1 /* 2 * Copyright (c) 2011-2018 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 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression) 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in 17 * the documentation and/or other materials provided with the 18 * distribution. 19 * 3. Neither the name of The DragonFly Project nor the names of its 20 * contributors may be used to endorse or promote products derived 21 * from this software without specific, prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 /* 37 * Kernel Filesystem interface 38 * 39 * NOTE! local ipdata pointers must be reloaded on any modifying operation 40 * to the inode as its underlying chain may have changed. 41 */ 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/fcntl.h> 47 #include <sys/buf.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/vnode.h> 51 #include <sys/mountctl.h> 52 #include <sys/dirent.h> 53 #include <sys/uio.h> 54 #include <sys/objcache.h> 55 #include <sys/event.h> 56 #include <sys/file.h> 57 #include <vfs/fifofs/fifo.h> 58 59 #include "hammer2.h" 60 61 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, 62 int seqcount); 63 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio, 64 int ioflag, int seqcount); 65 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize); 66 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize); 67 68 struct objcache *cache_xops; 69 70 static __inline 71 void 72 hammer2_knote(struct vnode *vp, int flags) 73 { 74 if (flags) 75 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 76 } 77 78 /* 79 * Last reference to a vnode is going away but it is still cached. 80 */ 81 static 82 int 83 hammer2_vop_inactive(struct vop_inactive_args *ap) 84 { 85 hammer2_inode_t *ip; 86 struct vnode *vp; 87 88 vp = ap->a_vp; 89 ip = VTOI(vp); 90 91 /* 92 * Degenerate case 93 */ 94 if (ip == NULL) { 95 vrecycle(vp); 96 return (0); 97 } 98 99 /* 100 * Check for deleted inodes and recycle immediately on the last 101 * release. Be sure to destroy any left-over buffer cache buffers 102 * so we do not waste time trying to flush them. 103 * 104 * Note that deleting the file block chains under the inode chain 105 * would just be a waste of energy, so don't do it. 106 * 107 * WARNING: nvtruncbuf() can only be safely called without the inode 108 * lock held due to the way our write thread works. 109 */ 110 if (ip->flags & HAMMER2_INODE_ISUNLINKED) { 111 hammer2_key_t lbase; 112 int nblksize; 113 114 /* 115 * Detect updates to the embedded data which may be 116 * synchronized by the strategy code. Simply mark the 117 * inode modified so it gets picked up by our normal flush. 118 */ 119 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL); 120 nvtruncbuf(vp, 0, nblksize, 0, 0); 121 vrecycle(vp); 122 } 123 return (0); 124 } 125 126 /* 127 * Reclaim a vnode so that it can be reused; after the inode is 128 * disassociated, the filesystem must manage it alone. 129 */ 130 static 131 int 132 hammer2_vop_reclaim(struct vop_reclaim_args *ap) 133 { 134 hammer2_inode_t *ip; 135 hammer2_pfs_t *pmp; 136 struct vnode *vp; 137 138 vp = ap->a_vp; 139 ip = VTOI(vp); 140 if (ip == NULL) { 141 return(0); 142 } 143 pmp = ip->pmp; 144 145 /* 146 * The final close of a deleted file or directory marks it for 147 * destruction. The DELETED flag allows the flusher to shortcut 148 * any modified blocks still unflushed (that is, just ignore them). 149 * 150 * HAMMER2 usually does not try to optimize the freemap by returning 151 * deleted blocks to it as it does not usually know how many snapshots 152 * might be referencing portions of the file/dir. 153 */ 154 vp->v_data = NULL; 155 ip->vp = NULL; 156 157 /* 158 * NOTE! We do not attempt to flush chains here, flushing is 159 * really fragile and could also deadlock. 160 */ 161 vclrisdirty(vp); 162 163 /* 164 * Modified inodes will already be on SIDEQ or SYNCQ. However, 165 * unlinked-but-open inodes may already have been synced and might 166 * still require deletion-on-reclaim. 167 */ 168 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED | 169 HAMMER2_INODE_DELETING)) == 170 HAMMER2_INODE_ISUNLINKED) { 171 hammer2_inode_lock(ip, 0); 172 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED | 173 HAMMER2_INODE_DELETING)) == 174 HAMMER2_INODE_ISUNLINKED) { 175 atomic_set_int(&ip->flags, HAMMER2_INODE_DELETING); 176 hammer2_inode_delayed_sideq(ip); 177 } 178 hammer2_inode_unlock(ip); 179 } 180 181 /* 182 * Modified inodes will already be on SIDEQ or SYNCQ, no further 183 * action is needed. 184 * 185 * We cannot safely synchronize the inode from inside the reclaim 186 * due to potentially deep locks held as-of when the reclaim occurs. 187 * Interactions and potential deadlocks abound. We also can't do it 188 * here without desynchronizing from the related directory entrie(s). 189 */ 190 hammer2_inode_drop(ip); /* vp ref */ 191 192 /* 193 * XXX handle background sync when ip dirty, kernel will no longer 194 * notify us regarding this inode because there is no longer a 195 * vnode attached to it. 196 */ 197 198 return (0); 199 } 200 201 /* 202 * Currently this function synchronizes the front-end inode state to the 203 * backend chain topology, then flushes the inode's chain and sub-topology 204 * to backend media. This function does not flush the root topology down to 205 * the inode. 206 */ 207 static 208 int 209 hammer2_vop_fsync(struct vop_fsync_args *ap) 210 { 211 hammer2_inode_t *ip; 212 struct vnode *vp; 213 int error1; 214 int error2; 215 216 vp = ap->a_vp; 217 ip = VTOI(vp); 218 error1 = 0; 219 220 hammer2_trans_init(ip->pmp, 0); 221 222 /* 223 * Flush dirty buffers in the file's logical buffer cache. 224 * It is best to wait for the strategy code to commit the 225 * buffers to the device's backing buffer cache before 226 * then trying to flush the inode. 227 * 228 * This should be quick, but certain inode modifications cached 229 * entirely in the hammer2_inode structure may not trigger a 230 * buffer read until the flush so the fsync can wind up also 231 * doing scattered reads. 232 */ 233 vfsync(vp, ap->a_waitfor, 1, NULL, NULL); 234 bio_track_wait(&vp->v_track_write, 0, 0); 235 236 /* 237 * Flush any inode changes 238 */ 239 hammer2_inode_lock(ip, 0); 240 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MODIFIED)) 241 error1 = hammer2_inode_chain_sync(ip); 242 243 /* 244 * Flush dirty chains related to the inode. 245 * 246 * NOTE! We are not in a flush transaction. The inode remains on 247 * the sideq so the filesystem syncer can synchronize it to 248 * the volume root. 249 */ 250 error2 = hammer2_inode_chain_flush(ip, HAMMER2_XOP_INODE_STOP); 251 if (error2) 252 error1 = error2; 253 254 /* 255 * We may be able to clear the vnode dirty flag. The 256 * hammer2_pfs_moderate() code depends on this usually working. 257 */ 258 if ((ip->flags & (HAMMER2_INODE_MODIFIED | 259 HAMMER2_INODE_RESIZED | 260 HAMMER2_INODE_DIRTYDATA)) == 0 && 261 RB_EMPTY(&vp->v_rbdirty_tree) && 262 !bio_track_active(&vp->v_track_write)) { 263 vclrisdirty(vp); 264 } 265 hammer2_inode_unlock(ip); 266 hammer2_trans_done(ip->pmp, 0); 267 268 return (error1); 269 } 270 271 /* 272 * No lock needed, just handle ip->update 273 */ 274 static 275 int 276 hammer2_vop_access(struct vop_access_args *ap) 277 { 278 hammer2_inode_t *ip = VTOI(ap->a_vp); 279 uid_t uid; 280 gid_t gid; 281 mode_t mode; 282 uint32_t uflags; 283 int error; 284 int update; 285 286 retry: 287 update = spin_access_start(&ip->cluster_spin); 288 289 /*hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);*/ 290 uid = hammer2_to_unix_xid(&ip->meta.uid); 291 gid = hammer2_to_unix_xid(&ip->meta.gid); 292 mode = ip->meta.mode; 293 uflags = ip->meta.uflags; 294 /*hammer2_inode_unlock(ip);*/ 295 296 if (__predict_false(spin_access_end(&ip->cluster_spin, update))) 297 goto retry; 298 299 error = vop_helper_access(ap, uid, gid, mode, uflags); 300 301 return (error); 302 } 303 304 static 305 int 306 hammer2_vop_getattr(struct vop_getattr_args *ap) 307 { 308 hammer2_pfs_t *pmp; 309 hammer2_inode_t *ip; 310 struct vnode *vp; 311 struct vattr *vap; 312 hammer2_chain_t *chain; 313 int update; 314 int i; 315 316 vp = ap->a_vp; 317 vap = ap->a_vap; 318 319 ip = VTOI(vp); 320 pmp = ip->pmp; 321 322 retry: 323 update = spin_access_start(&ip->cluster_spin); 324 325 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0]; 326 vap->va_fileid = ip->meta.inum; 327 vap->va_mode = ip->meta.mode; 328 vap->va_nlink = ip->meta.nlinks; 329 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid); 330 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid); 331 vap->va_rmajor = 0; 332 vap->va_rminor = 0; 333 vap->va_size = ip->meta.size; /* protected by shared lock */ 334 vap->va_blocksize = HAMMER2_PBUFSIZE; 335 vap->va_flags = ip->meta.uflags; 336 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime); 337 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime); 338 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime); 339 vap->va_gen = 1; 340 vap->va_bytes = 0; 341 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) { 342 /* 343 * Can't really calculate directory use sans the files under 344 * it, just assume one block for now. 345 */ 346 vap->va_bytes += HAMMER2_INODE_BYTES; 347 } else { 348 for (i = 0; i < ip->cluster.nchains; ++i) { 349 if ((chain = ip->cluster.array[i].chain) != NULL) { 350 if (vap->va_bytes < 351 chain->bref.embed.stats.data_count) { 352 vap->va_bytes = 353 chain->bref.embed.stats.data_count; 354 } 355 } 356 } 357 } 358 vap->va_type = hammer2_get_vtype(ip->meta.type); 359 vap->va_filerev = 0; 360 vap->va_uid_uuid = ip->meta.uid; 361 vap->va_gid_uuid = ip->meta.gid; 362 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID | 363 VA_FSID_UUID_VALID; 364 365 if (__predict_false(spin_access_end(&ip->cluster_spin, update))) 366 goto retry; 367 368 return (0); 369 } 370 371 static 372 int 373 hammer2_vop_getattr_quick(struct vop_getattr_args *ap) 374 { 375 hammer2_pfs_t *pmp; 376 hammer2_inode_t *ip; 377 struct vnode *vp; 378 struct vattr *vap; 379 int update; 380 381 vp = ap->a_vp; 382 vap = ap->a_vap; 383 384 ip = VTOI(vp); 385 pmp = ip->pmp; 386 387 retry: 388 update = spin_access_start(&ip->cluster_spin); 389 390 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0]; 391 vap->va_fileid = ip->meta.inum; 392 vap->va_mode = ip->meta.mode; 393 vap->va_nlink = ip->meta.nlinks; 394 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid); 395 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid); 396 vap->va_rmajor = 0; 397 vap->va_rminor = 0; 398 vap->va_size = -1; 399 vap->va_blocksize = HAMMER2_PBUFSIZE; 400 vap->va_flags = ip->meta.uflags; 401 vap->va_type = hammer2_get_vtype(ip->meta.type); 402 vap->va_filerev = 0; 403 vap->va_uid_uuid = ip->meta.uid; 404 vap->va_gid_uuid = ip->meta.gid; 405 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID | 406 VA_FSID_UUID_VALID; 407 408 if (__predict_false(spin_access_end(&ip->cluster_spin, update))) 409 goto retry; 410 411 return (0); 412 } 413 414 static 415 int 416 hammer2_vop_setattr(struct vop_setattr_args *ap) 417 { 418 hammer2_inode_t *ip; 419 struct vnode *vp; 420 struct vattr *vap; 421 int error; 422 int kflags = 0; 423 uint64_t ctime; 424 425 vp = ap->a_vp; 426 vap = ap->a_vap; 427 hammer2_update_time(&ctime); 428 429 ip = VTOI(vp); 430 431 if (ip->pmp->ronly) 432 return (EROFS); 433 434 /* 435 * Normally disallow setattr if there is no space, unless we 436 * are in emergency mode (might be needed to chflags -R noschg 437 * files prior to removal). 438 */ 439 if ((ip->pmp->flags & HAMMER2_PMPF_EMERG) == 0 && 440 hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1) { 441 return (ENOSPC); 442 } 443 444 hammer2_trans_init(ip->pmp, 0); 445 hammer2_inode_lock(ip, 0); 446 error = 0; 447 448 if (vap->va_flags != VNOVAL) { 449 uint32_t flags; 450 451 flags = ip->meta.uflags; 452 error = vop_helper_setattr_flags(&flags, vap->va_flags, 453 hammer2_to_unix_xid(&ip->meta.uid), 454 ap->a_cred); 455 if (error == 0) { 456 if (ip->meta.uflags != flags) { 457 hammer2_inode_modify(ip); 458 spin_lock_update(&ip->cluster_spin); 459 ip->meta.uflags = flags; 460 ip->meta.ctime = ctime; 461 spin_unlock_update(&ip->cluster_spin); 462 kflags |= NOTE_ATTRIB; 463 } 464 if (ip->meta.uflags & (IMMUTABLE | APPEND)) { 465 error = 0; 466 goto done; 467 } 468 } 469 goto done; 470 } 471 if (ip->meta.uflags & (IMMUTABLE | APPEND)) { 472 error = EPERM; 473 goto done; 474 } 475 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 476 mode_t cur_mode = ip->meta.mode; 477 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid); 478 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid); 479 uuid_t uuid_uid; 480 uuid_t uuid_gid; 481 482 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid, 483 ap->a_cred, 484 &cur_uid, &cur_gid, &cur_mode); 485 if (error == 0) { 486 hammer2_guid_to_uuid(&uuid_uid, cur_uid); 487 hammer2_guid_to_uuid(&uuid_gid, cur_gid); 488 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) || 489 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) || 490 ip->meta.mode != cur_mode 491 ) { 492 hammer2_inode_modify(ip); 493 spin_lock_update(&ip->cluster_spin); 494 ip->meta.uid = uuid_uid; 495 ip->meta.gid = uuid_gid; 496 ip->meta.mode = cur_mode; 497 ip->meta.ctime = ctime; 498 spin_unlock_update(&ip->cluster_spin); 499 } 500 kflags |= NOTE_ATTRIB; 501 } 502 } 503 504 /* 505 * Resize the file 506 */ 507 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) { 508 switch(vp->v_type) { 509 case VREG: 510 if (vap->va_size == ip->meta.size) 511 break; 512 if (vap->va_size < ip->meta.size) { 513 hammer2_mtx_ex(&ip->truncate_lock); 514 hammer2_truncate_file(ip, vap->va_size); 515 hammer2_mtx_unlock(&ip->truncate_lock); 516 kflags |= NOTE_WRITE; 517 } else { 518 hammer2_extend_file(ip, vap->va_size); 519 kflags |= NOTE_WRITE | NOTE_EXTEND; 520 } 521 hammer2_inode_modify(ip); 522 ip->meta.mtime = ctime; 523 vclrflags(vp, VLASTWRITETS); 524 break; 525 default: 526 error = EINVAL; 527 goto done; 528 } 529 } 530 #if 0 531 /* atime not supported */ 532 if (vap->va_atime.tv_sec != VNOVAL) { 533 hammer2_inode_modify(ip); 534 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime); 535 kflags |= NOTE_ATTRIB; 536 } 537 #endif 538 if (vap->va_mode != (mode_t)VNOVAL) { 539 mode_t cur_mode = ip->meta.mode; 540 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid); 541 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid); 542 543 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred, 544 cur_uid, cur_gid, &cur_mode); 545 if (error == 0 && ip->meta.mode != cur_mode) { 546 hammer2_inode_modify(ip); 547 spin_lock_update(&ip->cluster_spin); 548 ip->meta.mode = cur_mode; 549 ip->meta.ctime = ctime; 550 spin_unlock_update(&ip->cluster_spin); 551 kflags |= NOTE_ATTRIB; 552 } 553 } 554 555 if (vap->va_mtime.tv_sec != VNOVAL) { 556 hammer2_inode_modify(ip); 557 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime); 558 kflags |= NOTE_ATTRIB; 559 vclrflags(vp, VLASTWRITETS); 560 } 561 562 done: 563 /* 564 * If a truncation occurred we must call chain_sync() now in order 565 * to trim the related data chains, otherwise a later expansion can 566 * cause havoc. 567 * 568 * If an extend occured that changed the DIRECTDATA state, we must 569 * call inode_fsync now in order to prepare the inode's indirect 570 * block table. 571 * 572 * WARNING! This means we are making an adjustment to the inode's 573 * chain outside of sync/fsync, and not just to inode->meta, which 574 * may result in some consistency issues if a crash were to occur 575 * at just the wrong time. 576 */ 577 if (ip->flags & HAMMER2_INODE_RESIZED) 578 hammer2_inode_chain_sync(ip); 579 580 /* 581 * Cleanup. 582 */ 583 hammer2_inode_unlock(ip); 584 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_SIDEQ); 585 hammer2_knote(ip->vp, kflags); 586 587 return (error); 588 } 589 590 static 591 int 592 hammer2_vop_readdir(struct vop_readdir_args *ap) 593 { 594 hammer2_xop_readdir_t *xop; 595 hammer2_blockref_t bref; 596 hammer2_inode_t *ip; 597 hammer2_tid_t inum; 598 hammer2_key_t lkey; 599 struct uio *uio; 600 off_t *cookies; 601 off_t saveoff; 602 int cookie_index; 603 int ncookies; 604 int error; 605 int eofflag; 606 int r; 607 608 ip = VTOI(ap->a_vp); 609 uio = ap->a_uio; 610 saveoff = uio->uio_offset; 611 eofflag = 0; 612 error = 0; 613 614 /* 615 * Setup cookies directory entry cookies if requested 616 */ 617 if (ap->a_ncookies) { 618 ncookies = uio->uio_resid / 16 + 1; 619 if (ncookies > 1024) 620 ncookies = 1024; 621 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 622 } else { 623 ncookies = -1; 624 cookies = NULL; 625 } 626 cookie_index = 0; 627 628 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 629 630 /* 631 * Handle artificial entries. To ensure that only positive 64 bit 632 * quantities are returned to userland we always strip off bit 63. 633 * The hash code is designed such that codes 0x0000-0x7FFF are not 634 * used, allowing us to use these codes for articial entries. 635 * 636 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not 637 * allow '..' to cross the mount point into (e.g.) the super-root. 638 */ 639 if (saveoff == 0) { 640 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK; 641 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, "."); 642 if (r) 643 goto done; 644 if (cookies) 645 cookies[cookie_index] = saveoff; 646 ++saveoff; 647 ++cookie_index; 648 if (cookie_index == ncookies) 649 goto done; 650 } 651 652 if (saveoff == 1) { 653 /* 654 * Be careful with lockorder when accessing ".." 655 * 656 * (ip is the current dir. xip is the parent dir). 657 */ 658 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK; 659 if (ip != ip->pmp->iroot) 660 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK; 661 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, ".."); 662 if (r) 663 goto done; 664 if (cookies) 665 cookies[cookie_index] = saveoff; 666 ++saveoff; 667 ++cookie_index; 668 if (cookie_index == ncookies) 669 goto done; 670 } 671 672 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE; 673 if (hammer2_debug & 0x0020) 674 kprintf("readdir: lkey %016jx\n", lkey); 675 if (error) 676 goto done; 677 678 /* 679 * Use XOP for cluster scan. 680 * 681 * parent is the inode cluster, already locked for us. Don't 682 * double lock shared locks as this will screw up upgrades. 683 */ 684 xop = hammer2_xop_alloc(ip, 0); 685 xop->lkey = lkey; 686 hammer2_xop_start(&xop->head, &hammer2_readdir_desc); 687 688 for (;;) { 689 const hammer2_inode_data_t *ripdata; 690 const char *dname; 691 int dtype; 692 693 error = hammer2_xop_collect(&xop->head, 0); 694 error = hammer2_error_to_errno(error); 695 if (error) { 696 break; 697 } 698 if (cookie_index == ncookies) 699 break; 700 if (hammer2_debug & 0x0020) 701 kprintf("cluster chain %p %p\n", 702 xop->head.cluster.focus, 703 (xop->head.cluster.focus ? 704 xop->head.cluster.focus->data : (void *)-1)); 705 hammer2_cluster_bref(&xop->head.cluster, &bref); 706 707 if (bref.type == HAMMER2_BREF_TYPE_INODE) { 708 ripdata = &hammer2_xop_gdata(&xop->head)->ipdata; 709 dtype = hammer2_get_dtype(ripdata->meta.type); 710 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 711 r = vop_write_dirent(&error, uio, 712 ripdata->meta.inum & 713 HAMMER2_DIRHASH_USERMSK, 714 dtype, 715 ripdata->meta.name_len, 716 ripdata->filename); 717 hammer2_xop_pdata(&xop->head); 718 if (r) 719 break; 720 if (cookies) 721 cookies[cookie_index] = saveoff; 722 ++cookie_index; 723 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) { 724 uint16_t namlen; 725 726 dtype = hammer2_get_dtype(bref.embed.dirent.type); 727 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 728 namlen = bref.embed.dirent.namlen; 729 if (namlen <= sizeof(bref.check.buf)) { 730 dname = bref.check.buf; 731 } else { 732 dname = hammer2_xop_gdata(&xop->head)->buf; 733 } 734 r = vop_write_dirent(&error, uio, 735 bref.embed.dirent.inum, dtype, 736 namlen, dname); 737 if (namlen > sizeof(bref.check.buf)) 738 hammer2_xop_pdata(&xop->head); 739 if (r) 740 break; 741 if (cookies) 742 cookies[cookie_index] = saveoff; 743 ++cookie_index; 744 } else { 745 /* XXX chain error */ 746 kprintf("bad chain type readdir %d\n", bref.type); 747 } 748 } 749 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 750 if (error == ENOENT) { 751 error = 0; 752 eofflag = 1; 753 saveoff = (hammer2_key_t)-1; 754 } else { 755 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 756 } 757 done: 758 hammer2_inode_unlock(ip); 759 if (ap->a_eofflag) 760 *ap->a_eofflag = eofflag; 761 if (hammer2_debug & 0x0020) 762 kprintf("readdir: done at %016jx\n", saveoff); 763 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE; 764 if (error && cookie_index == 0) { 765 if (cookies) { 766 kfree(cookies, M_TEMP); 767 *ap->a_ncookies = 0; 768 *ap->a_cookies = NULL; 769 } 770 } else { 771 if (cookies) { 772 *ap->a_ncookies = cookie_index; 773 *ap->a_cookies = cookies; 774 } 775 } 776 return (error); 777 } 778 779 /* 780 * hammer2_vop_readlink { vp, uio, cred } 781 */ 782 static 783 int 784 hammer2_vop_readlink(struct vop_readlink_args *ap) 785 { 786 struct vnode *vp; 787 hammer2_inode_t *ip; 788 int error; 789 790 vp = ap->a_vp; 791 if (vp->v_type != VLNK) 792 return (EINVAL); 793 ip = VTOI(vp); 794 795 error = hammer2_read_file(ip, ap->a_uio, 0); 796 return (error); 797 } 798 799 static 800 int 801 hammer2_vop_read(struct vop_read_args *ap) 802 { 803 struct vnode *vp; 804 hammer2_inode_t *ip; 805 struct uio *uio; 806 int error; 807 int seqcount; 808 int bigread; 809 810 /* 811 * Read operations supported on this vnode? 812 */ 813 vp = ap->a_vp; 814 if (vp->v_type != VREG) 815 return (EINVAL); 816 817 /* 818 * Misc 819 */ 820 ip = VTOI(vp); 821 uio = ap->a_uio; 822 error = 0; 823 824 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 825 bigread = (uio->uio_resid > 100 * 1024 * 1024); 826 827 error = hammer2_read_file(ip, uio, seqcount); 828 return (error); 829 } 830 831 static 832 int 833 hammer2_vop_write(struct vop_write_args *ap) 834 { 835 hammer2_inode_t *ip; 836 thread_t td; 837 struct vnode *vp; 838 struct uio *uio; 839 int error; 840 int seqcount; 841 int ioflag; 842 843 /* 844 * Read operations supported on this vnode? 845 */ 846 vp = ap->a_vp; 847 if (vp->v_type != VREG) 848 return (EINVAL); 849 850 /* 851 * Misc 852 */ 853 ip = VTOI(vp); 854 ioflag = ap->a_ioflag; 855 uio = ap->a_uio; 856 error = 0; 857 if (ip->pmp->ronly || (ip->pmp->flags & HAMMER2_PMPF_EMERG)) 858 return (EROFS); 859 switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) { 860 case 2: 861 return (ENOSPC); 862 case 1: 863 ioflag |= IO_DIRECT; /* semi-synchronous */ 864 /* fall through */ 865 default: 866 break; 867 } 868 869 seqcount = ioflag >> IO_SEQSHIFT; 870 871 /* 872 * Check resource limit 873 */ 874 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc && 875 uio->uio_offset + uio->uio_resid > 876 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 877 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ); 878 return (EFBIG); 879 } 880 881 /* 882 * The transaction interlocks against flush initiations 883 * (note: but will run concurrently with the actual flush). 884 * 885 * To avoid deadlocking against the VM system, we must flag any 886 * transaction related to the buffer cache or other direct 887 * VM page manipulation. 888 */ 889 if (uio->uio_segflg == UIO_NOCOPY) { 890 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE); 891 } else { 892 hammer2_trans_init(ip->pmp, 0); 893 } 894 error = hammer2_write_file(ip, uio, ioflag, seqcount); 895 if (uio->uio_segflg == UIO_NOCOPY) 896 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_BUFCACHE | 897 HAMMER2_TRANS_SIDEQ); 898 else 899 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_SIDEQ); 900 901 return (error); 902 } 903 904 /* 905 * Perform read operations on a file or symlink given an UNLOCKED 906 * inode and uio. 907 * 908 * The passed ip is not locked. 909 */ 910 static 911 int 912 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount) 913 { 914 hammer2_off_t size; 915 struct buf *bp; 916 int error; 917 918 error = 0; 919 920 /* 921 * UIO read loop. 922 * 923 * WARNING! Assumes that the kernel interlocks size changes at the 924 * vnode level. 925 */ 926 hammer2_mtx_sh(&ip->lock); 927 hammer2_mtx_sh(&ip->truncate_lock); 928 size = ip->meta.size; 929 hammer2_mtx_unlock(&ip->lock); 930 931 while (uio->uio_resid > 0 && uio->uio_offset < size) { 932 hammer2_key_t lbase; 933 hammer2_key_t leof; 934 int lblksize; 935 int loff; 936 int n; 937 938 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 939 &lbase, &leof); 940 941 #if 1 942 bp = NULL; 943 error = cluster_readx(ip->vp, leof, lbase, lblksize, 944 B_NOTMETA | B_KVABIO, 945 uio->uio_resid, 946 seqcount * MAXBSIZE, 947 &bp); 948 #else 949 if (uio->uio_segflg == UIO_NOCOPY) { 950 bp = getblk(ip->vp, lbase, lblksize, 951 GETBLK_BHEAVY | GETBLK_KVABIO, 0); 952 if (bp->b_flags & B_CACHE) { 953 int i; 954 int j = 0; 955 if (bp->b_xio.xio_npages != 16) 956 kprintf("NPAGES BAD\n"); 957 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 958 vm_page_t m; 959 m = bp->b_xio.xio_pages[i]; 960 if (m == NULL || m->valid == 0) { 961 kprintf("bp %016jx %016jx pg %d inv", 962 lbase, leof, i); 963 if (m) 964 kprintf("m->object %p/%p", m->object, ip->vp->v_object); 965 kprintf("\n"); 966 j = 1; 967 } 968 } 969 if (j) 970 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error); 971 } 972 bqrelse(bp); 973 } 974 error = bread_kvabio(ip->vp, lbase, lblksize, &bp); 975 #endif 976 if (error) { 977 brelse(bp); 978 break; 979 } 980 bkvasync(bp); 981 loff = (int)(uio->uio_offset - lbase); 982 n = lblksize - loff; 983 if (n > uio->uio_resid) 984 n = uio->uio_resid; 985 if (n > size - uio->uio_offset) 986 n = (int)(size - uio->uio_offset); 987 bp->b_flags |= B_AGE; 988 uiomovebp(bp, (char *)bp->b_data + loff, n, uio); 989 bqrelse(bp); 990 } 991 hammer2_mtx_unlock(&ip->truncate_lock); 992 993 return (error); 994 } 995 996 /* 997 * Write to the file represented by the inode via the logical buffer cache. 998 * The inode may represent a regular file or a symlink. 999 * 1000 * The inode must not be locked. 1001 */ 1002 static 1003 int 1004 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio, 1005 int ioflag, int seqcount) 1006 { 1007 hammer2_key_t old_eof; 1008 hammer2_key_t new_eof; 1009 struct buf *bp; 1010 int kflags; 1011 int error; 1012 int modified; 1013 1014 /* 1015 * Setup if append 1016 * 1017 * WARNING! Assumes that the kernel interlocks size changes at the 1018 * vnode level. 1019 */ 1020 hammer2_mtx_ex(&ip->lock); 1021 hammer2_mtx_sh(&ip->truncate_lock); 1022 if (ioflag & IO_APPEND) 1023 uio->uio_offset = ip->meta.size; 1024 old_eof = ip->meta.size; 1025 1026 /* 1027 * Extend the file if necessary. If the write fails at some point 1028 * we will truncate it back down to cover as much as we were able 1029 * to write. 1030 * 1031 * Doing this now makes it easier to calculate buffer sizes in 1032 * the loop. 1033 */ 1034 kflags = 0; 1035 error = 0; 1036 modified = 0; 1037 1038 if (uio->uio_offset + uio->uio_resid > old_eof) { 1039 new_eof = uio->uio_offset + uio->uio_resid; 1040 modified = 1; 1041 hammer2_extend_file(ip, new_eof); 1042 kflags |= NOTE_EXTEND; 1043 } else { 1044 new_eof = old_eof; 1045 } 1046 hammer2_mtx_unlock(&ip->lock); 1047 1048 /* 1049 * UIO write loop 1050 */ 1051 while (uio->uio_resid > 0) { 1052 hammer2_key_t lbase; 1053 int trivial; 1054 int endofblk; 1055 int lblksize; 1056 int loff; 1057 int n; 1058 1059 /* 1060 * Don't allow the buffer build to blow out the buffer 1061 * cache. 1062 */ 1063 if ((ioflag & IO_RECURSE) == 0) 1064 bwillwrite(HAMMER2_PBUFSIZE); 1065 1066 /* 1067 * This nominally tells us how much we can cluster and 1068 * what the logical buffer size needs to be. Currently 1069 * we don't try to cluster the write and just handle one 1070 * block at a time. 1071 */ 1072 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 1073 &lbase, NULL); 1074 loff = (int)(uio->uio_offset - lbase); 1075 1076 KKASSERT(lblksize <= MAXBSIZE); 1077 1078 /* 1079 * Calculate bytes to copy this transfer and whether the 1080 * copy completely covers the buffer or not. 1081 */ 1082 trivial = 0; 1083 n = lblksize - loff; 1084 if (n > uio->uio_resid) { 1085 n = uio->uio_resid; 1086 if (loff == lbase && uio->uio_offset + n == new_eof) 1087 trivial = 1; 1088 endofblk = 0; 1089 } else { 1090 if (loff == 0) 1091 trivial = 1; 1092 endofblk = 1; 1093 } 1094 if (lbase >= new_eof) 1095 trivial = 1; 1096 1097 /* 1098 * Get the buffer 1099 */ 1100 if (uio->uio_segflg == UIO_NOCOPY) { 1101 /* 1102 * Issuing a write with the same data backing the 1103 * buffer. Instantiate the buffer to collect the 1104 * backing vm pages, then read-in any missing bits. 1105 * 1106 * This case is used by vop_stdputpages(). 1107 */ 1108 bp = getblk(ip->vp, lbase, lblksize, 1109 GETBLK_BHEAVY | GETBLK_KVABIO, 0); 1110 if ((bp->b_flags & B_CACHE) == 0) { 1111 bqrelse(bp); 1112 error = bread_kvabio(ip->vp, lbase, 1113 lblksize, &bp); 1114 } 1115 } else if (trivial) { 1116 /* 1117 * Even though we are entirely overwriting the buffer 1118 * we may still have to zero it out to avoid a 1119 * mmap/write visibility issue. 1120 */ 1121 bp = getblk(ip->vp, lbase, lblksize, 1122 GETBLK_BHEAVY | GETBLK_KVABIO, 0); 1123 if ((bp->b_flags & B_CACHE) == 0) 1124 vfs_bio_clrbuf(bp); 1125 } else { 1126 /* 1127 * Partial overwrite, read in any missing bits then 1128 * replace the portion being written. 1129 * 1130 * (The strategy code will detect zero-fill physical 1131 * blocks for this case). 1132 */ 1133 error = bread_kvabio(ip->vp, lbase, lblksize, &bp); 1134 if (error == 0) 1135 bheavy(bp); 1136 } 1137 1138 if (error) { 1139 brelse(bp); 1140 break; 1141 } 1142 1143 /* 1144 * Ok, copy the data in 1145 */ 1146 bkvasync(bp); 1147 error = uiomovebp(bp, bp->b_data + loff, n, uio); 1148 kflags |= NOTE_WRITE; 1149 modified = 1; 1150 if (error) { 1151 brelse(bp); 1152 break; 1153 } 1154 1155 /* 1156 * WARNING: Pageout daemon will issue UIO_NOCOPY writes 1157 * with IO_SYNC or IO_ASYNC set. These writes 1158 * must be handled as the pageout daemon expects. 1159 * 1160 * NOTE! H2 relies on cluster_write() here because it 1161 * cannot preallocate disk blocks at the logical 1162 * level due to not knowing what the compression 1163 * size will be at this time. 1164 * 1165 * We must use cluster_write() here and we depend 1166 * on the write-behind feature to flush buffers 1167 * appropriately. If we let the buffer daemons do 1168 * it the block allocations will be all over the 1169 * map. 1170 */ 1171 if (ioflag & IO_SYNC) { 1172 bwrite(bp); 1173 } else if ((ioflag & IO_DIRECT) && endofblk) { 1174 bawrite(bp); 1175 } else if (ioflag & IO_ASYNC) { 1176 bawrite(bp); 1177 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) { 1178 bdwrite(bp); 1179 } else { 1180 #if 1 1181 bp->b_flags |= B_CLUSTEROK; 1182 cluster_write(bp, new_eof, lblksize, seqcount); 1183 #else 1184 bp->b_flags |= B_CLUSTEROK; 1185 bdwrite(bp); 1186 #endif 1187 } 1188 } 1189 1190 /* 1191 * Cleanup. If we extended the file EOF but failed to write through 1192 * the entire write is a failure and we have to back-up. 1193 */ 1194 if (error && new_eof != old_eof) { 1195 hammer2_mtx_unlock(&ip->truncate_lock); 1196 hammer2_mtx_ex(&ip->lock); /* note lock order */ 1197 hammer2_mtx_ex(&ip->truncate_lock); /* note lock order */ 1198 hammer2_truncate_file(ip, old_eof); 1199 if (ip->flags & HAMMER2_INODE_MODIFIED) 1200 hammer2_inode_chain_sync(ip); 1201 hammer2_mtx_unlock(&ip->lock); 1202 } else if (modified) { 1203 struct vnode *vp = ip->vp; 1204 1205 hammer2_mtx_ex(&ip->lock); 1206 hammer2_inode_modify(ip); 1207 if (uio->uio_segflg == UIO_NOCOPY) { 1208 if (vp->v_flag & VLASTWRITETS) { 1209 ip->meta.mtime = 1210 (unsigned long)vp->v_lastwrite_ts.tv_sec * 1211 1000000 + 1212 vp->v_lastwrite_ts.tv_nsec / 1000; 1213 } 1214 } else { 1215 hammer2_update_time(&ip->meta.mtime); 1216 vclrflags(vp, VLASTWRITETS); 1217 } 1218 1219 #if 0 1220 /* 1221 * REMOVED - handled by hammer2_extend_file(). Do not issue 1222 * a chain_sync() outside of a sync/fsync except for DIRECTDATA 1223 * state changes. 1224 * 1225 * Under normal conditions we only issue a chain_sync if 1226 * the inode's DIRECTDATA state changed. 1227 */ 1228 if (ip->flags & HAMMER2_INODE_RESIZED) 1229 hammer2_inode_chain_sync(ip); 1230 #endif 1231 hammer2_mtx_unlock(&ip->lock); 1232 hammer2_knote(ip->vp, kflags); 1233 } 1234 hammer2_trans_assert_strategy(ip->pmp); 1235 hammer2_mtx_unlock(&ip->truncate_lock); 1236 1237 return error; 1238 } 1239 1240 /* 1241 * Truncate the size of a file. The inode must be locked. 1242 * 1243 * We must unconditionally set HAMMER2_INODE_RESIZED to properly 1244 * ensure that any on-media data beyond the new file EOF has been destroyed. 1245 * 1246 * WARNING: nvtruncbuf() can only be safely called without the inode lock 1247 * held due to the way our write thread works. If the truncation 1248 * occurs in the middle of a buffer, nvtruncbuf() is responsible 1249 * for dirtying that buffer and zeroing out trailing bytes. 1250 * 1251 * WARNING! Assumes that the kernel interlocks size changes at the 1252 * vnode level. 1253 * 1254 * WARNING! Caller assumes responsibility for removing dead blocks 1255 * if INODE_RESIZED is set. 1256 */ 1257 static 1258 void 1259 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1260 { 1261 hammer2_key_t lbase; 1262 int nblksize; 1263 1264 hammer2_mtx_unlock(&ip->lock); 1265 if (ip->vp) { 1266 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1267 nvtruncbuf(ip->vp, nsize, 1268 nblksize, (int)nsize & (nblksize - 1), 1269 0); 1270 } 1271 hammer2_mtx_ex(&ip->lock); 1272 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0); 1273 ip->osize = ip->meta.size; 1274 ip->meta.size = nsize; 1275 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1276 hammer2_inode_modify(ip); 1277 } 1278 1279 /* 1280 * Extend the size of a file. The inode must be locked. 1281 * 1282 * Even though the file size is changing, we do not have to set the 1283 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES 1284 * boundary. When this occurs a hammer2_inode_chain_sync() is required 1285 * to prepare the inode cluster's indirect block table, otherwise 1286 * async execution of the strategy code will implode on us. 1287 * 1288 * WARNING! Assumes that the kernel interlocks size changes at the 1289 * vnode level. 1290 * 1291 * WARNING! Caller assumes responsibility for transitioning out 1292 * of the inode DIRECTDATA mode if INODE_RESIZED is set. 1293 */ 1294 static 1295 void 1296 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1297 { 1298 hammer2_key_t lbase; 1299 hammer2_key_t osize; 1300 int oblksize; 1301 int nblksize; 1302 int error; 1303 1304 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0); 1305 hammer2_inode_modify(ip); 1306 osize = ip->meta.size; 1307 ip->osize = osize; 1308 ip->meta.size = nsize; 1309 1310 /* 1311 * We must issue a chain_sync() when the DIRECTDATA state changes 1312 * to prevent confusion between the flush code and the in-memory 1313 * state. This is not perfect because we are doing it outside of 1314 * a sync/fsync operation, so it might not be fully synchronized 1315 * with the meta-data topology flush. 1316 * 1317 * We must retain and re-dirty the buffer cache buffer containing 1318 * the direct data so it can be written to a real block. It should 1319 * not be possible for a bread error to occur since the original data 1320 * is extracted from the inode structure directly. 1321 */ 1322 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) { 1323 if (osize) { 1324 struct buf *bp; 1325 1326 oblksize = hammer2_calc_logical(ip, 0, NULL, NULL); 1327 error = bread_kvabio(ip->vp, 0, oblksize, &bp); 1328 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1329 hammer2_inode_chain_sync(ip); 1330 if (error == 0) { 1331 bheavy(bp); 1332 bdwrite(bp); 1333 } else { 1334 brelse(bp); 1335 } 1336 } else { 1337 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1338 hammer2_inode_chain_sync(ip); 1339 } 1340 } 1341 hammer2_mtx_unlock(&ip->lock); 1342 if (ip->vp) { 1343 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL); 1344 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1345 nvextendbuf(ip->vp, 1346 osize, nsize, 1347 oblksize, nblksize, 1348 -1, -1, 0); 1349 } 1350 hammer2_mtx_ex(&ip->lock); 1351 } 1352 1353 static 1354 int 1355 hammer2_vop_nresolve(struct vop_nresolve_args *ap) 1356 { 1357 hammer2_xop_nresolve_t *xop; 1358 hammer2_inode_t *ip; 1359 hammer2_inode_t *dip; 1360 struct namecache *ncp; 1361 struct vnode *vp; 1362 int error; 1363 1364 dip = VTOI(ap->a_dvp); 1365 xop = hammer2_xop_alloc(dip, 0); 1366 1367 ncp = ap->a_nch->ncp; 1368 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1369 1370 /* 1371 * Note: In DragonFly the kernel handles '.' and '..'. 1372 */ 1373 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1374 hammer2_xop_start(&xop->head, &hammer2_nresolve_desc); 1375 1376 error = hammer2_xop_collect(&xop->head, 0); 1377 error = hammer2_error_to_errno(error); 1378 if (error) { 1379 ip = NULL; 1380 } else { 1381 ip = hammer2_inode_get(dip->pmp, &xop->head, -1, -1); 1382 } 1383 hammer2_inode_unlock(dip); 1384 1385 /* 1386 * Acquire the related vnode 1387 * 1388 * NOTE: For error processing, only ENOENT resolves the namecache 1389 * entry to NULL, otherwise we just return the error and 1390 * leave the namecache unresolved. 1391 * 1392 * NOTE: multiple hammer2_inode structures can be aliased to the 1393 * same chain element, for example for hardlinks. This 1394 * use case does not 'reattach' inode associations that 1395 * might already exist, but always allocates a new one. 1396 * 1397 * WARNING: inode structure is locked exclusively via inode_get 1398 * but chain was locked shared. inode_unlock() 1399 * will handle it properly. 1400 */ 1401 if (ip) { 1402 vp = hammer2_igetv(ip, &error); /* error set to UNIX error */ 1403 if (error == 0) { 1404 vn_unlock(vp); 1405 cache_setvp(ap->a_nch, vp); 1406 } else if (error == ENOENT) { 1407 cache_setvp(ap->a_nch, NULL); 1408 } 1409 hammer2_inode_unlock(ip); 1410 1411 /* 1412 * The vp should not be released until after we've disposed 1413 * of our locks, because it might cause vop_inactive() to 1414 * be called. 1415 */ 1416 if (vp) 1417 vrele(vp); 1418 } else { 1419 error = ENOENT; 1420 cache_setvp(ap->a_nch, NULL); 1421 } 1422 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1423 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL, 1424 ("resolve error %d/%p ap %p\n", 1425 error, ap->a_nch->ncp->nc_vp, ap)); 1426 1427 return error; 1428 } 1429 1430 static 1431 int 1432 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 1433 { 1434 hammer2_inode_t *dip; 1435 hammer2_tid_t inum; 1436 int error; 1437 1438 dip = VTOI(ap->a_dvp); 1439 inum = dip->meta.iparent; 1440 *ap->a_vpp = NULL; 1441 1442 if (inum) { 1443 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL, 1444 inum, ap->a_vpp); 1445 } else { 1446 error = ENOENT; 1447 } 1448 return error; 1449 } 1450 1451 static 1452 int 1453 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap) 1454 { 1455 hammer2_inode_t *dip; 1456 hammer2_inode_t *nip; 1457 struct namecache *ncp; 1458 const uint8_t *name; 1459 size_t name_len; 1460 hammer2_tid_t inum; 1461 int error; 1462 1463 dip = VTOI(ap->a_dvp); 1464 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG)) 1465 return (EROFS); 1466 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1467 return (ENOSPC); 1468 1469 ncp = ap->a_nch->ncp; 1470 name = ncp->nc_name; 1471 name_len = ncp->nc_nlen; 1472 1473 hammer2_trans_init(dip->pmp, 0); 1474 1475 inum = hammer2_trans_newinum(dip->pmp); 1476 1477 /* 1478 * Create the actual inode as a hidden file in the iroot, then 1479 * create the directory entry. The creation of the actual inode 1480 * sets its nlinks to 1 which is the value we desire. 1481 * 1482 * dip must be locked before nip to avoid deadlock. 1483 */ 1484 hammer2_inode_lock(dip, 0); 1485 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred, 1486 inum, &error); 1487 if (error) { 1488 error = hammer2_error_to_errno(error); 1489 } else { 1490 error = hammer2_dirent_create(dip, name, name_len, 1491 nip->meta.inum, nip->meta.type); 1492 /* returns UNIX error code */ 1493 } 1494 if (error) { 1495 if (nip) { 1496 hammer2_inode_unlink_finisher(nip, 0); 1497 hammer2_inode_unlock(nip); 1498 nip = NULL; 1499 } 1500 *ap->a_vpp = NULL; 1501 } else { 1502 /* 1503 * inode_depend() must occur before the igetv() because 1504 * the igetv() can temporarily release the inode lock. 1505 */ 1506 hammer2_inode_depend(dip, nip); /* before igetv */ 1507 *ap->a_vpp = hammer2_igetv(nip, &error); 1508 hammer2_inode_unlock(nip); 1509 } 1510 1511 /* 1512 * Update dip's mtime 1513 * 1514 * We can use a shared inode lock and allow the meta.mtime update 1515 * SMP race. hammer2_inode_modify() is MPSAFE w/a shared lock. 1516 */ 1517 if (error == 0) { 1518 uint64_t mtime; 1519 1520 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/ 1521 hammer2_update_time(&mtime); 1522 hammer2_inode_modify(dip); 1523 dip->meta.mtime = mtime; 1524 /*hammer2_inode_unlock(dip);*/ 1525 } 1526 hammer2_inode_unlock(dip); 1527 1528 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 1529 1530 if (error == 0) { 1531 cache_setunresolved(ap->a_nch); 1532 cache_setvp(ap->a_nch, *ap->a_vpp); 1533 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK); 1534 } 1535 return error; 1536 } 1537 1538 static 1539 int 1540 hammer2_vop_open(struct vop_open_args *ap) 1541 { 1542 return vop_stdopen(ap); 1543 } 1544 1545 /* 1546 * hammer2_vop_advlock { vp, id, op, fl, flags } 1547 */ 1548 static 1549 int 1550 hammer2_vop_advlock(struct vop_advlock_args *ap) 1551 { 1552 hammer2_inode_t *ip = VTOI(ap->a_vp); 1553 hammer2_off_t size; 1554 1555 size = ip->meta.size; 1556 return (lf_advlock(ap, &ip->advlock, size)); 1557 } 1558 1559 static 1560 int 1561 hammer2_vop_close(struct vop_close_args *ap) 1562 { 1563 return vop_stdclose(ap); 1564 } 1565 1566 /* 1567 * hammer2_vop_nlink { nch, dvp, vp, cred } 1568 * 1569 * Create a hardlink from (vp) to {dvp, nch}. 1570 */ 1571 static 1572 int 1573 hammer2_vop_nlink(struct vop_nlink_args *ap) 1574 { 1575 hammer2_inode_t *tdip; /* target directory to create link in */ 1576 hammer2_inode_t *ip; /* inode we are hardlinking to */ 1577 struct namecache *ncp; 1578 const uint8_t *name; 1579 size_t name_len; 1580 int error; 1581 1582 if (ap->a_dvp->v_mount != ap->a_vp->v_mount) 1583 return(EXDEV); 1584 1585 tdip = VTOI(ap->a_dvp); 1586 if (tdip->pmp->ronly || (tdip->pmp->flags & HAMMER2_PMPF_EMERG)) 1587 return (EROFS); 1588 if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1) 1589 return (ENOSPC); 1590 1591 ncp = ap->a_nch->ncp; 1592 name = ncp->nc_name; 1593 name_len = ncp->nc_nlen; 1594 1595 /* 1596 * ip represents the file being hardlinked. The file could be a 1597 * normal file or a hardlink target if it has already been hardlinked. 1598 * (with the new semantics, it will almost always be a hardlink 1599 * target). 1600 * 1601 * Bump nlinks and potentially also create or move the hardlink 1602 * target in the parent directory common to (ip) and (tdip). The 1603 * consolidation code can modify ip->cluster. The returned cluster 1604 * is locked. 1605 */ 1606 ip = VTOI(ap->a_vp); 1607 KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp)); 1608 hammer2_trans_init(ip->pmp, 0); 1609 1610 /* 1611 * Target should be an indexed inode or there's no way we will ever 1612 * be able to find it! 1613 */ 1614 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0); 1615 1616 error = 0; 1617 1618 /* 1619 * Can return NULL and error == EXDEV if the common parent 1620 * crosses a directory with the xlink flag set. 1621 */ 1622 hammer2_inode_lock4(tdip, ip, NULL, NULL); 1623 1624 /* 1625 * Create the directory entry and bump nlinks. 1626 */ 1627 if (error == 0) { 1628 error = hammer2_dirent_create(tdip, name, name_len, 1629 ip->meta.inum, ip->meta.type); 1630 hammer2_inode_modify(ip); 1631 ++ip->meta.nlinks; 1632 } 1633 if (error == 0) { 1634 /* 1635 * Update dip's mtime 1636 */ 1637 uint64_t mtime; 1638 1639 hammer2_update_time(&mtime); 1640 hammer2_inode_modify(tdip); 1641 tdip->meta.mtime = mtime; 1642 1643 cache_setunresolved(ap->a_nch); 1644 cache_setvp(ap->a_nch, ap->a_vp); 1645 } 1646 hammer2_inode_unlock(ip); 1647 hammer2_inode_unlock(tdip); 1648 1649 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_SIDEQ); 1650 hammer2_knote(ap->a_vp, NOTE_LINK); 1651 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1652 1653 return error; 1654 } 1655 1656 /* 1657 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap } 1658 * 1659 * The operating system has already ensured that the directory entry 1660 * does not exist and done all appropriate namespace locking. 1661 */ 1662 static 1663 int 1664 hammer2_vop_ncreate(struct vop_ncreate_args *ap) 1665 { 1666 hammer2_inode_t *dip; 1667 hammer2_inode_t *nip; 1668 struct namecache *ncp; 1669 const uint8_t *name; 1670 size_t name_len; 1671 hammer2_tid_t inum; 1672 int error; 1673 1674 dip = VTOI(ap->a_dvp); 1675 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG)) 1676 return (EROFS); 1677 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1678 return (ENOSPC); 1679 1680 ncp = ap->a_nch->ncp; 1681 name = ncp->nc_name; 1682 name_len = ncp->nc_nlen; 1683 hammer2_trans_init(dip->pmp, 0); 1684 1685 inum = hammer2_trans_newinum(dip->pmp); 1686 1687 /* 1688 * Create the actual inode as a hidden file in the iroot, then 1689 * create the directory entry. The creation of the actual inode 1690 * sets its nlinks to 1 which is the value we desire. 1691 * 1692 * dip must be locked before nip to avoid deadlock. 1693 */ 1694 hammer2_inode_lock(dip, 0); 1695 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred, 1696 inum, &error); 1697 1698 if (error) { 1699 error = hammer2_error_to_errno(error); 1700 } else { 1701 error = hammer2_dirent_create(dip, name, name_len, 1702 nip->meta.inum, nip->meta.type); 1703 } 1704 if (error) { 1705 if (nip) { 1706 hammer2_inode_unlink_finisher(nip, 0); 1707 hammer2_inode_unlock(nip); 1708 nip = NULL; 1709 } 1710 *ap->a_vpp = NULL; 1711 } else { 1712 hammer2_inode_depend(dip, nip); /* before igetv */ 1713 *ap->a_vpp = hammer2_igetv(nip, &error); 1714 hammer2_inode_unlock(nip); 1715 } 1716 1717 /* 1718 * Update dip's mtime 1719 */ 1720 if (error == 0) { 1721 uint64_t mtime; 1722 1723 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/ 1724 hammer2_update_time(&mtime); 1725 hammer2_inode_modify(dip); 1726 dip->meta.mtime = mtime; 1727 /*hammer2_inode_unlock(dip);*/ 1728 } 1729 hammer2_inode_unlock(dip); 1730 1731 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 1732 1733 if (error == 0) { 1734 cache_setunresolved(ap->a_nch); 1735 cache_setvp(ap->a_nch, *ap->a_vpp); 1736 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1737 } 1738 return error; 1739 } 1740 1741 /* 1742 * Make a device node (typically a fifo) 1743 */ 1744 static 1745 int 1746 hammer2_vop_nmknod(struct vop_nmknod_args *ap) 1747 { 1748 hammer2_inode_t *dip; 1749 hammer2_inode_t *nip; 1750 struct namecache *ncp; 1751 const uint8_t *name; 1752 size_t name_len; 1753 hammer2_tid_t inum; 1754 int error; 1755 1756 dip = VTOI(ap->a_dvp); 1757 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG)) 1758 return (EROFS); 1759 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1760 return (ENOSPC); 1761 1762 ncp = ap->a_nch->ncp; 1763 name = ncp->nc_name; 1764 name_len = ncp->nc_nlen; 1765 hammer2_trans_init(dip->pmp, 0); 1766 1767 /* 1768 * Create the device inode and then create the directory entry. 1769 * 1770 * dip must be locked before nip to avoid deadlock. 1771 */ 1772 inum = hammer2_trans_newinum(dip->pmp); 1773 1774 hammer2_inode_lock(dip, 0); 1775 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred, 1776 inum, &error); 1777 if (error == 0) { 1778 error = hammer2_dirent_create(dip, name, name_len, 1779 nip->meta.inum, nip->meta.type); 1780 } 1781 if (error) { 1782 if (nip) { 1783 hammer2_inode_unlink_finisher(nip, 0); 1784 hammer2_inode_unlock(nip); 1785 nip = NULL; 1786 } 1787 *ap->a_vpp = NULL; 1788 } else { 1789 hammer2_inode_depend(dip, nip); /* before igetv */ 1790 *ap->a_vpp = hammer2_igetv(nip, &error); 1791 hammer2_inode_unlock(nip); 1792 } 1793 1794 /* 1795 * Update dip's mtime 1796 */ 1797 if (error == 0) { 1798 uint64_t mtime; 1799 1800 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/ 1801 hammer2_update_time(&mtime); 1802 hammer2_inode_modify(dip); 1803 dip->meta.mtime = mtime; 1804 /*hammer2_inode_unlock(dip);*/ 1805 } 1806 hammer2_inode_unlock(dip); 1807 1808 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 1809 1810 if (error == 0) { 1811 cache_setunresolved(ap->a_nch); 1812 cache_setvp(ap->a_nch, *ap->a_vpp); 1813 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1814 } 1815 return error; 1816 } 1817 1818 /* 1819 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target } 1820 */ 1821 static 1822 int 1823 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap) 1824 { 1825 hammer2_inode_t *dip; 1826 hammer2_inode_t *nip; 1827 struct namecache *ncp; 1828 const uint8_t *name; 1829 size_t name_len; 1830 hammer2_tid_t inum; 1831 int error; 1832 1833 dip = VTOI(ap->a_dvp); 1834 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG)) 1835 return (EROFS); 1836 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1837 return (ENOSPC); 1838 1839 ncp = ap->a_nch->ncp; 1840 name = ncp->nc_name; 1841 name_len = ncp->nc_nlen; 1842 hammer2_trans_init(dip->pmp, 0); 1843 1844 ap->a_vap->va_type = VLNK; /* enforce type */ 1845 1846 /* 1847 * Create the softlink as an inode and then create the directory 1848 * entry. 1849 * 1850 * dip must be locked before nip to avoid deadlock. 1851 */ 1852 inum = hammer2_trans_newinum(dip->pmp); 1853 1854 hammer2_inode_lock(dip, 0); 1855 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred, 1856 inum, &error); 1857 if (error == 0) { 1858 error = hammer2_dirent_create(dip, name, name_len, 1859 nip->meta.inum, nip->meta.type); 1860 } 1861 if (error) { 1862 if (nip) { 1863 hammer2_inode_unlink_finisher(nip, 0); 1864 hammer2_inode_unlock(nip); 1865 nip = NULL; 1866 } 1867 *ap->a_vpp = NULL; 1868 hammer2_inode_unlock(dip); 1869 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 1870 return error; 1871 } 1872 hammer2_inode_depend(dip, nip); /* before igetv */ 1873 *ap->a_vpp = hammer2_igetv(nip, &error); 1874 1875 /* 1876 * Build the softlink (~like file data) and finalize the namecache. 1877 */ 1878 if (error == 0) { 1879 size_t bytes; 1880 struct uio auio; 1881 struct iovec aiov; 1882 1883 bytes = strlen(ap->a_target); 1884 1885 hammer2_inode_unlock(nip); 1886 bzero(&auio, sizeof(auio)); 1887 bzero(&aiov, sizeof(aiov)); 1888 auio.uio_iov = &aiov; 1889 auio.uio_segflg = UIO_SYSSPACE; 1890 auio.uio_rw = UIO_WRITE; 1891 auio.uio_resid = bytes; 1892 auio.uio_iovcnt = 1; 1893 auio.uio_td = curthread; 1894 aiov.iov_base = ap->a_target; 1895 aiov.iov_len = bytes; 1896 error = hammer2_write_file(nip, &auio, IO_APPEND, 0); 1897 /* XXX handle error */ 1898 error = 0; 1899 } else { 1900 hammer2_inode_unlock(nip); 1901 } 1902 1903 /* 1904 * Update dip's mtime 1905 */ 1906 if (error == 0) { 1907 uint64_t mtime; 1908 1909 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/ 1910 hammer2_update_time(&mtime); 1911 hammer2_inode_modify(dip); 1912 dip->meta.mtime = mtime; 1913 /*hammer2_inode_unlock(dip);*/ 1914 } 1915 hammer2_inode_unlock(dip); 1916 1917 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 1918 1919 /* 1920 * Finalize namecache 1921 */ 1922 if (error == 0) { 1923 cache_setunresolved(ap->a_nch); 1924 cache_setvp(ap->a_nch, *ap->a_vpp); 1925 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1926 } 1927 return error; 1928 } 1929 1930 /* 1931 * hammer2_vop_nremove { nch, dvp, cred } 1932 */ 1933 static 1934 int 1935 hammer2_vop_nremove(struct vop_nremove_args *ap) 1936 { 1937 hammer2_xop_unlink_t *xop; 1938 hammer2_inode_t *dip; 1939 hammer2_inode_t *ip; 1940 struct namecache *ncp; 1941 int error; 1942 int isopen; 1943 1944 dip = VTOI(ap->a_dvp); 1945 if (dip->pmp->ronly) 1946 return (EROFS); 1947 #if 0 1948 /* allow removals, except user to also bulkfree */ 1949 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1950 return (ENOSPC); 1951 #endif 1952 1953 ncp = ap->a_nch->ncp; 1954 1955 if (hammer2_debug_inode && dip->meta.inum == hammer2_debug_inode) { 1956 kprintf("hammer2: attempt to delete inside debug inode: %s\n", 1957 ncp->nc_name); 1958 while (hammer2_debug_inode && 1959 dip->meta.inum == hammer2_debug_inode) { 1960 tsleep(&hammer2_debug_inode, 0, "h2debug", hz*5); 1961 } 1962 } 1963 1964 hammer2_trans_init(dip->pmp, 0); 1965 hammer2_inode_lock(dip, 0); 1966 1967 /* 1968 * The unlink XOP unlinks the path from the directory and 1969 * locates and returns the cluster associated with the real inode. 1970 * We have to handle nlinks here on the frontend. 1971 */ 1972 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING); 1973 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1974 1975 /* 1976 * The namecache entry is locked so nobody can use this namespace. 1977 * Calculate isopen to determine if this namespace has an open vp 1978 * associated with it and resolve the vp only if it does. 1979 * 1980 * We try to avoid resolving the vnode if nobody has it open, but 1981 * note that the test is via this namespace only. 1982 */ 1983 isopen = cache_isopen(ap->a_nch); 1984 xop->isdir = 0; 1985 xop->dopermanent = 0; 1986 hammer2_xop_start(&xop->head, &hammer2_unlink_desc); 1987 1988 /* 1989 * Collect the real inode and adjust nlinks, destroy the real 1990 * inode if nlinks transitions to 0 and it was the real inode 1991 * (else it has already been removed). 1992 */ 1993 error = hammer2_xop_collect(&xop->head, 0); 1994 error = hammer2_error_to_errno(error); 1995 1996 if (error == 0) { 1997 ip = hammer2_inode_get(dip->pmp, &xop->head, -1, -1); 1998 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1999 if (ip) { 2000 if (hammer2_debug_inode && 2001 ip->meta.inum == hammer2_debug_inode) { 2002 kprintf("hammer2: attempt to delete debug " 2003 "inode!\n"); 2004 while (hammer2_debug_inode && 2005 ip->meta.inum == hammer2_debug_inode) { 2006 tsleep(&hammer2_debug_inode, 0, 2007 "h2debug", hz*5); 2008 } 2009 } 2010 hammer2_inode_unlink_finisher(ip, isopen); 2011 hammer2_inode_depend(dip, ip); /* after modified */ 2012 hammer2_inode_unlock(ip); 2013 } 2014 } else { 2015 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 2016 } 2017 2018 /* 2019 * Update dip's mtime 2020 */ 2021 if (error == 0) { 2022 uint64_t mtime; 2023 2024 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/ 2025 hammer2_update_time(&mtime); 2026 hammer2_inode_modify(dip); 2027 dip->meta.mtime = mtime; 2028 /*hammer2_inode_unlock(dip);*/ 2029 } 2030 hammer2_inode_unlock(dip); 2031 2032 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 2033 if (error == 0) { 2034 cache_unlink(ap->a_nch); 2035 hammer2_knote(ap->a_dvp, NOTE_WRITE); 2036 } 2037 return (error); 2038 } 2039 2040 /* 2041 * hammer2_vop_nrmdir { nch, dvp, cred } 2042 */ 2043 static 2044 int 2045 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap) 2046 { 2047 hammer2_xop_unlink_t *xop; 2048 hammer2_inode_t *dip; 2049 hammer2_inode_t *ip; 2050 struct namecache *ncp; 2051 int isopen; 2052 int error; 2053 2054 dip = VTOI(ap->a_dvp); 2055 if (dip->pmp->ronly) 2056 return (EROFS); 2057 #if 0 2058 /* allow removals, except user to also bulkfree */ 2059 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 2060 return (ENOSPC); 2061 #endif 2062 2063 hammer2_trans_init(dip->pmp, 0); 2064 hammer2_inode_lock(dip, 0); 2065 2066 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING); 2067 2068 ncp = ap->a_nch->ncp; 2069 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 2070 isopen = cache_isopen(ap->a_nch); 2071 xop->isdir = 1; 2072 xop->dopermanent = 0; 2073 hammer2_xop_start(&xop->head, &hammer2_unlink_desc); 2074 2075 /* 2076 * Collect the real inode and adjust nlinks, destroy the real 2077 * inode if nlinks transitions to 0 and it was the real inode 2078 * (else it has already been removed). 2079 */ 2080 error = hammer2_xop_collect(&xop->head, 0); 2081 error = hammer2_error_to_errno(error); 2082 2083 if (error == 0) { 2084 ip = hammer2_inode_get(dip->pmp, &xop->head, -1, -1); 2085 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 2086 if (ip) { 2087 hammer2_inode_unlink_finisher(ip, isopen); 2088 hammer2_inode_depend(dip, ip); /* after modified */ 2089 hammer2_inode_unlock(ip); 2090 } 2091 } else { 2092 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 2093 } 2094 2095 /* 2096 * Update dip's mtime 2097 */ 2098 if (error == 0) { 2099 uint64_t mtime; 2100 2101 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/ 2102 hammer2_update_time(&mtime); 2103 hammer2_inode_modify(dip); 2104 dip->meta.mtime = mtime; 2105 /*hammer2_inode_unlock(dip);*/ 2106 } 2107 hammer2_inode_unlock(dip); 2108 2109 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ); 2110 if (error == 0) { 2111 cache_unlink(ap->a_nch); 2112 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK); 2113 } 2114 return (error); 2115 } 2116 2117 /* 2118 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred } 2119 */ 2120 static 2121 int 2122 hammer2_vop_nrename(struct vop_nrename_args *ap) 2123 { 2124 struct namecache *fncp; 2125 struct namecache *tncp; 2126 hammer2_inode_t *fdip; /* source directory */ 2127 hammer2_inode_t *tdip; /* target directory */ 2128 hammer2_inode_t *ip; /* file being renamed */ 2129 hammer2_inode_t *tip; /* replaced target during rename or NULL */ 2130 const uint8_t *fname; 2131 size_t fname_len; 2132 const uint8_t *tname; 2133 size_t tname_len; 2134 int error; 2135 int update_tdip; 2136 int update_fdip; 2137 hammer2_key_t tlhc; 2138 2139 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount) 2140 return(EXDEV); 2141 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount) 2142 return(EXDEV); 2143 2144 fdip = VTOI(ap->a_fdvp); /* source directory */ 2145 tdip = VTOI(ap->a_tdvp); /* target directory */ 2146 2147 if (fdip->pmp->ronly || (fdip->pmp->flags & HAMMER2_PMPF_EMERG)) 2148 return (EROFS); 2149 if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1) 2150 return (ENOSPC); 2151 2152 fncp = ap->a_fnch->ncp; /* entry name in source */ 2153 fname = fncp->nc_name; 2154 fname_len = fncp->nc_nlen; 2155 2156 tncp = ap->a_tnch->ncp; /* entry name in target */ 2157 tname = tncp->nc_name; 2158 tname_len = tncp->nc_nlen; 2159 2160 hammer2_trans_init(tdip->pmp, 0); 2161 2162 update_tdip = 0; 2163 update_fdip = 0; 2164 2165 ip = VTOI(fncp->nc_vp); 2166 hammer2_inode_ref(ip); /* extra ref */ 2167 2168 /* 2169 * Lookup the target name to determine if a directory entry 2170 * is being overwritten. We only hold related inode locks 2171 * temporarily, the operating system is expected to protect 2172 * against rename races. 2173 */ 2174 tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL; 2175 if (tip) 2176 hammer2_inode_ref(tip); /* extra ref */ 2177 2178 /* 2179 * Can return NULL and error == EXDEV if the common parent 2180 * crosses a directory with the xlink flag set. 2181 * 2182 * For now try to avoid deadlocks with a simple pointer address 2183 * test. (tip) can be NULL. 2184 */ 2185 error = 0; 2186 { 2187 hammer2_inode_t *ip1 = fdip; 2188 hammer2_inode_t *ip2 = tdip; 2189 hammer2_inode_t *ip3 = ip; 2190 hammer2_inode_t *ip4 = tip; /* may be NULL */ 2191 2192 if (fdip > tdip) { 2193 ip1 = tdip; 2194 ip2 = fdip; 2195 } 2196 if (tip && ip > tip) { 2197 ip3 = tip; 2198 ip4 = ip; 2199 } 2200 hammer2_inode_lock4(ip1, ip2, ip3, ip4); 2201 } 2202 2203 /* 2204 * Resolve the collision space for (tdip, tname, tname_len) 2205 * 2206 * tdip must be held exclusively locked to prevent races since 2207 * multiple filenames can end up in the same collision space. 2208 */ 2209 { 2210 hammer2_xop_scanlhc_t *sxop; 2211 hammer2_tid_t lhcbase; 2212 2213 tlhc = hammer2_dirhash(tname, tname_len); 2214 lhcbase = tlhc; 2215 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING); 2216 sxop->lhc = tlhc; 2217 hammer2_xop_start(&sxop->head, &hammer2_scanlhc_desc); 2218 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) { 2219 if (tlhc != sxop->head.cluster.focus->bref.key) 2220 break; 2221 ++tlhc; 2222 } 2223 error = hammer2_error_to_errno(error); 2224 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP); 2225 2226 if (error) { 2227 if (error != ENOENT) 2228 goto done2; 2229 ++tlhc; 2230 error = 0; 2231 } 2232 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) { 2233 error = ENOSPC; 2234 goto done2; 2235 } 2236 } 2237 2238 /* 2239 * Ready to go, issue the rename to the backend. Note that meta-data 2240 * updates to the related inodes occur separately from the rename 2241 * operation. 2242 * 2243 * NOTE: While it is not necessary to update ip->meta.name*, doing 2244 * so aids catastrophic recovery and debugging. 2245 */ 2246 if (error == 0) { 2247 hammer2_xop_nrename_t *xop4; 2248 2249 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING); 2250 xop4->lhc = tlhc; 2251 xop4->ip_key = ip->meta.name_key; 2252 hammer2_xop_setip2(&xop4->head, ip); 2253 hammer2_xop_setip3(&xop4->head, tdip); 2254 hammer2_xop_setname(&xop4->head, fname, fname_len); 2255 hammer2_xop_setname2(&xop4->head, tname, tname_len); 2256 hammer2_xop_start(&xop4->head, &hammer2_nrename_desc); 2257 2258 error = hammer2_xop_collect(&xop4->head, 0); 2259 error = hammer2_error_to_errno(error); 2260 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP); 2261 2262 if (error == ENOENT) 2263 error = 0; 2264 2265 /* 2266 * Update inode meta-data. 2267 * 2268 * WARNING! The in-memory inode (ip) structure does not 2269 * maintain a copy of the inode's filename buffer. 2270 */ 2271 if (error == 0 && 2272 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) { 2273 hammer2_inode_modify(ip); 2274 ip->meta.name_len = tname_len; 2275 ip->meta.name_key = tlhc; 2276 } 2277 if (error == 0) { 2278 hammer2_inode_modify(ip); 2279 ip->meta.iparent = tdip->meta.inum; 2280 } 2281 update_fdip = 1; 2282 update_tdip = 1; 2283 } 2284 2285 done2: 2286 /* 2287 * If no error, the backend has replaced the target directory entry. 2288 * We must adjust nlinks on the original replace target if it exists. 2289 */ 2290 if (error == 0 && tip) { 2291 int isopen; 2292 2293 isopen = cache_isopen(ap->a_tnch); 2294 hammer2_inode_unlink_finisher(tip, isopen); 2295 } 2296 2297 /* 2298 * Update directory mtimes to represent the something changed. 2299 */ 2300 if (update_fdip || update_tdip) { 2301 uint64_t mtime; 2302 2303 hammer2_update_time(&mtime); 2304 if (update_fdip) { 2305 hammer2_inode_modify(fdip); 2306 fdip->meta.mtime = mtime; 2307 } 2308 if (update_tdip) { 2309 hammer2_inode_modify(tdip); 2310 tdip->meta.mtime = mtime; 2311 } 2312 } 2313 if (tip) { 2314 hammer2_inode_unlock(tip); 2315 hammer2_inode_drop(tip); 2316 } 2317 hammer2_inode_unlock(ip); 2318 hammer2_inode_unlock(tdip); 2319 hammer2_inode_unlock(fdip); 2320 hammer2_inode_drop(ip); 2321 hammer2_trans_done(tdip->pmp, HAMMER2_TRANS_SIDEQ); 2322 2323 /* 2324 * Issue the namecache update after unlocking all the internal 2325 * hammer2 structures, otherwise we might deadlock. 2326 * 2327 * WARNING! The target namespace must be updated atomically, 2328 * and we depend on cache_rename() to handle that for 2329 * us. Do not do a separate cache_unlink() because 2330 * that leaves a small window of opportunity for other 2331 * threads to allocate the target namespace before we 2332 * manage to complete our rename. 2333 * 2334 * WARNING! cache_rename() (and cache_unlink()) will properly 2335 * set VREF_FINALIZE on any attached vnode. Do not 2336 * call cache_setunresolved() manually before-hand as 2337 * this will prevent the flag from being set later via 2338 * cache_rename(). If VREF_FINALIZE is not properly set 2339 * and the inode is no longer in the topology, related 2340 * chains can remain dirty indefinitely. 2341 */ 2342 if (error == 0 && tip) { 2343 /*cache_unlink(ap->a_tnch); see above */ 2344 /*cache_setunresolved(ap->a_tnch); see above */ 2345 } 2346 if (error == 0) { 2347 cache_rename(ap->a_fnch, ap->a_tnch); 2348 hammer2_knote(ap->a_fdvp, NOTE_WRITE); 2349 hammer2_knote(ap->a_tdvp, NOTE_WRITE); 2350 hammer2_knote(fncp->nc_vp, NOTE_RENAME); 2351 } 2352 2353 return (error); 2354 } 2355 2356 /* 2357 * hammer2_vop_ioctl { vp, command, data, fflag, cred } 2358 */ 2359 static 2360 int 2361 hammer2_vop_ioctl(struct vop_ioctl_args *ap) 2362 { 2363 hammer2_inode_t *ip; 2364 int error; 2365 2366 ip = VTOI(ap->a_vp); 2367 2368 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data, 2369 ap->a_fflag, ap->a_cred); 2370 return (error); 2371 } 2372 2373 static 2374 int 2375 hammer2_vop_mountctl(struct vop_mountctl_args *ap) 2376 { 2377 struct mount *mp; 2378 hammer2_pfs_t *pmp; 2379 int rc; 2380 2381 switch (ap->a_op) { 2382 case (MOUNTCTL_SET_EXPORT): 2383 mp = ap->a_head.a_ops->head.vv_mount; 2384 pmp = MPTOPMP(mp); 2385 2386 if (ap->a_ctllen != sizeof(struct export_args)) 2387 rc = (EINVAL); 2388 else 2389 rc = vfs_export(mp, &pmp->export, 2390 (const struct export_args *)ap->a_ctl); 2391 break; 2392 default: 2393 rc = vop_stdmountctl(ap); 2394 break; 2395 } 2396 return (rc); 2397 } 2398 2399 /* 2400 * KQFILTER 2401 */ 2402 static void filt_hammer2detach(struct knote *kn); 2403 static int filt_hammer2read(struct knote *kn, long hint); 2404 static int filt_hammer2write(struct knote *kn, long hint); 2405 static int filt_hammer2vnode(struct knote *kn, long hint); 2406 2407 static struct filterops hammer2read_filtops = 2408 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2409 NULL, filt_hammer2detach, filt_hammer2read }; 2410 static struct filterops hammer2write_filtops = 2411 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2412 NULL, filt_hammer2detach, filt_hammer2write }; 2413 static struct filterops hammer2vnode_filtops = 2414 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2415 NULL, filt_hammer2detach, filt_hammer2vnode }; 2416 2417 static 2418 int 2419 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap) 2420 { 2421 struct vnode *vp = ap->a_vp; 2422 struct knote *kn = ap->a_kn; 2423 2424 switch (kn->kn_filter) { 2425 case EVFILT_READ: 2426 kn->kn_fop = &hammer2read_filtops; 2427 break; 2428 case EVFILT_WRITE: 2429 kn->kn_fop = &hammer2write_filtops; 2430 break; 2431 case EVFILT_VNODE: 2432 kn->kn_fop = &hammer2vnode_filtops; 2433 break; 2434 default: 2435 return (EOPNOTSUPP); 2436 } 2437 2438 kn->kn_hook = (caddr_t)vp; 2439 2440 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2441 2442 return(0); 2443 } 2444 2445 static void 2446 filt_hammer2detach(struct knote *kn) 2447 { 2448 struct vnode *vp = (void *)kn->kn_hook; 2449 2450 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2451 } 2452 2453 static int 2454 filt_hammer2read(struct knote *kn, long hint) 2455 { 2456 struct vnode *vp = (void *)kn->kn_hook; 2457 hammer2_inode_t *ip = VTOI(vp); 2458 off_t off; 2459 2460 if (hint == NOTE_REVOKE) { 2461 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2462 return(1); 2463 } 2464 off = ip->meta.size - kn->kn_fp->f_offset; 2465 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2466 if (kn->kn_sfflags & NOTE_OLDAPI) 2467 return(1); 2468 return (kn->kn_data != 0); 2469 } 2470 2471 2472 static int 2473 filt_hammer2write(struct knote *kn, long hint) 2474 { 2475 if (hint == NOTE_REVOKE) 2476 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2477 kn->kn_data = 0; 2478 return (1); 2479 } 2480 2481 static int 2482 filt_hammer2vnode(struct knote *kn, long hint) 2483 { 2484 if (kn->kn_sfflags & hint) 2485 kn->kn_fflags |= hint; 2486 if (hint == NOTE_REVOKE) { 2487 kn->kn_flags |= (EV_EOF | EV_NODATA); 2488 return (1); 2489 } 2490 return (kn->kn_fflags != 0); 2491 } 2492 2493 /* 2494 * FIFO VOPS 2495 */ 2496 static 2497 int 2498 hammer2_vop_markatime(struct vop_markatime_args *ap) 2499 { 2500 hammer2_inode_t *ip; 2501 struct vnode *vp; 2502 2503 vp = ap->a_vp; 2504 ip = VTOI(vp); 2505 2506 if (ip->pmp->ronly || (ip->pmp->flags & HAMMER2_PMPF_EMERG)) 2507 return (EROFS); 2508 return(0); 2509 } 2510 2511 static 2512 int 2513 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap) 2514 { 2515 int error; 2516 2517 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2518 if (error) 2519 error = hammer2_vop_kqfilter(ap); 2520 return(error); 2521 } 2522 2523 /* 2524 * VOPS vector 2525 */ 2526 struct vop_ops hammer2_vnode_vops = { 2527 .vop_default = vop_defaultop, 2528 .vop_fsync = hammer2_vop_fsync, 2529 .vop_getpages = vop_stdgetpages, 2530 .vop_putpages = vop_stdputpages, 2531 .vop_access = hammer2_vop_access, 2532 .vop_advlock = hammer2_vop_advlock, 2533 .vop_close = hammer2_vop_close, 2534 .vop_nlink = hammer2_vop_nlink, 2535 .vop_ncreate = hammer2_vop_ncreate, 2536 .vop_nsymlink = hammer2_vop_nsymlink, 2537 .vop_nremove = hammer2_vop_nremove, 2538 .vop_nrmdir = hammer2_vop_nrmdir, 2539 .vop_nrename = hammer2_vop_nrename, 2540 .vop_getattr = hammer2_vop_getattr, 2541 .vop_getattr_quick = hammer2_vop_getattr_quick, 2542 .vop_setattr = hammer2_vop_setattr, 2543 .vop_readdir = hammer2_vop_readdir, 2544 .vop_readlink = hammer2_vop_readlink, 2545 .vop_read = hammer2_vop_read, 2546 .vop_write = hammer2_vop_write, 2547 .vop_open = hammer2_vop_open, 2548 .vop_inactive = hammer2_vop_inactive, 2549 .vop_reclaim = hammer2_vop_reclaim, 2550 .vop_nresolve = hammer2_vop_nresolve, 2551 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot, 2552 .vop_nmkdir = hammer2_vop_nmkdir, 2553 .vop_nmknod = hammer2_vop_nmknod, 2554 .vop_ioctl = hammer2_vop_ioctl, 2555 .vop_mountctl = hammer2_vop_mountctl, 2556 .vop_bmap = hammer2_vop_bmap, 2557 .vop_strategy = hammer2_vop_strategy, 2558 .vop_kqfilter = hammer2_vop_kqfilter 2559 }; 2560 2561 struct vop_ops hammer2_spec_vops = { 2562 .vop_default = vop_defaultop, 2563 .vop_fsync = hammer2_vop_fsync, 2564 .vop_read = vop_stdnoread, 2565 .vop_write = vop_stdnowrite, 2566 .vop_access = hammer2_vop_access, 2567 .vop_close = hammer2_vop_close, 2568 .vop_markatime = hammer2_vop_markatime, 2569 .vop_getattr = hammer2_vop_getattr, 2570 .vop_inactive = hammer2_vop_inactive, 2571 .vop_reclaim = hammer2_vop_reclaim, 2572 .vop_setattr = hammer2_vop_setattr 2573 }; 2574 2575 struct vop_ops hammer2_fifo_vops = { 2576 .vop_default = fifo_vnoperate, 2577 .vop_fsync = hammer2_vop_fsync, 2578 #if 0 2579 .vop_read = hammer2_vop_fiforead, 2580 .vop_write = hammer2_vop_fifowrite, 2581 #endif 2582 .vop_access = hammer2_vop_access, 2583 #if 0 2584 .vop_close = hammer2_vop_fifoclose, 2585 #endif 2586 .vop_markatime = hammer2_vop_markatime, 2587 .vop_getattr = hammer2_vop_getattr, 2588 .vop_inactive = hammer2_vop_inactive, 2589 .vop_reclaim = hammer2_vop_reclaim, 2590 .vop_setattr = hammer2_vop_setattr, 2591 .vop_kqfilter = hammer2_vop_fifokqfilter 2592 }; 2593 2594