| 1 | /* |
| 2 | * Copyright (c) 2007 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_inode.c,v 1.19 2008/01/11 01:41:33 dillon Exp $ |
| 35 | */ |
| 36 | |
| 37 | #include "hammer.h" |
| 38 | #include <sys/buf.h> |
| 39 | #include <sys/buf2.h> |
| 40 | |
| 41 | /* |
| 42 | * The kernel is not actively referencing this vnode but is still holding |
| 43 | * it cached. |
| 44 | */ |
| 45 | int |
| 46 | hammer_vop_inactive(struct vop_inactive_args *ap) |
| 47 | { |
| 48 | struct hammer_inode *ip = VTOI(ap->a_vp); |
| 49 | |
| 50 | /* |
| 51 | * Degenerate case |
| 52 | */ |
| 53 | if (ip == NULL) { |
| 54 | vrecycle(ap->a_vp); |
| 55 | return(0); |
| 56 | } |
| 57 | |
| 58 | /* |
| 59 | * If the inode no longer has any references we recover its |
| 60 | * in-memory resources immediately. |
| 61 | */ |
| 62 | if (ip->ino_rec.ino_nlinks == 0) |
| 63 | vrecycle(ap->a_vp); |
| 64 | return(0); |
| 65 | } |
| 66 | |
| 67 | /* |
| 68 | * Release the vnode association. This is typically (but not always) |
| 69 | * the last reference on the inode and will flush the inode to the |
| 70 | * buffer cache. |
| 71 | * |
| 72 | * XXX Currently our sync code only runs through inodes with vnode |
| 73 | * associations, so we depend on hammer_rel_inode() to sync any inode |
| 74 | * record data to the block device prior to losing the association. |
| 75 | * Otherwise transactions that the user expected to be distinct by |
| 76 | * doing a manual sync may be merged. |
| 77 | */ |
| 78 | int |
| 79 | hammer_vop_reclaim(struct vop_reclaim_args *ap) |
| 80 | { |
| 81 | struct hammer_inode *ip; |
| 82 | struct vnode *vp; |
| 83 | |
| 84 | vp = ap->a_vp; |
| 85 | |
| 86 | if ((ip = vp->v_data) != NULL) { |
| 87 | vp->v_data = NULL; |
| 88 | ip->vp = NULL; |
| 89 | hammer_rel_inode(ip, 0); |
| 90 | } |
| 91 | return(0); |
| 92 | } |
| 93 | |
| 94 | /* |
| 95 | * Obtain a vnode for the specified inode number. An exclusively locked |
| 96 | * vnode is returned. |
| 97 | */ |
| 98 | int |
| 99 | hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) |
| 100 | { |
| 101 | struct hammer_mount *hmp = (void *)mp->mnt_data; |
| 102 | struct hammer_inode *ip; |
| 103 | int error; |
| 104 | |
| 105 | /* |
| 106 | * Get/allocate the hammer_inode structure. The structure must be |
| 107 | * unlocked while we manipulate the related vnode to avoid a |
| 108 | * deadlock. |
| 109 | */ |
| 110 | ip = hammer_get_inode(hmp, NULL, ino, hmp->asof, 0, &error); |
| 111 | if (ip == NULL) { |
| 112 | *vpp = NULL; |
| 113 | return(error); |
| 114 | } |
| 115 | error = hammer_get_vnode(ip, LK_EXCLUSIVE, vpp); |
| 116 | hammer_rel_inode(ip, 0); |
| 117 | return (error); |
| 118 | } |
| 119 | |
| 120 | /* |
| 121 | * Return a locked vnode for the specified inode. The inode must be |
| 122 | * referenced but NOT LOCKED on entry and will remain referenced on |
| 123 | * return. |
| 124 | */ |
| 125 | int |
| 126 | hammer_get_vnode(struct hammer_inode *ip, int lktype, struct vnode **vpp) |
| 127 | { |
| 128 | struct vnode *vp; |
| 129 | int error = 0; |
| 130 | |
| 131 | for (;;) { |
| 132 | if ((vp = ip->vp) == NULL) { |
| 133 | error = getnewvnode(VT_HAMMER, ip->hmp->mp, vpp, 0, 0); |
| 134 | if (error) |
| 135 | break; |
| 136 | hammer_lock_ex(&ip->lock); |
| 137 | if (ip->vp != NULL) { |
| 138 | hammer_unlock(&ip->lock); |
| 139 | vp->v_type = VBAD; |
| 140 | vx_put(vp); |
| 141 | continue; |
| 142 | } |
| 143 | hammer_ref(&ip->lock); |
| 144 | vp = *vpp; |
| 145 | ip->vp = vp; |
| 146 | vp->v_type = hammer_get_vnode_type( |
| 147 | ip->ino_rec.base.base.obj_type); |
| 148 | |
| 149 | switch(ip->ino_rec.base.base.obj_type) { |
| 150 | case HAMMER_OBJTYPE_CDEV: |
| 151 | case HAMMER_OBJTYPE_BDEV: |
| 152 | vp->v_ops = &ip->hmp->mp->mnt_vn_spec_ops; |
| 153 | addaliasu(vp, ip->ino_data.rmajor, |
| 154 | ip->ino_data.rminor); |
| 155 | break; |
| 156 | case HAMMER_OBJTYPE_FIFO: |
| 157 | vp->v_ops = &ip->hmp->mp->mnt_vn_fifo_ops; |
| 158 | break; |
| 159 | default: |
| 160 | break; |
| 161 | } |
| 162 | if (ip->obj_id == HAMMER_OBJID_ROOT) |
| 163 | vp->v_flag |= VROOT; |
| 164 | |
| 165 | vp->v_data = (void *)ip; |
| 166 | /* vnode locked by getnewvnode() */ |
| 167 | /* make related vnode dirty if inode dirty? */ |
| 168 | hammer_unlock(&ip->lock); |
| 169 | if (vp->v_type == VREG) |
| 170 | vinitvmio(vp, ip->ino_rec.ino_size); |
| 171 | break; |
| 172 | } |
| 173 | |
| 174 | /* |
| 175 | * loop if the vget fails (aka races), or if the vp |
| 176 | * no longer matches ip->vp. |
| 177 | */ |
| 178 | if (vget(vp, LK_EXCLUSIVE) == 0) { |
| 179 | if (vp == ip->vp) |
| 180 | break; |
| 181 | vput(vp); |
| 182 | } |
| 183 | } |
| 184 | *vpp = vp; |
| 185 | return(error); |
| 186 | } |
| 187 | |
| 188 | /* |
| 189 | * Acquire a HAMMER inode. The returned inode is not locked. These functions |
| 190 | * do not attach or detach the related vnode (use hammer_get_vnode() for |
| 191 | * that). |
| 192 | * |
| 193 | * The flags argument is only applied for newly created inodes, and only |
| 194 | * certain flags are inherited. |
| 195 | */ |
| 196 | struct hammer_inode * |
| 197 | hammer_get_inode(struct hammer_mount *hmp, struct hammer_node **cache, |
| 198 | u_int64_t obj_id, hammer_tid_t asof, int flags, int *errorp) |
| 199 | { |
| 200 | struct hammer_inode_info iinfo; |
| 201 | struct hammer_cursor cursor; |
| 202 | struct hammer_inode *ip; |
| 203 | |
| 204 | /* |
| 205 | * Determine if we already have an inode cached. If we do then |
| 206 | * we are golden. |
| 207 | */ |
| 208 | iinfo.obj_id = obj_id; |
| 209 | iinfo.obj_asof = asof; |
| 210 | loop: |
| 211 | ip = hammer_ino_rb_tree_RB_LOOKUP_INFO(&hmp->rb_inos_root, &iinfo); |
| 212 | if (ip) { |
| 213 | hammer_ref(&ip->lock); |
| 214 | *errorp = 0; |
| 215 | return(ip); |
| 216 | } |
| 217 | |
| 218 | ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO); |
| 219 | ++hammer_count_inodes; |
| 220 | ip->obj_id = obj_id; |
| 221 | ip->obj_asof = iinfo.obj_asof; |
| 222 | ip->hmp = hmp; |
| 223 | ip->flags = flags & HAMMER_INODE_RO; |
| 224 | if (hmp->ronly) |
| 225 | ip->flags |= HAMMER_INODE_RO; |
| 226 | RB_INIT(&ip->rec_tree); |
| 227 | |
| 228 | /* |
| 229 | * Locate the on-disk inode. |
| 230 | */ |
| 231 | hammer_init_cursor_hmp(&cursor, cache, hmp); |
| 232 | cursor.key_beg.obj_id = ip->obj_id; |
| 233 | cursor.key_beg.key = 0; |
| 234 | cursor.key_beg.create_tid = iinfo.obj_asof; |
| 235 | cursor.key_beg.delete_tid = 0; |
| 236 | cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE; |
| 237 | cursor.key_beg.obj_type = 0; |
| 238 | cursor.flags = HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_GET_DATA; |
| 239 | |
| 240 | *errorp = hammer_btree_lookup(&cursor); |
| 241 | |
| 242 | /* |
| 243 | * On success the B-Tree lookup will hold the appropriate |
| 244 | * buffer cache buffers and provide a pointer to the requested |
| 245 | * information. Copy the information to the in-memory inode |
| 246 | * and cache the B-Tree node to improve future operations. |
| 247 | */ |
| 248 | if (*errorp == 0) { |
| 249 | ip->ino_rec = cursor.record->inode; |
| 250 | ip->ino_data = cursor.data->inode; |
| 251 | hammer_cache_node(cursor.node, &ip->cache[0]); |
| 252 | if (cache) |
| 253 | hammer_cache_node(cursor.node, cache); |
| 254 | } |
| 255 | |
| 256 | /* |
| 257 | * On success load the inode's record and data and insert the |
| 258 | * inode into the B-Tree. It is possible to race another lookup |
| 259 | * insertion of the same inode so deal with that condition too. |
| 260 | * |
| 261 | * The cursor's locked node interlocks against others creating and |
| 262 | * destroying ip while we were blocked. |
| 263 | */ |
| 264 | if (*errorp == 0) { |
| 265 | hammer_ref(&ip->lock); |
| 266 | if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) { |
| 267 | hammer_uncache_node(&ip->cache[0]); |
| 268 | hammer_uncache_node(&ip->cache[1]); |
| 269 | hammer_unref(&ip->lock); |
| 270 | --hammer_count_inodes; |
| 271 | kfree(ip, M_HAMMER); |
| 272 | hammer_done_cursor(&cursor); |
| 273 | goto loop; |
| 274 | } |
| 275 | ip->flags |= HAMMER_INODE_ONDISK; |
| 276 | } else { |
| 277 | --hammer_count_inodes; |
| 278 | kfree(ip, M_HAMMER); |
| 279 | ip = NULL; |
| 280 | } |
| 281 | hammer_done_cursor(&cursor); |
| 282 | return (ip); |
| 283 | } |
| 284 | |
| 285 | /* |
| 286 | * Create a new filesystem object, returning the inode in *ipp. The |
| 287 | * returned inode will be referenced but not locked. |
| 288 | * |
| 289 | * The inode is created in-memory and will be delay-synchronized to the |
| 290 | * disk. |
| 291 | */ |
| 292 | int |
| 293 | hammer_create_inode(hammer_transaction_t trans, struct vattr *vap, |
| 294 | struct ucred *cred, hammer_inode_t dip, |
| 295 | struct hammer_inode **ipp) |
| 296 | { |
| 297 | hammer_mount_t hmp; |
| 298 | hammer_inode_t ip; |
| 299 | uid_t xuid; |
| 300 | |
| 301 | hmp = trans->hmp; |
| 302 | ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO); |
| 303 | ++hammer_count_inodes; |
| 304 | ip->obj_id = hammer_alloc_tid(trans); |
| 305 | KKASSERT(ip->obj_id != 0); |
| 306 | ip->obj_asof = hmp->asof; |
| 307 | ip->hmp = hmp; |
| 308 | ip->flags = HAMMER_INODE_DDIRTY | HAMMER_INODE_RDIRTY | |
| 309 | HAMMER_INODE_ITIMES; |
| 310 | ip->last_tid = trans->tid; |
| 311 | |
| 312 | RB_INIT(&ip->rec_tree); |
| 313 | |
| 314 | ip->ino_rec.ino_atime = trans->tid; |
| 315 | ip->ino_rec.ino_mtime = trans->tid; |
| 316 | ip->ino_rec.ino_size = 0; |
| 317 | ip->ino_rec.ino_nlinks = 0; |
| 318 | /* XXX */ |
| 319 | ip->ino_rec.base.rec_id = hammer_alloc_recid(trans); |
| 320 | KKASSERT(ip->ino_rec.base.rec_id != 0); |
| 321 | ip->ino_rec.base.base.obj_id = ip->obj_id; |
| 322 | ip->ino_rec.base.base.key = 0; |
| 323 | ip->ino_rec.base.base.create_tid = trans->tid; |
| 324 | ip->ino_rec.base.base.delete_tid = 0; |
| 325 | ip->ino_rec.base.base.rec_type = HAMMER_RECTYPE_INODE; |
| 326 | ip->ino_rec.base.base.obj_type = hammer_get_obj_type(vap->va_type); |
| 327 | |
| 328 | ip->ino_data.version = HAMMER_INODE_DATA_VERSION; |
| 329 | ip->ino_data.mode = vap->va_mode; |
| 330 | ip->ino_data.ctime = trans->tid; |
| 331 | ip->ino_data.parent_obj_id = (dip) ? dip->ino_rec.base.base.obj_id : 0; |
| 332 | |
| 333 | switch(ip->ino_rec.base.base.obj_type) { |
| 334 | case HAMMER_OBJTYPE_CDEV: |
| 335 | case HAMMER_OBJTYPE_BDEV: |
| 336 | ip->ino_data.rmajor = vap->va_rmajor; |
| 337 | ip->ino_data.rminor = vap->va_rminor; |
| 338 | break; |
| 339 | default: |
| 340 | break; |
| 341 | } |
| 342 | |
| 343 | /* |
| 344 | * Calculate default uid/gid and overwrite with information from |
| 345 | * the vap. |
| 346 | */ |
| 347 | xuid = hammer_to_unix_xid(&dip->ino_data.uid); |
| 348 | ip->ino_data.gid = dip->ino_data.gid; |
| 349 | xuid = vop_helper_create_uid(hmp->mp, dip->ino_data.mode, xuid, cred, |
| 350 | &vap->va_mode); |
| 351 | ip->ino_data.mode = vap->va_mode; |
| 352 | |
| 353 | if (vap->va_vaflags & VA_UID_UUID_VALID) |
| 354 | ip->ino_data.uid = vap->va_uid_uuid; |
| 355 | else if (vap->va_uid != (uid_t)VNOVAL) |
| 356 | hammer_guid_to_uuid(&ip->ino_data.uid, xuid); |
| 357 | if (vap->va_vaflags & VA_GID_UUID_VALID) |
| 358 | ip->ino_data.gid = vap->va_gid_uuid; |
| 359 | else if (vap->va_gid != (gid_t)VNOVAL) |
| 360 | hammer_guid_to_uuid(&ip->ino_data.gid, vap->va_gid); |
| 361 | |
| 362 | hammer_ref(&ip->lock); |
| 363 | if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) { |
| 364 | hammer_unref(&ip->lock); |
| 365 | panic("hammer_create_inode: duplicate obj_id %llx", ip->obj_id); |
| 366 | } |
| 367 | *ipp = ip; |
| 368 | return(0); |
| 369 | } |
| 370 | |
| 371 | /* |
| 372 | * Called by hammer_sync_inode(). |
| 373 | */ |
| 374 | static int |
| 375 | hammer_update_inode(hammer_inode_t ip) |
| 376 | { |
| 377 | struct hammer_cursor cursor; |
| 378 | struct hammer_cursor *spike = NULL; |
| 379 | hammer_record_t record; |
| 380 | int error; |
| 381 | hammer_tid_t last_tid; |
| 382 | |
| 383 | /* |
| 384 | * Locate the record on-disk and mark it as deleted. Both the B-Tree |
| 385 | * node and the record must be marked deleted. The record may or |
| 386 | * may not be physically deleted, depending on the retention policy. |
| 387 | * |
| 388 | * If the inode has already been deleted on-disk we have nothing |
| 389 | * to do. |
| 390 | * |
| 391 | * XXX Update the inode record and data in-place if the retention |
| 392 | * policy allows it. |
| 393 | */ |
| 394 | last_tid = ip->last_tid; |
| 395 | retry: |
| 396 | error = 0; |
| 397 | |
| 398 | if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) == |
| 399 | HAMMER_INODE_ONDISK) { |
| 400 | hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp); |
| 401 | cursor.key_beg.obj_id = ip->obj_id; |
| 402 | cursor.key_beg.key = 0; |
| 403 | cursor.key_beg.create_tid = ip->obj_asof; |
| 404 | cursor.key_beg.delete_tid = 0; |
| 405 | cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE; |
| 406 | cursor.key_beg.obj_type = 0; |
| 407 | cursor.flags = HAMMER_CURSOR_GET_RECORD; |
| 408 | |
| 409 | error = hammer_btree_lookup(&cursor); |
| 410 | |
| 411 | if (error == 0) { |
| 412 | error = hammer_ip_delete_record(&cursor, last_tid); |
| 413 | if (error == 0) |
| 414 | ip->flags |= HAMMER_INODE_DELONDISK; |
| 415 | } |
| 416 | hammer_cache_node(cursor.node, &ip->cache[0]); |
| 417 | hammer_done_cursor(&cursor); |
| 418 | } |
| 419 | |
| 420 | /* |
| 421 | * Write out a new record if the in-memory inode is not marked |
| 422 | * as having been deleted. Update our inode statistics if this |
| 423 | * is the first application of the inode on-disk. |
| 424 | * |
| 425 | * If the inode has been deleted permanently, HAMMER_INODE_DELONDISK |
| 426 | * will remain set and prevent further updates. |
| 427 | */ |
| 428 | if (error == 0 && (ip->flags & HAMMER_INODE_DELETED) == 0) { |
| 429 | record = hammer_alloc_mem_record(ip); |
| 430 | record->rec.inode = ip->ino_rec; |
| 431 | record->rec.inode.base.base.create_tid = last_tid; |
| 432 | record->rec.inode.base.data_len = sizeof(ip->ino_data); |
| 433 | record->data = (void *)&ip->ino_data; |
| 434 | error = hammer_ip_sync_record(record, &spike); |
| 435 | record->flags |= HAMMER_RECF_DELETED; |
| 436 | hammer_rel_mem_record(record); |
| 437 | if (error == ENOSPC) { |
| 438 | error = hammer_spike(&spike); |
| 439 | if (error == 0) |
| 440 | goto retry; |
| 441 | } |
| 442 | KKASSERT(spike == NULL); |
| 443 | if (error == 0) { |
| 444 | ip->flags &= ~(HAMMER_INODE_RDIRTY | |
| 445 | HAMMER_INODE_DDIRTY | |
| 446 | HAMMER_INODE_DELONDISK | |
| 447 | HAMMER_INODE_ITIMES); |
| 448 | if ((ip->flags & HAMMER_INODE_ONDISK) == 0) { |
| 449 | hammer_modify_volume(ip->hmp->rootvol); |
| 450 | ++ip->hmp->rootvol->ondisk->vol0_stat_inodes; |
| 451 | ip->flags |= HAMMER_INODE_ONDISK; |
| 452 | } |
| 453 | } |
| 454 | } |
| 455 | return(error); |
| 456 | } |
| 457 | |
| 458 | /* |
| 459 | * Update only the itimes fields. This is done no-historically. The |
| 460 | * record is updated in-place on the disk. |
| 461 | */ |
| 462 | static int |
| 463 | hammer_update_itimes(hammer_inode_t ip) |
| 464 | { |
| 465 | struct hammer_cursor cursor; |
| 466 | struct hammer_inode_record *rec; |
| 467 | int error; |
| 468 | |
| 469 | error = 0; |
| 470 | if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) == |
| 471 | HAMMER_INODE_ONDISK) { |
| 472 | hammer_init_cursor_hmp(&cursor, &ip->cache[0], ip->hmp); |
| 473 | cursor.key_beg.obj_id = ip->obj_id; |
| 474 | cursor.key_beg.key = 0; |
| 475 | cursor.key_beg.create_tid = ip->obj_asof; |
| 476 | cursor.key_beg.delete_tid = 0; |
| 477 | cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE; |
| 478 | cursor.key_beg.obj_type = 0; |
| 479 | cursor.flags = HAMMER_CURSOR_GET_RECORD; |
| 480 | |
| 481 | error = hammer_btree_lookup(&cursor); |
| 482 | |
| 483 | if (error == 0) { |
| 484 | rec = &cursor.record->inode; |
| 485 | hammer_modify_buffer(cursor.record_buffer); |
| 486 | rec->ino_atime = ip->ino_rec.ino_atime; |
| 487 | rec->ino_mtime = ip->ino_rec.ino_mtime; |
| 488 | ip->flags &= ~HAMMER_INODE_ITIMES; |
| 489 | /* XXX recalculate crc */ |
| 490 | } |
| 491 | hammer_cache_node(cursor.node, &ip->cache[0]); |
| 492 | hammer_done_cursor(&cursor); |
| 493 | } |
| 494 | return(error); |
| 495 | } |
| 496 | |
| 497 | /* |
| 498 | * Release a reference on an inode. If asked to flush the last release |
| 499 | * will flush the inode. |
| 500 | */ |
| 501 | void |
| 502 | hammer_rel_inode(struct hammer_inode *ip, int flush) |
| 503 | { |
| 504 | hammer_unref(&ip->lock); |
| 505 | if (flush) |
| 506 | ip->flags |= HAMMER_INODE_FLUSH; |
| 507 | if (ip->lock.refs == 0) { |
| 508 | if (ip->flags & HAMMER_INODE_FLUSH) |
| 509 | hammer_unload_inode(ip, (void *)MNT_WAIT); |
| 510 | else |
| 511 | hammer_unload_inode(ip, (void *)MNT_NOWAIT); |
| 512 | } |
| 513 | } |
| 514 | |
| 515 | /* |
| 516 | * Unload and destroy the specified inode. |
| 517 | * |
| 518 | * (called via RB_SCAN) |
| 519 | */ |
| 520 | int |
| 521 | hammer_unload_inode(struct hammer_inode *ip, void *data) |
| 522 | { |
| 523 | int error; |
| 524 | |
| 525 | KASSERT(ip->lock.refs == 0, |
| 526 | ("hammer_unload_inode: %d refs\n", ip->lock.refs)); |
| 527 | KKASSERT(ip->vp == NULL); |
| 528 | hammer_ref(&ip->lock); |
| 529 | |
| 530 | error = hammer_sync_inode(ip, (int)data, 1); |
| 531 | if (error) |
| 532 | kprintf("hammer_sync_inode failed error %d\n", error); |
| 533 | if (ip->lock.refs == 1) { |
| 534 | KKASSERT(RB_EMPTY(&ip->rec_tree)); |
| 535 | RB_REMOVE(hammer_ino_rb_tree, &ip->hmp->rb_inos_root, ip); |
| 536 | |
| 537 | hammer_uncache_node(&ip->cache[0]); |
| 538 | hammer_uncache_node(&ip->cache[1]); |
| 539 | --hammer_count_inodes; |
| 540 | kfree(ip, M_HAMMER); |
| 541 | } else { |
| 542 | hammer_unref(&ip->lock); |
| 543 | } |
| 544 | return(0); |
| 545 | } |
| 546 | |
| 547 | /* |
| 548 | * A transaction has modified an inode, requiring updates as specified by |
| 549 | * the passed flags. |
| 550 | * |
| 551 | * HAMMER_INODE_RDIRTY: Inode record has been updated |
| 552 | * HAMMER_INODE_DDIRTY: Inode data has been updated |
| 553 | * HAMMER_INODE_DELETED: Inode record/data must be deleted |
| 554 | * HAMMER_INODE_ITIMES: mtime/atime has been updated |
| 555 | * |
| 556 | * last_tid is the TID to use to generate the correct TID when the inode |
| 557 | * is synced to disk. |
| 558 | */ |
| 559 | void |
| 560 | hammer_modify_inode(struct hammer_transaction *trans, |
| 561 | struct hammer_inode *ip, int flags) |
| 562 | { |
| 563 | KKASSERT ((ip->flags & HAMMER_INODE_RO) == 0 || |
| 564 | (HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY| |
| 565 | HAMMER_INODE_DELETED|HAMMER_INODE_ITIMES) == 0); |
| 566 | |
| 567 | if (flags & |
| 568 | (HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|HAMMER_INODE_DELETED)) { |
| 569 | if (hammer_debug_tid) { |
| 570 | kprintf("hammer_modify_inode: %016llx (%08x)\n", |
| 571 | trans->tid, (int)(trans->tid / 1000000000LL)); |
| 572 | } |
| 573 | ip->last_tid = trans->tid; |
| 574 | } |
| 575 | ip->flags |= flags; |
| 576 | } |
| 577 | |
| 578 | /* |
| 579 | * Sync any dirty buffers and records associated with an inode. The |
| 580 | * inode's last_tid field is used as the transaction id for the sync, |
| 581 | * overriding any intermediate TIDs that were used for records. Note |
| 582 | * that the dirty buffer cache buffers do not have any knowledge of |
| 583 | * the transaction id they were modified under. |
| 584 | * |
| 585 | * If we can't sync due to a cluster becoming full the spike structure |
| 586 | * will be filled in and ENOSPC returned. We must return -ENOSPC to |
| 587 | * terminate the RB_SCAN. |
| 588 | */ |
| 589 | static int |
| 590 | hammer_sync_inode_callback(hammer_record_t rec, void *data) |
| 591 | { |
| 592 | struct hammer_cursor **spike = data; |
| 593 | int error; |
| 594 | |
| 595 | hammer_ref(&rec->lock); |
| 596 | error = hammer_ip_sync_record(rec, spike); |
| 597 | hammer_rel_mem_record(rec); |
| 598 | |
| 599 | if (error) { |
| 600 | error = -error; |
| 601 | if (error != -ENOSPC) { |
| 602 | kprintf("hammer_sync_inode_callback: sync failed rec " |
| 603 | "%p, error %d\n", rec, error); |
| 604 | } |
| 605 | } |
| 606 | return(error); |
| 607 | } |
| 608 | |
| 609 | /* |
| 610 | * XXX error handling |
| 611 | */ |
| 612 | int |
| 613 | hammer_sync_inode(hammer_inode_t ip, int waitfor, int handle_delete) |
| 614 | { |
| 615 | struct hammer_transaction trans; |
| 616 | struct hammer_cursor *spike = NULL; |
| 617 | int error; |
| 618 | |
| 619 | if ((ip->flags & HAMMER_INODE_MODMASK) == 0) { |
| 620 | return(0); |
| 621 | } |
| 622 | |
| 623 | hammer_lock_ex(&ip->lock); |
| 624 | |
| 625 | /* |
| 626 | * Use the transaction id of the last operation to sync. |
| 627 | */ |
| 628 | if (ip->last_tid) |
| 629 | hammer_start_transaction_tid(&trans, ip->hmp, ip->last_tid); |
| 630 | else |
| 631 | hammer_start_transaction(&trans, ip->hmp); |
| 632 | |
| 633 | /* |
| 634 | * If the inode has been deleted (nlinks == 0), and the OS no longer |
| 635 | * has any references to it (handle_delete != 0), clean up in-memory |
| 636 | * data. |
| 637 | * |
| 638 | * NOTE: We do not set the RDIRTY flag when updating the delete_tid, |
| 639 | * setting HAMMER_INODE_DELETED takes care of it. |
| 640 | * |
| 641 | * NOTE: Because we may sync records within this new transaction, |
| 642 | * force the inode update later on to use our transaction id or |
| 643 | * the delete_tid of the inode may be less then the create_tid of |
| 644 | * the inode update. XXX shouldn't happen but don't take the chance. |
| 645 | * |
| 646 | * NOTE: The call to hammer_ip_delete_range() cannot return ENOSPC |
| 647 | * so we can pass a NULL spike structure, because no partial data |
| 648 | * deletion can occur (yet). |
| 649 | */ |
| 650 | if (ip->ino_rec.ino_nlinks == 0 && handle_delete && |
| 651 | (ip->flags & HAMMER_INODE_GONE) == 0) { |
| 652 | ip->flags |= HAMMER_INODE_GONE; |
| 653 | if (ip->vp) |
| 654 | vtruncbuf(ip->vp, 0, HAMMER_BUFSIZE); |
| 655 | error = hammer_ip_delete_range_all(&trans, ip); |
| 656 | KKASSERT(RB_EMPTY(&ip->rec_tree)); |
| 657 | ip->ino_rec.base.base.delete_tid = trans.tid; |
| 658 | hammer_modify_inode(&trans, ip, HAMMER_INODE_DELETED); |
| 659 | hammer_modify_volume(ip->hmp->rootvol); |
| 660 | --ip->hmp->rootvol->ondisk->vol0_stat_inodes; |
| 661 | } |
| 662 | |
| 663 | /* |
| 664 | * Sync the buffer cache. |
| 665 | */ |
| 666 | if (ip->vp != NULL) { |
| 667 | error = vfsync(ip->vp, waitfor, 1, NULL, NULL); |
| 668 | if (RB_ROOT(&ip->vp->v_rbdirty_tree) == NULL) |
| 669 | ip->flags &= ~HAMMER_INODE_BUFS; |
| 670 | } else { |
| 671 | error = 0; |
| 672 | } |
| 673 | |
| 674 | |
| 675 | /* |
| 676 | * Now sync related records |
| 677 | */ |
| 678 | for (;;) { |
| 679 | error = RB_SCAN(hammer_rec_rb_tree, &ip->rec_tree, NULL, |
| 680 | hammer_sync_inode_callback, &spike); |
| 681 | KKASSERT(error <= 0); |
| 682 | if (error < 0) |
| 683 | error = -error; |
| 684 | if (error == ENOSPC) { |
| 685 | error = hammer_spike(&spike); |
| 686 | if (error == 0) |
| 687 | continue; |
| 688 | } |
| 689 | break; |
| 690 | } |
| 691 | if (RB_EMPTY(&ip->rec_tree)) |
| 692 | ip->flags &= ~HAMMER_INODE_XDIRTY; |
| 693 | |
| 694 | /* |
| 695 | * Now update the inode's on-disk inode-data and/or on-disk record. |
| 696 | */ |
| 697 | switch(ip->flags & (HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK)) { |
| 698 | case HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK: |
| 699 | /* |
| 700 | * If deleted and on-disk, don't set any additional flags. |
| 701 | * the delete flag takes care of things. |
| 702 | */ |
| 703 | break; |
| 704 | case HAMMER_INODE_DELETED: |
| 705 | /* |
| 706 | * Take care of the case where a deleted inode was never |
| 707 | * flushed to the disk in the first place. |
| 708 | */ |
| 709 | ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY| |
| 710 | HAMMER_INODE_XDIRTY|HAMMER_INODE_ITIMES); |
| 711 | while (RB_ROOT(&ip->rec_tree)) { |
| 712 | hammer_record_t rec = RB_ROOT(&ip->rec_tree); |
| 713 | hammer_ref(&rec->lock); |
| 714 | rec->flags |= HAMMER_RECF_DELETED; |
| 715 | hammer_rel_mem_record(rec); |
| 716 | } |
| 717 | break; |
| 718 | case HAMMER_INODE_ONDISK: |
| 719 | /* |
| 720 | * If already on-disk, do not set any additional flags. |
| 721 | */ |
| 722 | break; |
| 723 | default: |
| 724 | /* |
| 725 | * If not on-disk and not deleted, set both dirty flags |
| 726 | * to force an initial record to be written. |
| 727 | */ |
| 728 | ip->flags |= HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY; |
| 729 | break; |
| 730 | } |
| 731 | |
| 732 | /* |
| 733 | * If RDIRTY or DDIRTY is set, write out a new record. If the inode |
| 734 | * is already on-disk the old record is marked as deleted. |
| 735 | * |
| 736 | * If DELETED is set hammer_update_inode() will delete the existing |
| 737 | * record without writing out a new one. |
| 738 | * |
| 739 | * If *ONLY* the ITIMES flag is set we can update the record in-place. |
| 740 | */ |
| 741 | if ((ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY | |
| 742 | HAMMER_INODE_ITIMES | HAMMER_INODE_DELETED)) == |
| 743 | HAMMER_INODE_ITIMES) { |
| 744 | error = hammer_update_itimes(ip); |
| 745 | } else |
| 746 | if (ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY | |
| 747 | HAMMER_INODE_ITIMES | HAMMER_INODE_DELETED)) { |
| 748 | error = hammer_update_inode(ip); |
| 749 | } |
| 750 | hammer_commit_transaction(&trans); |
| 751 | hammer_unlock(&ip->lock); |
| 752 | return(error); |
| 753 | } |
| 754 | |
| 755 | /* |
| 756 | * Access the filesystem buffer containing the cluster-relative byte |
| 757 | * offset, validate the buffer type, load *bufferp and return a |
| 758 | * pointer to the requested data. The buffer is reference and locked on |
| 759 | * return. |
| 760 | * |
| 761 | * If buf_type is 0 the buffer is assumed to be a pure-data buffer and |
| 762 | * no type or crc check is performed. |
| 763 | * |
| 764 | * If *bufferp is not NULL on entry it is assumed to contain a locked |
| 765 | * and referenced buffer which will then be replaced. |
| 766 | * |
| 767 | * If the caller is holding another unrelated buffer locked it must be |
| 768 | * passed in reorderbuf so we can properly order buffer locks. |
| 769 | * |
| 770 | * XXX add a flag for the buffer type and check the CRC here XXX |
| 771 | */ |
| 772 | void * |
| 773 | hammer_bread(hammer_cluster_t cluster, int32_t cloff, |
| 774 | u_int64_t buf_type, int *errorp, |
| 775 | struct hammer_buffer **bufferp) |
| 776 | { |
| 777 | hammer_buffer_t buffer; |
| 778 | int32_t buf_no; |
| 779 | int32_t buf_off; |
| 780 | |
| 781 | /* |
| 782 | * Load the correct filesystem buffer, replacing *bufferp. |
| 783 | */ |
| 784 | buf_no = cloff / HAMMER_BUFSIZE; |
| 785 | buffer = *bufferp; |
| 786 | if (buffer == NULL || buffer->cluster != cluster || |
| 787 | buffer->buf_no != buf_no) { |
| 788 | if (buffer) { |
| 789 | /*hammer_unlock(&buffer->io.lock);*/ |
| 790 | hammer_rel_buffer(buffer, 0); |
| 791 | } |
| 792 | buffer = hammer_get_buffer(cluster, buf_no, 0, errorp); |
| 793 | *bufferp = buffer; |
| 794 | if (buffer == NULL) |
| 795 | return(NULL); |
| 796 | /*hammer_lock_ex(&buffer->io.lock);*/ |
| 797 | } |
| 798 | |
| 799 | /* |
| 800 | * Validate the buffer type |
| 801 | */ |
| 802 | buf_off = cloff & HAMMER_BUFMASK; |
| 803 | if (buf_type) { |
| 804 | if (buf_type != buffer->ondisk->head.buf_type) { |
| 805 | kprintf("BUFFER HEAD TYPE MISMATCH %llx %llx\n", |
| 806 | buf_type, buffer->ondisk->head.buf_type); |
| 807 | *errorp = EIO; |
| 808 | return(NULL); |
| 809 | } |
| 810 | if (buf_off < sizeof(buffer->ondisk->head)) { |
| 811 | kprintf("BUFFER OFFSET TOO LOW %d\n", buf_off); |
| 812 | *errorp = EIO; |
| 813 | return(NULL); |
| 814 | } |
| 815 | } |
| 816 | |
| 817 | /* |
| 818 | * Return a pointer to the buffer data. |
| 819 | */ |
| 820 | *errorp = 0; |
| 821 | return((char *)buffer->ondisk + buf_off); |
| 822 | } |
| 823 | |