1 /* 2 * Copyright (c) 1989, 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)mfs_vfsops.c 8.11 (Berkeley) 6/19/95 30 * $FreeBSD: src/sys/ufs/mfs/mfs_vfsops.c,v 1.81.2.3 2001/07/04 17:35:21 tegge Exp $ 31 */ 32 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/conf.h> 37 #include <sys/device.h> 38 #include <sys/kernel.h> 39 #include <sys/proc.h> 40 #include <sys/buf.h> 41 #include <sys/mount.h> 42 #include <sys/signalvar.h> 43 #include <sys/signal2.h> 44 #include <sys/spinlock2.h> 45 #include <sys/vnode.h> 46 #include <sys/malloc.h> 47 #include <sys/sysproto.h> 48 #include <sys/mman.h> 49 #include <sys/linker.h> 50 #include <sys/fcntl.h> 51 #include <sys/nlookup.h> 52 #include <sys/devfs.h> 53 54 #include <vm/vm.h> 55 #include <vm/vm_object.h> 56 #include <vm/vm_page.h> 57 #include <vm/vm_pager.h> 58 #include <vm/vnode_pager.h> 59 #include <vm/vm_extern.h> 60 61 #include <sys/buf2.h> 62 #include <sys/thread2.h> 63 64 #include <vfs/ufs/quota.h> 65 #include <vfs/ufs/inode.h> 66 #include <vfs/ufs/ufsmount.h> 67 #include <vfs/ufs/ufs_extern.h> 68 #include <vfs/ufs/fs.h> 69 #include <vfs/ufs/ffs_extern.h> 70 71 #include "mfsnode.h" 72 #include "mfs_extern.h" 73 74 MALLOC_DEFINE(M_MFSNODE, "MFS node", "MFS vnode private part"); 75 76 static int mfs_mount (struct mount *mp, 77 char *path, caddr_t data, struct ucred *td); 78 static int mfs_start (struct mount *mp, int flags); 79 static int mfs_statfs (struct mount *mp, struct statfs *sbp, 80 struct ucred *cred); 81 static int mfs_init (struct vfsconf *); 82 static void mfs_doio(struct bio *bio, struct mfsnode *mfsp); 83 84 d_open_t mfsopen; 85 d_close_t mfsclose; 86 d_strategy_t mfsstrategy; 87 88 static struct dev_ops mfs_ops = { 89 { "MFS", -1, D_DISK | D_NOEMERGPGR }, 90 .d_open = mfsopen, 91 .d_close = mfsclose, 92 .d_read = physread, 93 .d_write = physwrite, 94 .d_strategy = mfsstrategy, 95 }; 96 97 /* 98 * mfs vfs operations. 99 */ 100 static struct vfsops mfs_vfsops = { 101 .vfs_mount = mfs_mount, 102 .vfs_start = mfs_start, 103 .vfs_unmount = ffs_unmount, 104 .vfs_root = ufs_root, 105 .vfs_quotactl = ufs_quotactl, 106 .vfs_statfs = mfs_statfs, 107 .vfs_sync = ffs_sync, 108 .vfs_vget = ffs_vget, 109 .vfs_fhtovp = ffs_fhtovp, 110 .vfs_checkexp = ufs_check_export, 111 .vfs_vptofh = ffs_vptofh, 112 .vfs_init = mfs_init 113 }; 114 115 VFS_SET(mfs_vfsops, mfs, 0); 116 MODULE_VERSION(mfs, 1); 117 118 /* 119 * We allow the underlying MFS block device to be opened and read. 120 */ 121 int 122 mfsopen(struct dev_open_args *ap) 123 { 124 cdev_t dev = ap->a_head.a_dev; 125 126 #if 0 127 if (ap->a_oflags & FWRITE) 128 return(EROFS); 129 #endif 130 if (dev->si_drv1) 131 return(0); 132 return(ENXIO); 133 } 134 135 int 136 mfsclose(struct dev_close_args *ap) 137 { 138 cdev_t dev = ap->a_head.a_dev; 139 struct mfsnode *mfsp; 140 141 if ((mfsp = dev->si_drv1) == NULL) 142 return(0); 143 mfsp->mfs_active = 0; 144 wakeup((caddr_t)mfsp); 145 return(0); 146 } 147 148 int 149 mfsstrategy(struct dev_strategy_args *ap) 150 { 151 cdev_t dev = ap->a_head.a_dev; 152 struct bio *bio = ap->a_bio; 153 struct buf *bp = bio->bio_buf; 154 off_t boff = bio->bio_offset; 155 off_t eoff = boff + bp->b_bcount; 156 struct mfsnode *mfsp; 157 158 if ((mfsp = dev->si_drv1) == NULL) { 159 bp->b_error = ENXIO; 160 goto error; 161 } 162 if (boff < 0) 163 goto bad; 164 if (eoff > mfsp->mfs_size) { 165 if (boff > mfsp->mfs_size || (bp->b_flags & B_BNOCLIP)) 166 goto bad; 167 /* 168 * Return EOF by completing the I/O with 0 bytes transfered. 169 * Set B_INVAL to indicate that any data in the buffer is not 170 * valid. 171 */ 172 if (boff == mfsp->mfs_size) { 173 bp->b_resid = bp->b_bcount; 174 bp->b_flags |= B_INVAL; 175 goto done; 176 } 177 bp->b_bcount = mfsp->mfs_size - boff; 178 } 179 180 /* 181 * Initiate I/O 182 */ 183 if (mfsp->mfs_td == curthread) { 184 mfs_doio(bio, mfsp); 185 } else { 186 bioq_insert_tail(&mfsp->bio_queue, bio); 187 wakeup((caddr_t)mfsp); 188 } 189 return(0); 190 191 /* 192 * Failure conditions on bio 193 */ 194 bad: 195 bp->b_error = EINVAL; 196 error: 197 bp->b_flags |= B_ERROR | B_INVAL; 198 done: 199 biodone(bio); 200 return(0); 201 } 202 203 /* 204 * mfs_mount 205 * 206 * Called when mounting local physical media 207 * 208 * PARAMETERS: 209 * mountroot 210 * mp mount point structure 211 * path NULL (flag for root mount!!!) 212 * data <unused> 213 * ndp <unused> 214 * p process (user credentials check [statfs]) 215 * 216 * mount 217 * mp mount point structure 218 * path path to mount point 219 * data pointer to argument struct in user space 220 * ndp mount point namei() return (used for 221 * credentials on reload), reused to look 222 * up block device. 223 * p process (user credentials check) 224 * 225 * RETURNS: 0 Success 226 * !0 error number (errno.h) 227 * 228 * LOCK STATE: 229 * 230 * ENTRY 231 * mount point is locked 232 * EXIT 233 * mount point is locked 234 * 235 * NOTES: 236 * A NULL path can be used for a flag since the mount 237 * system call will fail with EFAULT in copyinstr in 238 * namei() if it is a genuine NULL from the user. 239 */ 240 /* ARGSUSED */ 241 static int 242 mfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred) 243 { 244 struct vnode *devvp; 245 struct mfs_args args; 246 struct ufsmount *ump; 247 struct fs *fs; 248 struct mfsnode *mfsp; 249 struct nlookupdata nd; 250 size_t size; 251 char devname[16]; 252 int flags; 253 int minnum; 254 int error; 255 cdev_t dev; 256 257 /* 258 * Use NULL path to flag a root mount 259 */ 260 if (path == NULL) { 261 /* 262 *** 263 * Mounting root file system 264 *** 265 */ 266 267 /* you lose */ 268 panic("mfs_mount: mount MFS as root: not configured!"); 269 } 270 271 mfsp = NULL; 272 273 /* 274 *** 275 * Mounting non-root file system or updating a file system 276 *** 277 */ 278 279 /* copy in user arguments*/ 280 error = copyin(data, (caddr_t)&args, sizeof (struct mfs_args)); 281 if (error) 282 goto error_1; 283 284 /* 285 * If updating, check whether changing from read-only to 286 * read/write; if there is no device name, that's all we do. 287 */ 288 if (mp->mnt_flag & MNT_UPDATE) { 289 /* 290 ******************** 291 * UPDATE 292 ******************** 293 */ 294 ump = VFSTOUFS(mp); 295 fs = ump->um_fs; 296 if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) { 297 flags = WRITECLOSE; 298 if (mp->mnt_flag & MNT_FORCE) 299 flags |= FORCECLOSE; 300 error = ffs_flushfiles(mp, flags); 301 if (error) 302 goto error_1; 303 } 304 if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) { 305 /* XXX reopen the device vnode read-write */ 306 fs->fs_ronly = 0; 307 } 308 /* if not updating name...*/ 309 if (args.fspec == 0) { 310 /* 311 * Process export requests. Jumping to "success" 312 * will return the vfs_export() error code. 313 */ 314 error = vfs_export(mp, &ump->um_export, &args.export); 315 goto success; 316 } 317 318 /* XXX MFS does not support name updating*/ 319 goto success; 320 } 321 322 /* 323 * Do the MALLOC before the make_dev since doing so afterward 324 * might cause a bogus v_data pointer to get dereferenced 325 * elsewhere if MALLOC should block. 326 */ 327 mfsp = kmalloc(sizeof *mfsp, M_MFSNODE, M_WAITOK | M_ZERO); 328 329 minnum = (int)curproc->p_pid; 330 331 dev = make_dev(&mfs_ops, minnum, UID_ROOT, GID_WHEEL, 0600, 332 "mfs%d", minnum); 333 /* It is not clear that these will get initialized otherwise */ 334 dev->si_bsize_phys = DEV_BSIZE; 335 dev->si_iosize_max = MAXPHYS; 336 dev->si_drv1 = mfsp; 337 mfsp->mfs_baseoff = args.base; 338 mfsp->mfs_size = args.size; 339 mfsp->mfs_dev = dev; 340 mfsp->mfs_td = curthread; 341 mfsp->mfs_active = 1; 342 bioq_init(&mfsp->bio_queue); 343 344 devfs_config(); /* sync devfs work */ 345 ksnprintf(devname, sizeof(devname), "/dev/mfs%d", minnum); 346 nlookup_init(&nd, devname, UIO_SYSSPACE, 0); 347 devvp = NULL; 348 error = nlookup(&nd); 349 if (error == 0) { 350 devvp = nd.nl_nch.ncp->nc_vp; 351 if (devvp == NULL) 352 error = ENOENT; 353 error = vget(devvp, LK_SHARED); 354 } 355 nlookup_done(&nd); 356 357 if (error) 358 goto error_1; 359 vn_unlock(devvp); 360 361 /* 362 * Our 'block' device must be backed by a VM object. Theoretically 363 * we could use the anonymous memory VM object supplied by userland, 364 * but it would be somewhat of a complex task to deal with it 365 * that way since it would result in I/O requests which supply 366 * the VM pages from our own object. 367 * 368 * vnode_pager_alloc() is typically called when a VM object is 369 * being referenced externally. We have to undo the refs for 370 * the self reference between vnode and object. 371 */ 372 vnode_pager_setsize(devvp, args.size); 373 374 /* Save "mounted from" info for mount point (NULL pad)*/ 375 copyinstr(args.fspec, /* device name*/ 376 mp->mnt_stat.f_mntfromname, /* save area*/ 377 MNAMELEN - 1, /* max size*/ 378 &size); /* real size*/ 379 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size); 380 /* vref is eaten by mount? */ 381 382 error = ffs_mountfs(devvp, mp, M_MFSNODE); 383 if (error) { 384 mfsp->mfs_active = 0; 385 goto error_2; 386 } 387 388 /* 389 * Initialize FS stat information in mount struct; uses 390 * mp->mnt_stat.f_mntfromname. 391 * 392 * This code is common to root and non-root mounts 393 */ 394 VFS_STATFS(mp, &mp->mnt_stat, cred); 395 396 /* 397 * Mark VFS_START MPSAFE; this is to avoid accessing 398 * per-mount token after VFS_START exits 399 */ 400 mp->mnt_kern_flag |= MNTK_ST_MPSAFE; 401 402 goto success; 403 404 error_2: /* error with devvp held*/ 405 vrele(devvp); 406 407 error_1: /* no state to back out*/ 408 if (mfsp) { 409 if (mfsp->mfs_dev) { 410 destroy_dev(mfsp->mfs_dev); 411 mfsp->mfs_dev = NULL; 412 } 413 kfree(mfsp, M_MFSNODE); 414 } 415 416 success: 417 return(error); 418 } 419 420 /* 421 * Used to grab the process and keep it in the kernel to service 422 * memory filesystem I/O requests. 423 * 424 * Loop servicing I/O requests. 425 * Copy the requested data into or out of the memory filesystem 426 * address space. 427 */ 428 /* ARGSUSED */ 429 static int 430 mfs_start(struct mount *mp, int flags) 431 { 432 struct vnode *vp = VFSTOUFS(mp)->um_devvp; 433 struct mfsnode *mfsp = vp->v_rdev->si_drv1; 434 struct bio *bio; 435 struct buf *bp; 436 int gotsig = 0, sig; 437 thread_t td = curthread; 438 439 /* 440 * We must prevent the system from trying to swap 441 * out or kill ( when swap space is low, see vm/pageout.c ) the 442 * process. A deadlock can occur if the process is swapped out, 443 * and the system can loop trying to kill the unkillable ( while 444 * references exist ) MFS process when swap space is low. 445 */ 446 KKASSERT(curproc); 447 PHOLD(curproc); 448 449 mfsp->mfs_td = td; 450 451 while (mfsp->mfs_active) { 452 crit_enter(); 453 454 while ((bio = bioq_takefirst(&mfsp->bio_queue)) != NULL) { 455 crit_exit(); 456 bp = bio->bio_buf; 457 mfs_doio(bio, mfsp); 458 wakeup(bp); 459 crit_enter(); 460 } 461 462 crit_exit(); 463 464 /* 465 * If a non-ignored signal is received, try to unmount. 466 * If that fails, clear the signal (it has been "processed"), 467 * otherwise we will loop here, as tsleep will always return 468 * EINTR/ERESTART. 469 */ 470 /* 471 * Note that dounmount() may fail if work was queued after 472 * we slept. We have to jump hoops here to make sure that we 473 * process any buffers after the sleep, before we dounmount() 474 */ 475 if (gotsig) { 476 gotsig = 0; 477 if (dounmount(mp, 0, 0) != 0) { 478 KKASSERT(td->td_proc); 479 lwkt_gettoken(&td->td_proc->p_token); 480 sig = CURSIG(td->td_lwp); 481 if (sig) { 482 spin_lock(&td->td_lwp->lwp_spin); 483 lwp_delsig(td->td_lwp, sig, 1); 484 spin_unlock(&td->td_lwp->lwp_spin); 485 } 486 lwkt_reltoken(&td->td_proc->p_token); 487 } 488 } 489 else if (tsleep((caddr_t)mfsp, PCATCH, "mfsidl", 0)) 490 gotsig++; /* try to unmount in next pass */ 491 } 492 PRELE(curproc); 493 if (mfsp->mfs_dev) { 494 destroy_dev(mfsp->mfs_dev); 495 mfsp->mfs_dev = NULL; 496 } 497 kfree(mfsp, M_MFSNODE); 498 return (EMOUNTEXIT); 499 } 500 501 /* 502 * Get file system statistics. 503 */ 504 static int 505 mfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 506 { 507 int error; 508 509 error = ffs_statfs(mp, sbp, cred); 510 sbp->f_type = mp->mnt_vfc->vfc_typenum; 511 return (error); 512 } 513 514 /* 515 * Memory based filesystem initialization. 516 */ 517 static int 518 mfs_init(struct vfsconf *vfsp) 519 { 520 return (0); 521 } 522 523 /* 524 * Memory file system I/O. 525 * 526 * Trivial on the HP since buffer has already been mapping into KVA space. 527 * 528 * Read and Write are handled with a simple copyin and copyout. 529 * 530 * We also partially support VOP_FREEBLKS(). We can't implement 531 * completely -- for example, on fragments or inode metadata, but we can 532 * implement it for page-aligned requests. 533 */ 534 static void 535 mfs_doio(struct bio *bio, struct mfsnode *mfsp) 536 { 537 struct buf *bp = bio->bio_buf; 538 caddr_t base = mfsp->mfs_baseoff + bio->bio_offset; 539 int bytes; 540 541 switch(bp->b_cmd) { 542 case BUF_CMD_FREEBLKS: 543 /* 544 * Implement FREEBLKS, which allows the filesystem to tell 545 * a block device when blocks are no longer needed (like when 546 * a file is deleted). We use the hook to MADV_FREE the VM. 547 * This makes an MFS filesystem work as well or better then 548 * a sun-style swap-mounted filesystem. 549 */ 550 bytes = bp->b_bcount; 551 552 if ((vm_offset_t)base & PAGE_MASK) { 553 int n = PAGE_SIZE - ((vm_offset_t)base & PAGE_MASK); 554 bytes -= n; 555 base += n; 556 } 557 if (bytes > 0) { 558 struct madvise_args uap; 559 560 bytes &= ~PAGE_MASK; 561 if (bytes != 0) { 562 bzero(&uap, sizeof(uap)); 563 uap.addr = base; 564 uap.len = bytes; 565 uap.behav = MADV_FREE; 566 sys_madvise(&uap); 567 } 568 } 569 bp->b_error = 0; 570 break; 571 case BUF_CMD_READ: 572 /* 573 * Read data from our 'memory' disk 574 */ 575 bp->b_error = copyin(base, bp->b_data, bp->b_bcount); 576 break; 577 case BUF_CMD_WRITE: 578 /* 579 * Write data to our 'memory' disk 580 */ 581 bp->b_error = copyout(bp->b_data, base, bp->b_bcount); 582 break; 583 default: 584 panic("mfs: bad b_cmd %d", bp->b_cmd); 585 } 586 if (bp->b_error) 587 bp->b_flags |= B_ERROR; 588 biodone(bio); 589 } 590