1 /* 2 * Copyright (c) 2011-2013 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@dragonflybsd.org> 6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * 3. Neither the name of The DragonFly Project nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific, prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 #include <sys/cdefs.h> 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/types.h> 39 #include <sys/lock.h> 40 #include <sys/uuid.h> 41 42 #include "hammer2.h" 43 44 RB_GENERATE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp, 45 hammer2_tid_t, inum); 46 47 int 48 hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2) 49 { 50 if (ip1->inum < ip2->inum) 51 return(-1); 52 if (ip1->inum > ip2->inum) 53 return(1); 54 return(0); 55 } 56 57 /* 58 * HAMMER2 inode locks 59 * 60 * HAMMER2 offers shared locks and exclusive locks on inodes. 61 * 62 * An inode's ip->chain pointer is resolved and stable while an inode is 63 * locked, and can be cleaned out at any time (become NULL) when an inode 64 * is not locked. 65 * 66 * This function handles duplication races and hardlink replacement races 67 * which can cause ip's cached chain to become stale. 68 * 69 * The underlying chain is also locked and returned. 70 * 71 * NOTE: We don't combine the inode/chain lock because putting away an 72 * inode would otherwise confuse multiple lock holders of the inode. 73 */ 74 hammer2_chain_t * 75 hammer2_inode_lock_ex(hammer2_inode_t *ip) 76 { 77 hammer2_chain_t *chain; 78 hammer2_chain_t *ochain; 79 hammer2_chain_core_t *core; 80 int error; 81 82 hammer2_inode_ref(ip); 83 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE); 84 85 chain = ip->chain; 86 core = chain->core; 87 for (;;) { 88 if (chain->flags & HAMMER2_CHAIN_DUPLICATED) { 89 spin_lock(&core->cst.spin); 90 while (chain->flags & HAMMER2_CHAIN_DUPLICATED) 91 chain = TAILQ_NEXT(chain, core_entry); 92 hammer2_chain_ref(chain); 93 spin_unlock(&core->cst.spin); 94 hammer2_inode_repoint(ip, NULL, chain); 95 hammer2_chain_drop(chain); 96 } 97 hammer2_chain_lock(chain, HAMMER2_RESOLVE_ALWAYS); 98 if ((chain->flags & HAMMER2_CHAIN_DUPLICATED) == 0) 99 break; 100 hammer2_chain_unlock(chain); 101 } 102 if (chain->data->ipdata.type == HAMMER2_OBJTYPE_HARDLINK) { 103 error = hammer2_hardlink_find(ip->pip, &chain, &ochain); 104 hammer2_chain_drop(ochain); 105 KKASSERT(error == 0); 106 /* XXX error handling */ 107 } 108 return (chain); 109 } 110 111 void 112 hammer2_inode_unlock_ex(hammer2_inode_t *ip, hammer2_chain_t *chain) 113 { 114 /* 115 * XXX this will catch parent directories too which we don't 116 * really want. 117 */ 118 if (chain) 119 hammer2_chain_unlock(chain); 120 ccms_thread_unlock(&ip->topo_cst); 121 hammer2_inode_drop(ip); 122 } 123 124 /* 125 * NOTE: We don't combine the inode/chain lock because putting away an 126 * inode would otherwise confuse multiple lock holders of the inode. 127 * 128 * Shared locks are especially sensitive to having too many shared 129 * lock counts (from the same thread) on certain paths which might 130 * need to upgrade them. Only one count of a shared lock can be 131 * upgraded. 132 */ 133 hammer2_chain_t * 134 hammer2_inode_lock_sh(hammer2_inode_t *ip) 135 { 136 hammer2_chain_t *chain; 137 138 hammer2_inode_ref(ip); 139 for (;;) { 140 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_SHARED); 141 142 chain = ip->chain; 143 KKASSERT(chain != NULL); /* for now */ 144 hammer2_chain_lock(chain, HAMMER2_RESOLVE_ALWAYS | 145 HAMMER2_RESOLVE_SHARED); 146 147 /* 148 * Resolve duplication races, resolve hardlinks by giving 149 * up and cycling an exclusive lock. 150 */ 151 if ((chain->flags & HAMMER2_CHAIN_DUPLICATED) == 0 && 152 chain->data->ipdata.type != HAMMER2_OBJTYPE_HARDLINK) { 153 break; 154 } 155 hammer2_chain_unlock(chain); 156 ccms_thread_unlock(&ip->topo_cst); 157 chain = hammer2_inode_lock_ex(ip); 158 hammer2_inode_unlock_ex(ip, chain); 159 } 160 return (chain); 161 } 162 163 void 164 hammer2_inode_unlock_sh(hammer2_inode_t *ip, hammer2_chain_t *chain) 165 { 166 if (chain) 167 hammer2_chain_unlock(chain); 168 ccms_thread_unlock(&ip->topo_cst); 169 hammer2_inode_drop(ip); 170 } 171 172 ccms_state_t 173 hammer2_inode_lock_temp_release(hammer2_inode_t *ip) 174 { 175 return(ccms_thread_lock_temp_release(&ip->topo_cst)); 176 } 177 178 void 179 hammer2_inode_lock_temp_restore(hammer2_inode_t *ip, ccms_state_t ostate) 180 { 181 ccms_thread_lock_temp_restore(&ip->topo_cst, ostate); 182 } 183 184 ccms_state_t 185 hammer2_inode_lock_upgrade(hammer2_inode_t *ip) 186 { 187 return(ccms_thread_lock_upgrade(&ip->topo_cst)); 188 } 189 190 void 191 hammer2_inode_lock_downgrade(hammer2_inode_t *ip, ccms_state_t ostate) 192 { 193 ccms_thread_lock_downgrade(&ip->topo_cst, ostate); 194 } 195 196 /* 197 * Lookup an inode by inode number 198 */ 199 hammer2_inode_t * 200 hammer2_inode_lookup(hammer2_pfsmount_t *pmp, hammer2_tid_t inum) 201 { 202 hammer2_inode_t *ip; 203 204 if (pmp) { 205 spin_lock(&pmp->inum_spin); 206 ip = RB_LOOKUP(hammer2_inode_tree, &pmp->inum_tree, inum); 207 if (ip) 208 hammer2_inode_ref(ip); 209 spin_unlock(&pmp->inum_spin); 210 } else { 211 ip = NULL; 212 } 213 return(ip); 214 } 215 216 /* 217 * Adding a ref to an inode is only legal if the inode already has at least 218 * one ref. 219 */ 220 void 221 hammer2_inode_ref(hammer2_inode_t *ip) 222 { 223 atomic_add_int(&ip->refs, 1); 224 } 225 226 /* 227 * Drop an inode reference, freeing the inode when the last reference goes 228 * away. 229 */ 230 void 231 hammer2_inode_drop(hammer2_inode_t *ip) 232 { 233 hammer2_pfsmount_t *pmp; 234 hammer2_inode_t *pip; 235 u_int refs; 236 237 while (ip) { 238 refs = ip->refs; 239 cpu_ccfence(); 240 if (refs == 1) { 241 /* 242 * Transition to zero, must interlock with 243 * the inode inumber lookup tree (if applicable). 244 * 245 * NOTE: The super-root inode has no pmp. 246 */ 247 pmp = ip->pmp; 248 if (pmp) 249 spin_lock(&pmp->inum_spin); 250 251 if (atomic_cmpset_int(&ip->refs, 1, 0)) { 252 KKASSERT(ip->topo_cst.count == 0); 253 if (ip->flags & HAMMER2_INODE_ONRBTREE) { 254 atomic_clear_int(&ip->flags, 255 HAMMER2_INODE_ONRBTREE); 256 RB_REMOVE(hammer2_inode_tree, 257 &pmp->inum_tree, ip); 258 } 259 if (pmp) 260 spin_unlock(&pmp->inum_spin); 261 262 pip = ip->pip; 263 ip->pip = NULL; 264 ip->pmp = NULL; 265 266 /* 267 * Cleaning out ip->chain isn't entirely 268 * trivial. 269 */ 270 hammer2_inode_repoint(ip, NULL, NULL); 271 272 /* 273 * We have to drop pip (if non-NULL) to 274 * dispose of our implied reference from 275 * ip->pip. We can simply loop on it. 276 */ 277 if (pmp) { 278 KKASSERT((ip->flags & 279 HAMMER2_INODE_SROOT) == 0); 280 kfree(ip, pmp->minode); 281 atomic_add_long(&pmp->inmem_inodes, -1); 282 } else { 283 KKASSERT(ip->flags & 284 HAMMER2_INODE_SROOT); 285 kfree(ip, M_HAMMER2); 286 } 287 ip = pip; 288 /* continue with pip (can be NULL) */ 289 } else { 290 if (pmp) 291 spin_unlock(&ip->pmp->inum_spin); 292 } 293 } else { 294 /* 295 * Non zero transition 296 */ 297 if (atomic_cmpset_int(&ip->refs, refs, refs - 1)) 298 break; 299 } 300 } 301 } 302 303 /* 304 * Get the vnode associated with the given inode, allocating the vnode if 305 * necessary. The vnode will be returned exclusively locked. 306 * 307 * The caller must lock the inode (shared or exclusive). 308 * 309 * Great care must be taken to avoid deadlocks and vnode acquisition/reclaim 310 * races. 311 */ 312 struct vnode * 313 hammer2_igetv(hammer2_inode_t *ip, int *errorp) 314 { 315 hammer2_inode_data_t *ipdata; 316 hammer2_pfsmount_t *pmp; 317 struct vnode *vp; 318 ccms_state_t ostate; 319 320 pmp = ip->pmp; 321 KKASSERT(pmp != NULL); 322 *errorp = 0; 323 ipdata = &ip->chain->data->ipdata; 324 325 for (;;) { 326 /* 327 * Attempt to reuse an existing vnode assignment. It is 328 * possible to race a reclaim so the vget() may fail. The 329 * inode must be unlocked during the vget() to avoid a 330 * deadlock against a reclaim. 331 */ 332 vp = ip->vp; 333 if (vp) { 334 /* 335 * Inode must be unlocked during the vget() to avoid 336 * possible deadlocks, but leave the ip ref intact. 337 * 338 * vnode is held to prevent destruction during the 339 * vget(). The vget() can still fail if we lost 340 * a reclaim race on the vnode. 341 */ 342 vhold(vp); 343 ostate = hammer2_inode_lock_temp_release(ip); 344 if (vget(vp, LK_EXCLUSIVE)) { 345 vdrop(vp); 346 hammer2_inode_lock_temp_restore(ip, ostate); 347 continue; 348 } 349 hammer2_inode_lock_temp_restore(ip, ostate); 350 vdrop(vp); 351 /* vp still locked and ref from vget */ 352 if (ip->vp != vp) { 353 kprintf("hammer2: igetv race %p/%p\n", 354 ip->vp, vp); 355 vput(vp); 356 continue; 357 } 358 *errorp = 0; 359 break; 360 } 361 362 /* 363 * No vnode exists, allocate a new vnode. Beware of 364 * allocation races. This function will return an 365 * exclusively locked and referenced vnode. 366 */ 367 *errorp = getnewvnode(VT_HAMMER2, pmp->mp, &vp, 0, 0); 368 if (*errorp) { 369 kprintf("hammer2: igetv getnewvnode failed %d\n", 370 *errorp); 371 vp = NULL; 372 break; 373 } 374 375 /* 376 * Lock the inode and check for an allocation race. 377 */ 378 ostate = hammer2_inode_lock_upgrade(ip); 379 if (ip->vp != NULL) { 380 vp->v_type = VBAD; 381 vx_put(vp); 382 hammer2_inode_lock_downgrade(ip, ostate); 383 continue; 384 } 385 386 switch (ipdata->type) { 387 case HAMMER2_OBJTYPE_DIRECTORY: 388 vp->v_type = VDIR; 389 break; 390 case HAMMER2_OBJTYPE_REGFILE: 391 vp->v_type = VREG; 392 vinitvmio(vp, ipdata->size, 393 HAMMER2_LBUFSIZE, 394 (int)ipdata->size & HAMMER2_LBUFMASK); 395 break; 396 case HAMMER2_OBJTYPE_SOFTLINK: 397 /* 398 * XXX for now we are using the generic file_read 399 * and file_write code so we need a buffer cache 400 * association. 401 */ 402 vp->v_type = VLNK; 403 vinitvmio(vp, ipdata->size, 404 HAMMER2_LBUFSIZE, 405 (int)ipdata->size & HAMMER2_LBUFMASK); 406 break; 407 case HAMMER2_OBJTYPE_CDEV: 408 vp->v_type = VCHR; 409 /* fall through */ 410 case HAMMER2_OBJTYPE_BDEV: 411 vp->v_ops = &pmp->mp->mnt_vn_spec_ops; 412 if (ipdata->type != HAMMER2_OBJTYPE_CDEV) 413 vp->v_type = VBLK; 414 addaliasu(vp, ipdata->rmajor, ipdata->rminor); 415 break; 416 case HAMMER2_OBJTYPE_FIFO: 417 vp->v_type = VFIFO; 418 vp->v_ops = &pmp->mp->mnt_vn_fifo_ops; 419 break; 420 default: 421 panic("hammer2: unhandled objtype %d", ipdata->type); 422 break; 423 } 424 425 if (ip == pmp->iroot) 426 vsetflags(vp, VROOT); 427 428 vp->v_data = ip; 429 ip->vp = vp; 430 hammer2_inode_ref(ip); /* vp association */ 431 hammer2_inode_lock_downgrade(ip, ostate); 432 break; 433 } 434 435 /* 436 * Return non-NULL vp and *errorp == 0, or NULL vp and *errorp != 0. 437 */ 438 if (hammer2_debug & 0x0002) { 439 kprintf("igetv vp %p refs 0x%08x aux 0x%08x\n", 440 vp, vp->v_refcnt, vp->v_auxrefs); 441 } 442 return (vp); 443 } 444 445 /* 446 * The passed-in chain must be locked and the returned inode will also be 447 * locked. This routine typically locates or allocates the inode, assigns 448 * ip->chain (adding a ref to chain if necessary), and returns the inode. 449 * 450 * The hammer2_inode structure regulates the interface between the high level 451 * kernel VNOPS API and the filesystem backend (the chains). 452 * 453 * WARNING! This routine sucks up the chain's lock (makes it part of the 454 * inode lock from the point of view of the inode lock API), 455 * so callers need to be careful. 456 * 457 * WARNING! The mount code is allowed to pass dip == NULL for iroot and 458 * is allowed to pass pmp == NULL and dip == NULL for sroot. 459 */ 460 hammer2_inode_t * 461 hammer2_inode_get(hammer2_pfsmount_t *pmp, hammer2_inode_t *dip, 462 hammer2_chain_t *chain) 463 { 464 hammer2_inode_t *nip; 465 466 KKASSERT(chain->bref.type == HAMMER2_BREF_TYPE_INODE); 467 468 /* 469 * Interlocked lookup/ref of the inode. This code is only needed 470 * when looking up inodes with nlinks != 0 (TODO: optimize out 471 * otherwise and test for duplicates). 472 */ 473 again: 474 for (;;) { 475 nip = hammer2_inode_lookup(pmp, chain->data->ipdata.inum); 476 if (nip == NULL) 477 break; 478 ccms_thread_lock(&nip->topo_cst, CCMS_STATE_EXCLUSIVE); 479 if ((nip->flags & HAMMER2_INODE_ONRBTREE) == 0) { /* race */ 480 ccms_thread_unlock(&nip->topo_cst); 481 hammer2_inode_drop(nip); 482 continue; 483 } 484 if (nip->chain != chain) 485 hammer2_inode_repoint(nip, NULL, chain); 486 487 /* 488 * Consolidated nip/nip->chain is locked (chain locked 489 * by caller). 490 */ 491 return nip; 492 } 493 494 /* 495 * We couldn't find the inode number, create a new inode. 496 */ 497 if (pmp) { 498 nip = kmalloc(sizeof(*nip), pmp->minode, M_WAITOK | M_ZERO); 499 atomic_add_long(&pmp->inmem_inodes, 1); 500 hammer2_chain_memory_inc(pmp); 501 hammer2_chain_memory_wakeup(pmp); 502 } else { 503 nip = kmalloc(sizeof(*nip), M_HAMMER2, M_WAITOK | M_ZERO); 504 nip->flags = HAMMER2_INODE_SROOT; 505 } 506 nip->inum = chain->data->ipdata.inum; 507 nip->size = chain->data->ipdata.size; 508 nip->mtime = chain->data->ipdata.mtime; 509 hammer2_inode_repoint(nip, NULL, chain); 510 nip->pip = dip; /* can be NULL */ 511 if (dip) 512 hammer2_inode_ref(dip); /* ref dip for nip->pip */ 513 514 nip->pmp = pmp; 515 516 /* 517 * ref and lock on nip gives it state compatible to after a 518 * hammer2_inode_lock_ex() call. 519 */ 520 nip->refs = 1; 521 ccms_cst_init(&nip->topo_cst, &nip->chain); 522 ccms_thread_lock(&nip->topo_cst, CCMS_STATE_EXCLUSIVE); 523 /* combination of thread lock and chain lock == inode lock */ 524 525 /* 526 * Attempt to add the inode. If it fails we raced another inode 527 * get. Undo all the work and try again. 528 */ 529 if (pmp) { 530 spin_lock(&pmp->inum_spin); 531 if (RB_INSERT(hammer2_inode_tree, &pmp->inum_tree, nip)) { 532 spin_unlock(&pmp->inum_spin); 533 ccms_thread_unlock(&nip->topo_cst); 534 hammer2_inode_drop(nip); 535 goto again; 536 } 537 atomic_set_int(&nip->flags, HAMMER2_INODE_ONRBTREE); 538 spin_unlock(&pmp->inum_spin); 539 } 540 541 return (nip); 542 } 543 544 /* 545 * Create a new inode in the specified directory using the vattr to 546 * figure out the type of inode. 547 * 548 * If no error occurs the new inode with its chain locked is returned in 549 * *nipp, otherwise an error is returned and *nipp is set to NULL. 550 * 551 * If vap and/or cred are NULL the related fields are not set and the 552 * inode type defaults to a directory. This is used when creating PFSs 553 * under the super-root, so the inode number is set to 1 in this case. 554 * 555 * dip is not locked on entry. 556 */ 557 hammer2_inode_t * 558 hammer2_inode_create(hammer2_trans_t *trans, hammer2_inode_t *dip, 559 struct vattr *vap, struct ucred *cred, 560 const uint8_t *name, size_t name_len, 561 hammer2_chain_t **chainp, int *errorp) 562 { 563 hammer2_inode_data_t *dipdata; 564 hammer2_inode_data_t *nipdata; 565 hammer2_chain_t *chain; 566 hammer2_chain_t *parent; 567 hammer2_inode_t *nip; 568 hammer2_key_t key_dummy; 569 hammer2_key_t lhc; 570 int error; 571 uid_t xuid; 572 uuid_t dip_uid; 573 uuid_t dip_gid; 574 uint32_t dip_mode; 575 uint8_t dip_algo; 576 int cache_index = -1; 577 578 lhc = hammer2_dirhash(name, name_len); 579 *errorp = 0; 580 581 /* 582 * Locate the inode or indirect block to create the new 583 * entry in. At the same time check for key collisions 584 * and iterate until we don't get one. 585 * 586 * NOTE: hidden inodes do not have iterators. 587 */ 588 retry: 589 parent = hammer2_inode_lock_ex(dip); 590 dipdata = &dip->chain->data->ipdata; 591 dip_uid = dipdata->uid; 592 dip_gid = dipdata->gid; 593 dip_mode = dipdata->mode; 594 dip_algo = dipdata->comp_algo; 595 596 error = 0; 597 while (error == 0) { 598 chain = hammer2_chain_lookup(&parent, &key_dummy, 599 lhc, lhc, &cache_index, 0); 600 if (chain == NULL) 601 break; 602 if ((lhc & HAMMER2_DIRHASH_VISIBLE) == 0) 603 error = ENOSPC; 604 if ((lhc & HAMMER2_DIRHASH_LOMASK) == HAMMER2_DIRHASH_LOMASK) 605 error = ENOSPC; 606 hammer2_chain_unlock(chain); 607 chain = NULL; 608 ++lhc; 609 } 610 611 if (error == 0) { 612 error = hammer2_chain_create(trans, &parent, &chain, 613 lhc, 0, 614 HAMMER2_BREF_TYPE_INODE, 615 HAMMER2_INODE_BYTES); 616 } 617 618 /* 619 * Cleanup and handle retries. 620 */ 621 if (error == EAGAIN) { 622 hammer2_chain_ref(parent); 623 hammer2_inode_unlock_ex(dip, parent); 624 hammer2_chain_wait(parent); 625 hammer2_chain_drop(parent); 626 goto retry; 627 } 628 hammer2_inode_unlock_ex(dip, parent); 629 630 if (error) { 631 KKASSERT(chain == NULL); 632 *errorp = error; 633 return (NULL); 634 } 635 636 /* 637 * Set up the new inode. 638 * 639 * NOTE: *_get() integrates chain's lock into the inode lock. 640 * 641 * NOTE: Only one new inode can currently be created per 642 * transaction. If the need arises we can adjust 643 * hammer2_trans_init() to allow more. 644 * 645 * NOTE: nipdata will have chain's blockset data. 646 */ 647 chain->data->ipdata.inum = trans->inode_tid; 648 nip = hammer2_inode_get(dip->pmp, dip, chain); 649 nipdata = &chain->data->ipdata; 650 651 if (vap) { 652 KKASSERT(trans->inodes_created == 0); 653 nipdata->type = hammer2_get_obj_type(vap->va_type); 654 nipdata->inum = trans->inode_tid; 655 ++trans->inodes_created; 656 657 switch (nipdata->type) { 658 case HAMMER2_OBJTYPE_CDEV: 659 case HAMMER2_OBJTYPE_BDEV: 660 nipdata->rmajor = vap->va_rmajor; 661 nipdata->rminor = vap->va_rminor; 662 break; 663 default: 664 break; 665 } 666 } else { 667 nipdata->type = HAMMER2_OBJTYPE_DIRECTORY; 668 nipdata->inum = 1; 669 } 670 671 /* Inherit parent's inode compression mode. */ 672 nip->comp_heuristic = 0; 673 nipdata->comp_algo = dip_algo; 674 nipdata->version = HAMMER2_INODE_VERSION_ONE; 675 hammer2_update_time(&nipdata->ctime); 676 nipdata->mtime = nipdata->ctime; 677 if (vap) 678 nipdata->mode = vap->va_mode; 679 nipdata->nlinks = 1; 680 if (vap) { 681 if (dip && dip->pmp) { 682 xuid = hammer2_to_unix_xid(&dip_uid); 683 xuid = vop_helper_create_uid(dip->pmp->mp, 684 dip_mode, 685 xuid, 686 cred, 687 &vap->va_mode); 688 } else { 689 /* super-root has no dip and/or pmp */ 690 xuid = 0; 691 } 692 if (vap->va_vaflags & VA_UID_UUID_VALID) 693 nipdata->uid = vap->va_uid_uuid; 694 else if (vap->va_uid != (uid_t)VNOVAL) 695 hammer2_guid_to_uuid(&nipdata->uid, vap->va_uid); 696 else 697 hammer2_guid_to_uuid(&nipdata->uid, xuid); 698 699 if (vap->va_vaflags & VA_GID_UUID_VALID) 700 nipdata->gid = vap->va_gid_uuid; 701 else if (vap->va_gid != (gid_t)VNOVAL) 702 hammer2_guid_to_uuid(&nipdata->gid, vap->va_gid); 703 else if (dip) 704 nipdata->gid = dip_gid; 705 } 706 707 /* 708 * Regular files and softlinks allow a small amount of data to be 709 * directly embedded in the inode. This flag will be cleared if 710 * the size is extended past the embedded limit. 711 */ 712 if (nipdata->type == HAMMER2_OBJTYPE_REGFILE || 713 nipdata->type == HAMMER2_OBJTYPE_SOFTLINK) { 714 nipdata->op_flags |= HAMMER2_OPFLAG_DIRECTDATA; 715 } 716 717 KKASSERT(name_len < HAMMER2_INODE_MAXNAME); 718 bcopy(name, nipdata->filename, name_len); 719 nipdata->name_key = lhc; 720 nipdata->name_len = name_len; 721 *chainp = chain; 722 723 return (nip); 724 } 725 726 /* 727 * chain may have been moved around by the create. 728 */ 729 void 730 hammer2_chain_refactor(hammer2_chain_t **chainp) 731 { 732 hammer2_chain_t *chain = *chainp; 733 hammer2_chain_core_t *core; 734 735 core = chain->core; 736 while (chain->flags & HAMMER2_CHAIN_DUPLICATED) { 737 spin_lock(&core->cst.spin); 738 chain = TAILQ_NEXT(chain, core_entry); 739 while (chain->flags & HAMMER2_CHAIN_DUPLICATED) 740 chain = TAILQ_NEXT(chain, core_entry); 741 hammer2_chain_ref(chain); 742 spin_unlock(&core->cst.spin); 743 KKASSERT(chain->core == core); 744 745 hammer2_chain_unlock(*chainp); 746 hammer2_chain_lock(chain, HAMMER2_RESOLVE_ALWAYS | 747 HAMMER2_RESOLVE_NOREF); /* eat ref */ 748 *chainp = chain; 749 } 750 } 751 752 /* 753 * Shift *chainp up to the specified directory, change the filename 754 * to "0xINODENUMBER", and adjust the key. The chain becomes the 755 * invisible hardlink target. 756 * 757 * The original *chainp has already been marked deleted. 758 */ 759 static 760 void 761 hammer2_hardlink_shiftup(hammer2_trans_t *trans, hammer2_chain_t **chainp, 762 hammer2_inode_t *dip, int nlinks, int *errorp) 763 { 764 hammer2_inode_data_t *nipdata; 765 hammer2_chain_t *parent; 766 hammer2_chain_t *chain; 767 hammer2_chain_t *xchain; 768 hammer2_key_t key_dummy; 769 hammer2_key_t lhc; 770 hammer2_blockref_t bref; 771 int cache_index = -1; 772 773 chain = *chainp; 774 lhc = chain->data->ipdata.inum; 775 KKASSERT((lhc & HAMMER2_DIRHASH_VISIBLE) == 0); 776 777 /* 778 * Locate the inode or indirect block to create the new 779 * entry in. lhc represents the inode number so there is 780 * no collision iteration. 781 * 782 * There should be no key collisions with invisible inode keys. 783 * 784 * WARNING! Must use inode_lock_ex() on dip to handle a stale 785 * dip->chain cache. 786 */ 787 retry: 788 *errorp = 0; 789 parent = hammer2_inode_lock_ex(dip); 790 /*parent = hammer2_chain_lookup_init(dip->chain, 0);*/ 791 xchain = hammer2_chain_lookup(&parent, &key_dummy, 792 lhc, lhc, &cache_index, 0); 793 if (xchain) { 794 kprintf("X3 chain %p parent %p dip %p dip->chain %p\n", 795 xchain, parent, dip, dip->chain); 796 hammer2_chain_unlock(xchain); 797 xchain = NULL; 798 *errorp = ENOSPC; 799 #if 0 800 Debugger("X3"); 801 #endif 802 } 803 804 /* 805 * Create entry in common parent directory using the seek position 806 * calculated above. 807 * 808 * We must refactor chain because it might have been shifted into 809 * an indirect chain by the create. 810 */ 811 if (*errorp == 0) { 812 KKASSERT(xchain == NULL); 813 #if 0 814 *errorp = hammer2_chain_create(trans, &parent, &xchain, 815 lhc, 0, 816 HAMMER2_BREF_TYPE_INODE,/* n/a */ 817 HAMMER2_INODE_BYTES); /* n/a */ 818 #endif 819 /*XXX this somehow isn't working on chain XXX*/ 820 /*KKASSERT(xxx)*/ 821 } 822 823 /* 824 * Cleanup and handle retries. 825 */ 826 if (*errorp == EAGAIN) { 827 hammer2_chain_ref(parent); 828 /* hammer2_chain_lookup_done(parent); */ 829 hammer2_inode_unlock_ex(dip, parent); 830 hammer2_chain_wait(parent); 831 hammer2_chain_drop(parent); 832 goto retry; 833 } 834 835 /* 836 * Handle the error case 837 */ 838 if (*errorp) { 839 panic("error2"); 840 KKASSERT(xchain == NULL); 841 hammer2_inode_unlock_ex(dip, parent); 842 /*hammer2_chain_lookup_done(parent);*/ 843 return; 844 } 845 846 /* 847 * Use xchain as a placeholder for (lhc). Duplicate chain to the 848 * same target bref as xchain and then delete xchain. The duplication 849 * occurs after xchain in flush order even though xchain is deleted 850 * after the duplication. XXX 851 * 852 * WARNING! Duplications (to a different parent) can cause indirect 853 * blocks to be inserted, refactor xchain. 854 */ 855 bref = chain->bref; 856 bref.key = lhc; /* invisible dir entry key */ 857 bref.keybits = 0; 858 #if 0 859 hammer2_chain_delete(trans, xchain, 0); 860 #endif 861 hammer2_chain_duplicate(trans, &parent, &chain, &bref, 0, 2); 862 #if 0 863 hammer2_chain_refactor(&xchain); 864 /*hammer2_chain_delete(trans, xchain, 0);*/ 865 #endif 866 867 hammer2_inode_unlock_ex(dip, parent); 868 /*hammer2_chain_lookup_done(parent);*/ 869 #if 0 870 hammer2_chain_unlock(xchain); /* no longer needed */ 871 #endif 872 873 /* 874 * chain is now 'live' again.. adjust the filename. 875 * 876 * Directory entries are inodes but this is a hidden hardlink 877 * target. The name isn't used but to ease debugging give it 878 * a name after its inode number. 879 */ 880 hammer2_chain_modify(trans, &chain, 0); 881 nipdata = &chain->data->ipdata; 882 ksnprintf(nipdata->filename, sizeof(nipdata->filename), 883 "0x%016jx", (intmax_t)nipdata->inum); 884 nipdata->name_len = strlen(nipdata->filename); 885 nipdata->name_key = lhc; 886 nipdata->nlinks += nlinks; 887 888 *chainp = chain; 889 } 890 891 /* 892 * Connect the target inode represented by (*chainp) to the media topology 893 * at (dip, name, len). 894 * 895 * If hlink is TRUE this function creates an OBJTYPE_HARDLINK directory 896 * entry instead of connecting (*chainp). 897 * 898 * If hlink is FALSE this function uses chain_duplicate() to make a copy 899 * if (*chainp) in the directory entry. (*chainp) is likely to be deleted 900 * by the caller in this case (e.g. rename). 901 */ 902 int 903 hammer2_inode_connect(hammer2_trans_t *trans, int hlink, 904 hammer2_inode_t *dip, hammer2_chain_t **chainp, 905 const uint8_t *name, size_t name_len) 906 { 907 hammer2_inode_data_t *ipdata; 908 hammer2_chain_t *nchain; 909 hammer2_chain_t *parent; 910 hammer2_chain_t *ochain; 911 hammer2_key_t key_dummy; 912 hammer2_key_t lhc; 913 int cache_index = -1; 914 int error; 915 916 ochain = *chainp; 917 918 /* 919 * Since ochain is either disconnected from the topology or represents 920 * a hardlink terminus which is always a parent of or equal to dip, 921 * we should be able to safely lock dip->chain for our setup. 922 * 923 * WARNING! Must use inode_lock_ex() on dip to handle a stale 924 * dip->chain cache. 925 */ 926 parent = hammer2_inode_lock_ex(dip); 927 /*parent = hammer2_chain_lookup_init(dip->chain, 0);*/ 928 929 lhc = hammer2_dirhash(name, name_len); 930 931 /* 932 * Locate the inode or indirect block to create the new 933 * entry in. At the same time check for key collisions 934 * and iterate until we don't get one. 935 */ 936 error = 0; 937 while (error == 0) { 938 nchain = hammer2_chain_lookup(&parent, &key_dummy, 939 lhc, lhc, &cache_index, 0); 940 if (nchain == NULL) 941 break; 942 if ((lhc & HAMMER2_DIRHASH_LOMASK) == HAMMER2_DIRHASH_LOMASK) 943 error = ENOSPC; 944 hammer2_chain_unlock(nchain); 945 nchain = NULL; 946 ++lhc; 947 } 948 949 if (error == 0) { 950 if (hlink) { 951 /* 952 * Hardlink pointer needed, create totally fresh 953 * directory entry. 954 * 955 * We must refactor ochain because it might have 956 * been shifted into an indirect chain by the 957 * create. 958 */ 959 KKASSERT(nchain == NULL); 960 error = hammer2_chain_create(trans, &parent, &nchain, 961 lhc, 0, 962 HAMMER2_BREF_TYPE_INODE, 963 HAMMER2_INODE_BYTES); 964 hammer2_chain_refactor(&ochain); 965 } else { 966 /* 967 * Reconnect the original chain and rename. Use 968 * chain_duplicate(). The caller will likely delete 969 * or has already deleted the original chain in 970 * this case. 971 * 972 * NOTE: chain_duplicate() generates a new chain 973 * with CHAIN_DELETED cleared (ochain typically 974 * has it set from the file unlink). 975 * 976 * WARNING! Can cause held-over chains to require a 977 * refactor. Fortunately we have none (our 978 * locked chains are passed into and 979 * modified by the call). 980 */ 981 nchain = ochain; 982 ochain = NULL; 983 hammer2_chain_duplicate(trans, NULL, &nchain, NULL, 984 0, 3); 985 error = hammer2_chain_create(trans, &parent, &nchain, 986 lhc, 0, 987 HAMMER2_BREF_TYPE_INODE, 988 HAMMER2_INODE_BYTES); 989 } 990 } 991 992 /* 993 * Unlock stuff. 994 */ 995 KKASSERT(error != EAGAIN); 996 hammer2_inode_unlock_ex(dip, parent); 997 /*hammer2_chain_lookup_done(parent);*/ 998 parent = NULL; 999 1000 /* 1001 * nchain should be NULL on error, leave ochain (== *chainp) alone. 1002 */ 1003 if (error) { 1004 KKASSERT(nchain == NULL); 1005 return (error); 1006 } 1007 1008 /* 1009 * Directory entries are inodes so if the name has changed we have 1010 * to update the inode. 1011 * 1012 * When creating an OBJTYPE_HARDLINK entry remember to unlock the 1013 * chain, the caller will access the hardlink via the actual hardlink 1014 * target file and not the hardlink pointer entry, so we must still 1015 * return ochain. 1016 */ 1017 if (hlink && hammer2_hardlink_enable >= 0) { 1018 /* 1019 * Create the HARDLINK pointer. oip represents the hardlink 1020 * target in this situation. 1021 * 1022 * We will return ochain (the hardlink target). 1023 */ 1024 hammer2_chain_modify(trans, &nchain, 0); 1025 KKASSERT(name_len < HAMMER2_INODE_MAXNAME); 1026 ipdata = &nchain->data->ipdata; 1027 bcopy(name, ipdata->filename, name_len); 1028 ipdata->name_key = lhc; 1029 ipdata->name_len = name_len; 1030 ipdata->target_type = ochain->data->ipdata.type; 1031 ipdata->type = HAMMER2_OBJTYPE_HARDLINK; 1032 ipdata->inum = ochain->data->ipdata.inum; 1033 ipdata->nlinks = 1; 1034 hammer2_chain_unlock(nchain); 1035 nchain = ochain; 1036 ochain = NULL; 1037 } else if (hlink && hammer2_hardlink_enable < 0) { 1038 /* 1039 * Create a snapshot (hardlink fake mode for debugging). 1040 * (ochain already flushed above so we can just copy the 1041 * bref XXX). 1042 * 1043 * Since this is a snapshot we return nchain in the fake 1044 * hardlink case. 1045 */ 1046 hammer2_chain_modify(trans, &nchain, 0); 1047 KKASSERT(name_len < HAMMER2_INODE_MAXNAME); 1048 ipdata = &nchain->data->ipdata; 1049 *ipdata = ochain->data->ipdata; 1050 bcopy(name, ipdata->filename, name_len); 1051 ipdata->name_key = lhc; 1052 ipdata->name_len = name_len; 1053 atomic_clear_int(&nchain->core->flags, 1054 HAMMER2_CORE_COUNTEDBREFS); 1055 kprintf("created fake hardlink %*.*s\n", 1056 (int)name_len, (int)name_len, name); 1057 } else { 1058 /* 1059 * nchain is a duplicate of ochain at the new location. 1060 * We must fixup the name stored in oip. The bref key 1061 * has already been set up. 1062 */ 1063 hammer2_chain_modify(trans, &nchain, 0); 1064 ipdata = &nchain->data->ipdata; 1065 1066 KKASSERT(name_len < HAMMER2_INODE_MAXNAME); 1067 bcopy(name, ipdata->filename, name_len); 1068 ipdata->name_key = lhc; 1069 ipdata->name_len = name_len; 1070 ipdata->nlinks = 1; 1071 } 1072 1073 /* 1074 * We are replacing ochain with nchain, unlock ochain. In the 1075 * case where ochain is left unchanged the code above sets 1076 * nchain to ochain and ochain to NULL, resulting in a NOP here. 1077 */ 1078 if (ochain) 1079 hammer2_chain_unlock(ochain); 1080 *chainp = nchain; 1081 1082 return (0); 1083 } 1084 1085 /* 1086 * Repoint ip->chain to nchain. Caller must hold the inode exclusively 1087 * locked. 1088 * 1089 * ip->chain is set to nchain. The prior chain in ip->chain is dropped 1090 * and nchain is ref'd. 1091 */ 1092 void 1093 hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip, 1094 hammer2_chain_t *nchain) 1095 { 1096 hammer2_chain_t *ochain; 1097 hammer2_inode_t *opip; 1098 1099 /* 1100 * Repoint ip->chain if requested. 1101 */ 1102 ochain = ip->chain; 1103 ip->chain = nchain; 1104 if (nchain) 1105 hammer2_chain_ref(nchain); 1106 if (ochain) 1107 hammer2_chain_drop(ochain); 1108 1109 /* 1110 * Repoint ip->pip if requested (non-NULL pip). 1111 */ 1112 if (pip && ip->pip != pip) { 1113 opip = ip->pip; 1114 hammer2_inode_ref(pip); 1115 ip->pip = pip; 1116 if (opip) 1117 hammer2_inode_drop(opip); 1118 } 1119 } 1120 1121 /* 1122 * Unlink the file from the specified directory inode. The directory inode 1123 * does not need to be locked. 1124 * 1125 * isdir determines whether a directory/non-directory check should be made. 1126 * No check is made if isdir is set to -1. 1127 * 1128 * NOTE! The underlying file can still be active with open descriptors 1129 * or if the chain is being manually held (e.g. for rename). 1130 * 1131 * The caller is responsible for fixing up ip->chain if e.g. a 1132 * rename occurs (see chain_duplicate()). 1133 */ 1134 int 1135 hammer2_unlink_file(hammer2_trans_t *trans, hammer2_inode_t *dip, 1136 const uint8_t *name, size_t name_len, 1137 int isdir, int *hlinkp) 1138 { 1139 hammer2_inode_data_t *ipdata; 1140 hammer2_chain_t *parent; 1141 hammer2_chain_t *ochain; 1142 hammer2_chain_t *chain; 1143 hammer2_chain_t *dparent; 1144 hammer2_chain_t *dchain; 1145 hammer2_key_t key_dummy; 1146 hammer2_key_t key_next; 1147 hammer2_key_t lhc; 1148 int error; 1149 int cache_index = -1; 1150 uint8_t type; 1151 1152 error = 0; 1153 ochain = NULL; 1154 lhc = hammer2_dirhash(name, name_len); 1155 1156 /* 1157 * Search for the filename in the directory 1158 */ 1159 if (hlinkp) 1160 *hlinkp = 0; 1161 parent = hammer2_inode_lock_ex(dip); 1162 chain = hammer2_chain_lookup(&parent, &key_next, 1163 lhc, lhc + HAMMER2_DIRHASH_LOMASK, 1164 &cache_index, 0); 1165 while (chain) { 1166 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && 1167 name_len == chain->data->ipdata.name_len && 1168 bcmp(name, chain->data->ipdata.filename, name_len) == 0) { 1169 break; 1170 } 1171 chain = hammer2_chain_next(&parent, chain, &key_next, 1172 key_next, 1173 lhc + HAMMER2_DIRHASH_LOMASK, 1174 &cache_index, 0); 1175 } 1176 hammer2_inode_unlock_ex(dip, NULL); /* retain parent */ 1177 1178 /* 1179 * Not found or wrong type (isdir < 0 disables the type check). 1180 * If a hardlink pointer, type checks use the hardlink target. 1181 */ 1182 if (chain == NULL) { 1183 error = ENOENT; 1184 goto done; 1185 } 1186 if ((type = chain->data->ipdata.type) == HAMMER2_OBJTYPE_HARDLINK) { 1187 if (hlinkp) 1188 *hlinkp = 1; 1189 type = chain->data->ipdata.target_type; 1190 } 1191 1192 if (type == HAMMER2_OBJTYPE_DIRECTORY && isdir == 0) { 1193 error = ENOTDIR; 1194 goto done; 1195 } 1196 if (type != HAMMER2_OBJTYPE_DIRECTORY && isdir >= 1) { 1197 error = EISDIR; 1198 goto done; 1199 } 1200 1201 /* 1202 * Hardlink must be resolved. We can't hold parent locked while we 1203 * do this or we could deadlock. 1204 * 1205 * On success chain will be adjusted to point at the hardlink target 1206 * and ochain will point to the hardlink pointer in the original 1207 * directory. Otherwise chain remains pointing to the original. 1208 */ 1209 if (chain->data->ipdata.type == HAMMER2_OBJTYPE_HARDLINK) { 1210 hammer2_chain_unlock(parent); 1211 parent = NULL; 1212 error = hammer2_hardlink_find(dip, &chain, &ochain); 1213 } 1214 1215 /* 1216 * If this is a directory the directory must be empty. However, if 1217 * isdir < 0 we are doing a rename and the directory does not have 1218 * to be empty, and if isdir > 1 we are deleting a PFS/snapshot 1219 * and the directory does not have to be empty. 1220 * 1221 * NOTE: We check the full key range here which covers both visible 1222 * and invisible entries. Theoretically there should be no 1223 * invisible (hardlink target) entries if there are no visible 1224 * entries. 1225 */ 1226 if (type == HAMMER2_OBJTYPE_DIRECTORY && isdir == 1) { 1227 dparent = hammer2_chain_lookup_init(chain, 0); 1228 dchain = hammer2_chain_lookup(&dparent, &key_dummy, 1229 0, (hammer2_key_t)-1, 1230 &cache_index, 1231 HAMMER2_LOOKUP_NODATA); 1232 if (dchain) { 1233 hammer2_chain_unlock(dchain); 1234 hammer2_chain_lookup_done(dparent); 1235 error = ENOTEMPTY; 1236 goto done; 1237 } 1238 hammer2_chain_lookup_done(dparent); 1239 dparent = NULL; 1240 /* dchain NULL */ 1241 } 1242 1243 /* 1244 * Ok, we can now unlink the chain. We always decrement nlinks even 1245 * if the entry can be deleted in case someone has the file open and 1246 * does an fstat(). 1247 * 1248 * The chain itself will no longer be in the on-media topology but 1249 * can still be flushed to the media (e.g. if an open descriptor 1250 * remains). When the last vnode/ip ref goes away the chain will 1251 * be marked unmodified, avoiding any further (now unnecesary) I/O. 1252 * 1253 * A non-NULL ochain indicates a hardlink. 1254 */ 1255 if (ochain) { 1256 /* 1257 * Delete the original hardlink pointer. 1258 * 1259 * NOTE: parent from above is NULL when ochain != NULL 1260 * so we can reuse it. 1261 */ 1262 hammer2_chain_lock(ochain, HAMMER2_RESOLVE_ALWAYS); 1263 hammer2_chain_delete(trans, ochain, 0); 1264 hammer2_chain_unlock(ochain); 1265 1266 /* 1267 * Then decrement nlinks on hardlink target, deleting 1268 * the target when nlinks drops to 0. 1269 */ 1270 hammer2_chain_modify(trans, &chain, 0); 1271 --chain->data->ipdata.nlinks; 1272 if (chain->data->ipdata.nlinks == 0) 1273 hammer2_chain_delete(trans, chain, 0); 1274 } else { 1275 /* 1276 * Otherwise this was not a hardlink and we can just 1277 * remove the entry and decrement nlinks. 1278 * 1279 * NOTE: *_get() integrates chain's lock into the inode lock. 1280 */ 1281 hammer2_chain_modify(trans, &chain, 0); 1282 ipdata = &chain->data->ipdata; 1283 --ipdata->nlinks; 1284 hammer2_chain_delete(trans, chain, 0); 1285 } 1286 1287 error = 0; 1288 done: 1289 if (chain) 1290 hammer2_chain_unlock(chain); 1291 if (parent) 1292 hammer2_chain_lookup_done(parent); 1293 if (ochain) 1294 hammer2_chain_drop(ochain); 1295 1296 return error; 1297 } 1298 1299 /* 1300 * Given an exclusively locked inode and chain we consolidate its chain 1301 * for hardlink creation, adding (nlinks) to the file's link count and 1302 * potentially relocating the inode to a directory common to ip->pip and tdip. 1303 * 1304 * Replaces (*chainp) if consolidation occurred, unlocking the old chain 1305 * and returning a new locked chain. 1306 * 1307 * NOTE! This function will also replace ip->chain. 1308 */ 1309 int 1310 hammer2_hardlink_consolidate(hammer2_trans_t *trans, hammer2_inode_t *ip, 1311 hammer2_chain_t **chainp, 1312 hammer2_inode_t *tdip, int nlinks) 1313 { 1314 hammer2_inode_data_t *ipdata; 1315 hammer2_inode_t *fdip; 1316 hammer2_inode_t *cdip; 1317 hammer2_chain_t *chain; 1318 hammer2_chain_t *nchain; 1319 int error; 1320 1321 chain = *chainp; 1322 if (nlinks == 0 && /* no hardlink needed */ 1323 (chain->data->ipdata.name_key & HAMMER2_DIRHASH_VISIBLE)) { 1324 return (0); 1325 } 1326 if (hammer2_hardlink_enable < 0) { /* fake hardlinks */ 1327 return (0); 1328 } 1329 1330 if (hammer2_hardlink_enable == 0) { /* disallow hardlinks */ 1331 hammer2_chain_unlock(chain); 1332 *chainp = NULL; 1333 return (ENOTSUP); 1334 } 1335 1336 /* 1337 * cdip will be returned with a ref, but not locked. 1338 */ 1339 fdip = ip->pip; 1340 cdip = hammer2_inode_common_parent(fdip, tdip); 1341 1342 /* 1343 * If no change in the hardlink's target directory is required and 1344 * this is already a hardlink target, all we need to do is adjust 1345 * the link count. 1346 * 1347 * XXX The common parent is a big wiggly due to duplication from 1348 * renames. Compare the core (RBTREE) pointer instead of the 1349 * ip's. 1350 */ 1351 if (cdip == fdip && 1352 (chain->data->ipdata.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) { 1353 if (nlinks) { 1354 hammer2_chain_modify(trans, &chain, 0); 1355 chain->data->ipdata.nlinks += nlinks; 1356 } 1357 error = 0; 1358 goto done; 1359 } 1360 1361 1362 /* 1363 * chain is the real inode. If it's visible we have to convert it 1364 * to a hardlink pointer. If it is not visible then it is already 1365 * a hardlink target and only needs to be deleted. 1366 */ 1367 KKASSERT((chain->flags & HAMMER2_CHAIN_DELETED) == 0); 1368 KKASSERT(chain->data->ipdata.type != HAMMER2_OBJTYPE_HARDLINK); 1369 if (chain->data->ipdata.name_key & HAMMER2_DIRHASH_VISIBLE) { 1370 /* 1371 * We are going to duplicate chain later, causing its 1372 * media block to be shifted to the duplicate. Even though 1373 * we are delete-duplicating nchain here it might decide not 1374 * to reallocate the block. Set FORCECOW to force it to. 1375 */ 1376 nchain = chain; 1377 hammer2_chain_lock(nchain, HAMMER2_RESOLVE_ALWAYS); 1378 atomic_set_int(&nchain->flags, HAMMER2_CHAIN_FORCECOW); 1379 hammer2_chain_delete_duplicate(trans, &nchain, 1380 HAMMER2_DELDUP_RECORE); 1381 KKASSERT((chain->flags & HAMMER2_CHAIN_DUPLICATED) == 0); 1382 1383 ipdata = &nchain->data->ipdata; 1384 ipdata->target_type = ipdata->type; 1385 ipdata->type = HAMMER2_OBJTYPE_HARDLINK; 1386 ipdata->uflags = 0; 1387 ipdata->rmajor = 0; 1388 ipdata->rminor = 0; 1389 ipdata->ctime = 0; 1390 ipdata->mtime = 0; 1391 ipdata->atime = 0; 1392 ipdata->btime = 0; 1393 bzero(&ipdata->uid, sizeof(ipdata->uid)); 1394 bzero(&ipdata->gid, sizeof(ipdata->gid)); 1395 ipdata->op_flags = HAMMER2_OPFLAG_DIRECTDATA; 1396 ipdata->cap_flags = 0; 1397 ipdata->mode = 0; 1398 ipdata->size = 0; 1399 ipdata->nlinks = 1; 1400 ipdata->iparent = 0; /* XXX */ 1401 ipdata->pfs_type = 0; 1402 ipdata->pfs_inum = 0; 1403 bzero(&ipdata->pfs_clid, sizeof(ipdata->pfs_clid)); 1404 bzero(&ipdata->pfs_fsid, sizeof(ipdata->pfs_fsid)); 1405 ipdata->data_quota = 0; 1406 ipdata->data_count = 0; 1407 ipdata->inode_quota = 0; 1408 ipdata->inode_count = 0; 1409 ipdata->attr_tid = 0; 1410 ipdata->dirent_tid = 0; 1411 bzero(&ipdata->u, sizeof(ipdata->u)); 1412 /* XXX transaction ids */ 1413 } else { 1414 hammer2_chain_delete(trans, chain, 0); 1415 nchain = NULL; 1416 } 1417 1418 /* 1419 * chain represents the hardlink target and is now flagged deleted. 1420 * duplicate it to the parent directory and adjust nlinks. 1421 * 1422 * WARNING! The shiftup() call can cause nchain to be moved into 1423 * an indirect block, and our nchain will wind up pointing 1424 * to the older/original version. 1425 */ 1426 KKASSERT(chain->flags & HAMMER2_CHAIN_DELETED); 1427 hammer2_hardlink_shiftup(trans, &chain, cdip, nlinks, &error); 1428 1429 if (error == 0) 1430 hammer2_inode_repoint(ip, cdip, chain); 1431 1432 /* 1433 * Unlock the original chain last as the lock blocked races against 1434 * the creation of the new hardlink target. 1435 */ 1436 if (nchain) 1437 hammer2_chain_unlock(nchain); 1438 1439 done: 1440 /* 1441 * Cleanup, chain/nchain already dealt with. 1442 */ 1443 *chainp = chain; 1444 hammer2_inode_drop(cdip); 1445 1446 return (error); 1447 } 1448 1449 /* 1450 * If (*ochainp) is non-NULL it points to the forward OBJTYPE_HARDLINK 1451 * inode while (*chainp) points to the resolved (hidden hardlink 1452 * target) inode. In this situation when nlinks is 1 we wish to 1453 * deconsolidate the hardlink, moving it back to the directory that now 1454 * represents the only remaining link. 1455 */ 1456 int 1457 hammer2_hardlink_deconsolidate(hammer2_trans_t *trans, 1458 hammer2_inode_t *dip, 1459 hammer2_chain_t **chainp, 1460 hammer2_chain_t **ochainp) 1461 { 1462 if (*ochainp == NULL) 1463 return (0); 1464 /* XXX */ 1465 return (0); 1466 } 1467 1468 /* 1469 * The caller presents a locked *chainp pointing to a HAMMER2_BREF_TYPE_INODE 1470 * with an obj_type of HAMMER2_OBJTYPE_HARDLINK. This routine will gobble 1471 * the *chainp and return a new locked *chainp representing the file target 1472 * (the original *chainp will be unlocked). 1473 * 1474 * When a match is found the chain representing the original HARDLINK 1475 * will be returned in *ochainp with a ref, but not locked. 1476 * 1477 * When no match is found *chainp is set to NULL and EIO is returned. 1478 * (*ochainp) will still be set to the original chain with a ref but not 1479 * locked. 1480 */ 1481 int 1482 hammer2_hardlink_find(hammer2_inode_t *dip, hammer2_chain_t **chainp, 1483 hammer2_chain_t **ochainp) 1484 { 1485 hammer2_chain_t *chain = *chainp; 1486 hammer2_chain_t *parent; 1487 hammer2_inode_t *ip; 1488 hammer2_inode_t *pip; 1489 hammer2_key_t key_dummy; 1490 hammer2_key_t lhc; 1491 int cache_index = -1; 1492 1493 pip = dip; 1494 hammer2_inode_ref(pip); /* for loop */ 1495 hammer2_chain_ref(chain); /* for (*ochainp) */ 1496 *ochainp = chain; 1497 1498 /* 1499 * Locate the hardlink. pip is referenced and not locked, 1500 * ipp. 1501 * 1502 * chain is reused. 1503 */ 1504 lhc = chain->data->ipdata.inum; 1505 hammer2_chain_unlock(chain); 1506 chain = NULL; 1507 1508 while ((ip = pip) != NULL) { 1509 parent = hammer2_inode_lock_ex(ip); 1510 hammer2_inode_drop(ip); /* loop */ 1511 KKASSERT(parent->bref.type == HAMMER2_BREF_TYPE_INODE); 1512 chain = hammer2_chain_lookup(&parent, &key_dummy, 1513 lhc, lhc, &cache_index, 0); 1514 hammer2_chain_lookup_done(parent); /* discard parent */ 1515 if (chain) 1516 break; 1517 pip = ip->pip; /* safe, ip held locked */ 1518 if (pip) 1519 hammer2_inode_ref(pip); /* loop */ 1520 hammer2_inode_unlock_ex(ip, NULL); 1521 } 1522 1523 /* 1524 * chain is locked, ip is locked. Unlock ip, return the locked 1525 * chain. *ipp is already set w/a ref count and not locked. 1526 * 1527 * (parent is already unlocked). 1528 */ 1529 if (ip) 1530 hammer2_inode_unlock_ex(ip, NULL); 1531 *chainp = chain; 1532 if (chain) { 1533 KKASSERT(chain->bref.type == HAMMER2_BREF_TYPE_INODE); 1534 /* already locked */ 1535 return (0); 1536 } else { 1537 return (EIO); 1538 } 1539 } 1540 1541 /* 1542 * Find the directory common to both fdip and tdip, hold and return 1543 * its inode. 1544 */ 1545 hammer2_inode_t * 1546 hammer2_inode_common_parent(hammer2_inode_t *fdip, hammer2_inode_t *tdip) 1547 { 1548 hammer2_inode_t *scan1; 1549 hammer2_inode_t *scan2; 1550 1551 /* 1552 * We used to have a depth field but it complicated matters too 1553 * much for directory renames. So now its ugly. Check for 1554 * simple cases before giving up and doing it the expensive way. 1555 * 1556 * XXX need a bottom-up topology stability lock 1557 */ 1558 if (fdip == tdip || fdip == tdip->pip) { 1559 hammer2_inode_ref(fdip); 1560 return(fdip); 1561 } 1562 if (fdip->pip == tdip) { 1563 hammer2_inode_ref(tdip); 1564 return(tdip); 1565 } 1566 1567 /* 1568 * XXX not MPSAFE 1569 */ 1570 for (scan1 = fdip; scan1->pmp == fdip->pmp; scan1 = scan1->pip) { 1571 scan2 = tdip; 1572 while (scan2->pmp == tdip->pmp) { 1573 if (scan1 == scan2) { 1574 hammer2_inode_ref(scan1); 1575 return(scan1); 1576 } 1577 scan2 = scan2->pip; 1578 if (scan2 == NULL) 1579 break; 1580 } 1581 } 1582 panic("hammer2_inode_common_parent: no common parent %p %p\n", 1583 fdip, tdip); 1584 /* NOT REACHED */ 1585 return(NULL); 1586 } 1587 1588 /* 1589 * Synchronize the inode's frontend state with the chain state prior 1590 * to any explicit flush of the inode or any strategy write call. 1591 * 1592 * Called with a locked inode. 1593 */ 1594 void 1595 hammer2_inode_fsync(hammer2_trans_t *trans, hammer2_inode_t *ip, 1596 hammer2_chain_t **chainp) 1597 { 1598 hammer2_inode_data_t *ipdata; 1599 hammer2_chain_t *parent; 1600 hammer2_chain_t *chain; 1601 hammer2_key_t lbase; 1602 hammer2_key_t key_next; 1603 int cache_index; 1604 1605 ipdata = &ip->chain->data->ipdata; 1606 1607 if (ip->flags & HAMMER2_INODE_MTIME) { 1608 ipdata = hammer2_chain_modify_ip(trans, ip, chainp, 0); 1609 atomic_clear_int(&ip->flags, HAMMER2_INODE_MTIME); 1610 ipdata->mtime = ip->mtime; 1611 } 1612 if ((ip->flags & HAMMER2_INODE_RESIZED) && ip->size < ipdata->size) { 1613 ipdata = hammer2_chain_modify_ip(trans, ip, chainp, 0); 1614 ipdata->size = ip->size; 1615 atomic_clear_int(&ip->flags, HAMMER2_INODE_RESIZED); 1616 1617 /* 1618 * We must delete any chains beyond the EOF. The chain 1619 * straddling the EOF will be pending in the bioq. 1620 */ 1621 lbase = (ipdata->size + HAMMER2_PBUFMASK64) & 1622 ~HAMMER2_PBUFMASK64; 1623 parent = hammer2_chain_lookup_init(ip->chain, 0); 1624 chain = hammer2_chain_lookup(&parent, &key_next, 1625 lbase, (hammer2_key_t)-1, 1626 &cache_index, 1627 HAMMER2_LOOKUP_NODATA); 1628 while (chain) { 1629 /* 1630 * Degenerate embedded case, nothing to loop on 1631 */ 1632 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) { 1633 hammer2_chain_unlock(chain); 1634 break; 1635 } 1636 if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) { 1637 hammer2_chain_delete(trans, chain, 0); 1638 } 1639 chain = hammer2_chain_next(&parent, chain, &key_next, 1640 key_next, (hammer2_key_t)-1, 1641 &cache_index, 1642 HAMMER2_LOOKUP_NODATA); 1643 } 1644 hammer2_chain_lookup_done(parent); 1645 } else 1646 if ((ip->flags & HAMMER2_INODE_RESIZED) && ip->size > ipdata->size) { 1647 ipdata = hammer2_chain_modify_ip(trans, ip, chainp, 0); 1648 ipdata->size = ip->size; 1649 atomic_clear_int(&ip->flags, HAMMER2_INODE_RESIZED); 1650 1651 /* 1652 * When resizing larger we may not have any direct-data 1653 * available. 1654 */ 1655 if ((ipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) && 1656 ip->size > HAMMER2_EMBEDDED_BYTES) { 1657 ipdata->op_flags &= ~HAMMER2_OPFLAG_DIRECTDATA; 1658 bzero(&ipdata->u.blockset, sizeof(ipdata->u.blockset)); 1659 } 1660 } 1661 } 1662