fss.c revision 1.90
1 /* $NetBSD: fss.c,v 1.90 2014/07/25 08:02:19 dholland Exp $ */ 2 3 /*- 4 * Copyright (c) 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Juergen Hannken-Illjes. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 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 the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * File system snapshot disk driver. 34 * 35 * Block/character interface to the snapshot of a mounted file system. 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.90 2014/07/25 08:02:19 dholland Exp $"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/namei.h> 44 #include <sys/proc.h> 45 #include <sys/errno.h> 46 #include <sys/malloc.h> 47 #include <sys/buf.h> 48 #include <sys/ioctl.h> 49 #include <sys/disklabel.h> 50 #include <sys/device.h> 51 #include <sys/disk.h> 52 #include <sys/stat.h> 53 #include <sys/mount.h> 54 #include <sys/vnode.h> 55 #include <sys/file.h> 56 #include <sys/uio.h> 57 #include <sys/conf.h> 58 #include <sys/kthread.h> 59 #include <sys/fstrans.h> 60 #include <sys/vfs_syscalls.h> /* For do_sys_unlink(). */ 61 62 #include <miscfs/specfs/specdev.h> 63 64 #include <dev/fssvar.h> 65 66 #include <uvm/uvm.h> 67 68 void fssattach(int); 69 70 dev_type_open(fss_open); 71 dev_type_close(fss_close); 72 dev_type_read(fss_read); 73 dev_type_write(fss_write); 74 dev_type_ioctl(fss_ioctl); 75 dev_type_strategy(fss_strategy); 76 dev_type_dump(fss_dump); 77 dev_type_size(fss_size); 78 79 static void fss_unmount_hook(struct mount *); 80 static int fss_copy_on_write(void *, struct buf *, bool); 81 static inline void fss_error(struct fss_softc *, const char *); 82 static int fss_create_files(struct fss_softc *, struct fss_set *, 83 off_t *, struct lwp *); 84 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 85 struct lwp *); 86 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 87 static int fss_softc_alloc(struct fss_softc *); 88 static void fss_softc_free(struct fss_softc *); 89 static int fss_read_cluster(struct fss_softc *, u_int32_t); 90 static void fss_bs_thread(void *); 91 static int fss_bs_io(struct fss_softc *, fss_io_type, 92 u_int32_t, off_t, int, void *); 93 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 94 95 static kmutex_t fss_device_lock; /* Protect all units. */ 96 static int fss_num_attached = 0; /* Number of attached devices. */ 97 static struct vfs_hooks fss_vfs_hooks = { 98 .vh_unmount = fss_unmount_hook 99 }; 100 101 const struct bdevsw fss_bdevsw = { 102 .d_open = fss_open, 103 .d_close = fss_close, 104 .d_strategy = fss_strategy, 105 .d_ioctl = fss_ioctl, 106 .d_dump = fss_dump, 107 .d_psize = fss_size, 108 .d_discard = nodiscard, 109 .d_flag = D_DISK | D_MPSAFE 110 }; 111 112 const struct cdevsw fss_cdevsw = { 113 .d_open = fss_open, 114 .d_close = fss_close, 115 .d_read = fss_read, 116 .d_write = fss_write, 117 .d_ioctl = fss_ioctl, 118 .d_stop = nostop, 119 .d_tty = notty, 120 .d_poll = nopoll, 121 .d_mmap = nommap, 122 .d_kqfilter = nokqfilter, 123 .d_flag = D_DISK | D_MPSAFE 124 }; 125 126 static int fss_match(device_t, cfdata_t, void *); 127 static void fss_attach(device_t, device_t, void *); 128 static int fss_detach(device_t, int); 129 130 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 131 fss_match, fss_attach, fss_detach, NULL); 132 extern struct cfdriver fss_cd; 133 134 void 135 fssattach(int num) 136 { 137 138 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 139 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 140 aprint_error("%s: unable to register\n", fss_cd.cd_name); 141 } 142 143 static int 144 fss_match(device_t self, cfdata_t cfdata, void *aux) 145 { 146 return 1; 147 } 148 149 static void 150 fss_attach(device_t parent, device_t self, void *aux) 151 { 152 struct fss_softc *sc = device_private(self); 153 154 sc->sc_dev = self; 155 sc->sc_bdev = NODEV; 156 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 157 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 158 cv_init(&sc->sc_work_cv, "fssbs"); 159 cv_init(&sc->sc_cache_cv, "cowwait"); 160 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 161 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 162 sc->sc_dkdev->dk_info = NULL; 163 disk_init(sc->sc_dkdev, device_xname(self), NULL); 164 if (!pmf_device_register(self, NULL, NULL)) 165 aprint_error_dev(self, "couldn't establish power handler\n"); 166 167 if (fss_num_attached++ == 0) 168 vfs_hooks_attach(&fss_vfs_hooks); 169 } 170 171 static int 172 fss_detach(device_t self, int flags) 173 { 174 struct fss_softc *sc = device_private(self); 175 176 if (sc->sc_flags & FSS_ACTIVE) 177 return EBUSY; 178 179 if (--fss_num_attached == 0) 180 vfs_hooks_detach(&fss_vfs_hooks); 181 182 pmf_device_deregister(self); 183 mutex_destroy(&sc->sc_slock); 184 mutex_destroy(&sc->sc_lock); 185 cv_destroy(&sc->sc_work_cv); 186 cv_destroy(&sc->sc_cache_cv); 187 bufq_drain(sc->sc_bufq); 188 bufq_free(sc->sc_bufq); 189 disk_destroy(sc->sc_dkdev); 190 free(sc->sc_dkdev, M_DEVBUF); 191 192 return 0; 193 } 194 195 int 196 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 197 { 198 int mflag; 199 cfdata_t cf; 200 struct fss_softc *sc; 201 202 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 203 204 mutex_enter(&fss_device_lock); 205 206 sc = device_lookup_private(&fss_cd, minor(dev)); 207 if (sc == NULL) { 208 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 209 cf->cf_name = fss_cd.cd_name; 210 cf->cf_atname = fss_cd.cd_name; 211 cf->cf_unit = minor(dev); 212 cf->cf_fstate = FSTATE_STAR; 213 sc = device_private(config_attach_pseudo(cf)); 214 if (sc == NULL) { 215 mutex_exit(&fss_device_lock); 216 return ENOMEM; 217 } 218 } 219 220 mutex_enter(&sc->sc_slock); 221 222 sc->sc_flags |= mflag; 223 224 mutex_exit(&sc->sc_slock); 225 mutex_exit(&fss_device_lock); 226 227 return 0; 228 } 229 230 int 231 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 232 { 233 int mflag, error; 234 cfdata_t cf; 235 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 236 237 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 238 error = 0; 239 240 mutex_enter(&fss_device_lock); 241 restart: 242 mutex_enter(&sc->sc_slock); 243 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 244 sc->sc_flags &= ~mflag; 245 mutex_exit(&sc->sc_slock); 246 mutex_exit(&fss_device_lock); 247 return 0; 248 } 249 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 250 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 251 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 252 mutex_exit(&sc->sc_slock); 253 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 254 goto restart; 255 } 256 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 257 mutex_exit(&sc->sc_slock); 258 mutex_exit(&fss_device_lock); 259 return error; 260 } 261 262 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 263 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 264 mutex_exit(&sc->sc_slock); 265 cf = device_cfdata(sc->sc_dev); 266 error = config_detach(sc->sc_dev, DETACH_QUIET); 267 if (! error) 268 free(cf, M_DEVBUF); 269 mutex_exit(&fss_device_lock); 270 271 return error; 272 } 273 274 void 275 fss_strategy(struct buf *bp) 276 { 277 const bool write = ((bp->b_flags & B_READ) != B_READ); 278 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 279 280 mutex_enter(&sc->sc_slock); 281 282 if (write || !FSS_ISVALID(sc)) { 283 284 mutex_exit(&sc->sc_slock); 285 286 bp->b_error = (write ? EROFS : ENXIO); 287 bp->b_resid = bp->b_bcount; 288 biodone(bp); 289 return; 290 } 291 292 bp->b_rawblkno = bp->b_blkno; 293 bufq_put(sc->sc_bufq, bp); 294 cv_signal(&sc->sc_work_cv); 295 296 mutex_exit(&sc->sc_slock); 297 } 298 299 int 300 fss_read(dev_t dev, struct uio *uio, int flags) 301 { 302 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 303 } 304 305 int 306 fss_write(dev_t dev, struct uio *uio, int flags) 307 { 308 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 309 } 310 311 int 312 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 313 { 314 int error; 315 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 316 struct fss_set _fss; 317 struct fss_set *fss = (struct fss_set *)data; 318 struct fss_set50 *fss50 = (struct fss_set50 *)data; 319 struct fss_get *fsg = (struct fss_get *)data; 320 #ifndef _LP64 321 struct fss_get50 *fsg50 = (struct fss_get50 *)data; 322 #endif 323 324 switch (cmd) { 325 case FSSIOCSET50: 326 fss = &_fss; 327 fss->fss_mount = fss50->fss_mount; 328 fss->fss_bstore = fss50->fss_bstore; 329 fss->fss_csize = fss50->fss_csize; 330 fss->fss_flags = 0; 331 /* Fall through */ 332 case FSSIOCSET: 333 mutex_enter(&sc->sc_lock); 334 if ((flag & FWRITE) == 0) 335 error = EPERM; 336 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 337 error = EBUSY; 338 else 339 error = fss_create_snapshot(sc, fss, l); 340 if (error == 0) 341 sc->sc_uflags = fss->fss_flags; 342 mutex_exit(&sc->sc_lock); 343 break; 344 345 case FSSIOCCLR: 346 mutex_enter(&sc->sc_lock); 347 if ((flag & FWRITE) == 0) 348 error = EPERM; 349 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 350 error = ENXIO; 351 else 352 error = fss_delete_snapshot(sc, l); 353 mutex_exit(&sc->sc_lock); 354 break; 355 356 #ifndef _LP64 357 case FSSIOCGET50: 358 mutex_enter(&sc->sc_lock); 359 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 360 case FSS_ACTIVE: 361 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 362 fsg50->fsg_csize = FSS_CLSIZE(sc); 363 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 364 fsg50->fsg_mount_size = sc->sc_clcount; 365 fsg50->fsg_bs_size = sc->sc_clnext; 366 error = 0; 367 break; 368 case FSS_PERSISTENT | FSS_ACTIVE: 369 memcpy(fsg50->fsg_mount, sc->sc_mntname, MNAMELEN); 370 fsg50->fsg_csize = 0; 371 timeval_to_timeval50(&sc->sc_time, &fsg50->fsg_time); 372 fsg50->fsg_mount_size = 0; 373 fsg50->fsg_bs_size = 0; 374 error = 0; 375 break; 376 default: 377 error = ENXIO; 378 break; 379 } 380 mutex_exit(&sc->sc_lock); 381 break; 382 #endif /* _LP64 */ 383 384 case FSSIOCGET: 385 mutex_enter(&sc->sc_lock); 386 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 387 case FSS_ACTIVE: 388 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 389 fsg->fsg_csize = FSS_CLSIZE(sc); 390 fsg->fsg_time = sc->sc_time; 391 fsg->fsg_mount_size = sc->sc_clcount; 392 fsg->fsg_bs_size = sc->sc_clnext; 393 error = 0; 394 break; 395 case FSS_PERSISTENT | FSS_ACTIVE: 396 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 397 fsg->fsg_csize = 0; 398 fsg->fsg_time = sc->sc_time; 399 fsg->fsg_mount_size = 0; 400 fsg->fsg_bs_size = 0; 401 error = 0; 402 break; 403 default: 404 error = ENXIO; 405 break; 406 } 407 mutex_exit(&sc->sc_lock); 408 break; 409 410 case FSSIOFSET: 411 mutex_enter(&sc->sc_slock); 412 sc->sc_uflags = *(int *)data; 413 mutex_exit(&sc->sc_slock); 414 error = 0; 415 break; 416 417 case FSSIOFGET: 418 mutex_enter(&sc->sc_slock); 419 *(int *)data = sc->sc_uflags; 420 mutex_exit(&sc->sc_slock); 421 error = 0; 422 break; 423 424 default: 425 error = EINVAL; 426 break; 427 } 428 429 return error; 430 } 431 432 int 433 fss_size(dev_t dev) 434 { 435 return -1; 436 } 437 438 int 439 fss_dump(dev_t dev, daddr_t blkno, void *va, 440 size_t size) 441 { 442 return EROFS; 443 } 444 445 /* 446 * An error occurred reading or writing the snapshot or backing store. 447 * If it is the first error log to console. 448 * The caller holds the mutex. 449 */ 450 static inline void 451 fss_error(struct fss_softc *sc, const char *msg) 452 { 453 454 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 455 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 456 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 457 sc->sc_flags |= FSS_ERROR; 458 } 459 460 /* 461 * Allocate the variable sized parts of the softc and 462 * fork the kernel thread. 463 * 464 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 465 * must be initialized. 466 */ 467 static int 468 fss_softc_alloc(struct fss_softc *sc) 469 { 470 int i, error; 471 472 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 473 sc->sc_copied = 474 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 475 if (sc->sc_copied == NULL) 476 return(ENOMEM); 477 478 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 479 sizeof(struct fss_cache), KM_SLEEP); 480 if (sc->sc_cache == NULL) 481 return(ENOMEM); 482 483 for (i = 0; i < sc->sc_cache_size; i++) { 484 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 485 sc->sc_cache[i].fc_data = 486 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 487 if (sc->sc_cache[i].fc_data == NULL) 488 return(ENOMEM); 489 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 490 } 491 492 sc->sc_indir_valid = 493 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 494 if (sc->sc_indir_valid == NULL) 495 return(ENOMEM); 496 497 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 498 if (sc->sc_indir_data == NULL) 499 return(ENOMEM); 500 } else { 501 sc->sc_copied = NULL; 502 sc->sc_cache = NULL; 503 sc->sc_indir_valid = NULL; 504 sc->sc_indir_data = NULL; 505 } 506 507 sc->sc_flags |= FSS_BS_THREAD; 508 if ((error = kthread_create(PRI_BIO, KTHREAD_MUSTJOIN, NULL, 509 fss_bs_thread, sc, &sc->sc_bs_lwp, 510 "%s", device_xname(sc->sc_dev))) != 0) { 511 sc->sc_flags &= ~FSS_BS_THREAD; 512 return error; 513 } 514 515 disk_attach(sc->sc_dkdev); 516 517 return 0; 518 } 519 520 /* 521 * Free the variable sized parts of the softc. 522 */ 523 static void 524 fss_softc_free(struct fss_softc *sc) 525 { 526 int i; 527 528 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 529 mutex_enter(&sc->sc_slock); 530 sc->sc_flags &= ~FSS_BS_THREAD; 531 cv_signal(&sc->sc_work_cv); 532 mutex_exit(&sc->sc_slock); 533 kthread_join(sc->sc_bs_lwp); 534 535 disk_detach(sc->sc_dkdev); 536 } 537 538 if (sc->sc_copied != NULL) 539 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 540 sc->sc_copied = NULL; 541 542 if (sc->sc_cache != NULL) { 543 for (i = 0; i < sc->sc_cache_size; i++) 544 if (sc->sc_cache[i].fc_data != NULL) { 545 cv_destroy(&sc->sc_cache[i].fc_state_cv); 546 kmem_free(sc->sc_cache[i].fc_data, 547 FSS_CLSIZE(sc)); 548 } 549 kmem_free(sc->sc_cache, 550 sc->sc_cache_size*sizeof(struct fss_cache)); 551 } 552 sc->sc_cache = NULL; 553 554 if (sc->sc_indir_valid != NULL) 555 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 556 sc->sc_indir_valid = NULL; 557 558 if (sc->sc_indir_data != NULL) 559 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 560 sc->sc_indir_data = NULL; 561 } 562 563 /* 564 * Set all active snapshots on this file system into ERROR state. 565 */ 566 static void 567 fss_unmount_hook(struct mount *mp) 568 { 569 int i; 570 struct fss_softc *sc; 571 572 mutex_enter(&fss_device_lock); 573 for (i = 0; i < fss_cd.cd_ndevs; i++) { 574 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 575 continue; 576 mutex_enter(&sc->sc_slock); 577 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 578 sc->sc_mount == mp) 579 fss_error(sc, "forced unmount"); 580 mutex_exit(&sc->sc_slock); 581 } 582 mutex_exit(&fss_device_lock); 583 } 584 585 /* 586 * A buffer is written to the snapshotted block device. Copy to 587 * backing store if needed. 588 */ 589 static int 590 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 591 { 592 int error; 593 u_int32_t cl, ch, c; 594 struct fss_softc *sc = v; 595 596 mutex_enter(&sc->sc_slock); 597 if (!FSS_ISVALID(sc)) { 598 mutex_exit(&sc->sc_slock); 599 return 0; 600 } 601 602 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 603 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 604 error = 0; 605 if (curlwp == uvm.pagedaemon_lwp) { 606 for (c = cl; c <= ch; c++) 607 if (isclr(sc->sc_copied, c)) { 608 error = ENOMEM; 609 break; 610 } 611 } 612 mutex_exit(&sc->sc_slock); 613 614 if (error == 0) 615 for (c = cl; c <= ch; c++) { 616 error = fss_read_cluster(sc, c); 617 if (error) 618 break; 619 } 620 621 return error; 622 } 623 624 /* 625 * Lookup and open needed files. 626 * 627 * For file system internal snapshot initializes sc_mntname, sc_mount, 628 * sc_bs_vp and sc_time. 629 * 630 * Otherwise returns dev and size of the underlying block device. 631 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 632 */ 633 static int 634 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 635 off_t *bsize, struct lwp *l) 636 { 637 int error, bits, fsbsize; 638 uint64_t numsec; 639 unsigned int secsize; 640 struct timespec ts; 641 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 642 struct pathbuf *pb2; 643 struct nameidata nd2; 644 struct vnode *vp; 645 646 /* 647 * Get the mounted file system. 648 */ 649 650 error = namei_simple_user(fss->fss_mount, 651 NSM_FOLLOW_NOEMULROOT, &vp); 652 if (error != 0) 653 return error; 654 655 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 656 vrele(vp); 657 return EINVAL; 658 } 659 660 sc->sc_mount = vp->v_mount; 661 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 662 663 vrele(vp); 664 665 /* 666 * Check for file system internal snapshot. 667 */ 668 669 error = namei_simple_user(fss->fss_bstore, 670 NSM_FOLLOW_NOEMULROOT, &vp); 671 if (error != 0) 672 return error; 673 674 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 675 sc->sc_flags |= FSS_PERSISTENT; 676 sc->sc_bs_vp = vp; 677 678 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 679 bits = sizeof(sc->sc_bs_bshift)*NBBY; 680 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 681 sc->sc_bs_bshift++) 682 if (FSS_FSBSIZE(sc) == fsbsize) 683 break; 684 if (sc->sc_bs_bshift >= bits) 685 return EINVAL; 686 687 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 688 sc->sc_clshift = 0; 689 690 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 691 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 692 if (error) 693 return error; 694 } 695 error = vn_lock(vp, LK_EXCLUSIVE); 696 if (error != 0) 697 return error; 698 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 699 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 700 701 VOP_UNLOCK(sc->sc_bs_vp); 702 703 return error; 704 } 705 vrele(vp); 706 707 /* 708 * Get the block device it is mounted on and its size. 709 */ 710 711 error = spec_node_lookup_by_mount(sc->sc_mount, &vp); 712 if (error) 713 return error; 714 sc->sc_bdev = vp->v_rdev; 715 716 error = getdisksize(vp, &numsec, &secsize); 717 vrele(vp); 718 if (error) 719 return error; 720 721 *bsize = (off_t)numsec*secsize; 722 723 /* 724 * Get the backing store 725 */ 726 727 error = pathbuf_copyin(fss->fss_bstore, &pb2); 728 if (error) { 729 return error; 730 } 731 NDINIT(&nd2, LOOKUP, FOLLOW, pb2); 732 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) { 733 pathbuf_destroy(pb2); 734 return error; 735 } 736 VOP_UNLOCK(nd2.ni_vp); 737 738 sc->sc_bs_vp = nd2.ni_vp; 739 740 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) { 741 pathbuf_destroy(pb2); 742 return EINVAL; 743 } 744 pathbuf_destroy(pb2); 745 746 if ((fss->fss_flags & FSS_UNLINK_ON_CREATE) != 0) { 747 error = do_sys_unlink(fss->fss_bstore, UIO_USERSPACE); 748 if (error) 749 return error; 750 } 751 if (sc->sc_bs_vp->v_type == VREG) { 752 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 753 if (fsbsize & (fsbsize-1)) /* No power of two */ 754 return EINVAL; 755 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 756 sc->sc_bs_bshift++) 757 if (FSS_FSBSIZE(sc) == fsbsize) 758 break; 759 if (sc->sc_bs_bshift >= 32) 760 return EINVAL; 761 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 762 } else { 763 sc->sc_bs_bshift = DEV_BSHIFT; 764 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 765 } 766 767 return 0; 768 } 769 770 /* 771 * Create a snapshot. 772 */ 773 static int 774 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 775 { 776 int len, error; 777 u_int32_t csize; 778 off_t bsize; 779 780 bsize = 0; /* XXX gcc */ 781 782 /* 783 * Open needed files. 784 */ 785 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 786 goto bad; 787 788 if (sc->sc_flags & FSS_PERSISTENT) { 789 fss_softc_alloc(sc); 790 sc->sc_flags |= FSS_ACTIVE; 791 return 0; 792 } 793 794 /* 795 * Set cluster size. Must be a power of two and 796 * a multiple of backing store block size. 797 */ 798 if (fss->fss_csize <= 0) 799 csize = MAXPHYS; 800 else 801 csize = fss->fss_csize; 802 if (bsize/csize > FSS_CLUSTER_MAX) 803 csize = bsize/FSS_CLUSTER_MAX+1; 804 805 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 806 sc->sc_clshift++) 807 if (FSS_CLSIZE(sc) >= csize) 808 break; 809 if (sc->sc_clshift >= 32) { 810 error = EINVAL; 811 goto bad; 812 } 813 sc->sc_clmask = FSS_CLSIZE(sc)-1; 814 815 /* 816 * Set number of cache slots. 817 */ 818 if (FSS_CLSIZE(sc) <= 8192) 819 sc->sc_cache_size = 32; 820 else if (FSS_CLSIZE(sc) <= 65536) 821 sc->sc_cache_size = 8; 822 else 823 sc->sc_cache_size = 4; 824 825 /* 826 * Set number of clusters and size of last cluster. 827 */ 828 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 829 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 830 831 /* 832 * Set size of indirect table. 833 */ 834 len = sc->sc_clcount*sizeof(u_int32_t); 835 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 836 sc->sc_clnext = sc->sc_indir_size; 837 sc->sc_indir_cur = 0; 838 839 if ((error = fss_softc_alloc(sc)) != 0) 840 goto bad; 841 842 /* 843 * Activate the snapshot. 844 */ 845 846 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 847 goto bad; 848 849 microtime(&sc->sc_time); 850 851 error = fscow_establish(sc->sc_mount, fss_copy_on_write, sc); 852 if (error == 0) 853 sc->sc_flags |= FSS_ACTIVE; 854 855 vfs_resume(sc->sc_mount); 856 857 if (error != 0) 858 goto bad; 859 860 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 861 aprint_debug_dev(sc->sc_dev, 862 "%u clusters of %u, %u cache slots, %u indir clusters\n", 863 sc->sc_clcount, FSS_CLSIZE(sc), 864 sc->sc_cache_size, sc->sc_indir_size); 865 866 return 0; 867 868 bad: 869 fss_softc_free(sc); 870 if (sc->sc_bs_vp != NULL) { 871 if (sc->sc_flags & FSS_PERSISTENT) 872 vrele(sc->sc_bs_vp); 873 else 874 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 875 } 876 sc->sc_bs_vp = NULL; 877 878 return error; 879 } 880 881 /* 882 * Delete a snapshot. 883 */ 884 static int 885 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 886 { 887 888 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 889 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 890 891 mutex_enter(&sc->sc_slock); 892 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 893 sc->sc_mount = NULL; 894 sc->sc_bdev = NODEV; 895 mutex_exit(&sc->sc_slock); 896 897 fss_softc_free(sc); 898 if (sc->sc_flags & FSS_PERSISTENT) 899 vrele(sc->sc_bs_vp); 900 else 901 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 902 sc->sc_bs_vp = NULL; 903 sc->sc_flags &= ~FSS_PERSISTENT; 904 905 return 0; 906 } 907 908 /* 909 * Read a cluster from the snapshotted block device to the cache. 910 */ 911 static int 912 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 913 { 914 int error, todo, offset, len; 915 daddr_t dblk; 916 struct buf *bp, *mbp; 917 struct fss_cache *scp, *scl; 918 919 /* 920 * Get a free cache slot. 921 */ 922 scl = sc->sc_cache+sc->sc_cache_size; 923 924 mutex_enter(&sc->sc_slock); 925 926 restart: 927 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 928 mutex_exit(&sc->sc_slock); 929 return 0; 930 } 931 932 for (scp = sc->sc_cache; scp < scl; scp++) 933 if (scp->fc_cluster == cl) { 934 if (scp->fc_type == FSS_CACHE_VALID) { 935 mutex_exit(&sc->sc_slock); 936 return 0; 937 } else if (scp->fc_type == FSS_CACHE_BUSY) { 938 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 939 goto restart; 940 } 941 } 942 943 for (scp = sc->sc_cache; scp < scl; scp++) 944 if (scp->fc_type == FSS_CACHE_FREE) { 945 scp->fc_type = FSS_CACHE_BUSY; 946 scp->fc_cluster = cl; 947 break; 948 } 949 if (scp >= scl) { 950 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 951 goto restart; 952 } 953 954 mutex_exit(&sc->sc_slock); 955 956 /* 957 * Start the read. 958 */ 959 dblk = btodb(FSS_CLTOB(sc, cl)); 960 if (cl == sc->sc_clcount-1) { 961 todo = sc->sc_clresid; 962 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 963 } else 964 todo = FSS_CLSIZE(sc); 965 offset = 0; 966 mbp = getiobuf(NULL, true); 967 mbp->b_bufsize = todo; 968 mbp->b_data = scp->fc_data; 969 mbp->b_resid = mbp->b_bcount = todo; 970 mbp->b_flags = B_READ; 971 mbp->b_cflags = BC_BUSY; 972 mbp->b_dev = sc->sc_bdev; 973 while (todo > 0) { 974 len = todo; 975 if (len > MAXPHYS) 976 len = MAXPHYS; 977 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 978 bp = mbp; 979 else { 980 bp = getiobuf(NULL, true); 981 nestiobuf_setup(mbp, bp, offset, len); 982 } 983 bp->b_lblkno = 0; 984 bp->b_blkno = dblk; 985 bdev_strategy(bp); 986 dblk += btodb(len); 987 offset += len; 988 todo -= len; 989 } 990 error = biowait(mbp); 991 putiobuf(mbp); 992 993 mutex_enter(&sc->sc_slock); 994 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 995 cv_broadcast(&scp->fc_state_cv); 996 if (error == 0) { 997 setbit(sc->sc_copied, scp->fc_cluster); 998 cv_signal(&sc->sc_work_cv); 999 } 1000 mutex_exit(&sc->sc_slock); 1001 1002 return error; 1003 } 1004 1005 /* 1006 * Read/write clusters from/to backing store. 1007 * For persistent snapshots must be called with cl == 0. off is the 1008 * offset into the snapshot. 1009 */ 1010 static int 1011 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 1012 u_int32_t cl, off_t off, int len, void *data) 1013 { 1014 int error; 1015 1016 off += FSS_CLTOB(sc, cl); 1017 1018 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 1019 1020 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 1021 data, len, off, UIO_SYSSPACE, 1022 IO_ADV_ENCODE(POSIX_FADV_NOREUSE) | IO_NODELOCKED, 1023 sc->sc_bs_lwp->l_cred, NULL, NULL); 1024 if (error == 0) { 1025 mutex_enter(sc->sc_bs_vp->v_interlock); 1026 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 1027 round_page(off+len), PGO_CLEANIT | PGO_FREE | PGO_SYNCIO); 1028 } 1029 1030 VOP_UNLOCK(sc->sc_bs_vp); 1031 1032 return error; 1033 } 1034 1035 /* 1036 * Get a pointer to the indirect slot for this cluster. 1037 */ 1038 static u_int32_t * 1039 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 1040 { 1041 u_int32_t icl; 1042 int ioff; 1043 1044 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1045 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 1046 1047 if (sc->sc_indir_cur == icl) 1048 return &sc->sc_indir_data[ioff]; 1049 1050 if (sc->sc_indir_dirty) { 1051 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 1052 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1053 return NULL; 1054 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 1055 } 1056 1057 sc->sc_indir_dirty = 0; 1058 sc->sc_indir_cur = icl; 1059 1060 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 1061 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 1062 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1063 return NULL; 1064 } else 1065 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1066 1067 return &sc->sc_indir_data[ioff]; 1068 } 1069 1070 /* 1071 * The kernel thread (one for every active snapshot). 1072 * 1073 * After wakeup it cleans the cache and runs the I/O requests. 1074 */ 1075 static void 1076 fss_bs_thread(void *arg) 1077 { 1078 bool thread_idle, is_valid; 1079 int error, i, todo, len, crotor, is_read; 1080 long off; 1081 char *addr; 1082 u_int32_t c, cl, ch, *indirp; 1083 struct buf *bp, *nbp; 1084 struct fss_softc *sc; 1085 struct fss_cache *scp, *scl; 1086 1087 sc = arg; 1088 scl = sc->sc_cache+sc->sc_cache_size; 1089 crotor = 0; 1090 thread_idle = false; 1091 1092 mutex_enter(&sc->sc_slock); 1093 1094 for (;;) { 1095 if (thread_idle) 1096 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1097 thread_idle = true; 1098 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1099 mutex_exit(&sc->sc_slock); 1100 kthread_exit(0); 1101 } 1102 1103 /* 1104 * Process I/O requests (persistent) 1105 */ 1106 1107 if (sc->sc_flags & FSS_PERSISTENT) { 1108 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1109 continue; 1110 is_valid = FSS_ISVALID(sc); 1111 is_read = (bp->b_flags & B_READ); 1112 thread_idle = false; 1113 mutex_exit(&sc->sc_slock); 1114 1115 if (is_valid) { 1116 disk_busy(sc->sc_dkdev); 1117 error = fss_bs_io(sc, FSS_READ, 0, 1118 dbtob(bp->b_blkno), bp->b_bcount, 1119 bp->b_data); 1120 disk_unbusy(sc->sc_dkdev, 1121 (error ? 0 : bp->b_bcount), is_read); 1122 } else 1123 error = ENXIO; 1124 1125 bp->b_error = error; 1126 bp->b_resid = (error ? bp->b_bcount : 0); 1127 biodone(bp); 1128 1129 mutex_enter(&sc->sc_slock); 1130 continue; 1131 } 1132 1133 /* 1134 * Clean the cache 1135 */ 1136 for (i = 0; i < sc->sc_cache_size; i++) { 1137 crotor = (crotor + 1) % sc->sc_cache_size; 1138 scp = sc->sc_cache + crotor; 1139 if (scp->fc_type != FSS_CACHE_VALID) 1140 continue; 1141 mutex_exit(&sc->sc_slock); 1142 1143 thread_idle = false; 1144 indirp = fss_bs_indir(sc, scp->fc_cluster); 1145 if (indirp != NULL) { 1146 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1147 0, FSS_CLSIZE(sc), scp->fc_data); 1148 } else 1149 error = EIO; 1150 1151 mutex_enter(&sc->sc_slock); 1152 if (error == 0) { 1153 *indirp = sc->sc_clnext++; 1154 sc->sc_indir_dirty = 1; 1155 } else 1156 fss_error(sc, "write error on backing store"); 1157 1158 scp->fc_type = FSS_CACHE_FREE; 1159 cv_broadcast(&sc->sc_cache_cv); 1160 break; 1161 } 1162 1163 /* 1164 * Process I/O requests 1165 */ 1166 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1167 continue; 1168 is_valid = FSS_ISVALID(sc); 1169 is_read = (bp->b_flags & B_READ); 1170 thread_idle = false; 1171 1172 if (!is_valid) { 1173 mutex_exit(&sc->sc_slock); 1174 1175 bp->b_error = ENXIO; 1176 bp->b_resid = bp->b_bcount; 1177 biodone(bp); 1178 1179 mutex_enter(&sc->sc_slock); 1180 continue; 1181 } 1182 1183 disk_busy(sc->sc_dkdev); 1184 1185 /* 1186 * First read from the snapshotted block device unless 1187 * this request is completely covered by backing store. 1188 */ 1189 1190 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1191 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1192 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1193 error = 0; 1194 bp->b_resid = 0; 1195 bp->b_error = 0; 1196 for (c = cl; c <= ch; c++) { 1197 if (isset(sc->sc_copied, c)) 1198 continue; 1199 mutex_exit(&sc->sc_slock); 1200 1201 /* Not on backing store, read from device. */ 1202 nbp = getiobuf(NULL, true); 1203 nbp->b_flags = B_READ; 1204 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1205 nbp->b_bufsize = bp->b_bcount; 1206 nbp->b_data = bp->b_data; 1207 nbp->b_blkno = bp->b_blkno; 1208 nbp->b_lblkno = 0; 1209 nbp->b_dev = sc->sc_bdev; 1210 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1211 1212 bdev_strategy(nbp); 1213 1214 error = biowait(nbp); 1215 if (error != 0) { 1216 bp->b_resid = bp->b_bcount; 1217 bp->b_error = nbp->b_error; 1218 disk_unbusy(sc->sc_dkdev, 0, is_read); 1219 biodone(bp); 1220 } 1221 putiobuf(nbp); 1222 1223 mutex_enter(&sc->sc_slock); 1224 break; 1225 } 1226 if (error) 1227 continue; 1228 1229 /* 1230 * Replace those parts that have been saved to backing store. 1231 */ 1232 1233 addr = bp->b_data; 1234 todo = bp->b_bcount; 1235 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1236 len = FSS_CLSIZE(sc)-off; 1237 if (len > todo) 1238 len = todo; 1239 if (isclr(sc->sc_copied, c)) 1240 continue; 1241 mutex_exit(&sc->sc_slock); 1242 1243 indirp = fss_bs_indir(sc, c); 1244 if (indirp == NULL || *indirp == 0) { 1245 /* 1246 * Not on backing store. Either in cache 1247 * or hole in the snapshotted block device. 1248 */ 1249 1250 mutex_enter(&sc->sc_slock); 1251 for (scp = sc->sc_cache; scp < scl; scp++) 1252 if (scp->fc_type == FSS_CACHE_VALID && 1253 scp->fc_cluster == c) 1254 break; 1255 if (scp < scl) 1256 memcpy(addr, (char *)scp->fc_data+off, 1257 len); 1258 else 1259 memset(addr, 0, len); 1260 continue; 1261 } 1262 1263 /* 1264 * Read from backing store. 1265 */ 1266 error = 1267 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1268 1269 mutex_enter(&sc->sc_slock); 1270 if (error) { 1271 bp->b_resid = bp->b_bcount; 1272 bp->b_error = error; 1273 break; 1274 } 1275 } 1276 mutex_exit(&sc->sc_slock); 1277 1278 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1279 biodone(bp); 1280 1281 mutex_enter(&sc->sc_slock); 1282 } 1283 } 1284 1285 #ifdef _MODULE 1286 1287 #include <sys/module.h> 1288 1289 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1290 CFDRIVER_DECL(fss, DV_DISK, NULL); 1291 1292 static int 1293 fss_modcmd(modcmd_t cmd, void *arg) 1294 { 1295 int bmajor = -1, cmajor = -1, error = 0; 1296 1297 switch (cmd) { 1298 case MODULE_CMD_INIT: 1299 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1300 error = config_cfdriver_attach(&fss_cd); 1301 if (error) { 1302 mutex_destroy(&fss_device_lock); 1303 break; 1304 } 1305 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1306 if (error) { 1307 config_cfdriver_detach(&fss_cd); 1308 mutex_destroy(&fss_device_lock); 1309 break; 1310 } 1311 error = devsw_attach(fss_cd.cd_name, 1312 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1313 if (error == EEXIST) 1314 error = 0; 1315 if (error) { 1316 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1317 config_cfdriver_detach(&fss_cd); 1318 mutex_destroy(&fss_device_lock); 1319 break; 1320 } 1321 break; 1322 1323 case MODULE_CMD_FINI: 1324 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1325 if (error) 1326 break; 1327 config_cfdriver_detach(&fss_cd); 1328 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1329 mutex_destroy(&fss_device_lock); 1330 break; 1331 1332 default: 1333 error = ENOTTY; 1334 break; 1335 } 1336 1337 return error; 1338 } 1339 1340 #endif /* _MODULE */ 1341
Indexes created Wed Nov 11 00:26:22 UTC 2026