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 * 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/namei.h> 50 #include <sys/mount.h> 51 #include <sys/vnode.h> 52 #include <sys/mountctl.h> 53 #include <sys/dirent.h> 54 #include <sys/uio.h> 55 #include <sys/objcache.h> 56 #include <sys/event.h> 57 #include <sys/file.h> 58 #include <vfs/fifofs/fifo.h> 59 60 #include "hammer2.h" 61 #include "hammer2_lz4.h" 62 63 #include "zlib/hammer2_zlib.h" 64 65 #define ZFOFFSET (-2LL) 66 67 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, 68 int seqcount); 69 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio, 70 int ioflag, int seqcount); 71 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize); 72 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize); 73 static void hammer2_decompress_LZ4_callback(hammer2_io_t *dio, 74 hammer2_chain_t *arg_c, 75 void *arg_p, off_t arg_o); 76 static void hammer2_decompress_ZLIB_callback(hammer2_io_t *dio, 77 hammer2_chain_t *arg_c, 78 void *arg_p, off_t arg_o); 79 80 struct objcache *cache_buffer_read; 81 struct objcache *cache_buffer_write; 82 83 /* 84 * Callback used in read path in case that a block is compressed with LZ4. 85 */ 86 static 87 void 88 hammer2_decompress_LZ4_callback(hammer2_io_t *dio, hammer2_chain_t *arg_c, 89 void *arg_p, off_t arg_o) 90 { 91 struct buf *obp; 92 struct bio *obio = arg_p; 93 char *bdata; 94 int bytes = 1 << (int)(arg_o & HAMMER2_OFF_MASK_RADIX); 95 96 /* 97 * If BIO_DONE is already set the device buffer was already 98 * fully valid (B_CACHE). If it is not set then I/O was issued 99 * and we have to run I/O completion as the last bio. 100 * 101 * Nobody is waiting for our device I/O to complete, we are 102 * responsible for bqrelse()ing it which means we also have to do 103 * the equivalent of biowait() and clear BIO_DONE (which breadcb() 104 * may have set). 105 * 106 * Any preexisting device buffer should match the requested size, 107 * but due to bigblock recycling and other factors there is some 108 * fragility there, so we assert that the device buffer covers 109 * the request. 110 */ 111 obp = obio->bio_buf; 112 113 if (dio->bp->b_flags & B_ERROR) { 114 obp->b_flags |= B_ERROR; 115 obp->b_error = dio->bp->b_error; 116 #if 0 117 } else if (obio->bio_caller_info2.index && 118 obio->bio_caller_info1.uvalue32 != 119 crc32(bp->b_data, bp->b_bufsize)) { 120 obp->b_flags |= B_ERROR; 121 obp->b_error = EIO; 122 #endif 123 } else { 124 char *compressed_buffer; 125 int *compressed_size; 126 int result; 127 128 KKASSERT(obp->b_bufsize <= HAMMER2_PBUFSIZE); 129 bdata = hammer2_io_data(dio, arg_o); 130 compressed_size = (int *)bdata; 131 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT); 132 KKASSERT((unsigned int)*compressed_size <= HAMMER2_PBUFSIZE); 133 result = LZ4_decompress_safe(&bdata[sizeof(int)], 134 compressed_buffer, 135 *compressed_size, 136 obp->b_bufsize); 137 if (result < 0) { 138 kprintf("READ PATH: Error during decompression." 139 "bio %016jx/%d log %016jx/%d\n", 140 (intmax_t)dio->pbase, dio->psize, 141 (intmax_t)arg_o, bytes); 142 /* make sure it isn't random garbage */ 143 bzero(compressed_buffer, obp->b_bufsize); 144 } 145 KKASSERT(result <= obp->b_bufsize); 146 bcopy(compressed_buffer, obp->b_data, obp->b_bufsize); 147 if (result < obp->b_bufsize) 148 bzero(obp->b_data + result, obp->b_bufsize - result); 149 objcache_put(cache_buffer_read, compressed_buffer); 150 obp->b_resid = 0; 151 obp->b_flags |= B_AGE; 152 } 153 biodone(obio); 154 } 155 156 /* 157 * Callback used in read path in case that a block is compressed with ZLIB. 158 * It is almost identical to LZ4 callback, so in theory they can be unified, 159 * but we didn't want to make changes in bio structure for that. 160 */ 161 static 162 void 163 hammer2_decompress_ZLIB_callback(hammer2_io_t *dio, hammer2_chain_t *arg_c, 164 void *arg_p, off_t arg_o) 165 { 166 struct buf *obp; 167 struct bio *obio = arg_p; 168 char *bdata; 169 int bytes = 1 << (int)(arg_o & HAMMER2_OFF_MASK_RADIX); 170 171 /* 172 * If BIO_DONE is already set the device buffer was already 173 * fully valid (B_CACHE). If it is not set then I/O was issued 174 * and we have to run I/O completion as the last bio. 175 * 176 * Nobody is waiting for our device I/O to complete, we are 177 * responsible for bqrelse()ing it which means we also have to do 178 * the equivalent of biowait() and clear BIO_DONE (which breadcb() 179 * may have set). 180 * 181 * Any preexisting device buffer should match the requested size, 182 * but due to bigblock recycling and other factors there is some 183 * fragility there, so we assert that the device buffer covers 184 * the request. 185 */ 186 obp = obio->bio_buf; 187 188 if (dio->bp->b_flags & B_ERROR) { 189 obp->b_flags |= B_ERROR; 190 obp->b_error = dio->bp->b_error; 191 #if 0 192 } else if (obio->bio_caller_info2.index && 193 obio->bio_caller_info1.uvalue32 != 194 crc32(bp->b_data, bp->b_bufsize)) { 195 obp->b_flags |= B_ERROR; 196 obp->b_error = EIO; 197 #endif 198 } else { 199 char *compressed_buffer; 200 z_stream strm_decompress; 201 int result; 202 int ret; 203 204 KKASSERT(obp->b_bufsize <= HAMMER2_PBUFSIZE); 205 strm_decompress.avail_in = 0; 206 strm_decompress.next_in = Z_NULL; 207 208 ret = inflateInit(&strm_decompress); 209 210 if (ret != Z_OK) 211 kprintf("HAMMER2 ZLIB: Fatal error in inflateInit.\n"); 212 213 bdata = hammer2_io_data(dio, arg_o); 214 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT); 215 strm_decompress.next_in = bdata; 216 217 /* XXX supply proper size, subset of device bp */ 218 strm_decompress.avail_in = bytes; 219 strm_decompress.next_out = compressed_buffer; 220 strm_decompress.avail_out = obp->b_bufsize; 221 222 ret = inflate(&strm_decompress, Z_FINISH); 223 if (ret != Z_STREAM_END) { 224 kprintf("HAMMER2 ZLIB: Fatar error during decompression.\n"); 225 bzero(compressed_buffer, obp->b_bufsize); 226 } 227 bcopy(compressed_buffer, obp->b_data, obp->b_bufsize); 228 result = obp->b_bufsize - strm_decompress.avail_out; 229 if (result < obp->b_bufsize) 230 bzero(obp->b_data + result, strm_decompress.avail_out); 231 objcache_put(cache_buffer_read, compressed_buffer); 232 obp->b_resid = 0; 233 obp->b_flags |= B_AGE; 234 ret = inflateEnd(&strm_decompress); 235 } 236 biodone(obio); 237 } 238 239 static __inline 240 void 241 hammer2_knote(struct vnode *vp, int flags) 242 { 243 if (flags) 244 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 245 } 246 247 /* 248 * Last reference to a vnode is going away but it is still cached. 249 */ 250 static 251 int 252 hammer2_vop_inactive(struct vop_inactive_args *ap) 253 { 254 hammer2_inode_t *ip; 255 hammer2_chain_t *parent; 256 struct vnode *vp; 257 258 vp = ap->a_vp; 259 ip = VTOI(vp); 260 261 /* 262 * Degenerate case 263 */ 264 if (ip == NULL) { 265 vrecycle(vp); 266 return (0); 267 } 268 269 /* 270 * Detect updates to the embedded data which may be synchronized by 271 * the strategy code. Simply mark the inode modified so it gets 272 * picked up by our normal flush. 273 */ 274 parent = hammer2_inode_lock_ex(ip); 275 KKASSERT(parent); 276 277 /* 278 * Check for deleted inodes and recycle immediately. 279 */ 280 if (parent->flags & HAMMER2_CHAIN_UNLINKED) { 281 hammer2_inode_unlock_ex(ip, parent); 282 vrecycle(vp); 283 } else { 284 hammer2_inode_unlock_ex(ip, parent); 285 } 286 return (0); 287 } 288 289 /* 290 * Reclaim a vnode so that it can be reused; after the inode is 291 * disassociated, the filesystem must manage it alone. 292 */ 293 static 294 int 295 hammer2_vop_reclaim(struct vop_reclaim_args *ap) 296 { 297 hammer2_chain_t *chain; 298 hammer2_inode_t *ip; 299 hammer2_trans_t trans; 300 struct vnode *vp; 301 302 vp = ap->a_vp; 303 ip = VTOI(vp); 304 if (ip == NULL) 305 return(0); 306 307 /* 308 * Set update_hi so we can detect and propagate the DELETED 309 * bit in the flush code. 310 * 311 * ip->chain might be stale, correct it before checking as older 312 * versions of the chain are likely marked deleted even if the 313 * file hasn't been. XXX ip->chain should never be stale on 314 * reclaim. 315 */ 316 chain = hammer2_inode_lock_ex(ip); 317 #if 0 318 if (chain->next_parent) 319 kprintf("RECLAIM DUPLINKED IP: %p ip->ch=%p ch=%p np=%p\n", 320 ip, ip->chain, chain, chain->next_parent); 321 #endif 322 323 /* 324 * The final close of a deleted file or directory marks it for 325 * destruction. The DELETED flag allows the flusher to shortcut 326 * any modified blocks still unflushed (that is, just ignore them). 327 * 328 * HAMMER2 usually does not try to optimize the freemap by returning 329 * deleted blocks to it as it does not usually know how many snapshots 330 * might be referencing portions of the file/dir. XXX TODO. 331 * 332 * XXX TODO - However, any modified file as-of when a snapshot is made 333 * cannot use this optimization as some of the modifications 334 * may wind up being part of the snapshot. 335 */ 336 vp->v_data = NULL; 337 ip->vp = NULL; 338 if (chain->flags & HAMMER2_CHAIN_UNLINKED) { 339 kprintf("unlink on reclaim: %s\n", 340 chain->data->ipdata.filename); 341 hammer2_trans_init(&trans, ip->pmp, NULL, 342 HAMMER2_TRANS_BUFCACHE); 343 hammer2_chain_delete(&trans, chain, 0); 344 hammer2_chain_setsubmod(&trans, chain); 345 spin_lock(&chain->core->cst.spin); 346 if (chain->core->update_hi < trans.sync_tid) 347 chain->core->update_hi = trans.sync_tid; /* needed? */ 348 spin_unlock(&chain->core->cst.spin); 349 hammer2_trans_done(&trans); 350 } 351 352 /* 353 * NOTE! We do not attempt to flush chains here, flushing is 354 * really fragile and could also deadlock. 355 */ 356 vclrisdirty(vp); 357 hammer2_inode_unlock_ex(ip, chain); /* unlock */ 358 hammer2_inode_drop(ip); /* vp ref */ 359 /* chain no longer referenced */ 360 /* chain = NULL; not needed */ 361 362 /* 363 * XXX handle background sync when ip dirty, kernel will no longer 364 * notify us regarding this inode because there is no longer a 365 * vnode attached to it. 366 */ 367 368 return (0); 369 } 370 371 static 372 int 373 hammer2_vop_fsync(struct vop_fsync_args *ap) 374 { 375 hammer2_inode_t *ip; 376 hammer2_trans_t trans; 377 hammer2_chain_t *chain; 378 struct vnode *vp; 379 380 vp = ap->a_vp; 381 ip = VTOI(vp); 382 383 #if 0 384 /* XXX can't do this yet */ 385 hammer2_trans_init(&trans, ip->pmp, NULL, HAMMER2_TRANS_ISFLUSH); 386 vfsync(vp, ap->a_waitfor, 1, NULL, NULL); 387 hammer2_trans_clear_invfsync(&trans); 388 #endif 389 hammer2_trans_init(&trans, ip->pmp, NULL, 0); 390 vfsync(vp, ap->a_waitfor, 1, NULL, NULL); 391 392 /* 393 * Calling chain_flush here creates a lot of duplicative 394 * COW operations due to non-optimal vnode ordering. 395 * 396 * Only do it for an actual fsync() syscall. The other forms 397 * which call this function will eventually call chain_flush 398 * on the volume root as a catch-all, which is far more optimal. 399 */ 400 chain = hammer2_inode_lock_ex(ip); 401 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED); 402 vclrisdirty(vp); 403 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MTIME)) 404 hammer2_inode_fsync(&trans, ip, &chain); 405 406 #if 0 407 /* 408 * XXX creates discontinuity w/modify_tid 409 */ 410 if (ap->a_flags & VOP_FSYNC_SYSCALL) { 411 hammer2_chain_flush(&trans, &chain); 412 } 413 #endif 414 hammer2_inode_unlock_ex(ip, chain); 415 hammer2_trans_done(&trans); 416 417 return (0); 418 } 419 420 static 421 int 422 hammer2_vop_access(struct vop_access_args *ap) 423 { 424 hammer2_inode_t *ip = VTOI(ap->a_vp); 425 hammer2_inode_data_t *ipdata; 426 hammer2_chain_t *chain; 427 uid_t uid; 428 gid_t gid; 429 int error; 430 431 chain = hammer2_inode_lock_sh(ip); 432 ipdata = &chain->data->ipdata; 433 uid = hammer2_to_unix_xid(&ipdata->uid); 434 gid = hammer2_to_unix_xid(&ipdata->gid); 435 error = vop_helper_access(ap, uid, gid, ipdata->mode, ipdata->uflags); 436 hammer2_inode_unlock_sh(ip, chain); 437 438 return (error); 439 } 440 441 static 442 int 443 hammer2_vop_getattr(struct vop_getattr_args *ap) 444 { 445 hammer2_inode_data_t *ipdata; 446 hammer2_chain_t *chain; 447 hammer2_pfsmount_t *pmp; 448 hammer2_inode_t *ip; 449 struct vnode *vp; 450 struct vattr *vap; 451 452 vp = ap->a_vp; 453 vap = ap->a_vap; 454 455 ip = VTOI(vp); 456 pmp = ip->pmp; 457 458 chain = hammer2_inode_lock_sh(ip); 459 ipdata = &chain->data->ipdata; 460 461 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0]; 462 vap->va_fileid = ipdata->inum; 463 vap->va_mode = ipdata->mode; 464 vap->va_nlink = ipdata->nlinks; 465 vap->va_uid = hammer2_to_unix_xid(&ipdata->uid); 466 vap->va_gid = hammer2_to_unix_xid(&ipdata->gid); 467 vap->va_rmajor = 0; 468 vap->va_rminor = 0; 469 vap->va_size = ip->size; /* protected by shared lock */ 470 vap->va_blocksize = HAMMER2_PBUFSIZE; 471 vap->va_flags = ipdata->uflags; 472 hammer2_time_to_timespec(ipdata->ctime, &vap->va_ctime); 473 hammer2_time_to_timespec(ipdata->mtime, &vap->va_mtime); 474 hammer2_time_to_timespec(ipdata->mtime, &vap->va_atime); 475 vap->va_gen = 1; 476 vap->va_bytes = vap->va_size; /* XXX */ 477 vap->va_type = hammer2_get_vtype(chain); 478 vap->va_filerev = 0; 479 vap->va_uid_uuid = ipdata->uid; 480 vap->va_gid_uuid = ipdata->gid; 481 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID | 482 VA_FSID_UUID_VALID; 483 484 hammer2_inode_unlock_sh(ip, chain); 485 486 return (0); 487 } 488 489 static 490 int 491 hammer2_vop_setattr(struct vop_setattr_args *ap) 492 { 493 hammer2_inode_data_t *ipdata; 494 hammer2_inode_t *ip; 495 hammer2_chain_t *chain; 496 hammer2_trans_t trans; 497 struct vnode *vp; 498 struct vattr *vap; 499 int error; 500 int kflags = 0; 501 int domtime = 0; 502 uint64_t ctime; 503 504 vp = ap->a_vp; 505 vap = ap->a_vap; 506 hammer2_update_time(&ctime); 507 508 ip = VTOI(vp); 509 510 if (ip->pmp->ronly) 511 return(EROFS); 512 513 hammer2_chain_memory_wait(ip->pmp); 514 hammer2_trans_init(&trans, ip->pmp, NULL, 0); 515 chain = hammer2_inode_lock_ex(ip); 516 ipdata = &chain->data->ipdata; 517 error = 0; 518 519 if (vap->va_flags != VNOVAL) { 520 u_int32_t flags; 521 522 flags = ipdata->uflags; 523 error = vop_helper_setattr_flags(&flags, vap->va_flags, 524 hammer2_to_unix_xid(&ipdata->uid), 525 ap->a_cred); 526 if (error == 0) { 527 if (ipdata->uflags != flags) { 528 ipdata = hammer2_chain_modify_ip(&trans, ip, 529 &chain, 0); 530 ipdata->uflags = flags; 531 ipdata->ctime = ctime; 532 kflags |= NOTE_ATTRIB; 533 } 534 if (ipdata->uflags & (IMMUTABLE | APPEND)) { 535 error = 0; 536 goto done; 537 } 538 } 539 goto done; 540 } 541 if (ipdata->uflags & (IMMUTABLE | APPEND)) { 542 error = EPERM; 543 goto done; 544 } 545 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 546 mode_t cur_mode = ipdata->mode; 547 uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid); 548 gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid); 549 uuid_t uuid_uid; 550 uuid_t uuid_gid; 551 552 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid, 553 ap->a_cred, 554 &cur_uid, &cur_gid, &cur_mode); 555 if (error == 0) { 556 hammer2_guid_to_uuid(&uuid_uid, cur_uid); 557 hammer2_guid_to_uuid(&uuid_gid, cur_gid); 558 if (bcmp(&uuid_uid, &ipdata->uid, sizeof(uuid_uid)) || 559 bcmp(&uuid_gid, &ipdata->gid, sizeof(uuid_gid)) || 560 ipdata->mode != cur_mode 561 ) { 562 ipdata = hammer2_chain_modify_ip(&trans, ip, 563 &chain, 0); 564 ipdata->uid = uuid_uid; 565 ipdata->gid = uuid_gid; 566 ipdata->mode = cur_mode; 567 ipdata->ctime = ctime; 568 } 569 kflags |= NOTE_ATTRIB; 570 } 571 } 572 573 /* 574 * Resize the file 575 */ 576 if (vap->va_size != VNOVAL && ip->size != vap->va_size) { 577 switch(vp->v_type) { 578 case VREG: 579 if (vap->va_size == ip->size) 580 break; 581 hammer2_inode_unlock_ex(ip, chain); 582 if (vap->va_size < ip->size) { 583 hammer2_truncate_file(ip, vap->va_size); 584 } else { 585 hammer2_extend_file(ip, vap->va_size); 586 } 587 chain = hammer2_inode_lock_ex(ip); 588 ipdata = &chain->data->ipdata; /* RELOAD */ 589 domtime = 1; 590 break; 591 default: 592 error = EINVAL; 593 goto done; 594 } 595 } 596 #if 0 597 /* atime not supported */ 598 if (vap->va_atime.tv_sec != VNOVAL) { 599 ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0); 600 ipdata->atime = hammer2_timespec_to_time(&vap->va_atime); 601 kflags |= NOTE_ATTRIB; 602 } 603 #endif 604 if (vap->va_mtime.tv_sec != VNOVAL) { 605 ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0); 606 ipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime); 607 kflags |= NOTE_ATTRIB; 608 domtime = 0; 609 } 610 if (vap->va_mode != (mode_t)VNOVAL) { 611 mode_t cur_mode = ipdata->mode; 612 uid_t cur_uid = hammer2_to_unix_xid(&ipdata->uid); 613 gid_t cur_gid = hammer2_to_unix_xid(&ipdata->gid); 614 615 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred, 616 cur_uid, cur_gid, &cur_mode); 617 if (error == 0 && ipdata->mode != cur_mode) { 618 ipdata = hammer2_chain_modify_ip(&trans, ip, &chain, 0); 619 ipdata->mode = cur_mode; 620 ipdata->ctime = ctime; 621 kflags |= NOTE_ATTRIB; 622 } 623 } 624 625 /* 626 * If a truncation occurred we must call inode_fsync() now in order 627 * to trim the related data chains, otherwise a later expansion can 628 * cause havoc. 629 */ 630 hammer2_inode_fsync(&trans, ip, &chain); 631 632 /* 633 * Cleanup. If domtime is set an additional inode modification 634 * must be flagged. All other modifications will have already 635 * set INODE_MODIFIED and called vsetisdirty(). 636 */ 637 done: 638 if (domtime) { 639 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED | 640 HAMMER2_INODE_MTIME); 641 vsetisdirty(ip->vp); 642 } 643 hammer2_inode_unlock_ex(ip, chain); 644 hammer2_trans_done(&trans); 645 hammer2_knote(ip->vp, kflags); 646 647 return (error); 648 } 649 650 static 651 int 652 hammer2_vop_readdir(struct vop_readdir_args *ap) 653 { 654 hammer2_inode_data_t *ipdata; 655 hammer2_inode_t *ip; 656 hammer2_inode_t *xip; 657 hammer2_chain_t *parent; 658 hammer2_chain_t *chain; 659 hammer2_chain_t *xchain; 660 hammer2_tid_t inum; 661 hammer2_key_t key_next; 662 hammer2_key_t lkey; 663 struct uio *uio; 664 off_t *cookies; 665 off_t saveoff; 666 int cookie_index; 667 int cache_index = -1; 668 int ncookies; 669 int error; 670 int dtype; 671 int r; 672 673 ip = VTOI(ap->a_vp); 674 uio = ap->a_uio; 675 saveoff = uio->uio_offset; 676 677 /* 678 * Setup cookies directory entry cookies if requested 679 */ 680 if (ap->a_ncookies) { 681 ncookies = uio->uio_resid / 16 + 1; 682 if (ncookies > 1024) 683 ncookies = 1024; 684 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 685 } else { 686 ncookies = -1; 687 cookies = NULL; 688 } 689 cookie_index = 0; 690 691 parent = hammer2_inode_lock_sh(ip); 692 ipdata = &parent->data->ipdata; 693 694 /* 695 * Handle artificial entries. To ensure that only positive 64 bit 696 * quantities are returned to userland we always strip off bit 63. 697 * The hash code is designed such that codes 0x0000-0x7FFF are not 698 * used, allowing us to use these codes for articial entries. 699 * 700 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not 701 * allow '..' to cross the mount point into (e.g.) the super-root. 702 */ 703 error = 0; 704 chain = (void *)(intptr_t)-1; /* non-NULL for early goto done case */ 705 706 if (saveoff == 0) { 707 inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK; 708 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, "."); 709 if (r) 710 goto done; 711 if (cookies) 712 cookies[cookie_index] = saveoff; 713 ++saveoff; 714 ++cookie_index; 715 if (cookie_index == ncookies) 716 goto done; 717 } 718 719 if (saveoff == 1) { 720 /* 721 * Be careful with lockorder when accessing ".." 722 * 723 * (ip is the current dir. xip is the parent dir). 724 */ 725 inum = ipdata->inum & HAMMER2_DIRHASH_USERMSK; 726 while (ip->pip != NULL && ip != ip->pmp->iroot) { 727 xip = ip->pip; 728 hammer2_inode_ref(xip); 729 hammer2_inode_unlock_sh(ip, parent); 730 xchain = hammer2_inode_lock_sh(xip); 731 parent = hammer2_inode_lock_sh(ip); 732 hammer2_inode_drop(xip); 733 if (xip == ip->pip) { 734 inum = xchain->data->ipdata.inum & 735 HAMMER2_DIRHASH_USERMSK; 736 hammer2_inode_unlock_sh(xip, xchain); 737 break; 738 } 739 hammer2_inode_unlock_sh(xip, xchain); 740 } 741 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, ".."); 742 if (r) 743 goto done; 744 if (cookies) 745 cookies[cookie_index] = saveoff; 746 ++saveoff; 747 ++cookie_index; 748 if (cookie_index == ncookies) 749 goto done; 750 } 751 752 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE; 753 if (hammer2_debug & 0x0020) 754 kprintf("readdir: lkey %016jx\n", lkey); 755 756 /* 757 * parent is the inode chain, already locked for us. Don't 758 * double lock shared locks as this will screw up upgrades. 759 */ 760 if (error) { 761 goto done; 762 } 763 chain = hammer2_chain_lookup(&parent, &key_next, lkey, lkey, 764 &cache_index, HAMMER2_LOOKUP_SHARED); 765 if (chain == NULL) { 766 chain = hammer2_chain_lookup(&parent, &key_next, 767 lkey, (hammer2_key_t)-1, 768 &cache_index, 769 HAMMER2_LOOKUP_SHARED); 770 } 771 while (chain) { 772 if (hammer2_debug & 0x0020) 773 kprintf("readdir: p=%p chain=%p %016jx (next %016jx)\n", 774 parent, chain, chain->bref.key, key_next); 775 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) { 776 dtype = hammer2_get_dtype(chain); 777 saveoff = chain->bref.key & HAMMER2_DIRHASH_USERMSK; 778 r = vop_write_dirent(&error, uio, 779 chain->data->ipdata.inum & 780 HAMMER2_DIRHASH_USERMSK, 781 dtype, 782 chain->data->ipdata.name_len, 783 chain->data->ipdata.filename); 784 if (r) 785 break; 786 if (cookies) 787 cookies[cookie_index] = saveoff; 788 ++cookie_index; 789 } else { 790 /* XXX chain error */ 791 kprintf("bad chain type readdir %d\n", 792 chain->bref.type); 793 } 794 795 /* 796 * Keys may not be returned in order so once we have a 797 * placemarker (chain) the scan must allow the full range 798 * or some entries will be missed. 799 */ 800 chain = hammer2_chain_next(&parent, chain, &key_next, 801 key_next, (hammer2_key_t)-1, 802 &cache_index, HAMMER2_LOOKUP_SHARED); 803 if (chain) { 804 saveoff = (chain->bref.key & 805 HAMMER2_DIRHASH_USERMSK) + 1; 806 } else { 807 saveoff = (hammer2_key_t)-1; 808 } 809 if (cookie_index == ncookies) 810 break; 811 } 812 if (chain) 813 hammer2_chain_unlock(chain); 814 done: 815 hammer2_inode_unlock_sh(ip, parent); 816 if (ap->a_eofflag) 817 *ap->a_eofflag = (chain == NULL); 818 if (hammer2_debug & 0x0020) 819 kprintf("readdir: done at %016jx\n", saveoff); 820 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE; 821 if (error && cookie_index == 0) { 822 if (cookies) { 823 kfree(cookies, M_TEMP); 824 *ap->a_ncookies = 0; 825 *ap->a_cookies = NULL; 826 } 827 } else { 828 if (cookies) { 829 *ap->a_ncookies = cookie_index; 830 *ap->a_cookies = cookies; 831 } 832 } 833 return (error); 834 } 835 836 /* 837 * hammer2_vop_readlink { vp, uio, cred } 838 */ 839 static 840 int 841 hammer2_vop_readlink(struct vop_readlink_args *ap) 842 { 843 struct vnode *vp; 844 hammer2_inode_t *ip; 845 int error; 846 847 vp = ap->a_vp; 848 if (vp->v_type != VLNK) 849 return (EINVAL); 850 ip = VTOI(vp); 851 852 error = hammer2_read_file(ip, ap->a_uio, 0); 853 return (error); 854 } 855 856 static 857 int 858 hammer2_vop_read(struct vop_read_args *ap) 859 { 860 struct vnode *vp; 861 hammer2_inode_t *ip; 862 struct uio *uio; 863 int error; 864 int seqcount; 865 int bigread; 866 867 /* 868 * Read operations supported on this vnode? 869 */ 870 vp = ap->a_vp; 871 if (vp->v_type != VREG) 872 return (EINVAL); 873 874 /* 875 * Misc 876 */ 877 ip = VTOI(vp); 878 uio = ap->a_uio; 879 error = 0; 880 881 seqcount = ap->a_ioflag >> 16; 882 bigread = (uio->uio_resid > 100 * 1024 * 1024); 883 884 error = hammer2_read_file(ip, uio, seqcount); 885 return (error); 886 } 887 888 static 889 int 890 hammer2_vop_write(struct vop_write_args *ap) 891 { 892 hammer2_inode_t *ip; 893 hammer2_trans_t trans; 894 thread_t td; 895 struct vnode *vp; 896 struct uio *uio; 897 int error; 898 int seqcount; 899 int bigwrite; 900 901 /* 902 * Read operations supported on this vnode? 903 */ 904 vp = ap->a_vp; 905 if (vp->v_type != VREG) 906 return (EINVAL); 907 908 /* 909 * Misc 910 */ 911 ip = VTOI(vp); 912 uio = ap->a_uio; 913 error = 0; 914 if (ip->pmp->ronly) 915 return (EROFS); 916 917 seqcount = ap->a_ioflag >> 16; 918 bigwrite = (uio->uio_resid > 100 * 1024 * 1024); 919 920 /* 921 * Check resource limit 922 */ 923 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc && 924 uio->uio_offset + uio->uio_resid > 925 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 926 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ); 927 return (EFBIG); 928 } 929 930 bigwrite = (uio->uio_resid > 100 * 1024 * 1024); 931 932 /* 933 * The transaction interlocks against flushes initiations 934 * (note: but will run concurrently with the actual flush). 935 */ 936 hammer2_trans_init(&trans, ip->pmp, NULL, 0); 937 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount); 938 hammer2_trans_done(&trans); 939 940 return (error); 941 } 942 943 /* 944 * Perform read operations on a file or symlink given an UNLOCKED 945 * inode and uio. 946 * 947 * The passed ip is not locked. 948 */ 949 static 950 int 951 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount) 952 { 953 hammer2_off_t size; 954 struct buf *bp; 955 int error; 956 957 error = 0; 958 959 /* 960 * UIO read loop. 961 */ 962 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE); 963 size = ip->size; 964 ccms_thread_unlock(&ip->topo_cst); 965 966 while (uio->uio_resid > 0 && uio->uio_offset < size) { 967 hammer2_key_t lbase; 968 hammer2_key_t leof; 969 int lblksize; 970 int loff; 971 int n; 972 973 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 974 &lbase, &leof); 975 976 error = cluster_read(ip->vp, leof, lbase, lblksize, 977 uio->uio_resid, seqcount * BKVASIZE, 978 &bp); 979 980 if (error) 981 break; 982 loff = (int)(uio->uio_offset - lbase); 983 n = lblksize - loff; 984 if (n > uio->uio_resid) 985 n = uio->uio_resid; 986 if (n > size - uio->uio_offset) 987 n = (int)(size - uio->uio_offset); 988 bp->b_flags |= B_AGE; 989 uiomove((char *)bp->b_data + loff, n, uio); 990 bqrelse(bp); 991 } 992 return (error); 993 } 994 995 /* 996 * Write to the file represented by the inode via the logical buffer cache. 997 * The inode may represent a regular file or a symlink. 998 * 999 * The inode must not be locked. 1000 */ 1001 static 1002 int 1003 hammer2_write_file(hammer2_inode_t *ip, 1004 struct uio *uio, int ioflag, int seqcount) 1005 { 1006 hammer2_key_t old_eof; 1007 hammer2_key_t new_eof; 1008 struct buf *bp; 1009 int kflags; 1010 int error; 1011 int modified; 1012 1013 /* 1014 * Setup if append 1015 */ 1016 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE); 1017 if (ioflag & IO_APPEND) 1018 uio->uio_offset = ip->size; 1019 old_eof = ip->size; 1020 ccms_thread_unlock(&ip->topo_cst); 1021 1022 /* 1023 * Extend the file if necessary. If the write fails at some point 1024 * we will truncate it back down to cover as much as we were able 1025 * to write. 1026 * 1027 * Doing this now makes it easier to calculate buffer sizes in 1028 * the loop. 1029 */ 1030 kflags = 0; 1031 error = 0; 1032 modified = 0; 1033 1034 if (uio->uio_offset + uio->uio_resid > old_eof) { 1035 new_eof = uio->uio_offset + uio->uio_resid; 1036 modified = 1; 1037 hammer2_extend_file(ip, new_eof); 1038 kflags |= NOTE_EXTEND; 1039 } else { 1040 new_eof = old_eof; 1041 } 1042 1043 /* 1044 * UIO write loop 1045 */ 1046 while (uio->uio_resid > 0) { 1047 hammer2_key_t lbase; 1048 int trivial; 1049 int endofblk; 1050 int lblksize; 1051 int loff; 1052 int n; 1053 1054 /* 1055 * Don't allow the buffer build to blow out the buffer 1056 * cache. 1057 */ 1058 if ((ioflag & IO_RECURSE) == 0) 1059 bwillwrite(HAMMER2_PBUFSIZE); 1060 1061 /* 1062 * This nominally tells us how much we can cluster and 1063 * what the logical buffer size needs to be. Currently 1064 * we don't try to cluster the write and just handle one 1065 * block at a time. 1066 */ 1067 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 1068 &lbase, NULL); 1069 loff = (int)(uio->uio_offset - lbase); 1070 1071 KKASSERT(lblksize <= 65536); 1072 1073 /* 1074 * Calculate bytes to copy this transfer and whether the 1075 * copy completely covers the buffer or not. 1076 */ 1077 trivial = 0; 1078 n = lblksize - loff; 1079 if (n > uio->uio_resid) { 1080 n = uio->uio_resid; 1081 if (loff == lbase && uio->uio_offset + n == new_eof) 1082 trivial = 1; 1083 endofblk = 0; 1084 } else { 1085 if (loff == 0) 1086 trivial = 1; 1087 endofblk = 1; 1088 } 1089 1090 /* 1091 * Get the buffer 1092 */ 1093 if (uio->uio_segflg == UIO_NOCOPY) { 1094 /* 1095 * Issuing a write with the same data backing the 1096 * buffer. Instantiate the buffer to collect the 1097 * backing vm pages, then read-in any missing bits. 1098 * 1099 * This case is used by vop_stdputpages(). 1100 */ 1101 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0); 1102 if ((bp->b_flags & B_CACHE) == 0) { 1103 bqrelse(bp); 1104 error = bread(ip->vp, lbase, lblksize, &bp); 1105 } 1106 } else if (trivial) { 1107 /* 1108 * Even though we are entirely overwriting the buffer 1109 * we may still have to zero it out to avoid a 1110 * mmap/write visibility issue. 1111 */ 1112 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0); 1113 if ((bp->b_flags & B_CACHE) == 0) 1114 vfs_bio_clrbuf(bp); 1115 } else { 1116 /* 1117 * Partial overwrite, read in any missing bits then 1118 * replace the portion being written. 1119 * 1120 * (The strategy code will detect zero-fill physical 1121 * blocks for this case). 1122 */ 1123 error = bread(ip->vp, lbase, lblksize, &bp); 1124 if (error == 0) 1125 bheavy(bp); 1126 } 1127 1128 if (error) { 1129 brelse(bp); 1130 break; 1131 } 1132 1133 /* 1134 * Ok, copy the data in 1135 */ 1136 error = uiomove(bp->b_data + loff, n, uio); 1137 kflags |= NOTE_WRITE; 1138 modified = 1; 1139 if (error) { 1140 brelse(bp); 1141 break; 1142 } 1143 1144 /* 1145 * WARNING: Pageout daemon will issue UIO_NOCOPY writes 1146 * with IO_SYNC or IO_ASYNC set. These writes 1147 * must be handled as the pageout daemon expects. 1148 */ 1149 if (ioflag & IO_SYNC) { 1150 bwrite(bp); 1151 } else if ((ioflag & IO_DIRECT) && endofblk) { 1152 bawrite(bp); 1153 } else if (ioflag & IO_ASYNC) { 1154 bawrite(bp); 1155 } else { 1156 bdwrite(bp); 1157 } 1158 } 1159 1160 /* 1161 * Cleanup. If we extended the file EOF but failed to write through 1162 * the entire write is a failure and we have to back-up. 1163 */ 1164 if (error && new_eof != old_eof) { 1165 hammer2_truncate_file(ip, old_eof); 1166 } else if (modified) { 1167 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE); 1168 hammer2_update_time(&ip->mtime); 1169 atomic_set_int(&ip->flags, HAMMER2_INODE_MTIME); 1170 ccms_thread_unlock(&ip->topo_cst); 1171 } 1172 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED); 1173 hammer2_knote(ip->vp, kflags); 1174 vsetisdirty(ip->vp); 1175 1176 return error; 1177 } 1178 1179 /* 1180 * Truncate the size of a file. The inode must not be locked. 1181 * 1182 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED 1183 */ 1184 static 1185 void 1186 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1187 { 1188 hammer2_key_t lbase; 1189 int nblksize; 1190 1191 if (ip->vp) { 1192 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1193 nvtruncbuf(ip->vp, nsize, 1194 nblksize, (int)nsize & (nblksize - 1), 1195 0); 1196 } 1197 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE); 1198 ip->size = nsize; 1199 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1200 ccms_thread_unlock(&ip->topo_cst); 1201 } 1202 1203 /* 1204 * Extend the size of a file. The inode must not be locked. 1205 * 1206 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED 1207 */ 1208 static 1209 void 1210 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1211 { 1212 hammer2_key_t lbase; 1213 hammer2_key_t osize; 1214 int oblksize; 1215 int nblksize; 1216 1217 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE); 1218 osize = ip->size; 1219 ip->size = nsize; 1220 ccms_thread_unlock(&ip->topo_cst); 1221 1222 if (ip->vp) { 1223 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL); 1224 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1225 nvextendbuf(ip->vp, 1226 osize, nsize, 1227 oblksize, nblksize, 1228 -1, -1, 0); 1229 } 1230 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1231 } 1232 1233 static 1234 int 1235 hammer2_vop_nresolve(struct vop_nresolve_args *ap) 1236 { 1237 hammer2_inode_t *ip; 1238 hammer2_inode_t *dip; 1239 hammer2_chain_t *parent; 1240 hammer2_chain_t *chain; 1241 hammer2_chain_t *ochain; 1242 hammer2_trans_t trans; 1243 hammer2_key_t key_next; 1244 hammer2_key_t lhc; 1245 struct namecache *ncp; 1246 const uint8_t *name; 1247 size_t name_len; 1248 int error = 0; 1249 int cache_index = -1; 1250 struct vnode *vp; 1251 1252 dip = VTOI(ap->a_dvp); 1253 ncp = ap->a_nch->ncp; 1254 name = ncp->nc_name; 1255 name_len = ncp->nc_nlen; 1256 lhc = hammer2_dirhash(name, name_len); 1257 1258 /* 1259 * Note: In DragonFly the kernel handles '.' and '..'. 1260 */ 1261 parent = hammer2_inode_lock_sh(dip); 1262 chain = hammer2_chain_lookup(&parent, &key_next, 1263 lhc, lhc + HAMMER2_DIRHASH_LOMASK, 1264 &cache_index, HAMMER2_LOOKUP_SHARED); 1265 while (chain) { 1266 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && 1267 name_len == chain->data->ipdata.name_len && 1268 bcmp(name, chain->data->ipdata.filename, name_len) == 0) { 1269 break; 1270 } 1271 chain = hammer2_chain_next(&parent, chain, &key_next, 1272 key_next, 1273 lhc + HAMMER2_DIRHASH_LOMASK, 1274 &cache_index, HAMMER2_LOOKUP_SHARED); 1275 } 1276 hammer2_inode_unlock_sh(dip, parent); 1277 1278 /* 1279 * If the inode represents a forwarding entry for a hardlink we have 1280 * to locate the actual inode. The original ip is saved for possible 1281 * deconsolidation. (ip) will only be set to non-NULL when we have 1282 * to locate the real file via a hardlink. ip will be referenced but 1283 * not locked in that situation. chain is passed in locked and 1284 * returned locked. 1285 * 1286 * XXX what kind of chain lock? 1287 */ 1288 ochain = NULL; 1289 if (chain && chain->data->ipdata.type == HAMMER2_OBJTYPE_HARDLINK) { 1290 error = hammer2_hardlink_find(dip, &chain, &ochain); 1291 if (error) { 1292 kprintf("hammer2: unable to find hardlink\n"); 1293 if (chain) { 1294 hammer2_chain_unlock(chain); 1295 chain = NULL; 1296 } 1297 goto failed; 1298 } 1299 } 1300 1301 /* 1302 * Deconsolidate any hardlink whos nlinks == 1. Ignore errors. 1303 * If an error occurs chain and ip are left alone. 1304 * 1305 * XXX upgrade shared lock? 1306 */ 1307 if (ochain && chain && 1308 chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) { 1309 kprintf("hammer2: need to unconsolidate hardlink for %s\n", 1310 chain->data->ipdata.filename); 1311 /* XXX retain shared lock on dip? (currently not held) */ 1312 hammer2_trans_init(&trans, dip->pmp, NULL, 0); 1313 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain); 1314 hammer2_trans_done(&trans); 1315 } 1316 1317 /* 1318 * Acquire the related vnode 1319 * 1320 * NOTE: For error processing, only ENOENT resolves the namecache 1321 * entry to NULL, otherwise we just return the error and 1322 * leave the namecache unresolved. 1323 * 1324 * NOTE: multiple hammer2_inode structures can be aliased to the 1325 * same chain element, for example for hardlinks. This 1326 * use case does not 'reattach' inode associations that 1327 * might already exist, but always allocates a new one. 1328 * 1329 * WARNING: inode structure is locked exclusively via inode_get 1330 * but chain was locked shared. inode_unlock_ex() 1331 * will handle it properly. 1332 */ 1333 if (chain) { 1334 ip = hammer2_inode_get(dip->pmp, dip, chain); 1335 vp = hammer2_igetv(ip, &error); 1336 if (error == 0) { 1337 vn_unlock(vp); 1338 cache_setvp(ap->a_nch, vp); 1339 } else if (error == ENOENT) { 1340 cache_setvp(ap->a_nch, NULL); 1341 } 1342 hammer2_inode_unlock_ex(ip, chain); 1343 1344 /* 1345 * The vp should not be released until after we've disposed 1346 * of our locks, because it might cause vop_inactive() to 1347 * be called. 1348 */ 1349 if (vp) 1350 vrele(vp); 1351 } else { 1352 error = ENOENT; 1353 cache_setvp(ap->a_nch, NULL); 1354 } 1355 failed: 1356 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL, 1357 ("resolve error %d/%p chain %p ap %p\n", 1358 error, ap->a_nch->ncp->nc_vp, chain, ap)); 1359 if (ochain) 1360 hammer2_chain_drop(ochain); 1361 return error; 1362 } 1363 1364 static 1365 int 1366 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 1367 { 1368 hammer2_inode_t *dip; 1369 hammer2_inode_t *ip; 1370 hammer2_chain_t *parent; 1371 int error; 1372 1373 dip = VTOI(ap->a_dvp); 1374 1375 if ((ip = dip->pip) == NULL) { 1376 *ap->a_vpp = NULL; 1377 return ENOENT; 1378 } 1379 parent = hammer2_inode_lock_ex(ip); 1380 *ap->a_vpp = hammer2_igetv(ip, &error); 1381 hammer2_inode_unlock_ex(ip, parent); 1382 1383 return error; 1384 } 1385 1386 static 1387 int 1388 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap) 1389 { 1390 hammer2_inode_t *dip; 1391 hammer2_inode_t *nip; 1392 hammer2_trans_t trans; 1393 hammer2_chain_t *chain; 1394 struct namecache *ncp; 1395 const uint8_t *name; 1396 size_t name_len; 1397 int error; 1398 1399 dip = VTOI(ap->a_dvp); 1400 if (dip->pmp->ronly) 1401 return (EROFS); 1402 1403 ncp = ap->a_nch->ncp; 1404 name = ncp->nc_name; 1405 name_len = ncp->nc_nlen; 1406 1407 hammer2_chain_memory_wait(dip->pmp); 1408 hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE); 1409 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred, 1410 name, name_len, &chain, &error); 1411 chain->inode_reason = 1; 1412 if (error) { 1413 KKASSERT(nip == NULL); 1414 *ap->a_vpp = NULL; 1415 } else { 1416 *ap->a_vpp = hammer2_igetv(nip, &error); 1417 hammer2_inode_unlock_ex(nip, chain); 1418 } 1419 hammer2_trans_done(&trans); 1420 1421 if (error == 0) { 1422 cache_setunresolved(ap->a_nch); 1423 cache_setvp(ap->a_nch, *ap->a_vpp); 1424 } 1425 return error; 1426 } 1427 1428 /* 1429 * Return the largest contiguous physical disk range for the logical 1430 * request, in bytes. 1431 * 1432 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb) 1433 * 1434 * Basically disabled, the logical buffer write thread has to deal with 1435 * buffers one-at-a-time. 1436 */ 1437 static 1438 int 1439 hammer2_vop_bmap(struct vop_bmap_args *ap) 1440 { 1441 *ap->a_doffsetp = NOOFFSET; 1442 if (ap->a_runp) 1443 *ap->a_runp = 0; 1444 if (ap->a_runb) 1445 *ap->a_runb = 0; 1446 return (EOPNOTSUPP); 1447 } 1448 1449 static 1450 int 1451 hammer2_vop_open(struct vop_open_args *ap) 1452 { 1453 return vop_stdopen(ap); 1454 } 1455 1456 /* 1457 * hammer2_vop_advlock { vp, id, op, fl, flags } 1458 */ 1459 static 1460 int 1461 hammer2_vop_advlock(struct vop_advlock_args *ap) 1462 { 1463 hammer2_inode_t *ip = VTOI(ap->a_vp); 1464 hammer2_chain_t *parent; 1465 hammer2_off_t size; 1466 1467 parent = hammer2_inode_lock_sh(ip); 1468 size = parent->data->ipdata.size; 1469 hammer2_inode_unlock_sh(ip, parent); 1470 return (lf_advlock(ap, &ip->advlock, size)); 1471 } 1472 1473 1474 static 1475 int 1476 hammer2_vop_close(struct vop_close_args *ap) 1477 { 1478 return vop_stdclose(ap); 1479 } 1480 1481 /* 1482 * hammer2_vop_nlink { nch, dvp, vp, cred } 1483 * 1484 * Create a hardlink from (vp) to {dvp, nch}. 1485 */ 1486 static 1487 int 1488 hammer2_vop_nlink(struct vop_nlink_args *ap) 1489 { 1490 hammer2_inode_t *fdip; /* target directory to create link in */ 1491 hammer2_inode_t *tdip; /* target directory to create link in */ 1492 hammer2_inode_t *cdip; /* common parent directory */ 1493 hammer2_inode_t *ip; /* inode we are hardlinking to */ 1494 hammer2_chain_t *chain; 1495 hammer2_chain_t *fdchain; 1496 hammer2_chain_t *tdchain; 1497 hammer2_chain_t *cdchain; 1498 hammer2_trans_t trans; 1499 struct namecache *ncp; 1500 const uint8_t *name; 1501 size_t name_len; 1502 int error; 1503 1504 tdip = VTOI(ap->a_dvp); 1505 if (tdip->pmp->ronly) 1506 return (EROFS); 1507 1508 ncp = ap->a_nch->ncp; 1509 name = ncp->nc_name; 1510 name_len = ncp->nc_nlen; 1511 1512 /* 1513 * ip represents the file being hardlinked. The file could be a 1514 * normal file or a hardlink target if it has already been hardlinked. 1515 * If ip is a hardlinked target then ip->pip represents the location 1516 * of the hardlinked target, NOT the location of the hardlink pointer. 1517 * 1518 * Bump nlinks and potentially also create or move the hardlink 1519 * target in the parent directory common to (ip) and (tdip). The 1520 * consolidation code can modify ip->chain and ip->pip. The 1521 * returned chain is locked. 1522 */ 1523 ip = VTOI(ap->a_vp); 1524 hammer2_chain_memory_wait(ip->pmp); 1525 hammer2_trans_init(&trans, ip->pmp, NULL, HAMMER2_TRANS_NEWINODE); 1526 1527 /* 1528 * The common parent directory must be locked first to avoid deadlocks. 1529 * Also note that fdip and/or tdip might match cdip. 1530 */ 1531 fdip = ip->pip; 1532 cdip = hammer2_inode_common_parent(fdip, tdip); 1533 cdchain = hammer2_inode_lock_ex(cdip); 1534 fdchain = hammer2_inode_lock_ex(fdip); 1535 tdchain = hammer2_inode_lock_ex(tdip); 1536 chain = hammer2_inode_lock_ex(ip); 1537 error = hammer2_hardlink_consolidate(&trans, ip, &chain, 1538 cdip, &cdchain, 1); 1539 if (error) 1540 goto done; 1541 1542 /* 1543 * Create a directory entry connected to the specified chain. 1544 * The hardlink consolidation code has already adjusted ip->pip 1545 * to the common parent directory containing the actual hardlink 1546 * 1547 * (which may be different from dip where we created our hardlink 1548 * entry. ip->chain always represents the actual hardlink and not 1549 * any of the pointers to the actual hardlink). 1550 * 1551 * WARNING! chain can get moved by the connect (indirectly due to 1552 * potential indirect block creation). 1553 */ 1554 error = hammer2_inode_connect(&trans, &chain, 1, 1555 tdip, &tdchain, 1556 name, name_len, 0); 1557 if (error == 0) { 1558 cache_setunresolved(ap->a_nch); 1559 cache_setvp(ap->a_nch, ap->a_vp); 1560 } 1561 done: 1562 hammer2_inode_unlock_ex(ip, chain); 1563 hammer2_inode_unlock_ex(tdip, tdchain); 1564 hammer2_inode_unlock_ex(fdip, fdchain); 1565 hammer2_inode_unlock_ex(cdip, cdchain); 1566 hammer2_trans_done(&trans); 1567 1568 return error; 1569 } 1570 1571 /* 1572 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap } 1573 * 1574 * The operating system has already ensured that the directory entry 1575 * does not exist and done all appropriate namespace locking. 1576 */ 1577 static 1578 int 1579 hammer2_vop_ncreate(struct vop_ncreate_args *ap) 1580 { 1581 hammer2_inode_t *dip; 1582 hammer2_inode_t *nip; 1583 hammer2_trans_t trans; 1584 hammer2_chain_t *nchain; 1585 struct namecache *ncp; 1586 const uint8_t *name; 1587 size_t name_len; 1588 int error; 1589 1590 dip = VTOI(ap->a_dvp); 1591 if (dip->pmp->ronly) 1592 return (EROFS); 1593 1594 ncp = ap->a_nch->ncp; 1595 name = ncp->nc_name; 1596 name_len = ncp->nc_nlen; 1597 hammer2_chain_memory_wait(dip->pmp); 1598 hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE); 1599 1600 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred, 1601 name, name_len, &nchain, &error); 1602 nchain->inode_reason = 2; 1603 if (error) { 1604 KKASSERT(nip == NULL); 1605 *ap->a_vpp = NULL; 1606 } else { 1607 *ap->a_vpp = hammer2_igetv(nip, &error); 1608 hammer2_inode_unlock_ex(nip, nchain); 1609 } 1610 hammer2_trans_done(&trans); 1611 1612 if (error == 0) { 1613 cache_setunresolved(ap->a_nch); 1614 cache_setvp(ap->a_nch, *ap->a_vpp); 1615 } 1616 return error; 1617 } 1618 1619 /* 1620 * 1621 */ 1622 static 1623 int 1624 hammer2_vop_nmknod(struct vop_nmknod_args *ap) 1625 { 1626 hammer2_inode_t *dip; 1627 hammer2_inode_t *nip; 1628 hammer2_trans_t trans; 1629 hammer2_chain_t *nchain; 1630 struct namecache *ncp; 1631 const uint8_t *name; 1632 size_t name_len; 1633 int error; 1634 1635 dip = VTOI(ap->a_dvp); 1636 if (dip->pmp->ronly) 1637 return (EROFS); 1638 1639 ncp = ap->a_nch->ncp; 1640 name = ncp->nc_name; 1641 name_len = ncp->nc_nlen; 1642 hammer2_chain_memory_wait(dip->pmp); 1643 hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE); 1644 1645 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred, 1646 name, name_len, &nchain, &error); 1647 nchain->inode_reason = 3; 1648 if (error) { 1649 KKASSERT(nip == NULL); 1650 *ap->a_vpp = NULL; 1651 } else { 1652 *ap->a_vpp = hammer2_igetv(nip, &error); 1653 hammer2_inode_unlock_ex(nip, nchain); 1654 } 1655 hammer2_trans_done(&trans); 1656 1657 if (error == 0) { 1658 cache_setunresolved(ap->a_nch); 1659 cache_setvp(ap->a_nch, *ap->a_vpp); 1660 } 1661 return error; 1662 } 1663 1664 /* 1665 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target } 1666 */ 1667 static 1668 int 1669 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap) 1670 { 1671 hammer2_inode_t *dip; 1672 hammer2_inode_t *nip; 1673 hammer2_chain_t *nparent; 1674 hammer2_trans_t trans; 1675 struct namecache *ncp; 1676 const uint8_t *name; 1677 size_t name_len; 1678 int error; 1679 1680 dip = VTOI(ap->a_dvp); 1681 if (dip->pmp->ronly) 1682 return (EROFS); 1683 1684 ncp = ap->a_nch->ncp; 1685 name = ncp->nc_name; 1686 name_len = ncp->nc_nlen; 1687 hammer2_chain_memory_wait(dip->pmp); 1688 hammer2_trans_init(&trans, dip->pmp, NULL, HAMMER2_TRANS_NEWINODE); 1689 1690 ap->a_vap->va_type = VLNK; /* enforce type */ 1691 1692 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred, 1693 name, name_len, &nparent, &error); 1694 nparent->inode_reason = 4; 1695 if (error) { 1696 KKASSERT(nip == NULL); 1697 *ap->a_vpp = NULL; 1698 hammer2_trans_done(&trans); 1699 return error; 1700 } 1701 *ap->a_vpp = hammer2_igetv(nip, &error); 1702 1703 /* 1704 * Build the softlink (~like file data) and finalize the namecache. 1705 */ 1706 if (error == 0) { 1707 size_t bytes; 1708 struct uio auio; 1709 struct iovec aiov; 1710 hammer2_inode_data_t *nipdata; 1711 1712 nipdata = &nip->chain->data->ipdata; 1713 bytes = strlen(ap->a_target); 1714 1715 if (bytes <= HAMMER2_EMBEDDED_BYTES) { 1716 KKASSERT(nipdata->op_flags & 1717 HAMMER2_OPFLAG_DIRECTDATA); 1718 bcopy(ap->a_target, nipdata->u.data, bytes); 1719 nipdata->size = bytes; 1720 nip->size = bytes; 1721 hammer2_inode_unlock_ex(nip, nparent); 1722 } else { 1723 hammer2_inode_unlock_ex(nip, nparent); 1724 bzero(&auio, sizeof(auio)); 1725 bzero(&aiov, sizeof(aiov)); 1726 auio.uio_iov = &aiov; 1727 auio.uio_segflg = UIO_SYSSPACE; 1728 auio.uio_rw = UIO_WRITE; 1729 auio.uio_resid = bytes; 1730 auio.uio_iovcnt = 1; 1731 auio.uio_td = curthread; 1732 aiov.iov_base = ap->a_target; 1733 aiov.iov_len = bytes; 1734 error = hammer2_write_file(nip, &auio, IO_APPEND, 0); 1735 nipdata = &nip->chain->data->ipdata; /* RELOAD */ 1736 /* XXX handle error */ 1737 error = 0; 1738 } 1739 } else { 1740 hammer2_inode_unlock_ex(nip, nparent); 1741 } 1742 hammer2_trans_done(&trans); 1743 1744 /* 1745 * Finalize namecache 1746 */ 1747 if (error == 0) { 1748 cache_setunresolved(ap->a_nch); 1749 cache_setvp(ap->a_nch, *ap->a_vpp); 1750 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */ 1751 } 1752 return error; 1753 } 1754 1755 /* 1756 * hammer2_vop_nremove { nch, dvp, cred } 1757 */ 1758 static 1759 int 1760 hammer2_vop_nremove(struct vop_nremove_args *ap) 1761 { 1762 hammer2_inode_t *dip; 1763 hammer2_trans_t trans; 1764 struct namecache *ncp; 1765 const uint8_t *name; 1766 size_t name_len; 1767 int error; 1768 1769 dip = VTOI(ap->a_dvp); 1770 if (dip->pmp->ronly) 1771 return(EROFS); 1772 1773 ncp = ap->a_nch->ncp; 1774 name = ncp->nc_name; 1775 name_len = ncp->nc_nlen; 1776 1777 hammer2_chain_memory_wait(dip->pmp); 1778 hammer2_trans_init(&trans, dip->pmp, NULL, 0); 1779 error = hammer2_unlink_file(&trans, dip, name, name_len, 1780 0, NULL, ap->a_nch); 1781 hammer2_trans_done(&trans); 1782 if (error == 0) 1783 cache_unlink(ap->a_nch); 1784 return (error); 1785 } 1786 1787 /* 1788 * hammer2_vop_nrmdir { nch, dvp, cred } 1789 */ 1790 static 1791 int 1792 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap) 1793 { 1794 hammer2_inode_t *dip; 1795 hammer2_trans_t trans; 1796 struct namecache *ncp; 1797 const uint8_t *name; 1798 size_t name_len; 1799 int error; 1800 1801 dip = VTOI(ap->a_dvp); 1802 if (dip->pmp->ronly) 1803 return(EROFS); 1804 1805 ncp = ap->a_nch->ncp; 1806 name = ncp->nc_name; 1807 name_len = ncp->nc_nlen; 1808 1809 hammer2_chain_memory_wait(dip->pmp); 1810 hammer2_trans_init(&trans, dip->pmp, NULL, 0); 1811 error = hammer2_unlink_file(&trans, dip, name, name_len, 1812 1, NULL, ap->a_nch); 1813 hammer2_trans_done(&trans); 1814 if (error == 0) 1815 cache_unlink(ap->a_nch); 1816 return (error); 1817 } 1818 1819 /* 1820 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred } 1821 */ 1822 static 1823 int 1824 hammer2_vop_nrename(struct vop_nrename_args *ap) 1825 { 1826 struct namecache *fncp; 1827 struct namecache *tncp; 1828 hammer2_inode_t *cdip; 1829 hammer2_inode_t *fdip; 1830 hammer2_inode_t *tdip; 1831 hammer2_inode_t *ip; 1832 hammer2_chain_t *chain; 1833 hammer2_chain_t *fdchain; 1834 hammer2_chain_t *tdchain; 1835 hammer2_chain_t *cdchain; 1836 hammer2_trans_t trans; 1837 const uint8_t *fname; 1838 size_t fname_len; 1839 const uint8_t *tname; 1840 size_t tname_len; 1841 int error; 1842 int hlink; 1843 1844 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount) 1845 return(EXDEV); 1846 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount) 1847 return(EXDEV); 1848 1849 fdip = VTOI(ap->a_fdvp); /* source directory */ 1850 tdip = VTOI(ap->a_tdvp); /* target directory */ 1851 1852 if (fdip->pmp->ronly) 1853 return(EROFS); 1854 1855 fncp = ap->a_fnch->ncp; /* entry name in source */ 1856 fname = fncp->nc_name; 1857 fname_len = fncp->nc_nlen; 1858 1859 tncp = ap->a_tnch->ncp; /* entry name in target */ 1860 tname = tncp->nc_name; 1861 tname_len = tncp->nc_nlen; 1862 1863 hammer2_chain_memory_wait(tdip->pmp); 1864 hammer2_trans_init(&trans, tdip->pmp, NULL, 0); 1865 1866 /* 1867 * ip is the inode being renamed. If this is a hardlink then 1868 * ip represents the actual file and not the hardlink marker. 1869 */ 1870 ip = VTOI(fncp->nc_vp); 1871 chain = NULL; 1872 1873 1874 /* 1875 * The common parent directory must be locked first to avoid deadlocks. 1876 * Also note that fdip and/or tdip might match cdip. 1877 * 1878 * WARNING! fdip may not match ip->pip. That is, if the source file 1879 * is already a hardlink then what we are renaming is the 1880 * hardlink pointer, not the hardlink itself. The hardlink 1881 * directory (ip->pip) will already be at a common parent 1882 * of fdrip. 1883 * 1884 * Be sure to use ip->pip when finding the common parent 1885 * against tdip or we might accidently move the hardlink 1886 * target into a subdirectory that makes it inaccessible to 1887 * other pointers. 1888 */ 1889 cdip = hammer2_inode_common_parent(ip->pip, tdip); 1890 cdchain = hammer2_inode_lock_ex(cdip); 1891 fdchain = hammer2_inode_lock_ex(fdip); 1892 tdchain = hammer2_inode_lock_ex(tdip); 1893 1894 /* 1895 * Keep a tight grip on the inode so the temporary unlinking from 1896 * the source location prior to linking to the target location 1897 * does not cause the chain to be destroyed. 1898 * 1899 * NOTE: To avoid deadlocks we cannot lock (ip) while we are 1900 * unlinking elements from their directories. Locking 1901 * the nlinks field does not lock the whole inode. 1902 */ 1903 hammer2_inode_ref(ip); 1904 1905 /* 1906 * Remove target if it exists 1907 */ 1908 error = hammer2_unlink_file(&trans, tdip, tname, tname_len, 1909 -1, NULL, ap->a_tnch); 1910 if (error && error != ENOENT) 1911 goto done; 1912 cache_setunresolved(ap->a_tnch); 1913 1914 /* 1915 * When renaming a hardlinked file we may have to re-consolidate 1916 * the location of the hardlink target. Also adjust nlinks by +1 1917 * to counter-act the unlink below. 1918 * 1919 * If ip represents a regular file the consolidation code essentially 1920 * does nothing other than return the same locked chain that was 1921 * passed in. 1922 * 1923 * The returned chain will be locked. 1924 * 1925 * WARNING! We do not currently have a local copy of ipdata but 1926 * we do use one later remember that it must be reloaded 1927 * on any modification to the inode, including connects. 1928 */ 1929 chain = hammer2_inode_lock_ex(ip); 1930 error = hammer2_hardlink_consolidate(&trans, ip, &chain, 1931 cdip, &cdchain, 1); 1932 if (error) 1933 goto done; 1934 1935 /* 1936 * Disconnect (fdip, fname) from the source directory. This will 1937 * disconnect (ip) if it represents a direct file. If (ip) represents 1938 * a hardlink the HARDLINK pointer object will be removed but the 1939 * hardlink will stay intact. 1940 * 1941 * Always pass nch as NULL because we intend to reconnect the inode, 1942 * so we don't want hammer2_unlink_file() to rename it to the hidden 1943 * open-but-unlinked directory. 1944 * 1945 * The target chain may be marked DELETED but will not be destroyed 1946 * since we retain our hold on ip and chain. 1947 */ 1948 error = hammer2_unlink_file(&trans, fdip, fname, fname_len, 1949 -1, &hlink, NULL); 1950 KKASSERT(error != EAGAIN); 1951 if (error) 1952 goto done; 1953 1954 /* 1955 * Reconnect ip to target directory using chain. Chains cannot 1956 * actually be moved, so this will duplicate the chain in the new 1957 * spot and assign it to the ip, replacing the old chain. 1958 * 1959 * WARNING: Because recursive locks are allowed and we unlinked the 1960 * file that we have a chain-in-hand for just above, the 1961 * chain might have been delete-duplicated. We must refactor 1962 * the chain. 1963 * 1964 * WARNING: Chain locks can lock buffer cache buffers, to avoid 1965 * deadlocks we want to unlock before issuing a cache_*() 1966 * op (that might have to lock a vnode). 1967 */ 1968 hammer2_chain_refactor(&chain); 1969 error = hammer2_inode_connect(&trans, &chain, hlink, 1970 tdip, &tdchain, 1971 tname, tname_len, 0); 1972 chain->inode_reason = 5; 1973 if (error == 0) { 1974 KKASSERT(chain != NULL); 1975 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), chain); 1976 } 1977 done: 1978 hammer2_inode_unlock_ex(ip, chain); 1979 hammer2_inode_unlock_ex(tdip, tdchain); 1980 hammer2_inode_unlock_ex(fdip, fdchain); 1981 hammer2_inode_unlock_ex(cdip, cdchain); 1982 hammer2_inode_drop(ip); 1983 hammer2_trans_done(&trans); 1984 1985 /* 1986 * Issue the namecache update after unlocking all the internal 1987 * hammer structures, otherwise we might deadlock. 1988 */ 1989 if (error == 0) 1990 cache_rename(ap->a_fnch, ap->a_tnch); 1991 1992 return (error); 1993 } 1994 1995 /* 1996 * Strategy code 1997 * 1998 * WARNING: The strategy code cannot safely use hammer2 transactions 1999 * as this can deadlock against vfs_sync's vfsync() call 2000 * if multiple flushes are queued. 2001 */ 2002 static int hammer2_strategy_read(struct vop_strategy_args *ap); 2003 static int hammer2_strategy_write(struct vop_strategy_args *ap); 2004 static void hammer2_strategy_read_callback(hammer2_io_t *dio, 2005 hammer2_chain_t *chain, 2006 void *arg_p, off_t arg_o); 2007 2008 static 2009 int 2010 hammer2_vop_strategy(struct vop_strategy_args *ap) 2011 { 2012 struct bio *biop; 2013 struct buf *bp; 2014 int error; 2015 2016 biop = ap->a_bio; 2017 bp = biop->bio_buf; 2018 2019 switch(bp->b_cmd) { 2020 case BUF_CMD_READ: 2021 error = hammer2_strategy_read(ap); 2022 ++hammer2_iod_file_read; 2023 break; 2024 case BUF_CMD_WRITE: 2025 error = hammer2_strategy_write(ap); 2026 ++hammer2_iod_file_write; 2027 break; 2028 default: 2029 bp->b_error = error = EINVAL; 2030 bp->b_flags |= B_ERROR; 2031 biodone(biop); 2032 break; 2033 } 2034 2035 return (error); 2036 } 2037 2038 static 2039 int 2040 hammer2_strategy_read(struct vop_strategy_args *ap) 2041 { 2042 struct buf *bp; 2043 struct bio *bio; 2044 struct bio *nbio; 2045 hammer2_inode_t *ip; 2046 hammer2_chain_t *parent; 2047 hammer2_chain_t *chain; 2048 hammer2_key_t key_dummy; 2049 hammer2_key_t lbase; 2050 int cache_index = -1; 2051 2052 bio = ap->a_bio; 2053 bp = bio->bio_buf; 2054 ip = VTOI(ap->a_vp); 2055 nbio = push_bio(bio); 2056 2057 lbase = bio->bio_offset; 2058 chain = NULL; 2059 KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0); 2060 2061 parent = hammer2_inode_lock_sh(ip); 2062 chain = hammer2_chain_lookup(&parent, &key_dummy, 2063 lbase, lbase, 2064 &cache_index, 2065 HAMMER2_LOOKUP_NODATA | 2066 HAMMER2_LOOKUP_SHARED); 2067 2068 if (chain == NULL) { 2069 /* 2070 * Data is zero-fill 2071 */ 2072 bp->b_resid = 0; 2073 bp->b_error = 0; 2074 bzero(bp->b_data, bp->b_bcount); 2075 biodone(nbio); 2076 } else if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) { 2077 /* 2078 * Data is embedded in the inode (copy from inode). 2079 */ 2080 hammer2_chain_load_async(chain, 2081 hammer2_strategy_read_callback, 2082 nbio, 0); 2083 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) { 2084 /* 2085 * Data is on-media, issue device I/O and copy. 2086 * 2087 * XXX direct-IO shortcut could go here XXX. 2088 */ 2089 if (HAMMER2_DEC_COMP(chain->bref.methods) == HAMMER2_COMP_LZ4) { 2090 /* 2091 * Block compression is determined by bref.methods 2092 */ 2093 hammer2_blockref_t *bref; 2094 2095 bref = &chain->bref; 2096 hammer2_io_breadcb(chain->hmp, bref->data_off, 2097 chain->bytes, 2098 hammer2_decompress_LZ4_callback, 2099 NULL, nbio, bref->data_off); 2100 /* XXX async read dev blk not protected by chain lk */ 2101 hammer2_chain_unlock(chain); 2102 } else if (HAMMER2_DEC_COMP(chain->bref.methods) == 2103 HAMMER2_COMP_ZLIB) { 2104 hammer2_blockref_t *bref; 2105 2106 bref = &chain->bref; 2107 hammer2_io_breadcb(chain->hmp, bref->data_off, 2108 chain->bytes, 2109 hammer2_decompress_ZLIB_callback, 2110 NULL, nbio, bref->data_off); 2111 /* XXX async read dev blk not protected by chain lk */ 2112 hammer2_chain_unlock(chain); 2113 } else { 2114 hammer2_chain_load_async(chain, 2115 hammer2_strategy_read_callback, 2116 nbio, 0); 2117 } 2118 } else { 2119 panic("READ PATH: hammer2_strategy_read: unknown bref type"); 2120 chain = NULL; 2121 } 2122 hammer2_inode_unlock_sh(ip, parent); 2123 return (0); 2124 } 2125 2126 /* 2127 * Read callback for block that is not compressed. 2128 */ 2129 static 2130 void 2131 hammer2_strategy_read_callback(hammer2_io_t *dio, hammer2_chain_t *chain, 2132 void *arg_p, off_t arg_o __unused) 2133 { 2134 struct bio *nbio = arg_p; 2135 struct buf *bp = nbio->bio_buf; 2136 char *data; 2137 2138 if (dio) 2139 data = hammer2_io_data(dio, chain->bref.data_off); 2140 else 2141 data = (void *)chain->data; 2142 2143 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) { 2144 /* 2145 * Data is embedded in the inode (copy from inode). 2146 */ 2147 bcopy(((hammer2_inode_data_t *)data)->u.data, 2148 bp->b_data, HAMMER2_EMBEDDED_BYTES); 2149 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES, 2150 bp->b_bcount - HAMMER2_EMBEDDED_BYTES); 2151 bp->b_resid = 0; 2152 bp->b_error = 0; 2153 hammer2_chain_unlock(chain); 2154 biodone(nbio); 2155 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) { 2156 /* 2157 * Data is on-media, issue device I/O and copy. 2158 * 2159 * XXX direct-IO shortcut could go here XXX. 2160 */ 2161 KKASSERT(chain->bytes <= bp->b_bcount); 2162 bcopy(data, bp->b_data, chain->bytes); 2163 if (chain->bytes < bp->b_bcount); { 2164 bzero(bp->b_data + chain->bytes, 2165 bp->b_bcount - chain->bytes); 2166 } 2167 bp->b_flags |= B_NOTMETA; 2168 bp->b_resid = 0; 2169 bp->b_error = 0; 2170 hammer2_chain_unlock(chain); 2171 biodone(nbio); 2172 } else { 2173 /* bqrelse the dio to help stabilize the call to panic() */ 2174 if (dio) 2175 hammer2_io_bqrelse(&dio); 2176 panic("hammer2_strategy_read: unknown bref type"); 2177 /*hammer2_chain_unlock(chain);*/ 2178 /*chain = NULL;*/ 2179 } 2180 } 2181 2182 static 2183 int 2184 hammer2_strategy_write(struct vop_strategy_args *ap) 2185 { 2186 hammer2_pfsmount_t *pmp; 2187 struct bio *bio; 2188 struct buf *bp; 2189 hammer2_inode_t *ip; 2190 2191 bio = ap->a_bio; 2192 bp = bio->bio_buf; 2193 ip = VTOI(ap->a_vp); 2194 pmp = ip->pmp; 2195 2196 hammer2_lwinprog_ref(pmp); 2197 mtx_lock(&pmp->wthread_mtx); 2198 if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) { 2199 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio); 2200 mtx_unlock(&pmp->wthread_mtx); 2201 wakeup(&pmp->wthread_bioq); 2202 } else { 2203 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio); 2204 mtx_unlock(&pmp->wthread_mtx); 2205 } 2206 hammer2_lwinprog_wait(pmp); 2207 2208 return(0); 2209 } 2210 2211 /* 2212 * hammer2_vop_ioctl { vp, command, data, fflag, cred } 2213 */ 2214 static 2215 int 2216 hammer2_vop_ioctl(struct vop_ioctl_args *ap) 2217 { 2218 hammer2_inode_t *ip; 2219 int error; 2220 2221 ip = VTOI(ap->a_vp); 2222 2223 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data, 2224 ap->a_fflag, ap->a_cred); 2225 return (error); 2226 } 2227 2228 static 2229 int 2230 hammer2_vop_mountctl(struct vop_mountctl_args *ap) 2231 { 2232 struct mount *mp; 2233 hammer2_pfsmount_t *pmp; 2234 int rc; 2235 2236 switch (ap->a_op) { 2237 case (MOUNTCTL_SET_EXPORT): 2238 mp = ap->a_head.a_ops->head.vv_mount; 2239 pmp = MPTOPMP(mp); 2240 2241 if (ap->a_ctllen != sizeof(struct export_args)) 2242 rc = (EINVAL); 2243 else 2244 rc = vfs_export(mp, &pmp->export, 2245 (const struct export_args *)ap->a_ctl); 2246 break; 2247 default: 2248 rc = vop_stdmountctl(ap); 2249 break; 2250 } 2251 return (rc); 2252 } 2253 2254 /* 2255 * KQFILTER 2256 */ 2257 static void filt_hammer2detach(struct knote *kn); 2258 static int filt_hammer2read(struct knote *kn, long hint); 2259 static int filt_hammer2write(struct knote *kn, long hint); 2260 static int filt_hammer2vnode(struct knote *kn, long hint); 2261 2262 static struct filterops hammer2read_filtops = 2263 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2264 NULL, filt_hammer2detach, filt_hammer2read }; 2265 static struct filterops hammer2write_filtops = 2266 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2267 NULL, filt_hammer2detach, filt_hammer2write }; 2268 static struct filterops hammer2vnode_filtops = 2269 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2270 NULL, filt_hammer2detach, filt_hammer2vnode }; 2271 2272 static 2273 int 2274 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap) 2275 { 2276 struct vnode *vp = ap->a_vp; 2277 struct knote *kn = ap->a_kn; 2278 2279 switch (kn->kn_filter) { 2280 case EVFILT_READ: 2281 kn->kn_fop = &hammer2read_filtops; 2282 break; 2283 case EVFILT_WRITE: 2284 kn->kn_fop = &hammer2write_filtops; 2285 break; 2286 case EVFILT_VNODE: 2287 kn->kn_fop = &hammer2vnode_filtops; 2288 break; 2289 default: 2290 return (EOPNOTSUPP); 2291 } 2292 2293 kn->kn_hook = (caddr_t)vp; 2294 2295 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2296 2297 return(0); 2298 } 2299 2300 static void 2301 filt_hammer2detach(struct knote *kn) 2302 { 2303 struct vnode *vp = (void *)kn->kn_hook; 2304 2305 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2306 } 2307 2308 static int 2309 filt_hammer2read(struct knote *kn, long hint) 2310 { 2311 struct vnode *vp = (void *)kn->kn_hook; 2312 hammer2_inode_t *ip = VTOI(vp); 2313 off_t off; 2314 2315 if (hint == NOTE_REVOKE) { 2316 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2317 return(1); 2318 } 2319 off = ip->size - kn->kn_fp->f_offset; 2320 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2321 if (kn->kn_sfflags & NOTE_OLDAPI) 2322 return(1); 2323 return (kn->kn_data != 0); 2324 } 2325 2326 2327 static int 2328 filt_hammer2write(struct knote *kn, long hint) 2329 { 2330 if (hint == NOTE_REVOKE) 2331 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2332 kn->kn_data = 0; 2333 return (1); 2334 } 2335 2336 static int 2337 filt_hammer2vnode(struct knote *kn, long hint) 2338 { 2339 if (kn->kn_sfflags & hint) 2340 kn->kn_fflags |= hint; 2341 if (hint == NOTE_REVOKE) { 2342 kn->kn_flags |= (EV_EOF | EV_NODATA); 2343 return (1); 2344 } 2345 return (kn->kn_fflags != 0); 2346 } 2347 2348 /* 2349 * FIFO VOPS 2350 */ 2351 static 2352 int 2353 hammer2_vop_markatime(struct vop_markatime_args *ap) 2354 { 2355 hammer2_inode_t *ip; 2356 struct vnode *vp; 2357 2358 vp = ap->a_vp; 2359 ip = VTOI(vp); 2360 2361 if (ip->pmp->ronly) 2362 return(EROFS); 2363 return(0); 2364 } 2365 2366 static 2367 int 2368 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap) 2369 { 2370 int error; 2371 2372 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2373 if (error) 2374 error = hammer2_vop_kqfilter(ap); 2375 return(error); 2376 } 2377 2378 /* 2379 * VOPS vector 2380 */ 2381 struct vop_ops hammer2_vnode_vops = { 2382 .vop_default = vop_defaultop, 2383 .vop_fsync = hammer2_vop_fsync, 2384 .vop_getpages = vop_stdgetpages, 2385 .vop_putpages = vop_stdputpages, 2386 .vop_access = hammer2_vop_access, 2387 .vop_advlock = hammer2_vop_advlock, 2388 .vop_close = hammer2_vop_close, 2389 .vop_nlink = hammer2_vop_nlink, 2390 .vop_ncreate = hammer2_vop_ncreate, 2391 .vop_nsymlink = hammer2_vop_nsymlink, 2392 .vop_nremove = hammer2_vop_nremove, 2393 .vop_nrmdir = hammer2_vop_nrmdir, 2394 .vop_nrename = hammer2_vop_nrename, 2395 .vop_getattr = hammer2_vop_getattr, 2396 .vop_setattr = hammer2_vop_setattr, 2397 .vop_readdir = hammer2_vop_readdir, 2398 .vop_readlink = hammer2_vop_readlink, 2399 .vop_getpages = vop_stdgetpages, 2400 .vop_putpages = vop_stdputpages, 2401 .vop_read = hammer2_vop_read, 2402 .vop_write = hammer2_vop_write, 2403 .vop_open = hammer2_vop_open, 2404 .vop_inactive = hammer2_vop_inactive, 2405 .vop_reclaim = hammer2_vop_reclaim, 2406 .vop_nresolve = hammer2_vop_nresolve, 2407 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot, 2408 .vop_nmkdir = hammer2_vop_nmkdir, 2409 .vop_nmknod = hammer2_vop_nmknod, 2410 .vop_ioctl = hammer2_vop_ioctl, 2411 .vop_mountctl = hammer2_vop_mountctl, 2412 .vop_bmap = hammer2_vop_bmap, 2413 .vop_strategy = hammer2_vop_strategy, 2414 .vop_kqfilter = hammer2_vop_kqfilter 2415 }; 2416 2417 struct vop_ops hammer2_spec_vops = { 2418 .vop_default = vop_defaultop, 2419 .vop_fsync = hammer2_vop_fsync, 2420 .vop_read = vop_stdnoread, 2421 .vop_write = vop_stdnowrite, 2422 .vop_access = hammer2_vop_access, 2423 .vop_close = hammer2_vop_close, 2424 .vop_markatime = hammer2_vop_markatime, 2425 .vop_getattr = hammer2_vop_getattr, 2426 .vop_inactive = hammer2_vop_inactive, 2427 .vop_reclaim = hammer2_vop_reclaim, 2428 .vop_setattr = hammer2_vop_setattr 2429 }; 2430 2431 struct vop_ops hammer2_fifo_vops = { 2432 .vop_default = fifo_vnoperate, 2433 .vop_fsync = hammer2_vop_fsync, 2434 #if 0 2435 .vop_read = hammer2_vop_fiforead, 2436 .vop_write = hammer2_vop_fifowrite, 2437 #endif 2438 .vop_access = hammer2_vop_access, 2439 #if 0 2440 .vop_close = hammer2_vop_fifoclose, 2441 #endif 2442 .vop_markatime = hammer2_vop_markatime, 2443 .vop_getattr = hammer2_vop_getattr, 2444 .vop_inactive = hammer2_vop_inactive, 2445 .vop_reclaim = hammer2_vop_reclaim, 2446 .vop_setattr = hammer2_vop_setattr, 2447 .vop_kqfilter = hammer2_vop_fifokqfilter 2448 }; 2449 2450