("hfs: nr_delete: invalid NR (%p)", nrp);
 305                 LIST_REMOVE(nrp, nr_hash);
 306                 *nodecnt = nrp->nr_nodecnt;
 307                 bzero(nrp, sizeof(struct nreserve));
 308                 ++nrdeletes;
 309         } else {
 310                 *nodecnt = 0;
 311         }
 312         lck_mtx_unlock(&nr_mutex);
 313 }
 314
 315
 316 /*
 317  * Update a node reserve for any allocations that occurred.
 318  */
 319 static void
 320 nr_update(struct vnode * btvp, int nodecnt)
 321 {
 322         struct nodereserve *nrhead;
 323         struct nreserve *nrp;
 324         void* tag = NR_GET_TAG();
 325
 326         lck_mtx_lock(&nr_mutex);
 327
 328         nrhead = NR_HASH(btvp, tag);
 329         for (nrp = nrhead->lh_first; nrp; nrp = nrp->nr_hash.le_next) {
 330                 if ((nrp->nr_tag == tag) && (nrp->nr_btvp == btvp)) {
 331                         nrp->nr_newnodes += nodecnt;
 332                         break;
 333                 }
 334         }
 335         lck_mtx_unlock(&nr_mutex);
 336 }