1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ 27 /* All Rights Reserved */ 28 29 /* 30 * Portions of this source code were derived from Berkeley 4.3 BSD 31 * under license from the Regents of the University of California. 32 */ 33 34 #pragma ident "%Z%%M% %I% %E% SMI" 35 36 #include <sys/types.h> 37 #include <sys/t_lock.h> 38 #include <sys/ksynch.h> 39 #include <sys/param.h> 40 #include <sys/time.h> 41 #include <sys/systm.h> 42 #include <sys/sysmacros.h> 43 #include <sys/resource.h> 44 #include <sys/signal.h> 45 #include <sys/cred.h> 46 #include <sys/user.h> 47 #include <sys/buf.h> 48 #include <sys/vfs.h> 49 #include <sys/vfs_opreg.h> 50 #include <sys/vnode.h> 51 #include <sys/proc.h> 52 #include <sys/disp.h> 53 #include <sys/file.h> 54 #include <sys/fcntl.h> 55 #include <sys/flock.h> 56 #include <sys/atomic.h> 57 #include <sys/kmem.h> 58 #include <sys/uio.h> 59 #include <sys/dnlc.h> 60 #include <sys/conf.h> 61 #include <sys/mman.h> 62 #include <sys/pathname.h> 63 #include <sys/debug.h> 64 #include <sys/vmsystm.h> 65 #include <sys/cmn_err.h> 66 #include <sys/filio.h> 67 #include <sys/policy.h> 68 69 #include <sys/fs/ufs_fs.h> 70 #include <sys/fs/ufs_lockfs.h> 71 #include <sys/fs/ufs_filio.h> 72 #include <sys/fs/ufs_inode.h> 73 #include <sys/fs/ufs_fsdir.h> 74 #include <sys/fs/ufs_quota.h> 75 #include <sys/fs/ufs_log.h> 76 #include <sys/fs/ufs_snap.h> 77 #include <sys/fs/ufs_trans.h> 78 #include <sys/fs/ufs_panic.h> 79 #include <sys/fs/ufs_bio.h> 80 #include <sys/dirent.h> /* must be AFTER <sys/fs/fsdir.h>! */ 81 #include <sys/errno.h> 82 #include <sys/fssnap_if.h> 83 #include <sys/unistd.h> 84 #include <sys/sunddi.h> 85 86 #include <sys/filio.h> /* _FIOIO */ 87 88 #include <vm/hat.h> 89 #include <vm/page.h> 90 #include <vm/pvn.h> 91 #include <vm/as.h> 92 #include <vm/seg.h> 93 #include <vm/seg_map.h> 94 #include <vm/seg_vn.h> 95 #include <vm/seg_kmem.h> 96 #include <vm/rm.h> 97 #include <sys/swap.h> 98 99 #include <fs/fs_subr.h> 100 101 #include <sys/fs/decomp.h> 102 103 static struct instats ins; 104 105 static int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t); 106 static int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *, 107 caddr_t, struct page **, size_t, enum seg_rw, int); 108 static int ufs_open(struct vnode **, int, struct cred *, caller_context_t *); 109 static int ufs_close(struct vnode *, int, int, offset_t, struct cred *, 110 caller_context_t *); 111 static int ufs_read(struct vnode *, struct uio *, int, struct cred *, 112 struct caller_context *); 113 static int ufs_write(struct vnode *, struct uio *, int, struct cred *, 114 struct caller_context *); 115 static int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *, 116 int *, caller_context_t *); 117 static int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *, 118 caller_context_t *); 119 static int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *, 120 caller_context_t *); 121 static int ufs_access(struct vnode *, int, int, struct cred *, 122 caller_context_t *); 123 static int ufs_lookup(struct vnode *, char *, struct vnode **, 124 struct pathname *, int, struct vnode *, struct cred *, 125 caller_context_t *, int *, pathname_t *); 126 static int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl, 127 int, struct vnode **, struct cred *, int, 128 caller_context_t *, vsecattr_t *); 129 static int ufs_remove(struct vnode *, char *, struct cred *, 130 caller_context_t *, int); 131 static int ufs_link(struct vnode *, struct vnode *, char *, struct cred *, 132 caller_context_t *, int); 133 static int ufs_rename(struct vnode *, char *, struct vnode *, char *, 134 struct cred *, caller_context_t *, int); 135 static int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **, 136 struct cred *, caller_context_t *, int, vsecattr_t *); 137 static int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *, 138 caller_context_t *, int); 139 static int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *, 140 caller_context_t *, int); 141 static int ufs_symlink(struct vnode *, char *, struct vattr *, char *, 142 struct cred *, caller_context_t *, int); 143 static int ufs_readlink(struct vnode *, struct uio *, struct cred *, 144 caller_context_t *); 145 static int ufs_fsync(struct vnode *, int, struct cred *, caller_context_t *); 146 static void ufs_inactive(struct vnode *, struct cred *, caller_context_t *); 147 static int ufs_fid(struct vnode *, struct fid *, caller_context_t *); 148 static int ufs_rwlock(struct vnode *, int, caller_context_t *); 149 static void ufs_rwunlock(struct vnode *, int, caller_context_t *); 150 static int ufs_seek(struct vnode *, offset_t, offset_t *, caller_context_t *); 151 static int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t, 152 struct flk_callback *, struct cred *, 153 caller_context_t *); 154 static int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t, 155 cred_t *, caller_context_t *); 156 static int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *, 157 struct page **, size_t, struct seg *, caddr_t, 158 enum seg_rw, struct cred *, caller_context_t *); 159 static int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *, 160 caller_context_t *); 161 static int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *); 162 static int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t, 163 uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *); 164 static int ufs_addmap(struct vnode *, offset_t, struct as *, caddr_t, size_t, 165 uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *); 166 static int ufs_delmap(struct vnode *, offset_t, struct as *, caddr_t, size_t, 167 uint_t, uint_t, uint_t, struct cred *, caller_context_t *); 168 static int ufs_poll(vnode_t *, short, int, short *, struct pollhead **, 169 caller_context_t *); 170 static int ufs_dump(vnode_t *, caddr_t, offset_t, offset_t, 171 caller_context_t *); 172 static int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *, 173 caller_context_t *); 174 static int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int, 175 struct cred *, caller_context_t *); 176 static int ufs_dumpctl(vnode_t *, int, offset_t *, caller_context_t *); 177 static daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *, 178 daddr32_t *, int, int); 179 static int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *, 180 caller_context_t *); 181 static int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *, 182 caller_context_t *); 183 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *); 184 185 /* 186 * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions. 187 * 188 * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet. 189 */ 190 struct vnodeops *ufs_vnodeops; 191 192 /* NOTE: "not blkd" below means that the operation isn't blocked by lockfs */ 193 const fs_operation_def_t ufs_vnodeops_template[] = { 194 VOPNAME_OPEN, { .vop_open = ufs_open }, /* not blkd */ 195 VOPNAME_CLOSE, { .vop_close = ufs_close }, /* not blkd */ 196 VOPNAME_READ, { .vop_read = ufs_read }, 197 VOPNAME_WRITE, { .vop_write = ufs_write }, 198 VOPNAME_IOCTL, { .vop_ioctl = ufs_ioctl }, 199 VOPNAME_GETATTR, { .vop_getattr = ufs_getattr }, 200 VOPNAME_SETATTR, { .vop_setattr = ufs_setattr }, 201 VOPNAME_ACCESS, { .vop_access = ufs_access }, 202 VOPNAME_LOOKUP, { .vop_lookup = ufs_lookup }, 203 VOPNAME_CREATE, { .vop_create = ufs_create }, 204 VOPNAME_REMOVE, { .vop_remove = ufs_remove }, 205 VOPNAME_LINK, { .vop_link = ufs_link }, 206 VOPNAME_RENAME, { .vop_rename = ufs_rename }, 207 VOPNAME_MKDIR, { .vop_mkdir = ufs_mkdir }, 208 VOPNAME_RMDIR, { .vop_rmdir = ufs_rmdir }, 209 VOPNAME_READDIR, { .vop_readdir = ufs_readdir }, 210 VOPNAME_SYMLINK, { .vop_symlink = ufs_symlink }, 211 VOPNAME_READLINK, { .vop_readlink = ufs_readlink }, 212 VOPNAME_FSYNC, { .vop_fsync = ufs_fsync }, 213 VOPNAME_INACTIVE, { .vop_inactive = ufs_inactive }, /* not blkd */ 214 VOPNAME_FID, { .vop_fid = ufs_fid }, 215 VOPNAME_RWLOCK, { .vop_rwlock = ufs_rwlock }, /* not blkd */ 216 VOPNAME_RWUNLOCK, { .vop_rwunlock = ufs_rwunlock }, /* not blkd */ 217 VOPNAME_SEEK, { .vop_seek = ufs_seek }, 218 VOPNAME_FRLOCK, { .vop_frlock = ufs_frlock }, 219 VOPNAME_SPACE, { .vop_space = ufs_space }, 220 VOPNAME_GETPAGE, { .vop_getpage = ufs_getpage }, 221 VOPNAME_PUTPAGE, { .vop_putpage = ufs_putpage }, 222 VOPNAME_MAP, { .vop_map = ufs_map }, 223 VOPNAME_ADDMAP, { .vop_addmap = ufs_addmap }, /* not blkd */ 224 VOPNAME_DELMAP, { .vop_delmap = ufs_delmap }, /* not blkd */ 225 VOPNAME_POLL, { .vop_poll = ufs_poll }, /* not blkd */ 226 VOPNAME_DUMP, { .vop_dump = ufs_dump }, 227 VOPNAME_PATHCONF, { .vop_pathconf = ufs_l_pathconf }, 228 VOPNAME_PAGEIO, { .vop_pageio = ufs_pageio }, 229 VOPNAME_DUMPCTL, { .vop_dumpctl = ufs_dumpctl }, 230 VOPNAME_GETSECATTR, { .vop_getsecattr = ufs_getsecattr }, 231 VOPNAME_SETSECATTR, { .vop_setsecattr = ufs_setsecattr }, 232 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 233 NULL, NULL 234 }; 235 236 #define MAX_BACKFILE_COUNT 9999 237 238 /* 239 * Created by ufs_dumpctl() to store a file's disk block info into memory. 240 * Used by ufs_dump() to dump data to disk directly. 241 */ 242 struct dump { 243 struct inode *ip; /* the file we contain */ 244 daddr_t fsbs; /* number of blocks stored */ 245 struct timeval32 time; /* time stamp for the struct */ 246 daddr32_t dblk[1]; /* place holder for block info */ 247 }; 248 249 static struct dump *dump_info = NULL; 250 251 /* 252 * Previously there was no special action required for ordinary files. 253 * (Devices are handled through the device file system.) 254 * Now we support Large Files and Large File API requires open to 255 * fail if file is large. 256 * We could take care to prevent data corruption 257 * by doing an atomic check of size and truncate if file is opened with 258 * FTRUNC flag set but traditionally this is being done by the vfs/vnode 259 * layers. So taking care of truncation here is a change in the existing 260 * semantics of VOP_OPEN and therefore we chose not to implement any thing 261 * here. The check for the size of the file > 2GB is being done at the 262 * vfs layer in routine vn_open(). 263 */ 264 265 /* ARGSUSED */ 266 static int 267 ufs_open(struct vnode **vpp, int flag, struct cred *cr, caller_context_t *ct) 268 { 269 return (0); 270 } 271 272 /*ARGSUSED*/ 273 static int 274 ufs_close(struct vnode *vp, int flag, int count, offset_t offset, 275 struct cred *cr, caller_context_t *ct) 276 { 277 cleanlocks(vp, ttoproc(curthread)->p_pid, 0); 278 cleanshares(vp, ttoproc(curthread)->p_pid); 279 280 /* 281 * Push partially filled cluster at last close. 282 * ``last close'' is approximated because the dnlc 283 * may have a hold on the vnode. 284 * Checking for VBAD here will also act as a forced umount check. 285 */ 286 if (vp->v_count <= 2 && vp->v_type != VBAD) { 287 struct inode *ip = VTOI(vp); 288 if (ip->i_delaylen) { 289 ins.in_poc.value.ul++; 290 (void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen, 291 B_ASYNC | B_FREE, cr); 292 ip->i_delaylen = 0; 293 } 294 } 295 296 return (0); 297 } 298 299 /*ARGSUSED*/ 300 static int 301 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr, 302 struct caller_context *ct) 303 { 304 struct inode *ip = VTOI(vp); 305 struct ufsvfs *ufsvfsp; 306 struct ulockfs *ulp = NULL; 307 int error = 0; 308 int intrans = 0; 309 310 ASSERT(RW_READ_HELD(&ip->i_rwlock)); 311 312 /* 313 * Mandatory locking needs to be done before ufs_lockfs_begin() 314 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep. 315 */ 316 if (MANDLOCK(vp, ip->i_mode)) { 317 /* 318 * ufs_getattr ends up being called by chklock 319 */ 320 error = chklock(vp, FREAD, uiop->uio_loffset, 321 uiop->uio_resid, uiop->uio_fmode, ct); 322 if (error) 323 goto out; 324 } 325 326 ufsvfsp = ip->i_ufsvfs; 327 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK); 328 if (error) 329 goto out; 330 331 /* 332 * In the case that a directory is opened for reading as a file 333 * (eg "cat .") with the O_RSYNC, O_SYNC and O_DSYNC flags set. 334 * The locking order had to be changed to avoid a deadlock with 335 * an update taking place on that directory at the same time. 336 */ 337 if ((ip->i_mode & IFMT) == IFDIR) { 338 339 rw_enter(&ip->i_contents, RW_READER); 340 error = rdip(ip, uiop, ioflag, cr); 341 rw_exit(&ip->i_contents); 342 343 if (error) { 344 if (ulp) 345 ufs_lockfs_end(ulp); 346 goto out; 347 } 348 349 if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) && 350 TRANS_ISTRANS(ufsvfsp)) { 351 rw_exit(&ip->i_rwlock); 352 TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE, 353 error); 354 ASSERT(!error); 355 TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC, 356 TOP_READ_SIZE); 357 rw_enter(&ip->i_rwlock, RW_READER); 358 } 359 } else { 360 /* 361 * Only transact reads to files opened for sync-read and 362 * sync-write on a file system that is not write locked. 363 * 364 * The ``not write locked'' check prevents problems with 365 * enabling/disabling logging on a busy file system. E.g., 366 * logging exists at the beginning of the read but does not 367 * at the end. 368 * 369 */ 370 if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) && 371 TRANS_ISTRANS(ufsvfsp)) { 372 TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE, 373 error); 374 ASSERT(!error); 375 intrans = 1; 376 } 377 378 rw_enter(&ip->i_contents, RW_READER); 379 error = rdip(ip, uiop, ioflag, cr); 380 rw_exit(&ip->i_contents); 381 382 if (intrans) { 383 TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC, 384 TOP_READ_SIZE); 385 } 386 } 387 388 if (ulp) { 389 ufs_lockfs_end(ulp); 390 } 391 out: 392 393 return (error); 394 } 395 396 extern int ufs_HW; /* high water mark */ 397 extern int ufs_LW; /* low water mark */ 398 int ufs_WRITES = 1; /* XXX - enable/disable */ 399 int ufs_throttles = 0; /* throttling count */ 400 int ufs_allow_shared_writes = 1; /* directio shared writes */ 401 402 static int 403 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag) 404 { 405 int shared_write; 406 407 /* 408 * If the FDSYNC flag is set then ignore the global 409 * ufs_allow_shared_writes in this case. 410 */ 411 shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes; 412 413 /* 414 * Filter to determine if this request is suitable as a 415 * concurrent rewrite. This write must not allocate blocks 416 * by extending the file or filling in holes. No use trying 417 * through FSYNC descriptors as the inode will be synchronously 418 * updated after the write. The uio structure has not yet been 419 * checked for sanity, so assume nothing. 420 */ 421 return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) && 422 (uiop->uio_loffset >= (offset_t)0) && 423 (uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) && 424 ((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) && 425 !(ioflag & FSYNC) && !bmap_has_holes(ip) && 426 shared_write); 427 } 428 429 /*ARGSUSED*/ 430 static int 431 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr, 432 caller_context_t *ct) 433 { 434 struct inode *ip = VTOI(vp); 435 struct ufsvfs *ufsvfsp; 436 struct ulockfs *ulp; 437 int retry = 1; 438 int error, resv, resid = 0; 439 int directio_status; 440 int exclusive; 441 int rewriteflg; 442 long start_resid = uiop->uio_resid; 443 444 ASSERT(RW_LOCK_HELD(&ip->i_rwlock)); 445 446 retry_mandlock: 447 /* 448 * Mandatory locking needs to be done before ufs_lockfs_begin() 449 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep. 450 * Check for forced unmounts normally done in ufs_lockfs_begin(). 451 */ 452 if ((ufsvfsp = ip->i_ufsvfs) == NULL) { 453 error = EIO; 454 goto out; 455 } 456 if (MANDLOCK(vp, ip->i_mode)) { 457 458 ASSERT(RW_WRITE_HELD(&ip->i_rwlock)); 459 460 /* 461 * ufs_getattr ends up being called by chklock 462 */ 463 error = chklock(vp, FWRITE, uiop->uio_loffset, 464 uiop->uio_resid, uiop->uio_fmode, ct); 465 if (error) 466 goto out; 467 } 468 469 /* i_rwlock can change in chklock */ 470 exclusive = rw_write_held(&ip->i_rwlock); 471 rewriteflg = ufs_check_rewrite(ip, uiop, ioflag); 472 473 /* 474 * Check for fast-path special case of directio re-writes. 475 */ 476 if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) && 477 !exclusive && rewriteflg) { 478 479 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK); 480 if (error) 481 goto out; 482 483 rw_enter(&ip->i_contents, RW_READER); 484 error = ufs_directio_write(ip, uiop, ioflag, 1, cr, 485 &directio_status); 486 if (directio_status == DIRECTIO_SUCCESS) { 487 uint_t i_flag_save; 488 489 if (start_resid != uiop->uio_resid) 490 error = 0; 491 /* 492 * Special treatment of access times for re-writes. 493 * If IMOD is not already set, then convert it 494 * to IMODACC for this operation. This defers 495 * entering a delta into the log until the inode 496 * is flushed. This mimics what is done for read 497 * operations and inode access time. 498 */ 499 mutex_enter(&ip->i_tlock); 500 i_flag_save = ip->i_flag; 501 ip->i_flag |= IUPD | ICHG; 502 ip->i_seq++; 503 ITIMES_NOLOCK(ip); 504 if ((i_flag_save & IMOD) == 0) { 505 ip->i_flag &= ~IMOD; 506 ip->i_flag |= IMODACC; 507 } 508 mutex_exit(&ip->i_tlock); 509 rw_exit(&ip->i_contents); 510 if (ulp) 511 ufs_lockfs_end(ulp); 512 goto out; 513 } 514 rw_exit(&ip->i_contents); 515 if (ulp) 516 ufs_lockfs_end(ulp); 517 } 518 519 if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) { 520 rw_exit(&ip->i_rwlock); 521 rw_enter(&ip->i_rwlock, RW_WRITER); 522 /* 523 * Mandatory locking could have been enabled 524 * after dropping the i_rwlock. 525 */ 526 if (MANDLOCK(vp, ip->i_mode)) 527 goto retry_mandlock; 528 } 529 530 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK); 531 if (error) 532 goto out; 533 534 /* 535 * Amount of log space needed for this write 536 */ 537 if (!rewriteflg || !(ioflag & FDSYNC)) 538 TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid); 539 540 /* 541 * Throttle writes. 542 */ 543 if (ufs_WRITES && (ip->i_writes > ufs_HW)) { 544 mutex_enter(&ip->i_tlock); 545 while (ip->i_writes > ufs_HW) { 546 ufs_throttles++; 547 cv_wait(&ip->i_wrcv, &ip->i_tlock); 548 } 549 mutex_exit(&ip->i_tlock); 550 } 551 552 /* 553 * Enter Transaction 554 * 555 * If the write is a rewrite there is no need to open a transaction 556 * if the FDSYNC flag is set and not the FSYNC. In this case just 557 * set the IMODACC flag to modify do the update at a later time 558 * thus avoiding the overhead of the logging transaction that is 559 * not required. 560 */ 561 if (ioflag & (FSYNC|FDSYNC)) { 562 if (ulp) { 563 if (rewriteflg) { 564 uint_t i_flag_save; 565 566 rw_enter(&ip->i_contents, RW_READER); 567 mutex_enter(&ip->i_tlock); 568 i_flag_save = ip->i_flag; 569 ip->i_flag |= IUPD | ICHG; 570 ip->i_seq++; 571 ITIMES_NOLOCK(ip); 572 if ((i_flag_save & IMOD) == 0) { 573 ip->i_flag &= ~IMOD; 574 ip->i_flag |= IMODACC; 575 } 576 mutex_exit(&ip->i_tlock); 577 rw_exit(&ip->i_contents); 578 } else { 579 int terr = 0; 580 TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv, 581 terr); 582 ASSERT(!terr); 583 } 584 } 585 } else { 586 if (ulp) 587 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv); 588 } 589 590 /* 591 * Write the file 592 */ 593 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 594 rw_enter(&ip->i_contents, RW_WRITER); 595 if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) { 596 /* 597 * In append mode start at end of file. 598 */ 599 uiop->uio_loffset = ip->i_size; 600 } 601 602 /* 603 * Mild optimisation, don't call ufs_trans_write() unless we have to 604 * Also, suppress file system full messages if we will retry. 605 */ 606 if (retry) 607 ip->i_flag |= IQUIET; 608 if (resid) { 609 TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid); 610 } else { 611 error = wrip(ip, uiop, ioflag, cr); 612 } 613 ip->i_flag &= ~IQUIET; 614 615 rw_exit(&ip->i_contents); 616 rw_exit(&ufsvfsp->vfs_dqrwlock); 617 618 /* 619 * Leave Transaction 620 */ 621 if (ulp) { 622 if (ioflag & (FSYNC|FDSYNC)) { 623 if (!rewriteflg) { 624 int terr = 0; 625 626 TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC, 627 resv); 628 if (error == 0) 629 error = terr; 630 } 631 } else { 632 TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv); 633 } 634 ufs_lockfs_end(ulp); 635 } 636 out: 637 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) { 638 /* 639 * Any blocks tied up in pending deletes? 640 */ 641 ufs_delete_drain_wait(ufsvfsp, 1); 642 retry = 0; 643 goto retry_mandlock; 644 } 645 646 if (error == ENOSPC && (start_resid != uiop->uio_resid)) 647 error = 0; 648 649 return (error); 650 } 651 652 /* 653 * Don't cache write blocks to files with the sticky bit set. 654 * Used to keep swap files from blowing the page cache on a server. 655 */ 656 int stickyhack = 1; 657 658 /* 659 * Free behind hacks. The pager is busted. 660 * XXX - need to pass the information down to writedone() in a flag like B_SEQ 661 * or B_FREE_IF_TIGHT_ON_MEMORY. 662 */ 663 int freebehind = 1; 664 int smallfile = 0; 665 u_offset_t smallfile64 = 32 * 1024; 666 667 /* 668 * While we should, in most cases, cache the pages for write, we 669 * may also want to cache the pages for read as long as they are 670 * frequently re-usable. 671 * 672 * If cache_read_ahead = 1, the pages for read will go to the tail 673 * of the cache list when they are released, otherwise go to the head. 674 */ 675 int cache_read_ahead = 0; 676 677 /* 678 * Freebehind exists so that as we read large files sequentially we 679 * don't consume most of memory with pages from a few files. It takes 680 * longer to re-read from disk multiple small files as it does reading 681 * one large one sequentially. As system memory grows customers need 682 * to retain bigger chunks of files in memory. The advent of the 683 * cachelist opens up of the possibility freeing pages to the head or 684 * tail of the list. 685 * 686 * Not freeing a page is a bet that the page will be read again before 687 * it's segmap slot is needed for something else. If we loose the bet, 688 * it means some other thread is burdened with the page free we did 689 * not do. If we win we save a free and reclaim. 690 * 691 * Freeing it at the tail vs the head of cachelist is a bet that the 692 * page will survive until the next read. It's also saying that this 693 * page is more likely to be re-used than a page freed some time ago 694 * and never reclaimed. 695 * 696 * Freebehind maintains a range of file offset [smallfile1; smallfile2] 697 * 698 * 0 < offset < smallfile1 : pages are not freed. 699 * smallfile1 < offset < smallfile2 : pages freed to tail of cachelist. 700 * smallfile2 < offset : pages freed to head of cachelist. 701 * 702 * The range is computed at most once per second and depends on 703 * freemem and ncpus_online. Both parameters are bounded to be 704 * >= smallfile && >= smallfile64. 705 * 706 * smallfile1 = (free memory / ncpu) / 1000 707 * smallfile2 = (free memory / ncpu) / 10 708 * 709 * A few examples values: 710 * 711 * Free Mem (in Bytes) [smallfile1; smallfile2] [smallfile1; smallfile2] 712 * ncpus_online = 4 ncpus_online = 64 713 * ------------------ ----------------------- ----------------------- 714 * 1G [256K; 25M] [32K; 1.5M] 715 * 10G [2.5M; 250M] [156K; 15M] 716 * 100G [25M; 2.5G] [1.5M; 150M] 717 * 718 */ 719 720 #define SMALLFILE1_D 1000 721 #define SMALLFILE2_D 10 722 static u_offset_t smallfile1 = 32 * 1024; 723 static u_offset_t smallfile2 = 32 * 1024; 724 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */ 725 uint_t smallfile1_d = SMALLFILE1_D; 726 uint_t smallfile2_d = SMALLFILE2_D; 727 728 /* 729 * wrip does the real work of write requests for ufs. 730 */ 731 int 732 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr) 733 { 734 rlim64_t limit = uio->uio_llimit; 735 u_offset_t off; 736 u_offset_t old_i_size; 737 struct fs *fs; 738 struct vnode *vp; 739 struct ufsvfs *ufsvfsp; 740 caddr_t base; 741 long start_resid = uio->uio_resid; /* save starting resid */ 742 long premove_resid; /* resid before uiomove() */ 743 uint_t flags; 744 int newpage; 745 int iupdat_flag, directio_status; 746 int n, on, mapon; 747 int error, pagecreate; 748 int do_dqrwlock; /* drop/reacquire vfs_dqrwlock */ 749 int32_t iblocks; 750 int new_iblocks; 751 752 /* 753 * ip->i_size is incremented before the uiomove 754 * is done on a write. If the move fails (bad user 755 * address) reset ip->i_size. 756 * The better way would be to increment ip->i_size 757 * only if the uiomove succeeds. 758 */ 759 int i_size_changed = 0; 760 o_mode_t type; 761 int i_seq_needed = 0; 762 763 vp = ITOV(ip); 764 765 /* 766 * check for forced unmount - should not happen as 767 * the request passed the lockfs checks. 768 */ 769 if ((ufsvfsp = ip->i_ufsvfs) == NULL) 770 return (EIO); 771 772 fs = ip->i_fs; 773 774 ASSERT(RW_WRITE_HELD(&ip->i_contents)); 775 776 /* check for valid filetype */ 777 type = ip->i_mode & IFMT; 778 if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) && 779 (type != IFLNK) && (type != IFSHAD)) { 780 return (EIO); 781 } 782 783 /* 784 * the actual limit of UFS file size 785 * is UFS_MAXOFFSET_T 786 */ 787 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 788 limit = MAXOFFSET_T; 789 790 if (uio->uio_loffset >= limit) { 791 proc_t *p = ttoproc(curthread); 792 793 mutex_enter(&p->p_lock); 794 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls, 795 p, RCA_UNSAFE_SIGINFO); 796 mutex_exit(&p->p_lock); 797 return (EFBIG); 798 } 799 800 /* 801 * if largefiles are disallowed, the limit is 802 * the pre-largefiles value of 2GB 803 */ 804 if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES) 805 limit = MIN(UFS_MAXOFFSET_T, limit); 806 else 807 limit = MIN(MAXOFF32_T, limit); 808 809 if (uio->uio_loffset < (offset_t)0) { 810 return (EINVAL); 811 } 812 if (uio->uio_resid == 0) { 813 return (0); 814 } 815 816 if (uio->uio_loffset >= limit) 817 return (EFBIG); 818 819 ip->i_flag |= INOACC; /* don't update ref time in getpage */ 820 821 if (ioflag & (FSYNC|FDSYNC)) { 822 ip->i_flag |= ISYNC; 823 iupdat_flag = 1; 824 } 825 /* 826 * Try to go direct 827 */ 828 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) { 829 uio->uio_llimit = limit; 830 error = ufs_directio_write(ip, uio, ioflag, 0, cr, 831 &directio_status); 832 /* 833 * If ufs_directio wrote to the file or set the flags, 834 * we need to update i_seq, but it may be deferred. 835 */ 836 if (start_resid != uio->uio_resid || 837 (ip->i_flag & (ICHG|IUPD))) { 838 i_seq_needed = 1; 839 ip->i_flag |= ISEQ; 840 } 841 if (directio_status == DIRECTIO_SUCCESS) 842 goto out; 843 } 844 845 /* 846 * Behavior with respect to dropping/reacquiring vfs_dqrwlock: 847 * 848 * o shadow inodes: vfs_dqrwlock is not held at all 849 * o quota updates: vfs_dqrwlock is read or write held 850 * o other updates: vfs_dqrwlock is read held 851 * 852 * The first case is the only one where we do not hold 853 * vfs_dqrwlock at all while entering wrip(). 854 * We must make sure not to downgrade/drop vfs_dqrwlock if we 855 * have it as writer, i.e. if we are updating the quota inode. 856 * There is no potential deadlock scenario in this case as 857 * ufs_getpage() takes care of this and avoids reacquiring 858 * vfs_dqrwlock in that case. 859 * 860 * This check is done here since the above conditions do not change 861 * and we possibly loop below, so save a few cycles. 862 */ 863 if ((type == IFSHAD) || 864 (rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) { 865 do_dqrwlock = 0; 866 } else { 867 do_dqrwlock = 1; 868 } 869 870 /* 871 * Large Files: We cast MAXBMASK to offset_t 872 * inorder to mask out the higher bits. Since offset_t 873 * is a signed value, the high order bit set in MAXBMASK 874 * value makes it do the right thing by having all bits 1 875 * in the higher word. May be removed for _SOLARIS64_. 876 */ 877 878 fs = ip->i_fs; 879 do { 880 u_offset_t uoff = uio->uio_loffset; 881 off = uoff & (offset_t)MAXBMASK; 882 mapon = (int)(uoff & (offset_t)MAXBOFFSET); 883 on = (int)blkoff(fs, uoff); 884 n = (int)MIN(fs->fs_bsize - on, uio->uio_resid); 885 new_iblocks = 1; 886 887 if (type == IFREG && uoff + n >= limit) { 888 if (uoff >= limit) { 889 error = EFBIG; 890 goto out; 891 } 892 /* 893 * since uoff + n >= limit, 894 * therefore n >= limit - uoff, and n is an int 895 * so it is safe to cast it to an int 896 */ 897 n = (int)(limit - (rlim64_t)uoff); 898 } 899 if (uoff + n > ip->i_size) { 900 /* 901 * We are extending the length of the file. 902 * bmap is used so that we are sure that 903 * if we need to allocate new blocks, that it 904 * is done here before we up the file size. 905 */ 906 error = bmap_write(ip, uoff, (int)(on + n), 907 mapon == 0, NULL, cr); 908 /* 909 * bmap_write never drops i_contents so if 910 * the flags are set it changed the file. 911 */ 912 if (ip->i_flag & (ICHG|IUPD)) { 913 i_seq_needed = 1; 914 ip->i_flag |= ISEQ; 915 } 916 if (error) 917 break; 918 /* 919 * There is a window of vulnerability here. 920 * The sequence of operations: allocate file 921 * system blocks, uiomove the data into pages, 922 * and then update the size of the file in the 923 * inode, must happen atomically. However, due 924 * to current locking constraints, this can not 925 * be done. 926 */ 927 ASSERT(ip->i_writer == NULL); 928 ip->i_writer = curthread; 929 i_size_changed = 1; 930 /* 931 * If we are writing from the beginning of 932 * the mapping, we can just create the 933 * pages without having to read them. 934 */ 935 pagecreate = (mapon == 0); 936 } else if (n == MAXBSIZE) { 937 /* 938 * Going to do a whole mappings worth, 939 * so we can just create the pages w/o 940 * having to read them in. But before 941 * we do that, we need to make sure any 942 * needed blocks are allocated first. 943 */ 944 iblocks = ip->i_blocks; 945 error = bmap_write(ip, uoff, (int)(on + n), 946 BI_ALLOC_ONLY, NULL, cr); 947 /* 948 * bmap_write never drops i_contents so if 949 * the flags are set it changed the file. 950 */ 951 if (ip->i_flag & (ICHG|IUPD)) { 952 i_seq_needed = 1; 953 ip->i_flag |= ISEQ; 954 } 955 if (error) 956 break; 957 pagecreate = 1; 958 /* 959 * check if the new created page needed the 960 * allocation of new disk blocks. 961 */ 962 if (iblocks == ip->i_blocks) 963 new_iblocks = 0; /* no new blocks allocated */ 964 } else { 965 pagecreate = 0; 966 /* 967 * In sync mode flush the indirect blocks which 968 * may have been allocated and not written on 969 * disk. In above cases bmap_write will allocate 970 * in sync mode. 971 */ 972 if (ioflag & (FSYNC|FDSYNC)) { 973 error = ufs_indirblk_sync(ip, uoff); 974 if (error) 975 break; 976 } 977 } 978 979 /* 980 * At this point we can enter ufs_getpage() in one 981 * of two ways: 982 * 1) segmap_getmapflt() calls ufs_getpage() when the 983 * forcefault parameter is true (pagecreate == 0) 984 * 2) uiomove() causes a page fault. 985 * 986 * We have to drop the contents lock to prevent the VM 987 * system from trying to reacquire it in ufs_getpage() 988 * should the uiomove cause a pagefault. 989 * 990 * We have to drop the reader vfs_dqrwlock here as well. 991 */ 992 rw_exit(&ip->i_contents); 993 if (do_dqrwlock) { 994 ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock)); 995 ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock))); 996 rw_exit(&ufsvfsp->vfs_dqrwlock); 997 } 998 999 newpage = 0; 1000 premove_resid = uio->uio_resid; 1001 if (vpm_enable) { 1002 /* 1003 * Copy data. If new pages are created, part of 1004 * the page that is not written will be initizliazed 1005 * with zeros. 1006 */ 1007 error = vpm_data_copy(vp, (off + mapon), (uint_t)n, 1008 uio, !pagecreate, &newpage, 0, S_WRITE); 1009 } else { 1010 1011 base = segmap_getmapflt(segkmap, vp, (off + mapon), 1012 (uint_t)n, !pagecreate, S_WRITE); 1013 1014 /* 1015 * segmap_pagecreate() returns 1 if it calls 1016 * page_create_va() to allocate any pages. 1017 */ 1018 1019 if (pagecreate) 1020 newpage = segmap_pagecreate(segkmap, base, 1021 (size_t)n, 0); 1022 1023 error = uiomove(base + mapon, (long)n, UIO_WRITE, uio); 1024 } 1025 1026 /* 1027 * If "newpage" is set, then a new page was created and it 1028 * does not contain valid data, so it needs to be initialized 1029 * at this point. 1030 * Otherwise the page contains old data, which was overwritten 1031 * partially or as a whole in uiomove. 1032 * If there is only one iovec structure within uio, then 1033 * on error uiomove will not be able to update uio->uio_loffset 1034 * and we would zero the whole page here! 1035 * 1036 * If uiomove fails because of an error, the old valid data 1037 * is kept instead of filling the rest of the page with zero's. 1038 */ 1039 if (!vpm_enable && newpage && 1040 uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) { 1041 /* 1042 * We created pages w/o initializing them completely, 1043 * thus we need to zero the part that wasn't set up. 1044 * This happens on most EOF write cases and if 1045 * we had some sort of error during the uiomove. 1046 */ 1047 int nzero, nmoved; 1048 1049 nmoved = (int)(uio->uio_loffset - (off + mapon)); 1050 ASSERT(nmoved >= 0 && nmoved <= n); 1051 nzero = roundup(on + n, PAGESIZE) - nmoved; 1052 ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE); 1053 (void) kzero(base + mapon + nmoved, (uint_t)nzero); 1054 } 1055 1056 /* 1057 * Unlock the pages allocated by page_create_va() 1058 * in segmap_pagecreate() 1059 */ 1060 if (!vpm_enable && newpage) 1061 segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE); 1062 1063 /* 1064 * If the size of the file changed, then update the 1065 * size field in the inode now. This can't be done 1066 * before the call to segmap_pageunlock or there is 1067 * a potential deadlock with callers to ufs_putpage(). 1068 * They will be holding i_contents and trying to lock 1069 * a page, while this thread is holding a page locked 1070 * and trying to acquire i_contents. 1071 */ 1072 if (i_size_changed) { 1073 rw_enter(&ip->i_contents, RW_WRITER); 1074 old_i_size = ip->i_size; 1075 UFS_SET_ISIZE(uoff + n, ip); 1076 TRANS_INODE(ufsvfsp, ip); 1077 /* 1078 * file has grown larger than 2GB. Set flag 1079 * in superblock to indicate this, if it 1080 * is not already set. 1081 */ 1082 if ((ip->i_size > MAXOFF32_T) && 1083 !(fs->fs_flags & FSLARGEFILES)) { 1084 ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES); 1085 mutex_enter(&ufsvfsp->vfs_lock); 1086 fs->fs_flags |= FSLARGEFILES; 1087 ufs_sbwrite(ufsvfsp); 1088 mutex_exit(&ufsvfsp->vfs_lock); 1089 } 1090 mutex_enter(&ip->i_tlock); 1091 ip->i_writer = NULL; 1092 cv_broadcast(&ip->i_wrcv); 1093 mutex_exit(&ip->i_tlock); 1094 rw_exit(&ip->i_contents); 1095 } 1096 1097 if (error) { 1098 /* 1099 * If we failed on a write, we may have already 1100 * allocated file blocks as well as pages. It's 1101 * hard to undo the block allocation, but we must 1102 * be sure to invalidate any pages that may have 1103 * been allocated. 1104 * 1105 * If the page was created without initialization 1106 * then we must check if it should be possible 1107 * to destroy the new page and to keep the old data 1108 * on the disk. 1109 * 1110 * It is possible to destroy the page without 1111 * having to write back its contents only when 1112 * - the size of the file keeps unchanged 1113 * - bmap_write() did not allocate new disk blocks 1114 * it is possible to create big files using "seek" and 1115 * write to the end of the file. A "write" to a 1116 * position before the end of the file would not 1117 * change the size of the file but it would allocate 1118 * new disk blocks. 1119 * - uiomove intended to overwrite the whole page. 1120 * - a new page was created (newpage == 1). 1121 */ 1122 1123 if (i_size_changed == 0 && new_iblocks == 0 && 1124 newpage) { 1125 1126 /* unwind what uiomove eventually last did */ 1127 uio->uio_resid = premove_resid; 1128 1129 /* 1130 * destroy the page, do not write ambiguous 1131 * data to the disk. 1132 */ 1133 flags = SM_DESTROY; 1134 } else { 1135 /* 1136 * write the page back to the disk, if dirty, 1137 * and remove the page from the cache. 1138 */ 1139 flags = SM_INVAL; 1140 } 1141 1142 if (vpm_enable) { 1143 /* 1144 * Flush pages. 1145 */ 1146 (void) vpm_sync_pages(vp, off, n, flags); 1147 } else { 1148 (void) segmap_release(segkmap, base, flags); 1149 } 1150 } else { 1151 flags = 0; 1152 /* 1153 * Force write back for synchronous write cases. 1154 */ 1155 if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) { 1156 /* 1157 * If the sticky bit is set but the 1158 * execute bit is not set, we do a 1159 * synchronous write back and free 1160 * the page when done. We set up swap 1161 * files to be handled this way to 1162 * prevent servers from keeping around 1163 * the client's swap pages too long. 1164 * XXX - there ought to be a better way. 1165 */ 1166 if (IS_SWAPVP(vp)) { 1167 flags = SM_WRITE | SM_FREE | 1168 SM_DONTNEED; 1169 iupdat_flag = 0; 1170 } else { 1171 flags = SM_WRITE; 1172 } 1173 } else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) { 1174 /* 1175 * Have written a whole block. 1176 * Start an asynchronous write and 1177 * mark the buffer to indicate that 1178 * it won't be needed again soon. 1179 */ 1180 flags = SM_WRITE | SM_ASYNC | SM_DONTNEED; 1181 } 1182 if (vpm_enable) { 1183 /* 1184 * Flush pages. 1185 */ 1186 error = vpm_sync_pages(vp, off, n, flags); 1187 } else { 1188 error = segmap_release(segkmap, base, flags); 1189 } 1190 /* 1191 * If the operation failed and is synchronous, 1192 * then we need to unwind what uiomove() last 1193 * did so we can potentially return an error to 1194 * the caller. If this write operation was 1195 * done in two pieces and the first succeeded, 1196 * then we won't return an error for the second 1197 * piece that failed. However, we only want to 1198 * return a resid value that reflects what was 1199 * really done. 1200 * 1201 * Failures for non-synchronous operations can 1202 * be ignored since the page subsystem will 1203 * retry the operation until it succeeds or the 1204 * file system is unmounted. 1205 */ 1206 if (error) { 1207 if ((ioflag & (FSYNC | FDSYNC)) || 1208 type == IFDIR) { 1209 uio->uio_resid = premove_resid; 1210 } else { 1211 error = 0; 1212 } 1213 } 1214 } 1215 1216 /* 1217 * Re-acquire contents lock. 1218 * If it was dropped, reacquire reader vfs_dqrwlock as well. 1219 */ 1220 if (do_dqrwlock) 1221 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 1222 rw_enter(&ip->i_contents, RW_WRITER); 1223 1224 /* 1225 * If the uiomove() failed or if a synchronous 1226 * page push failed, fix up i_size. 1227 */ 1228 if (error) { 1229 if (i_size_changed) { 1230 /* 1231 * The uiomove failed, and we 1232 * allocated blocks,so get rid 1233 * of them. 1234 */ 1235 (void) ufs_itrunc(ip, old_i_size, 0, cr); 1236 } 1237 } else { 1238 /* 1239 * XXX - Can this be out of the loop? 1240 */ 1241 ip->i_flag |= IUPD | ICHG; 1242 /* 1243 * Only do one increase of i_seq for multiple 1244 * pieces. Because we drop locks, record 1245 * the fact that we changed the timestamp and 1246 * are deferring the increase in case another thread 1247 * pushes our timestamp update. 1248 */ 1249 i_seq_needed = 1; 1250 ip->i_flag |= ISEQ; 1251 if (i_size_changed) 1252 ip->i_flag |= IATTCHG; 1253 if ((ip->i_mode & (IEXEC | (IEXEC >> 3) | 1254 (IEXEC >> 6))) != 0 && 1255 (ip->i_mode & (ISUID | ISGID)) != 0 && 1256 secpolicy_vnode_setid_retain(cr, 1257 (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) { 1258 /* 1259 * Clear Set-UID & Set-GID bits on 1260 * successful write if not privileged 1261 * and at least one of the execute bits 1262 * is set. If we always clear Set-GID, 1263 * mandatory file and record locking is 1264 * unuseable. 1265 */ 1266 ip->i_mode &= ~(ISUID | ISGID); 1267 } 1268 } 1269 /* 1270 * In the case the FDSYNC flag is set and this is a 1271 * "rewrite" we won't log a delta. 1272 * The FSYNC flag overrides all cases. 1273 */ 1274 if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) { 1275 TRANS_INODE(ufsvfsp, ip); 1276 } 1277 } while (error == 0 && uio->uio_resid > 0 && n != 0); 1278 1279 out: 1280 /* 1281 * Make sure i_seq is increased at least once per write 1282 */ 1283 if (i_seq_needed) { 1284 ip->i_seq++; 1285 ip->i_flag &= ~ISEQ; /* no longer deferred */ 1286 } 1287 1288 /* 1289 * Inode is updated according to this table - 1290 * 1291 * FSYNC FDSYNC(posix.4) 1292 * -------------------------- 1293 * always@ IATTCHG|IBDWRITE 1294 * 1295 * @ - If we are doing synchronous write the only time we should 1296 * not be sync'ing the ip here is if we have the stickyhack 1297 * activated, the file is marked with the sticky bit and 1298 * no exec bit, the file length has not been changed and 1299 * no new blocks have been allocated during this write. 1300 */ 1301 1302 if ((ip->i_flag & ISYNC) != 0) { 1303 /* 1304 * we have eliminated nosync 1305 */ 1306 if ((ip->i_flag & (IATTCHG|IBDWRITE)) || 1307 ((ioflag & FSYNC) && iupdat_flag)) { 1308 ufs_iupdat(ip, 1); 1309 } 1310 } 1311 1312 /* 1313 * If we've already done a partial-write, terminate 1314 * the write but return no error unless the error is ENOSPC 1315 * because the caller can detect this and free resources and 1316 * try again. 1317 */ 1318 if ((start_resid != uio->uio_resid) && (error != ENOSPC)) 1319 error = 0; 1320 1321 ip->i_flag &= ~(INOACC | ISYNC); 1322 ITIMES_NOLOCK(ip); 1323 return (error); 1324 } 1325 1326 /* 1327 * rdip does the real work of read requests for ufs. 1328 */ 1329 int 1330 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr) 1331 { 1332 u_offset_t off; 1333 caddr_t base; 1334 struct fs *fs; 1335 struct ufsvfs *ufsvfsp; 1336 struct vnode *vp; 1337 long oresid = uio->uio_resid; 1338 u_offset_t n, on, mapon; 1339 int error = 0; 1340 int doupdate = 1; 1341 uint_t flags; 1342 int dofree, directio_status; 1343 krw_t rwtype; 1344 o_mode_t type; 1345 1346 vp = ITOV(ip); 1347 1348 ASSERT(RW_LOCK_HELD(&ip->i_contents)); 1349 1350 ufsvfsp = ip->i_ufsvfs; 1351 1352 if (ufsvfsp == NULL) 1353 return (EIO); 1354 1355 fs = ufsvfsp->vfs_fs; 1356 1357 /* check for valid filetype */ 1358 type = ip->i_mode & IFMT; 1359 if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) && 1360 (type != IFLNK) && (type != IFSHAD)) { 1361 return (EIO); 1362 } 1363 1364 if (uio->uio_loffset > UFS_MAXOFFSET_T) { 1365 error = 0; 1366 goto out; 1367 } 1368 if (uio->uio_loffset < (offset_t)0) { 1369 return (EINVAL); 1370 } 1371 if (uio->uio_resid == 0) { 1372 return (0); 1373 } 1374 1375 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) && 1376 (!ufsvfsp->vfs_noatime)) { 1377 mutex_enter(&ip->i_tlock); 1378 ip->i_flag |= IACC; 1379 mutex_exit(&ip->i_tlock); 1380 } 1381 /* 1382 * Try to go direct 1383 */ 1384 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) { 1385 error = ufs_directio_read(ip, uio, cr, &directio_status); 1386 if (directio_status == DIRECTIO_SUCCESS) 1387 goto out; 1388 } 1389 1390 rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER); 1391 1392 do { 1393 offset_t diff; 1394 u_offset_t uoff = uio->uio_loffset; 1395 off = uoff & (offset_t)MAXBMASK; 1396 mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET); 1397 on = (u_offset_t)blkoff(fs, uoff); 1398 n = MIN((u_offset_t)fs->fs_bsize - on, 1399 (u_offset_t)uio->uio_resid); 1400 1401 diff = ip->i_size - uoff; 1402 1403 if (diff <= (offset_t)0) { 1404 error = 0; 1405 goto out; 1406 } 1407 if (diff < (offset_t)n) 1408 n = (int)diff; 1409 1410 /* 1411 * We update smallfile2 and smallfile1 at most every second. 1412 */ 1413 if (lbolt >= smallfile_update) { 1414 uint64_t percpufreeb; 1415 if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D; 1416 if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D; 1417 percpufreeb = ptob((uint64_t)freemem) / ncpus_online; 1418 smallfile1 = percpufreeb / smallfile1_d; 1419 smallfile2 = percpufreeb / smallfile2_d; 1420 smallfile1 = MAX(smallfile1, smallfile); 1421 smallfile1 = MAX(smallfile1, smallfile64); 1422 smallfile2 = MAX(smallfile1, smallfile2); 1423 smallfile_update = lbolt + hz; 1424 } 1425 1426 dofree = freebehind && 1427 ip->i_nextr == (off & PAGEMASK) && off > smallfile1; 1428 1429 /* 1430 * At this point we can enter ufs_getpage() in one of two 1431 * ways: 1432 * 1) segmap_getmapflt() calls ufs_getpage() when the 1433 * forcefault parameter is true (value of 1 is passed) 1434 * 2) uiomove() causes a page fault. 1435 * 1436 * We cannot hold onto an i_contents reader lock without 1437 * risking deadlock in ufs_getpage() so drop a reader lock. 1438 * The ufs_getpage() dolock logic already allows for a 1439 * thread holding i_contents as writer to work properly 1440 * so we keep a writer lock. 1441 */ 1442 if (rwtype == RW_READER) 1443 rw_exit(&ip->i_contents); 1444 1445 if (vpm_enable) { 1446 /* 1447 * Copy data. 1448 */ 1449 error = vpm_data_copy(vp, (off + mapon), (uint_t)n, 1450 uio, 1, NULL, 0, S_READ); 1451 } else { 1452 base = segmap_getmapflt(segkmap, vp, (off + mapon), 1453 (uint_t)n, 1, S_READ); 1454 error = uiomove(base + mapon, (long)n, UIO_READ, uio); 1455 } 1456 1457 flags = 0; 1458 if (!error) { 1459 /* 1460 * If reading sequential we won't need this 1461 * buffer again soon. For offsets in range 1462 * [smallfile1, smallfile2] release the pages 1463 * at the tail of the cache list, larger 1464 * offsets are released at the head. 1465 */ 1466 if (dofree) { 1467 flags = SM_FREE | SM_ASYNC; 1468 if ((cache_read_ahead == 0) && 1469 (off > smallfile2)) 1470 flags |= SM_DONTNEED; 1471 } 1472 /* 1473 * In POSIX SYNC (FSYNC and FDSYNC) read mode, 1474 * we want to make sure that the page which has 1475 * been read, is written on disk if it is dirty. 1476 * And corresponding indirect blocks should also 1477 * be flushed out. 1478 */ 1479 if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) { 1480 flags &= ~SM_ASYNC; 1481 flags |= SM_WRITE; 1482 } 1483 if (vpm_enable) { 1484 error = vpm_sync_pages(vp, off, n, flags); 1485 } else { 1486 error = segmap_release(segkmap, base, flags); 1487 } 1488 } else { 1489 if (vpm_enable) { 1490 (void) vpm_sync_pages(vp, off, n, flags); 1491 } else { 1492 (void) segmap_release(segkmap, base, flags); 1493 } 1494 } 1495 1496 if (rwtype == RW_READER) 1497 rw_enter(&ip->i_contents, rwtype); 1498 } while (error == 0 && uio->uio_resid > 0 && n != 0); 1499 out: 1500 /* 1501 * Inode is updated according to this table if FRSYNC is set. 1502 * 1503 * FSYNC FDSYNC(posix.4) 1504 * -------------------------- 1505 * always IATTCHG|IBDWRITE 1506 */ 1507 /* 1508 * The inode is not updated if we're logging and the inode is a 1509 * directory with FRSYNC, FSYNC and FDSYNC flags set. 1510 */ 1511 if (ioflag & FRSYNC) { 1512 if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) { 1513 doupdate = 0; 1514 } 1515 if (doupdate) { 1516 if ((ioflag & FSYNC) || 1517 ((ioflag & FDSYNC) && 1518 (ip->i_flag & (IATTCHG|IBDWRITE)))) { 1519 ufs_iupdat(ip, 1); 1520 } 1521 } 1522 } 1523 /* 1524 * If we've already done a partial read, terminate 1525 * the read but return no error. 1526 */ 1527 if (oresid != uio->uio_resid) 1528 error = 0; 1529 ITIMES(ip); 1530 1531 return (error); 1532 } 1533 1534 /* ARGSUSED */ 1535 static int 1536 ufs_ioctl( 1537 struct vnode *vp, 1538 int cmd, 1539 intptr_t arg, 1540 int flag, 1541 struct cred *cr, 1542 int *rvalp, 1543 caller_context_t *ct) 1544 { 1545 struct lockfs lockfs, lockfs_out; 1546 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs; 1547 char *comment, *original_comment; 1548 struct fs *fs; 1549 struct ulockfs *ulp; 1550 offset_t off; 1551 extern int maxphys; 1552 int error; 1553 int issync; 1554 int trans_size; 1555 1556 1557 /* 1558 * forcibly unmounted 1559 */ 1560 if (ufsvfsp == NULL || vp->v_vfsp == NULL || 1561 vp->v_vfsp->vfs_flag & VFS_UNMOUNTED) 1562 return (EIO); 1563 fs = ufsvfsp->vfs_fs; 1564 1565 if (cmd == Q_QUOTACTL) { 1566 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK); 1567 if (error) 1568 return (error); 1569 1570 if (ulp) { 1571 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA, 1572 TOP_SETQUOTA_SIZE(fs)); 1573 } 1574 1575 error = quotactl(vp, arg, flag, cr); 1576 1577 if (ulp) { 1578 TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA, 1579 TOP_SETQUOTA_SIZE(fs)); 1580 ufs_lockfs_end(ulp); 1581 } 1582 return (error); 1583 } 1584 1585 switch (cmd) { 1586 case _FIOLFS: 1587 /* 1588 * file system locking 1589 */ 1590 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0) 1591 return (EPERM); 1592 1593 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) { 1594 if (copyin((caddr_t)arg, &lockfs, 1595 sizeof (struct lockfs))) 1596 return (EFAULT); 1597 } 1598 #ifdef _SYSCALL32_IMPL 1599 else { 1600 struct lockfs32 lockfs32; 1601 /* Translate ILP32 lockfs to LP64 lockfs */ 1602 if (copyin((caddr_t)arg, &lockfs32, 1603 sizeof (struct lockfs32))) 1604 return (EFAULT); 1605 lockfs.lf_lock = (ulong_t)lockfs32.lf_lock; 1606 lockfs.lf_flags = (ulong_t)lockfs32.lf_flags; 1607 lockfs.lf_key = (ulong_t)lockfs32.lf_key; 1608 lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen; 1609 lockfs.lf_comment = 1610 (caddr_t)(uintptr_t)lockfs32.lf_comment; 1611 } 1612 #endif /* _SYSCALL32_IMPL */ 1613 1614 if (lockfs.lf_comlen) { 1615 if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN) 1616 return (ENAMETOOLONG); 1617 comment = 1618 kmem_alloc(lockfs.lf_comlen, KM_SLEEP); 1619 if (copyin(lockfs.lf_comment, comment, 1620 lockfs.lf_comlen)) { 1621 kmem_free(comment, lockfs.lf_comlen); 1622 return (EFAULT); 1623 } 1624 original_comment = lockfs.lf_comment; 1625 lockfs.lf_comment = comment; 1626 } 1627 if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) { 1628 lockfs.lf_comment = original_comment; 1629 1630 if ((flag & DATAMODEL_MASK) == 1631 DATAMODEL_NATIVE) { 1632 (void) copyout(&lockfs, (caddr_t)arg, 1633 sizeof (struct lockfs)); 1634 } 1635 #ifdef _SYSCALL32_IMPL 1636 else { 1637 struct lockfs32 lockfs32; 1638 /* Translate LP64 to ILP32 lockfs */ 1639 lockfs32.lf_lock = 1640 (uint32_t)lockfs.lf_lock; 1641 lockfs32.lf_flags = 1642 (uint32_t)lockfs.lf_flags; 1643 lockfs32.lf_key = 1644 (uint32_t)lockfs.lf_key; 1645 lockfs32.lf_comlen = 1646 (uint32_t)lockfs.lf_comlen; 1647 lockfs32.lf_comment = 1648 (uint32_t)(uintptr_t) 1649 lockfs.lf_comment; 1650 (void) copyout(&lockfs32, (caddr_t)arg, 1651 sizeof (struct lockfs32)); 1652 } 1653 #endif /* _SYSCALL32_IMPL */ 1654 1655 } else { 1656 if (lockfs.lf_comlen) 1657 kmem_free(comment, lockfs.lf_comlen); 1658 } 1659 return (error); 1660 1661 case _FIOLFSS: 1662 /* 1663 * get file system locking status 1664 */ 1665 1666 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) { 1667 if (copyin((caddr_t)arg, &lockfs, 1668 sizeof (struct lockfs))) 1669 return (EFAULT); 1670 } 1671 #ifdef _SYSCALL32_IMPL 1672 else { 1673 struct lockfs32 lockfs32; 1674 /* Translate ILP32 lockfs to LP64 lockfs */ 1675 if (copyin((caddr_t)arg, &lockfs32, 1676 sizeof (struct lockfs32))) 1677 return (EFAULT); 1678 lockfs.lf_lock = (ulong_t)lockfs32.lf_lock; 1679 lockfs.lf_flags = (ulong_t)lockfs32.lf_flags; 1680 lockfs.lf_key = (ulong_t)lockfs32.lf_key; 1681 lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen; 1682 lockfs.lf_comment = 1683 (caddr_t)(uintptr_t)lockfs32.lf_comment; 1684 } 1685 #endif /* _SYSCALL32_IMPL */ 1686 1687 if (error = ufs_fiolfss(vp, &lockfs_out)) 1688 return (error); 1689 lockfs.lf_lock = lockfs_out.lf_lock; 1690 lockfs.lf_key = lockfs_out.lf_key; 1691 lockfs.lf_flags = lockfs_out.lf_flags; 1692 lockfs.lf_comlen = MIN(lockfs.lf_comlen, 1693 lockfs_out.lf_comlen); 1694 1695 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) { 1696 if (copyout(&lockfs, (caddr_t)arg, 1697 sizeof (struct lockfs))) 1698 return (EFAULT); 1699 } 1700 #ifdef _SYSCALL32_IMPL 1701 else { 1702 /* Translate LP64 to ILP32 lockfs */ 1703 struct lockfs32 lockfs32; 1704 lockfs32.lf_lock = (uint32_t)lockfs.lf_lock; 1705 lockfs32.lf_flags = (uint32_t)lockfs.lf_flags; 1706 lockfs32.lf_key = (uint32_t)lockfs.lf_key; 1707 lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen; 1708 lockfs32.lf_comment = 1709 (uint32_t)(uintptr_t)lockfs.lf_comment; 1710 if (copyout(&lockfs32, (caddr_t)arg, 1711 sizeof (struct lockfs32))) 1712 return (EFAULT); 1713 } 1714 #endif /* _SYSCALL32_IMPL */ 1715 1716 if (lockfs.lf_comlen && 1717 lockfs.lf_comment && lockfs_out.lf_comment) 1718 if (copyout(lockfs_out.lf_comment, 1719 lockfs.lf_comment, lockfs.lf_comlen)) 1720 return (EFAULT); 1721 return (0); 1722 1723 case _FIOSATIME: 1724 /* 1725 * set access time 1726 */ 1727 1728 /* 1729 * if mounted w/o atime, return quietly. 1730 * I briefly thought about returning ENOSYS, but 1731 * figured that most apps would consider this fatal 1732 * but the idea is to make this as seamless as poss. 1733 */ 1734 if (ufsvfsp->vfs_noatime) 1735 return (0); 1736 1737 error = ufs_lockfs_begin(ufsvfsp, &ulp, 1738 ULOCKFS_SETATTR_MASK); 1739 if (error) 1740 return (error); 1741 1742 if (ulp) { 1743 trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp)); 1744 TRANS_BEGIN_CSYNC(ufsvfsp, issync, 1745 TOP_SETATTR, trans_size); 1746 } 1747 1748 error = ufs_fiosatime(vp, (struct timeval *)arg, 1749 flag, cr); 1750 1751 if (ulp) { 1752 TRANS_END_CSYNC(ufsvfsp, error, issync, 1753 TOP_SETATTR, trans_size); 1754 ufs_lockfs_end(ulp); 1755 } 1756 return (error); 1757 1758 case _FIOSDIO: 1759 /* 1760 * set delayed-io 1761 */ 1762 return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr)); 1763 1764 case _FIOGDIO: 1765 /* 1766 * get delayed-io 1767 */ 1768 return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr)); 1769 1770 case _FIOIO: 1771 /* 1772 * inode open 1773 */ 1774 error = ufs_lockfs_begin(ufsvfsp, &ulp, 1775 ULOCKFS_VGET_MASK); 1776 if (error) 1777 return (error); 1778 1779 error = ufs_fioio(vp, (struct fioio *)arg, flag, cr); 1780 1781 if (ulp) { 1782 ufs_lockfs_end(ulp); 1783 } 1784 return (error); 1785 1786 case _FIOFFS: 1787 /* 1788 * file system flush (push w/invalidate) 1789 */ 1790 if ((caddr_t)arg != NULL) 1791 return (EINVAL); 1792 return (ufs_fioffs(vp, NULL, cr)); 1793 1794 case _FIOISBUSY: 1795 /* 1796 * Contract-private interface for Legato 1797 * Purge this vnode from the DNLC and decide 1798 * if this vnode is busy (*arg == 1) or not 1799 * (*arg == 0) 1800 */ 1801 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0) 1802 return (EPERM); 1803 error = ufs_fioisbusy(vp, (int *)arg, cr); 1804 return (error); 1805 1806 case _FIODIRECTIO: 1807 return (ufs_fiodirectio(vp, (int)arg, cr)); 1808 1809 case _FIOTUNE: 1810 /* 1811 * Tune the file system (aka setting fs attributes) 1812 */ 1813 error = ufs_lockfs_begin(ufsvfsp, &ulp, 1814 ULOCKFS_SETATTR_MASK); 1815 if (error) 1816 return (error); 1817 1818 error = ufs_fiotune(vp, (struct fiotune *)arg, cr); 1819 1820 if (ulp) 1821 ufs_lockfs_end(ulp); 1822 return (error); 1823 1824 case _FIOLOGENABLE: 1825 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0) 1826 return (EPERM); 1827 return (ufs_fiologenable(vp, (void *)arg, cr, flag)); 1828 1829 case _FIOLOGDISABLE: 1830 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0) 1831 return (EPERM); 1832 return (ufs_fiologdisable(vp, (void *)arg, cr, flag)); 1833 1834 case _FIOISLOG: 1835 return (ufs_fioislog(vp, (void *)arg, cr, flag)); 1836 1837 case _FIOSNAPSHOTCREATE_MULTI: 1838 { 1839 struct fiosnapcreate_multi fc, *fcp; 1840 size_t fcm_size; 1841 1842 if (copyin((void *)arg, &fc, sizeof (fc))) 1843 return (EFAULT); 1844 if (fc.backfilecount > MAX_BACKFILE_COUNT) 1845 return (EINVAL); 1846 fcm_size = sizeof (struct fiosnapcreate_multi) + 1847 (fc.backfilecount - 1) * sizeof (int); 1848 fcp = (struct fiosnapcreate_multi *) 1849 kmem_alloc(fcm_size, KM_SLEEP); 1850 if (copyin((void *)arg, fcp, fcm_size)) { 1851 kmem_free(fcp, fcm_size); 1852 return (EFAULT); 1853 } 1854 error = ufs_snap_create(vp, fcp, cr); 1855 /* 1856 * Do copyout even if there is an error because 1857 * the details of error is stored in fcp. 1858 */ 1859 if (copyout(fcp, (void *)arg, fcm_size)) 1860 error = EFAULT; 1861 kmem_free(fcp, fcm_size); 1862 return (error); 1863 } 1864 1865 case _FIOSNAPSHOTDELETE: 1866 { 1867 struct fiosnapdelete fc; 1868 1869 if (copyin((void *)arg, &fc, sizeof (fc))) 1870 return (EFAULT); 1871 error = ufs_snap_delete(vp, &fc, cr); 1872 if (!error && copyout(&fc, (void *)arg, sizeof (fc))) 1873 error = EFAULT; 1874 return (error); 1875 } 1876 1877 case _FIOGETSUPERBLOCK: 1878 if (copyout(fs, (void *)arg, SBSIZE)) 1879 return (EFAULT); 1880 return (0); 1881 1882 case _FIOGETMAXPHYS: 1883 if (copyout(&maxphys, (void *)arg, sizeof (maxphys))) 1884 return (EFAULT); 1885 return (0); 1886 1887 /* 1888 * The following 3 ioctls are for TSufs support 1889 * although could potentially be used elsewhere 1890 */ 1891 case _FIO_SET_LUFS_DEBUG: 1892 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0) 1893 return (EPERM); 1894 lufs_debug = (uint32_t)arg; 1895 return (0); 1896 1897 case _FIO_SET_LUFS_ERROR: 1898 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0) 1899 return (EPERM); 1900 TRANS_SETERROR(ufsvfsp); 1901 return (0); 1902 1903 case _FIO_GET_TOP_STATS: 1904 { 1905 fio_lufs_stats_t *ls; 1906 ml_unit_t *ul = ufsvfsp->vfs_log; 1907 1908 ls = kmem_zalloc(sizeof (*ls), KM_SLEEP); 1909 ls->ls_debug = ul->un_debug; /* return debug value */ 1910 /* Copy stucture if statistics are being kept */ 1911 if (ul->un_logmap->mtm_tops) { 1912 ls->ls_topstats = *(ul->un_logmap->mtm_tops); 1913 } 1914 error = 0; 1915 if (copyout(ls, (void *)arg, sizeof (*ls))) 1916 error = EFAULT; 1917 kmem_free(ls, sizeof (*ls)); 1918 return (error); 1919 } 1920 1921 case _FIO_SEEK_DATA: 1922 case _FIO_SEEK_HOLE: 1923 if (ddi_copyin((void *)arg, &off, sizeof (off), flag)) 1924 return (EFAULT); 1925 /* offset paramater is in/out */ 1926 error = ufs_fio_holey(vp, cmd, &off); 1927 if (error) 1928 return (error); 1929 if (ddi_copyout(&off, (void *)arg, sizeof (off), flag)) 1930 return (EFAULT); 1931 return (0); 1932 1933 case _FIO_COMPRESSED: 1934 { 1935 /* 1936 * This is a project private ufs ioctl() to mark 1937 * the inode as that belonging to a compressed 1938 * file. This is used to mark individual 1939 * files in a miniroot archive for SPARC boot. 1940 * The files compressed in this manner are 1941 * automatically decompressed by the dcfs filesystem 1942 * (via an interception in ufs_lookup - see decompvp()) 1943 * which is layered on top of ufs on a system running 1944 * the new archive booted SPARC system. See 1945 * uts/common/fs/dcfs for details. 1946 * This ioctl only marks the file as compressed - the 1947 * actual compression is done by fiocompress (a 1948 * userland utility) which invokes this ioctl(). 1949 */ 1950 struct inode *ip = VTOI(vp); 1951 1952 error = ufs_lockfs_begin(ufsvfsp, &ulp, 1953 ULOCKFS_SETATTR_MASK); 1954 if (error) 1955 return (error); 1956 1957 if (ulp) { 1958 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_IUPDAT, 1959 TOP_IUPDAT_SIZE(ip)); 1960 } 1961 1962 error = ufs_mark_compressed(vp); 1963 1964 if (ulp) { 1965 TRANS_END_ASYNC(ufsvfsp, TOP_IUPDAT, 1966 TOP_IUPDAT_SIZE(ip)); 1967 ufs_lockfs_end(ulp); 1968 } 1969 1970 return (error); 1971 1972 } 1973 1974 default: 1975 return (ENOTTY); 1976 } 1977 } 1978 1979 1980 /* ARGSUSED */ 1981 static int 1982 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags, 1983 struct cred *cr, caller_context_t *ct) 1984 { 1985 struct inode *ip = VTOI(vp); 1986 struct ufsvfs *ufsvfsp; 1987 int err; 1988 1989 if (vap->va_mask == AT_SIZE) { 1990 /* 1991 * for performance, if only the size is requested don't bother 1992 * with anything else. 1993 */ 1994 UFS_GET_ISIZE(&vap->va_size, ip); 1995 return (0); 1996 } 1997 1998 /* 1999 * inlined lockfs checks 2000 */ 2001 ufsvfsp = ip->i_ufsvfs; 2002 if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) { 2003 err = EIO; 2004 goto out; 2005 } 2006 2007 rw_enter(&ip->i_contents, RW_READER); 2008 /* 2009 * Return all the attributes. This should be refined so 2010 * that it only returns what's asked for. 2011 */ 2012 2013 /* 2014 * Copy from inode table. 2015 */ 2016 vap->va_type = vp->v_type; 2017 vap->va_mode = ip->i_mode & MODEMASK; 2018 /* 2019 * If there is an ACL and there is a mask entry, then do the 2020 * extra work that completes the equivalent of an acltomode(3) 2021 * call. According to POSIX P1003.1e, the acl mask should be 2022 * returned in the group permissions field. 2023 * 2024 * - start with the original permission and mode bits (from above) 2025 * - clear the group owner bits 2026 * - add in the mask bits. 2027 */ 2028 if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) { 2029 vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3); 2030 vap->va_mode |= 2031 (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3; 2032 } 2033 vap->va_uid = ip->i_uid; 2034 vap->va_gid = ip->i_gid; 2035 vap->va_fsid = ip->i_dev; 2036 vap->va_nodeid = (ino64_t)ip->i_number; 2037 vap->va_nlink = ip->i_nlink; 2038 vap->va_size = ip->i_size; 2039 if (vp->v_type == VCHR || vp->v_type == VBLK) 2040 vap->va_rdev = ip->i_rdev; 2041 else 2042 vap->va_rdev = 0; /* not a b/c spec. */ 2043 mutex_enter(&ip->i_tlock); 2044 ITIMES_NOLOCK(ip); /* mark correct time in inode */ 2045 vap->va_seq = ip->i_seq; 2046 vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec; 2047 vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000; 2048 vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec; 2049 vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000; 2050 vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec; 2051 vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000; 2052 mutex_exit(&ip->i_tlock); 2053 2054 switch (ip->i_mode & IFMT) { 2055 2056 case IFBLK: 2057 vap->va_blksize = MAXBSIZE; /* was BLKDEV_IOSIZE */ 2058 break; 2059 2060 case IFCHR: 2061 vap->va_blksize = MAXBSIZE; 2062 break; 2063 2064 default: 2065 vap->va_blksize = ip->i_fs->fs_bsize; 2066 break; 2067 } 2068 vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks; 2069 rw_exit(&ip->i_contents); 2070 err = 0; 2071 2072 out: 2073 return (err); 2074 } 2075 2076 /*ARGSUSED4*/ 2077 static int 2078 ufs_setattr( 2079 struct vnode *vp, 2080 struct vattr *vap, 2081 int flags, 2082 struct cred *cr, 2083 caller_context_t *ct) 2084 { 2085 struct inode *ip = VTOI(vp); 2086 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 2087 struct fs *fs; 2088 struct ulockfs *ulp; 2089 char *errmsg1; 2090 char *errmsg2; 2091 long blocks; 2092 long int mask = vap->va_mask; 2093 size_t len1, len2; 2094 int issync; 2095 int trans_size; 2096 int dotrans; 2097 int dorwlock; 2098 int error; 2099 int owner_change; 2100 int dodqlock; 2101 timestruc_t now; 2102 vattr_t oldva; 2103 int retry = 1; 2104 int indeadlock; 2105 2106 /* 2107 * Cannot set these attributes. 2108 */ 2109 if ((mask & AT_NOSET) || (mask & AT_XVATTR)) 2110 return (EINVAL); 2111 2112 /* 2113 * check for forced unmount 2114 */ 2115 if (ufsvfsp == NULL) 2116 return (EIO); 2117 2118 fs = ufsvfsp->vfs_fs; 2119 if (fs->fs_ronly != 0) 2120 return (EROFS); 2121 2122 again: 2123 errmsg1 = NULL; 2124 errmsg2 = NULL; 2125 dotrans = 0; 2126 dorwlock = 0; 2127 dodqlock = 0; 2128 2129 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK); 2130 if (error) 2131 goto out; 2132 2133 /* 2134 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file. 2135 * This follows the protocol for read()/write(). 2136 */ 2137 if (vp->v_type != VDIR) { 2138 /* 2139 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to 2140 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock 2141 * possible, retries the operation. 2142 */ 2143 ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file); 2144 if (indeadlock) { 2145 if (ulp) 2146 ufs_lockfs_end(ulp); 2147 goto again; 2148 } 2149 dorwlock = 1; 2150 } 2151 2152 /* 2153 * Truncate file. Must have write permission and not be a directory. 2154 */ 2155 if (mask & AT_SIZE) { 2156 rw_enter(&ip->i_contents, RW_WRITER); 2157 if (vp->v_type == VDIR) { 2158 error = EISDIR; 2159 goto update_inode; 2160 } 2161 if (error = ufs_iaccess(ip, IWRITE, cr)) 2162 goto update_inode; 2163 2164 rw_exit(&ip->i_contents); 2165 error = TRANS_ITRUNC(ip, vap->va_size, 0, cr); 2166 if (error) { 2167 rw_enter(&ip->i_contents, RW_WRITER); 2168 goto update_inode; 2169 } 2170 } 2171 2172 if (ulp) { 2173 trans_size = (int)TOP_SETATTR_SIZE(ip); 2174 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size); 2175 ++dotrans; 2176 } 2177 2178 /* 2179 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory. 2180 * This follows the protocol established by 2181 * ufs_link/create/remove/rename/mkdir/rmdir/symlink. 2182 */ 2183 if (vp->v_type == VDIR) { 2184 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR, 2185 retry_dir); 2186 if (indeadlock) 2187 goto again; 2188 dorwlock = 1; 2189 } 2190 2191 /* 2192 * Grab quota lock if we are changing the file's owner. 2193 */ 2194 if (mask & AT_UID) { 2195 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 2196 dodqlock = 1; 2197 } 2198 rw_enter(&ip->i_contents, RW_WRITER); 2199 2200 oldva.va_mode = ip->i_mode; 2201 oldva.va_uid = ip->i_uid; 2202 oldva.va_gid = ip->i_gid; 2203 2204 vap->va_mask &= ~AT_SIZE; 2205 /* 2206 * ufs_iaccess is "close enough"; that's because it doesn't 2207 * map the defines. 2208 */ 2209 error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2210 ufs_iaccess, ip); 2211 if (error) 2212 goto update_inode; 2213 2214 mask = vap->va_mask; 2215 2216 /* 2217 * Change file access modes. 2218 */ 2219 if (mask & AT_MODE) { 2220 ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT); 2221 TRANS_INODE(ufsvfsp, ip); 2222 ip->i_flag |= ICHG; 2223 if (stickyhack) { 2224 mutex_enter(&vp->v_lock); 2225 if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX) 2226 vp->v_flag |= VSWAPLIKE; 2227 else 2228 vp->v_flag &= ~VSWAPLIKE; 2229 mutex_exit(&vp->v_lock); 2230 } 2231 } 2232 if (mask & (AT_UID|AT_GID)) { 2233 if (mask & AT_UID) { 2234 /* 2235 * Don't change ownership of the quota inode. 2236 */ 2237 if (ufsvfsp->vfs_qinod == ip) { 2238 ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED); 2239 error = EINVAL; 2240 goto update_inode; 2241 } 2242 2243 /* 2244 * No real ownership change. 2245 */ 2246 if (ip->i_uid == vap->va_uid) { 2247 blocks = 0; 2248 owner_change = 0; 2249 } 2250 /* 2251 * Remove the blocks and the file, from the old user's 2252 * quota. 2253 */ 2254 else { 2255 blocks = ip->i_blocks; 2256 owner_change = 1; 2257 2258 (void) chkdq(ip, -blocks, /* force */ 1, cr, 2259 (char **)NULL, (size_t *)NULL); 2260 (void) chkiq(ufsvfsp, /* change */ -1, ip, 2261 (uid_t)ip->i_uid, /* force */ 1, cr, 2262 (char **)NULL, (size_t *)NULL); 2263 dqrele(ip->i_dquot); 2264 } 2265 2266 ip->i_uid = vap->va_uid; 2267 2268 /* 2269 * There is a real ownership change. 2270 */ 2271 if (owner_change) { 2272 /* 2273 * Add the blocks and the file to the new 2274 * user's quota. 2275 */ 2276 ip->i_dquot = getinoquota(ip); 2277 (void) chkdq(ip, blocks, /* force */ 1, cr, 2278 &errmsg1, &len1); 2279 (void) chkiq(ufsvfsp, /* change */ 1, 2280 (struct inode *)NULL, (uid_t)ip->i_uid, 2281 /* force */ 1, cr, &errmsg2, &len2); 2282 } 2283 } 2284 if (mask & AT_GID) { 2285 ip->i_gid = vap->va_gid; 2286 } 2287 TRANS_INODE(ufsvfsp, ip); 2288 ip->i_flag |= ICHG; 2289 } 2290 /* 2291 * Change file access or modified times. 2292 */ 2293 if (mask & (AT_ATIME|AT_MTIME)) { 2294 /* Check that the time value is within ufs range */ 2295 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2296 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2297 error = EOVERFLOW; 2298 goto update_inode; 2299 } 2300 2301 /* 2302 * if the "noaccess" mount option is set and only atime 2303 * update is requested, do nothing. No error is returned. 2304 */ 2305 if ((ufsvfsp->vfs_noatime) && 2306 ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME)) 2307 goto skip_atime; 2308 2309 if (mask & AT_ATIME) { 2310 ip->i_atime.tv_sec = vap->va_atime.tv_sec; 2311 ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000; 2312 ip->i_flag &= ~IACC; 2313 } 2314 if (mask & AT_MTIME) { 2315 ip->i_mtime.tv_sec = vap->va_mtime.tv_sec; 2316 ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000; 2317 gethrestime(&now); 2318 if (now.tv_sec > TIME32_MAX) { 2319 /* 2320 * In 2038, ctime sticks forever.. 2321 */ 2322 ip->i_ctime.tv_sec = TIME32_MAX; 2323 ip->i_ctime.tv_usec = 0; 2324 } else { 2325 ip->i_ctime.tv_sec = now.tv_sec; 2326 ip->i_ctime.tv_usec = now.tv_nsec / 1000; 2327 } 2328 ip->i_flag &= ~(IUPD|ICHG); 2329 ip->i_flag |= IMODTIME; 2330 } 2331 TRANS_INODE(ufsvfsp, ip); 2332 ip->i_flag |= IMOD; 2333 } 2334 2335 skip_atime: 2336 /* 2337 * The presence of a shadow inode may indicate an ACL, but does 2338 * not imply an ACL. Future FSD types should be handled here too 2339 * and check for the presence of the attribute-specific data 2340 * before referencing it. 2341 */ 2342 if (ip->i_shadow) { 2343 /* 2344 * XXX if ufs_iupdat is changed to sandbagged write fix 2345 * ufs_acl_setattr to push ip to keep acls consistent 2346 * 2347 * Suppress out of inodes messages if we will retry. 2348 */ 2349 if (retry) 2350 ip->i_flag |= IQUIET; 2351 error = ufs_acl_setattr(ip, vap, cr); 2352 ip->i_flag &= ~IQUIET; 2353 } 2354 2355 update_inode: 2356 /* 2357 * Setattr always increases the sequence number 2358 */ 2359 ip->i_seq++; 2360 2361 /* 2362 * if nfsd and not logging; push synchronously 2363 */ 2364 if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) { 2365 ufs_iupdat(ip, 1); 2366 } else { 2367 ITIMES_NOLOCK(ip); 2368 } 2369 2370 rw_exit(&ip->i_contents); 2371 if (dodqlock) { 2372 rw_exit(&ufsvfsp->vfs_dqrwlock); 2373 } 2374 if (dorwlock) 2375 rw_exit(&ip->i_rwlock); 2376 2377 if (ulp) { 2378 if (dotrans) { 2379 int terr = 0; 2380 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR, 2381 trans_size); 2382 if (error == 0) 2383 error = terr; 2384 } 2385 ufs_lockfs_end(ulp); 2386 } 2387 out: 2388 /* 2389 * If out of inodes or blocks, see if we can free something 2390 * up from the delete queue. 2391 */ 2392 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) { 2393 ufs_delete_drain_wait(ufsvfsp, 1); 2394 retry = 0; 2395 if (errmsg1 != NULL) 2396 kmem_free(errmsg1, len1); 2397 if (errmsg2 != NULL) 2398 kmem_free(errmsg2, len2); 2399 goto again; 2400 } 2401 if (errmsg1 != NULL) { 2402 uprintf(errmsg1); 2403 kmem_free(errmsg1, len1); 2404 } 2405 if (errmsg2 != NULL) { 2406 uprintf(errmsg2); 2407 kmem_free(errmsg2, len2); 2408 } 2409 return (error); 2410 } 2411 2412 /*ARGSUSED*/ 2413 static int 2414 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr, 2415 caller_context_t *ct) 2416 { 2417 struct inode *ip = VTOI(vp); 2418 int error; 2419 2420 if (ip->i_ufsvfs == NULL) 2421 return (EIO); 2422 2423 rw_enter(&ip->i_contents, RW_READER); 2424 2425 /* 2426 * The ufs_iaccess function wants to be called with 2427 * mode bits expressed as "ufs specific" bits. 2428 * I.e., VWRITE|VREAD|VEXEC do not make sense to 2429 * ufs_iaccess() but IWRITE|IREAD|IEXEC do. 2430 * But since they're the same we just pass the vnode mode 2431 * bit but just verify that assumption at compile time. 2432 */ 2433 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC 2434 #error "ufs_access needs to map Vmodes to Imodes" 2435 #endif 2436 error = ufs_iaccess(ip, mode, cr); 2437 2438 rw_exit(&ip->i_contents); 2439 2440 return (error); 2441 } 2442 2443 /* ARGSUSED */ 2444 static int 2445 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr, 2446 caller_context_t *ct) 2447 { 2448 struct inode *ip = VTOI(vp); 2449 struct ufsvfs *ufsvfsp; 2450 struct ulockfs *ulp; 2451 int error; 2452 int fastsymlink; 2453 2454 if (vp->v_type != VLNK) { 2455 error = EINVAL; 2456 goto nolockout; 2457 } 2458 2459 /* 2460 * If the symbolic link is empty there is nothing to read. 2461 * Fast-track these empty symbolic links 2462 */ 2463 if (ip->i_size == 0) { 2464 error = 0; 2465 goto nolockout; 2466 } 2467 2468 ufsvfsp = ip->i_ufsvfs; 2469 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK); 2470 if (error) 2471 goto nolockout; 2472 /* 2473 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK 2474 */ 2475 again: 2476 fastsymlink = 0; 2477 if (ip->i_flag & IFASTSYMLNK) { 2478 rw_enter(&ip->i_rwlock, RW_READER); 2479 rw_enter(&ip->i_contents, RW_READER); 2480 if (ip->i_flag & IFASTSYMLNK) { 2481 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && 2482 (ip->i_fs->fs_ronly == 0) && 2483 (!ufsvfsp->vfs_noatime)) { 2484 mutex_enter(&ip->i_tlock); 2485 ip->i_flag |= IACC; 2486 mutex_exit(&ip->i_tlock); 2487 } 2488 error = uiomove((caddr_t)&ip->i_db[1], 2489 MIN(ip->i_size, uiop->uio_resid), 2490 UIO_READ, uiop); 2491 ITIMES(ip); 2492 ++fastsymlink; 2493 } 2494 rw_exit(&ip->i_contents); 2495 rw_exit(&ip->i_rwlock); 2496 } 2497 if (!fastsymlink) { 2498 ssize_t size; /* number of bytes read */ 2499 caddr_t basep; /* pointer to input data */ 2500 ino_t ino; 2501 long igen; 2502 struct uio tuio; /* temp uio struct */ 2503 struct uio *tuiop; 2504 iovec_t tiov; /* temp iovec struct */ 2505 char kbuf[FSL_SIZE]; /* buffer to hold fast symlink */ 2506 int tflag = 0; /* flag to indicate temp vars used */ 2507 2508 ino = ip->i_number; 2509 igen = ip->i_gen; 2510 size = uiop->uio_resid; 2511 basep = uiop->uio_iov->iov_base; 2512 tuiop = uiop; 2513 2514 rw_enter(&ip->i_rwlock, RW_WRITER); 2515 rw_enter(&ip->i_contents, RW_WRITER); 2516 if (ip->i_flag & IFASTSYMLNK) { 2517 rw_exit(&ip->i_contents); 2518 rw_exit(&ip->i_rwlock); 2519 goto again; 2520 } 2521 2522 /* can this be a fast symlink and is it a user buffer? */ 2523 if (ip->i_size <= FSL_SIZE && 2524 (uiop->uio_segflg == UIO_USERSPACE || 2525 uiop->uio_segflg == UIO_USERISPACE)) { 2526 2527 bzero(&tuio, sizeof (struct uio)); 2528 /* 2529 * setup a kernel buffer to read link into. this 2530 * is to fix a race condition where the user buffer 2531 * got corrupted before copying it into the inode. 2532 */ 2533 size = ip->i_size; 2534 tiov.iov_len = size; 2535 tiov.iov_base = kbuf; 2536 tuio.uio_iov = &tiov; 2537 tuio.uio_iovcnt = 1; 2538 tuio.uio_offset = uiop->uio_offset; 2539 tuio.uio_segflg = UIO_SYSSPACE; 2540 tuio.uio_fmode = uiop->uio_fmode; 2541 tuio.uio_extflg = uiop->uio_extflg; 2542 tuio.uio_limit = uiop->uio_limit; 2543 tuio.uio_resid = size; 2544 2545 basep = tuio.uio_iov->iov_base; 2546 tuiop = &tuio; 2547 tflag = 1; 2548 } 2549 2550 error = rdip(ip, tuiop, 0, cr); 2551 if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) { 2552 rw_exit(&ip->i_contents); 2553 rw_exit(&ip->i_rwlock); 2554 goto out; 2555 } 2556 2557 if (tflag == 0) 2558 size -= uiop->uio_resid; 2559 2560 if ((tflag == 0 && ip->i_size <= FSL_SIZE && 2561 ip->i_size == size) || (tflag == 1 && 2562 tuio.uio_resid == 0)) { 2563 error = kcopy(basep, &ip->i_db[1], ip->i_size); 2564 if (error == 0) { 2565 ip->i_flag |= IFASTSYMLNK; 2566 /* 2567 * free page 2568 */ 2569 (void) VOP_PUTPAGE(ITOV(ip), 2570 (offset_t)0, PAGESIZE, 2571 (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC), 2572 cr, ct); 2573 } else { 2574 int i; 2575 /* error, clear garbage left behind */ 2576 for (i = 1; i < NDADDR; i++) 2577 ip->i_db[i] = 0; 2578 for (i = 0; i < NIADDR; i++) 2579 ip->i_ib[i] = 0; 2580 } 2581 } 2582 if (tflag == 1) { 2583 /* now, copy it into the user buffer */ 2584 error = uiomove((caddr_t)kbuf, 2585 MIN(size, uiop->uio_resid), 2586 UIO_READ, uiop); 2587 } 2588 rw_exit(&ip->i_contents); 2589 rw_exit(&ip->i_rwlock); 2590 } 2591 out: 2592 if (ulp) { 2593 ufs_lockfs_end(ulp); 2594 } 2595 nolockout: 2596 return (error); 2597 } 2598 2599 /* ARGSUSED */ 2600 static int 2601 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr, 2602 caller_context_t *ct) 2603 { 2604 struct inode *ip = VTOI(vp); 2605 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 2606 struct ulockfs *ulp; 2607 int error; 2608 2609 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK); 2610 if (error) 2611 return (error); 2612 2613 if (TRANS_ISTRANS(ufsvfsp)) { 2614 /* 2615 * First push out any data pages 2616 */ 2617 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2618 (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) { 2619 error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, 2620 0, CRED(), ct); 2621 if (error) 2622 goto out; 2623 } 2624 2625 /* 2626 * Delta any delayed inode times updates 2627 * and push inode to log. 2628 * All other inode deltas will have already been delta'd 2629 * and will be pushed during the commit. 2630 */ 2631 if (!(syncflag & FDSYNC) && 2632 ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) { 2633 if (ulp) { 2634 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC, 2635 TOP_SYNCIP_SIZE); 2636 } 2637 rw_enter(&ip->i_contents, RW_READER); 2638 mutex_enter(&ip->i_tlock); 2639 ip->i_flag &= ~IMODTIME; 2640 mutex_exit(&ip->i_tlock); 2641 ufs_iupdat(ip, I_SYNC); 2642 rw_exit(&ip->i_contents); 2643 if (ulp) { 2644 TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC, 2645 TOP_SYNCIP_SIZE); 2646 } 2647 } 2648 2649 /* 2650 * Commit the Moby transaction 2651 * 2652 * Deltas have already been made so we just need to 2653 * commit them with a synchronous transaction. 2654 * TRANS_BEGIN_SYNC() will return an error 2655 * if there are no deltas to commit, for an 2656 * empty transaction. 2657 */ 2658 if (ulp) { 2659 TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE, 2660 error); 2661 if (error) { 2662 error = 0; /* commit wasn't needed */ 2663 goto out; 2664 } 2665 TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC, 2666 TOP_COMMIT_SIZE); 2667 } 2668 } else { /* not logging */ 2669 if (!(IS_SWAPVP(vp))) 2670 if (syncflag & FNODSYNC) { 2671 /* Just update the inode only */ 2672 TRANS_IUPDAT(ip, 1); 2673 error = 0; 2674 } else if (syncflag & FDSYNC) 2675 /* Do data-synchronous writes */ 2676 error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC); 2677 else 2678 /* Do synchronous writes */ 2679 error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC); 2680 2681 rw_enter(&ip->i_contents, RW_WRITER); 2682 if (!error) 2683 error = ufs_sync_indir(ip); 2684 rw_exit(&ip->i_contents); 2685 } 2686 out: 2687 if (ulp) { 2688 ufs_lockfs_end(ulp); 2689 } 2690 return (error); 2691 } 2692 2693 /*ARGSUSED*/ 2694 static void 2695 ufs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct) 2696 { 2697 ufs_iinactive(VTOI(vp)); 2698 } 2699 2700 /* 2701 * Unix file system operations having to do with directory manipulation. 2702 */ 2703 int ufs_lookup_idle_count = 2; /* Number of inodes to idle each time */ 2704 /* ARGSUSED */ 2705 static int 2706 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp, 2707 struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr, 2708 caller_context_t *ct, int *direntflags, pathname_t *realpnp) 2709 { 2710 struct inode *ip; 2711 struct inode *sip; 2712 struct inode *xip; 2713 struct ufsvfs *ufsvfsp; 2714 struct ulockfs *ulp; 2715 struct vnode *vp; 2716 int error; 2717 2718 /* 2719 * Check flags for type of lookup (regular file or attribute file) 2720 */ 2721 2722 ip = VTOI(dvp); 2723 2724 if (flags & LOOKUP_XATTR) { 2725 2726 /* 2727 * If not mounted with XATTR support then return EINVAL 2728 */ 2729 2730 if (!(ip->i_ufsvfs->vfs_vfs->vfs_flag & VFS_XATTR)) 2731 return (EINVAL); 2732 /* 2733 * We don't allow recursive attributes... 2734 * Maybe someday we will. 2735 */ 2736 if ((ip->i_cflags & IXATTR)) { 2737 return (EINVAL); 2738 } 2739 2740 if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) { 2741 error = ufs_xattr_getattrdir(dvp, &sip, flags, cr); 2742 if (error) { 2743 *vpp = NULL; 2744 goto out; 2745 } 2746 2747 vp = ITOV(sip); 2748 dnlc_update(dvp, XATTR_DIR_NAME, vp); 2749 } 2750 2751 /* 2752 * Check accessibility of directory. 2753 */ 2754 if (vp == DNLC_NO_VNODE) { 2755 VN_RELE(vp); 2756 error = ENOENT; 2757 goto out; 2758 } 2759 if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr)) != 0) { 2760 VN_RELE(vp); 2761 goto out; 2762 } 2763 2764 *vpp = vp; 2765 return (0); 2766 } 2767 2768 /* 2769 * Check for a null component, which we should treat as 2770 * looking at dvp from within it's parent, so we don't 2771 * need a call to ufs_iaccess(), as it has already been 2772 * done. 2773 */ 2774 if (nm[0] == 0) { 2775 VN_HOLD(dvp); 2776 error = 0; 2777 *vpp = dvp; 2778 goto out; 2779 } 2780 2781 /* 2782 * Check for "." ie itself. this is a quick check and 2783 * avoids adding "." into the dnlc (which have been seen 2784 * to occupy >10% of the cache). 2785 */ 2786 if ((nm[0] == '.') && (nm[1] == 0)) { 2787 /* 2788 * Don't return without checking accessibility 2789 * of the directory. We only need the lock if 2790 * we are going to return it. 2791 */ 2792 if ((error = ufs_iaccess(ip, IEXEC, cr)) == 0) { 2793 VN_HOLD(dvp); 2794 *vpp = dvp; 2795 } 2796 goto out; 2797 } 2798 2799 /* 2800 * Fast path: Check the directory name lookup cache. 2801 */ 2802 if (vp = dnlc_lookup(dvp, nm)) { 2803 /* 2804 * Check accessibility of directory. 2805 */ 2806 if ((error = ufs_iaccess(ip, IEXEC, cr)) != 0) { 2807 VN_RELE(vp); 2808 goto out; 2809 } 2810 if (vp == DNLC_NO_VNODE) { 2811 VN_RELE(vp); 2812 error = ENOENT; 2813 goto out; 2814 } 2815 xip = VTOI(vp); 2816 ulp = NULL; 2817 goto fastpath; 2818 } 2819 2820 /* 2821 * Keep the idle queue from getting too long by 2822 * idling two inodes before attempting to allocate another. 2823 * This operation must be performed before entering 2824 * lockfs or a transaction. 2825 */ 2826 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat) 2827 if ((curthread->t_flag & T_DONTBLOCK) == 0) { 2828 ins.in_lidles.value.ul += ufs_lookup_idle_count; 2829 ufs_idle_some(ufs_lookup_idle_count); 2830 } 2831 2832 retry_lookup: 2833 ufsvfsp = ip->i_ufsvfs; 2834 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK); 2835 if (error) 2836 goto out; 2837 2838 error = ufs_dirlook(ip, nm, &xip, cr, 1); 2839 2840 fastpath: 2841 if (error == 0) { 2842 ip = xip; 2843 *vpp = ITOV(ip); 2844 2845 /* 2846 * If vnode is a device return special vnode instead. 2847 */ 2848 if (IS_DEVVP(*vpp)) { 2849 struct vnode *newvp; 2850 2851 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, 2852 cr); 2853 VN_RELE(*vpp); 2854 if (newvp == NULL) 2855 error = ENOSYS; 2856 else 2857 *vpp = newvp; 2858 } else if (ip->i_cflags & ICOMPRESS) { 2859 struct vnode *newvp; 2860 2861 /* 2862 * Compressed file, substitute dcfs vnode 2863 */ 2864 newvp = decompvp(*vpp, cr, ct); 2865 VN_RELE(*vpp); 2866 if (newvp == NULL) 2867 error = ENOSYS; 2868 else 2869 *vpp = newvp; 2870 } 2871 } 2872 if (ulp) { 2873 ufs_lockfs_end(ulp); 2874 } 2875 2876 if (error == EAGAIN) 2877 goto retry_lookup; 2878 2879 out: 2880 return (error); 2881 } 2882 2883 /*ARGSUSED*/ 2884 static int 2885 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl, 2886 int mode, struct vnode **vpp, struct cred *cr, int flag, 2887 caller_context_t *ct, vsecattr_t *vsecp) 2888 { 2889 struct inode *ip; 2890 struct inode *xip; 2891 struct inode *dip; 2892 struct vnode *xvp; 2893 struct ufsvfs *ufsvfsp; 2894 struct ulockfs *ulp; 2895 int error; 2896 int issync; 2897 int truncflag; 2898 int trans_size; 2899 int noentry; 2900 int defer_dip_seq_update = 0; /* need to defer update of dip->i_seq */ 2901 int retry = 1; 2902 int indeadlock; 2903 2904 again: 2905 ip = VTOI(dvp); 2906 ufsvfsp = ip->i_ufsvfs; 2907 truncflag = 0; 2908 2909 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK); 2910 if (error) 2911 goto out; 2912 2913 if (ulp) { 2914 trans_size = (int)TOP_CREATE_SIZE(ip); 2915 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size); 2916 } 2917 2918 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0) 2919 vap->va_mode &= ~VSVTX; 2920 2921 if (*name == '\0') { 2922 /* 2923 * Null component name refers to the directory itself. 2924 */ 2925 VN_HOLD(dvp); 2926 /* 2927 * Even though this is an error case, we need to grab the 2928 * quota lock since the error handling code below is common. 2929 */ 2930 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 2931 rw_enter(&ip->i_contents, RW_WRITER); 2932 error = EEXIST; 2933 } else { 2934 xip = NULL; 2935 noentry = 0; 2936 /* 2937 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK 2938 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock 2939 * possible, retries the operation. 2940 */ 2941 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE, 2942 retry_dir); 2943 if (indeadlock) 2944 goto again; 2945 2946 xvp = dnlc_lookup(dvp, name); 2947 if (xvp == DNLC_NO_VNODE) { 2948 noentry = 1; 2949 VN_RELE(xvp); 2950 xvp = NULL; 2951 } 2952 if (xvp) { 2953 rw_exit(&ip->i_rwlock); 2954 if (error = ufs_iaccess(ip, IEXEC, cr)) { 2955 VN_RELE(xvp); 2956 } else { 2957 error = EEXIST; 2958 xip = VTOI(xvp); 2959 } 2960 } else { 2961 /* 2962 * Suppress file system full message if we will retry 2963 */ 2964 error = ufs_direnter_cm(ip, name, DE_CREATE, 2965 vap, &xip, cr, (noentry | (retry ? IQUIET : 0))); 2966 if (error == EAGAIN) { 2967 if (ulp) { 2968 TRANS_END_CSYNC(ufsvfsp, error, issync, 2969 TOP_CREATE, trans_size); 2970 ufs_lockfs_end(ulp); 2971 } 2972 goto again; 2973 } 2974 rw_exit(&ip->i_rwlock); 2975 } 2976 ip = xip; 2977 if (ip != NULL) { 2978 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 2979 rw_enter(&ip->i_contents, RW_WRITER); 2980 } 2981 } 2982 2983 /* 2984 * If the file already exists and this is a non-exclusive create, 2985 * check permissions and allow access for non-directories. 2986 * Read-only create of an existing directory is also allowed. 2987 * We fail an exclusive create of anything which already exists. 2988 */ 2989 if (error == EEXIST) { 2990 dip = VTOI(dvp); 2991 if (excl == NONEXCL) { 2992 if ((((ip->i_mode & IFMT) == IFDIR) || 2993 ((ip->i_mode & IFMT) == IFATTRDIR)) && 2994 (mode & IWRITE)) 2995 error = EISDIR; 2996 else if (mode) 2997 error = ufs_iaccess(ip, mode, cr); 2998 else 2999 error = 0; 3000 } 3001 if (error) { 3002 rw_exit(&ip->i_contents); 3003 rw_exit(&ufsvfsp->vfs_dqrwlock); 3004 VN_RELE(ITOV(ip)); 3005 goto unlock; 3006 } 3007 /* 3008 * If the error EEXIST was set, then i_seq can not 3009 * have been updated. The sequence number interface 3010 * is defined such that a non-error VOP_CREATE must 3011 * increase the dir va_seq it by at least one. If we 3012 * have cleared the error, increase i_seq. Note that 3013 * we are increasing the dir i_seq and in rare cases 3014 * ip may actually be from the dvp, so we already have 3015 * the locks and it will not be subject to truncation. 3016 * In case we have to update i_seq of the parent 3017 * directory dip, we have to defer it till we have 3018 * released our locks on ip due to lock ordering requirements. 3019 */ 3020 if (ip != dip) 3021 defer_dip_seq_update = 1; 3022 else 3023 ip->i_seq++; 3024 3025 if (((ip->i_mode & IFMT) == IFREG) && 3026 (vap->va_mask & AT_SIZE) && vap->va_size == 0) { 3027 /* 3028 * Truncate regular files, if requested by caller. 3029 * Grab i_rwlock to make sure no one else is 3030 * currently writing to the file (we promised 3031 * bmap we would do this). 3032 * Must get the locks in the correct order. 3033 */ 3034 if (ip->i_size == 0) { 3035 ip->i_flag |= ICHG | IUPD; 3036 ip->i_seq++; 3037 TRANS_INODE(ufsvfsp, ip); 3038 } else { 3039 /* 3040 * Large Files: Why this check here? 3041 * Though we do it in vn_create() we really 3042 * want to guarantee that we do not destroy 3043 * Large file data by atomically checking 3044 * the size while holding the contents 3045 * lock. 3046 */ 3047 if (flag && !(flag & FOFFMAX) && 3048 ((ip->i_mode & IFMT) == IFREG) && 3049 (ip->i_size > (offset_t)MAXOFF32_T)) { 3050 rw_exit(&ip->i_contents); 3051 rw_exit(&ufsvfsp->vfs_dqrwlock); 3052 error = EOVERFLOW; 3053 goto unlock; 3054 } 3055 if (TRANS_ISTRANS(ufsvfsp)) 3056 truncflag++; 3057 else { 3058 rw_exit(&ip->i_contents); 3059 rw_exit(&ufsvfsp->vfs_dqrwlock); 3060 ufs_tryirwlock_trans(&ip->i_rwlock, 3061 RW_WRITER, TOP_CREATE, 3062 retry_file); 3063 if (indeadlock) { 3064 VN_RELE(ITOV(ip)); 3065 goto again; 3066 } 3067 rw_enter(&ufsvfsp->vfs_dqrwlock, 3068 RW_READER); 3069 rw_enter(&ip->i_contents, RW_WRITER); 3070 (void) ufs_itrunc(ip, (u_offset_t)0, 0, 3071 cr); 3072 rw_exit(&ip->i_rwlock); 3073 } 3074 3075 } 3076 if (error == 0) { 3077 vnevent_create(ITOV(ip), ct); 3078 } 3079 } 3080 } 3081 3082 if (error) { 3083 if (ip != NULL) { 3084 rw_exit(&ufsvfsp->vfs_dqrwlock); 3085 rw_exit(&ip->i_contents); 3086 } 3087 goto unlock; 3088 } 3089 3090 *vpp = ITOV(ip); 3091 ITIMES(ip); 3092 rw_exit(&ip->i_contents); 3093 rw_exit(&ufsvfsp->vfs_dqrwlock); 3094 3095 /* 3096 * If vnode is a device return special vnode instead. 3097 */ 3098 if (!error && IS_DEVVP(*vpp)) { 3099 struct vnode *newvp; 3100 3101 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 3102 VN_RELE(*vpp); 3103 if (newvp == NULL) { 3104 error = ENOSYS; 3105 goto unlock; 3106 } 3107 truncflag = 0; 3108 *vpp = newvp; 3109 } 3110 unlock: 3111 3112 /* 3113 * Do the deferred update of the parent directory's sequence 3114 * number now. 3115 */ 3116 if (defer_dip_seq_update == 1) { 3117 rw_enter(&dip->i_contents, RW_READER); 3118 mutex_enter(&dip->i_tlock); 3119 dip->i_seq++; 3120 mutex_exit(&dip->i_tlock); 3121 rw_exit(&dip->i_contents); 3122 } 3123 3124 if (ulp) { 3125 int terr = 0; 3126 3127 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE, 3128 trans_size); 3129 3130 /* 3131 * If we haven't had a more interesting failure 3132 * already, then anything that might've happened 3133 * here should be reported. 3134 */ 3135 if (error == 0) 3136 error = terr; 3137 } 3138 3139 if (!error && truncflag) { 3140 ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc); 3141 if (indeadlock) { 3142 if (ulp) 3143 ufs_lockfs_end(ulp); 3144 VN_RELE(ITOV(ip)); 3145 goto again; 3146 } 3147 (void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr); 3148 rw_exit(&ip->i_rwlock); 3149 } 3150 3151 if (ulp) 3152 ufs_lockfs_end(ulp); 3153 3154 /* 3155 * If no inodes available, try to free one up out of the 3156 * pending delete queue. 3157 */ 3158 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) { 3159 ufs_delete_drain_wait(ufsvfsp, 1); 3160 retry = 0; 3161 goto again; 3162 } 3163 3164 out: 3165 return (error); 3166 } 3167 3168 extern int ufs_idle_max; 3169 /*ARGSUSED*/ 3170 static int 3171 ufs_remove(struct vnode *vp, char *nm, struct cred *cr, 3172 caller_context_t *ct, int flags) 3173 { 3174 struct inode *ip = VTOI(vp); 3175 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 3176 struct ulockfs *ulp; 3177 vnode_t *rmvp = NULL; /* Vnode corresponding to name being removed */ 3178 int indeadlock; 3179 int error; 3180 int issync; 3181 int trans_size; 3182 3183 /* 3184 * don't let the delete queue get too long 3185 */ 3186 if (ufsvfsp == NULL) { 3187 error = EIO; 3188 goto out; 3189 } 3190 if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max) 3191 ufs_delete_drain(vp->v_vfsp, 1, 1); 3192 3193 retry_remove: 3194 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK); 3195 if (error) 3196 goto out; 3197 3198 if (ulp) 3199 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE, 3200 trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp))); 3201 3202 /* 3203 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK 3204 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock 3205 * possible, retries the operation. 3206 */ 3207 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry); 3208 if (indeadlock) 3209 goto retry_remove; 3210 error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0, 3211 DR_REMOVE, cr, &rmvp); 3212 rw_exit(&ip->i_rwlock); 3213 3214 if (ulp) { 3215 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size); 3216 ufs_lockfs_end(ulp); 3217 } 3218 3219 /* 3220 * This must be called after the remove transaction is closed. 3221 */ 3222 if (rmvp != NULL) { 3223 /* Only send the event if there were no errors */ 3224 if (error == 0) 3225 vnevent_remove(rmvp, vp, nm, ct); 3226 VN_RELE(rmvp); 3227 } 3228 out: 3229 return (error); 3230 } 3231 3232 /* 3233 * Link a file or a directory. Only privileged processes are allowed to 3234 * make links to directories. 3235 */ 3236 /*ARGSUSED*/ 3237 static int 3238 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr, 3239 caller_context_t *ct, int flags) 3240 { 3241 struct inode *sip; 3242 struct inode *tdp = VTOI(tdvp); 3243 struct ufsvfs *ufsvfsp = tdp->i_ufsvfs; 3244 struct ulockfs *ulp; 3245 struct vnode *realvp; 3246 int error; 3247 int issync; 3248 int trans_size; 3249 int isdev; 3250 int indeadlock; 3251 3252 retry_link: 3253 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK); 3254 if (error) 3255 goto out; 3256 3257 if (ulp) 3258 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK, 3259 trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp))); 3260 3261 if (VOP_REALVP(svp, &realvp, ct) == 0) 3262 svp = realvp; 3263 3264 /* 3265 * Make sure link for extended attributes is valid 3266 * We only support hard linking of attr in ATTRDIR to ATTRDIR 3267 * 3268 * Make certain we don't attempt to look at a device node as 3269 * a ufs inode. 3270 */ 3271 3272 isdev = IS_DEVVP(svp); 3273 if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) && 3274 ((tdp->i_mode & IFMT) == IFATTRDIR)) || 3275 ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) && 3276 ((tdp->i_mode & IFMT) == IFDIR))) { 3277 error = EINVAL; 3278 goto unlock; 3279 } 3280 3281 sip = VTOI(svp); 3282 if ((svp->v_type == VDIR && 3283 secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) || 3284 (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) { 3285 error = EPERM; 3286 goto unlock; 3287 } 3288 3289 /* 3290 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK 3291 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock 3292 * possible, retries the operation. 3293 */ 3294 ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry); 3295 if (indeadlock) 3296 goto retry_link; 3297 error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0, 3298 sip, cr, NULL); 3299 rw_exit(&tdp->i_rwlock); 3300 3301 unlock: 3302 if (ulp) { 3303 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size); 3304 ufs_lockfs_end(ulp); 3305 } 3306 3307 if (!error) { 3308 vnevent_link(svp, ct); 3309 } 3310 out: 3311 return (error); 3312 } 3313 3314 uint64_t ufs_rename_retry_cnt; 3315 uint64_t ufs_rename_upgrade_retry_cnt; 3316 uint64_t ufs_rename_dircheck_retry_cnt; 3317 clock_t ufs_rename_backoff_delay = 1; 3318 3319 /* 3320 * Rename a file or directory. 3321 * We are given the vnode and entry string of the source and the 3322 * vnode and entry string of the place we want to move the source 3323 * to (the target). The essential operation is: 3324 * unlink(target); 3325 * link(source, target); 3326 * unlink(source); 3327 * but "atomically". Can't do full commit without saving state in 3328 * the inode on disk, which isn't feasible at this time. Best we 3329 * can do is always guarantee that the TARGET exists. 3330 */ 3331 3332 /*ARGSUSED*/ 3333 static int 3334 ufs_rename( 3335 struct vnode *sdvp, /* old (source) parent vnode */ 3336 char *snm, /* old (source) entry name */ 3337 struct vnode *tdvp, /* new (target) parent vnode */ 3338 char *tnm, /* new (target) entry name */ 3339 struct cred *cr, 3340 caller_context_t *ct, 3341 int flags) 3342 { 3343 struct inode *sip = NULL; /* source inode */ 3344 struct inode *ip = NULL; /* check inode */ 3345 struct inode *sdp; /* old (source) parent inode */ 3346 struct inode *tdp; /* new (target) parent inode */ 3347 struct vnode *tvp = NULL; /* target vnode, if it exists */ 3348 struct vnode *realvp; 3349 struct ufsvfs *ufsvfsp; 3350 struct ulockfs *ulp; 3351 struct ufs_slot slot; 3352 timestruc_t now; 3353 int error; 3354 int issync; 3355 int trans_size; 3356 krwlock_t *first_lock; 3357 krwlock_t *second_lock; 3358 krwlock_t *reverse_lock; 3359 3360 sdp = VTOI(sdvp); 3361 slot.fbp = NULL; 3362 ufsvfsp = sdp->i_ufsvfs; 3363 retry_rename: 3364 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK); 3365 if (error) 3366 goto out; 3367 3368 if (ulp) 3369 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME, 3370 trans_size = (int)TOP_RENAME_SIZE(sdp)); 3371 3372 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3373 tdvp = realvp; 3374 3375 tdp = VTOI(tdvp); 3376 3377 3378 /* 3379 * We only allow renaming of attributes from ATTRDIR to ATTRDIR. 3380 */ 3381 if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) { 3382 error = EINVAL; 3383 goto unlock; 3384 } 3385 3386 /* 3387 * Look up inode of file we're supposed to rename. 3388 */ 3389 gethrestime(&now); 3390 if (error = ufs_dirlook(sdp, snm, &sip, cr, 0)) { 3391 if (error == EAGAIN) { 3392 if (ulp) { 3393 TRANS_END_CSYNC(ufsvfsp, error, issync, 3394 TOP_RENAME, trans_size); 3395 ufs_lockfs_end(ulp); 3396 } 3397 goto retry_rename; 3398 } 3399 3400 goto unlock; 3401 } 3402 3403 /* 3404 * Lock both the source and target directories (they may be 3405 * the same) to provide the atomicity semantics that was 3406 * previously provided by the per file system vfs_rename_lock 3407 * 3408 * with vfs_rename_lock removed to allow simultaneous renames 3409 * within a file system, ufs_dircheckpath can deadlock while 3410 * traversing back to ensure that source is not a parent directory 3411 * of target parent directory. This is because we get into 3412 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER. 3413 * If the tdp and sdp of the simultaneous renames happen to be 3414 * in the path of each other, it can lead to a deadlock. This 3415 * can be avoided by getting the locks as RW_READER here and then 3416 * upgrading to RW_WRITER after completing the ufs_dircheckpath. 3417 * 3418 * We hold the target directory's i_rwlock after calling 3419 * ufs_lockfs_begin but in many other operations (like ufs_readdir) 3420 * VOP_RWLOCK is explicitly called by the filesystem independent code 3421 * before calling the file system operation. In these cases the order 3422 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin 3423 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP 3424 * counter but with ufs_quiesce setting the SLOCK bit this becomes a 3425 * synchronizing object which might lead to a deadlock. So we use 3426 * rw_tryenter instead of rw_enter. If we fail to get this lock and 3427 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the 3428 * operation. 3429 */ 3430 retry: 3431 first_lock = &tdp->i_rwlock; 3432 second_lock = &sdp->i_rwlock; 3433 retry_firstlock: 3434 if (!rw_tryenter(first_lock, RW_READER)) { 3435 /* 3436 * We didn't get the lock. Check if the SLOCK is set in the 3437 * ufsvfs. If yes, we might be in a deadlock. Safer to give up 3438 * and wait for SLOCK to be cleared. 3439 */ 3440 3441 if (ulp && ULOCKFS_IS_SLOCK(ulp)) { 3442 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, 3443 trans_size); 3444 ufs_lockfs_end(ulp); 3445 goto retry_rename; 3446 3447 } else { 3448 /* 3449 * SLOCK isn't set so this is a genuine synchronization 3450 * case. Let's try again after giving them a breather. 3451 */ 3452 delay(RETRY_LOCK_DELAY); 3453 goto retry_firstlock; 3454 } 3455 } 3456 /* 3457 * Need to check if the tdp and sdp are same !!! 3458 */ 3459 if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) { 3460 /* 3461 * We didn't get the lock. Check if the SLOCK is set in the 3462 * ufsvfs. If yes, we might be in a deadlock. Safer to give up 3463 * and wait for SLOCK to be cleared. 3464 */ 3465 3466 rw_exit(first_lock); 3467 if (ulp && ULOCKFS_IS_SLOCK(ulp)) { 3468 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, 3469 trans_size); 3470 ufs_lockfs_end(ulp); 3471 goto retry_rename; 3472 3473 } else { 3474 /* 3475 * So we couldn't get the second level peer lock *and* 3476 * the SLOCK bit isn't set. Too bad we can be 3477 * contentding with someone wanting these locks otherway 3478 * round. Reverse the locks in case there is a heavy 3479 * contention for the second level lock. 3480 */ 3481 reverse_lock = first_lock; 3482 first_lock = second_lock; 3483 second_lock = reverse_lock; 3484 ufs_rename_retry_cnt++; 3485 goto retry_firstlock; 3486 } 3487 } 3488 3489 if (sip == tdp) { 3490 error = EINVAL; 3491 goto errout; 3492 } 3493 /* 3494 * Make sure we can delete the source entry. This requires 3495 * write permission on the containing directory. 3496 * Check for sticky directories. 3497 */ 3498 rw_enter(&sdp->i_contents, RW_READER); 3499 rw_enter(&sip->i_contents, RW_READER); 3500 if ((error = ufs_iaccess(sdp, IWRITE, cr)) != 0 || 3501 (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) { 3502 rw_exit(&sip->i_contents); 3503 rw_exit(&sdp->i_contents); 3504 goto errout; 3505 } 3506 3507 /* 3508 * If this is a rename of a directory and the parent is 3509 * different (".." must be changed), then the source 3510 * directory must not be in the directory hierarchy 3511 * above the target, as this would orphan everything 3512 * below the source directory. Also the user must have 3513 * write permission in the source so as to be able to 3514 * change "..". 3515 */ 3516 if ((((sip->i_mode & IFMT) == IFDIR) || 3517 ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) { 3518 ino_t inum; 3519 3520 if ((error = ufs_iaccess(sip, IWRITE, cr))) { 3521 rw_exit(&sip->i_contents); 3522 rw_exit(&sdp->i_contents); 3523 goto errout; 3524 } 3525 inum = sip->i_number; 3526 rw_exit(&sip->i_contents); 3527 rw_exit(&sdp->i_contents); 3528 if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) { 3529 /* 3530 * If we got EAGAIN ufs_dircheckpath detected a 3531 * potential deadlock and backed out. We need 3532 * to retry the operation since sdp and tdp have 3533 * to be released to avoid the deadlock. 3534 */ 3535 if (error == EAGAIN) { 3536 rw_exit(&tdp->i_rwlock); 3537 if (tdp != sdp) 3538 rw_exit(&sdp->i_rwlock); 3539 delay(ufs_rename_backoff_delay); 3540 ufs_rename_dircheck_retry_cnt++; 3541 goto retry; 3542 } 3543 goto errout; 3544 } 3545 } else { 3546 rw_exit(&sip->i_contents); 3547 rw_exit(&sdp->i_contents); 3548 } 3549 3550 3551 /* 3552 * Check for renaming '.' or '..' or alias of '.' 3553 */ 3554 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) { 3555 error = EINVAL; 3556 goto errout; 3557 } 3558 3559 /* 3560 * Simultaneous renames can deadlock in ufs_dircheckpath since it 3561 * tries to traverse back the file tree with both tdp and sdp held 3562 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks 3563 * as RW_READERS till ufs_dircheckpath is done. 3564 * Now that ufs_dircheckpath is done with, we can upgrade the locks 3565 * to RW_WRITER. 3566 */ 3567 if (!rw_tryupgrade(&tdp->i_rwlock)) { 3568 /* 3569 * The upgrade failed. We got to give away the lock 3570 * as to avoid deadlocking with someone else who is 3571 * waiting for writer lock. With the lock gone, we 3572 * cannot be sure the checks done above will hold 3573 * good when we eventually get them back as writer. 3574 * So if we can't upgrade we drop the locks and retry 3575 * everything again. 3576 */ 3577 rw_exit(&tdp->i_rwlock); 3578 if (tdp != sdp) 3579 rw_exit(&sdp->i_rwlock); 3580 delay(ufs_rename_backoff_delay); 3581 ufs_rename_upgrade_retry_cnt++; 3582 goto retry; 3583 } 3584 if (tdp != sdp) { 3585 if (!rw_tryupgrade(&sdp->i_rwlock)) { 3586 /* 3587 * The upgrade failed. We got to give away the lock 3588 * as to avoid deadlocking with someone else who is 3589 * waiting for writer lock. With the lock gone, we 3590 * cannot be sure the checks done above will hold 3591 * good when we eventually get them back as writer. 3592 * So if we can't upgrade we drop the locks and retry 3593 * everything again. 3594 */ 3595 rw_exit(&tdp->i_rwlock); 3596 rw_exit(&sdp->i_rwlock); 3597 delay(ufs_rename_backoff_delay); 3598 ufs_rename_upgrade_retry_cnt++; 3599 goto retry; 3600 } 3601 } 3602 3603 /* 3604 * Now that all the locks are held check to make sure another thread 3605 * didn't slip in and take out the sip. 3606 */ 3607 slot.status = NONE; 3608 if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec || 3609 sip->i_ctime.tv_sec > now.tv_sec) { 3610 rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER); 3611 rw_enter(&sdp->i_contents, RW_WRITER); 3612 error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot, 3613 &ip, cr, 0); 3614 rw_exit(&sdp->i_contents); 3615 rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock); 3616 if (error) { 3617 goto errout; 3618 } 3619 if (ip == NULL) { 3620 error = ENOENT; 3621 goto errout; 3622 } else { 3623 /* 3624 * If the inode was found need to drop the v_count 3625 * so as not to keep the filesystem from being 3626 * unmounted at a later time. 3627 */ 3628 VN_RELE(ITOV(ip)); 3629 } 3630 3631 /* 3632 * Release the slot.fbp that has the page mapped and 3633 * locked SE_SHARED, and could be used in in 3634 * ufs_direnter_lr() which needs to get the SE_EXCL lock 3635 * on said page. 3636 */ 3637 if (slot.fbp) { 3638 fbrelse(slot.fbp, S_OTHER); 3639 slot.fbp = NULL; 3640 } 3641 } 3642 3643 /* 3644 * Link source to the target. If a target exists, return its 3645 * vnode pointer in tvp. We'll release it after sending the 3646 * vnevent. 3647 */ 3648 if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr, &tvp)) { 3649 /* 3650 * ESAME isn't really an error; it indicates that the 3651 * operation should not be done because the source and target 3652 * are the same file, but that no error should be reported. 3653 */ 3654 if (error == ESAME) 3655 error = 0; 3656 goto errout; 3657 } 3658 3659 /* 3660 * Unlink the source. 3661 * Remove the source entry. ufs_dirremove() checks that the entry 3662 * still reflects sip, and returns an error if it doesn't. 3663 * If the entry has changed just forget about it. Release 3664 * the source inode. 3665 */ 3666 if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0, 3667 DR_RENAME, cr, NULL)) == ENOENT) 3668 error = 0; 3669 3670 errout: 3671 if (slot.fbp) 3672 fbrelse(slot.fbp, S_OTHER); 3673 3674 rw_exit(&tdp->i_rwlock); 3675 if (sdp != tdp) { 3676 rw_exit(&sdp->i_rwlock); 3677 } 3678 3679 unlock: 3680 if (ulp) { 3681 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size); 3682 ufs_lockfs_end(ulp); 3683 } 3684 3685 /* 3686 * If no errors, send the appropriate events on the source 3687 * and destination (a.k.a, target) vnodes, if they exist. 3688 * This has to be done after the rename transaction has closed. 3689 */ 3690 if (error == 0) { 3691 if (tvp != NULL) 3692 vnevent_rename_dest(tvp, tdvp, tnm, ct); 3693 3694 /* 3695 * Notify the target directory of the rename event 3696 * if source and target directories are not same. 3697 */ 3698 if (sdvp != tdvp) 3699 vnevent_rename_dest_dir(tdvp, ct); 3700 3701 /* 3702 * Note that if ufs_direnter_lr() returned ESAME then 3703 * this event will still be sent. This isn't expected 3704 * to be a problem for anticipated usage by consumers. 3705 */ 3706 if (sip != NULL) 3707 vnevent_rename_src(ITOV(sip), sdvp, snm, ct); 3708 } 3709 3710 if (tvp != NULL) 3711 VN_RELE(tvp); 3712 3713 if (sip != NULL) 3714 VN_RELE(ITOV(sip)); 3715 3716 out: 3717 return (error); 3718 } 3719 3720 /*ARGSUSED*/ 3721 static int 3722 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap, 3723 struct vnode **vpp, struct cred *cr, caller_context_t *ct, int flags, 3724 vsecattr_t *vsecp) 3725 { 3726 struct inode *ip; 3727 struct inode *xip; 3728 struct ufsvfs *ufsvfsp; 3729 struct ulockfs *ulp; 3730 int error; 3731 int issync; 3732 int trans_size; 3733 int indeadlock; 3734 int retry = 1; 3735 3736 ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 3737 3738 /* 3739 * Can't make directory in attr hidden dir 3740 */ 3741 if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR) 3742 return (EINVAL); 3743 3744 again: 3745 ip = VTOI(dvp); 3746 ufsvfsp = ip->i_ufsvfs; 3747 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK); 3748 if (error) 3749 goto out; 3750 if (ulp) 3751 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR, 3752 trans_size = (int)TOP_MKDIR_SIZE(ip)); 3753 3754 /* 3755 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK 3756 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock 3757 * possible, retries the operation. 3758 */ 3759 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry); 3760 if (indeadlock) 3761 goto again; 3762 3763 error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr, 3764 (retry ? IQUIET : 0)); 3765 if (error == EAGAIN) { 3766 if (ulp) { 3767 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR, 3768 trans_size); 3769 ufs_lockfs_end(ulp); 3770 } 3771 goto again; 3772 } 3773 3774 rw_exit(&ip->i_rwlock); 3775 if (error == 0) { 3776 ip = xip; 3777 *vpp = ITOV(ip); 3778 } else if (error == EEXIST) 3779 VN_RELE(ITOV(xip)); 3780 3781 if (ulp) { 3782 int terr = 0; 3783 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size); 3784 ufs_lockfs_end(ulp); 3785 if (error == 0) 3786 error = terr; 3787 } 3788 out: 3789 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) { 3790 ufs_delete_drain_wait(ufsvfsp, 1); 3791 retry = 0; 3792 goto again; 3793 } 3794 3795 return (error); 3796 } 3797 3798 /*ARGSUSED*/ 3799 static int 3800 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr, 3801 caller_context_t *ct, int flags) 3802 { 3803 struct inode *ip = VTOI(vp); 3804 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 3805 struct ulockfs *ulp; 3806 vnode_t *rmvp = NULL; /* Vnode of removed directory */ 3807 int error; 3808 int issync; 3809 int trans_size; 3810 int indeadlock; 3811 3812 /* 3813 * don't let the delete queue get too long 3814 */ 3815 if (ufsvfsp == NULL) { 3816 error = EIO; 3817 goto out; 3818 } 3819 if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max) 3820 ufs_delete_drain(vp->v_vfsp, 1, 1); 3821 3822 retry_rmdir: 3823 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK); 3824 if (error) 3825 goto out; 3826 3827 if (ulp) 3828 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR, 3829 trans_size = TOP_RMDIR_SIZE); 3830 3831 /* 3832 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK 3833 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock 3834 * possible, retries the operation. 3835 */ 3836 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry); 3837 if (indeadlock) 3838 goto retry_rmdir; 3839 error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr, 3840 &rmvp); 3841 rw_exit(&ip->i_rwlock); 3842 3843 if (ulp) { 3844 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR, 3845 trans_size); 3846 ufs_lockfs_end(ulp); 3847 } 3848 3849 /* 3850 * This must be done AFTER the rmdir transaction has closed. 3851 */ 3852 if (rmvp != NULL) { 3853 /* Only send the event if there were no errors */ 3854 if (error == 0) 3855 vnevent_rmdir(rmvp, vp, nm, ct); 3856 VN_RELE(rmvp); 3857 } 3858 out: 3859 return (error); 3860 } 3861 3862 /* ARGSUSED */ 3863 static int 3864 ufs_readdir( 3865 struct vnode *vp, 3866 struct uio *uiop, 3867 struct cred *cr, 3868 int *eofp, 3869 caller_context_t *ct, 3870 int flags) 3871 { 3872 struct iovec *iovp; 3873 struct inode *ip; 3874 struct direct *idp; 3875 struct dirent64 *odp; 3876 struct fbuf *fbp; 3877 struct ufsvfs *ufsvfsp; 3878 struct ulockfs *ulp; 3879 caddr_t outbuf; 3880 size_t bufsize; 3881 uint_t offset; 3882 uint_t bytes_wanted, total_bytes_wanted; 3883 int incount = 0; 3884 int outcount = 0; 3885 int error; 3886 3887 ip = VTOI(vp); 3888 ASSERT(RW_READ_HELD(&ip->i_rwlock)); 3889 3890 if (uiop->uio_loffset >= MAXOFF32_T) { 3891 if (eofp) 3892 *eofp = 1; 3893 return (0); 3894 } 3895 3896 /* 3897 * Check if we have been called with a valid iov_len 3898 * and bail out if not, otherwise we may potentially loop 3899 * forever further down. 3900 */ 3901 if (uiop->uio_iov->iov_len <= 0) { 3902 error = EINVAL; 3903 goto out; 3904 } 3905 3906 /* 3907 * Large Files: When we come here we are guaranteed that 3908 * uio_offset can be used safely. The high word is zero. 3909 */ 3910 3911 ufsvfsp = ip->i_ufsvfs; 3912 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK); 3913 if (error) 3914 goto out; 3915 3916 iovp = uiop->uio_iov; 3917 total_bytes_wanted = iovp->iov_len; 3918 3919 /* Large Files: directory files should not be "large" */ 3920 3921 ASSERT(ip->i_size <= MAXOFF32_T); 3922 3923 /* Force offset to be valid (to guard against bogus lseek() values) */ 3924 offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1); 3925 3926 /* Quit if at end of file or link count of zero (posix) */ 3927 if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) { 3928 if (eofp) 3929 *eofp = 1; 3930 error = 0; 3931 goto unlock; 3932 } 3933 3934 /* 3935 * Get space to change directory entries into fs independent format. 3936 * Do fast alloc for the most commonly used-request size (filesystem 3937 * block size). 3938 */ 3939 if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) { 3940 bufsize = total_bytes_wanted; 3941 outbuf = kmem_alloc(bufsize, KM_SLEEP); 3942 odp = (struct dirent64 *)outbuf; 3943 } else { 3944 bufsize = total_bytes_wanted; 3945 odp = (struct dirent64 *)iovp->iov_base; 3946 } 3947 3948 nextblk: 3949 bytes_wanted = total_bytes_wanted; 3950 3951 /* Truncate request to file size */ 3952 if (offset + bytes_wanted > (int)ip->i_size) 3953 bytes_wanted = (int)(ip->i_size - offset); 3954 3955 /* Comply with MAXBSIZE boundary restrictions of fbread() */ 3956 if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE) 3957 bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET); 3958 3959 /* 3960 * Read in the next chunk. 3961 * We are still holding the i_rwlock. 3962 */ 3963 error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp); 3964 3965 if (error) 3966 goto update_inode; 3967 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) && 3968 (!ufsvfsp->vfs_noatime)) { 3969 ip->i_flag |= IACC; 3970 } 3971 incount = 0; 3972 idp = (struct direct *)fbp->fb_addr; 3973 if (idp->d_ino == 0 && idp->d_reclen == 0 && idp->d_namlen == 0) { 3974 cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, " 3975 "fs = %s\n", 3976 (u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt); 3977 fbrelse(fbp, S_OTHER); 3978 error = ENXIO; 3979 goto update_inode; 3980 } 3981 /* Transform to file-system independent format */ 3982 while (incount < bytes_wanted) { 3983 /* 3984 * If the current directory entry is mangled, then skip 3985 * to the next block. It would be nice to set the FSBAD 3986 * flag in the super-block so that a fsck is forced on 3987 * next reboot, but locking is a problem. 3988 */ 3989 if (idp->d_reclen & 0x3) { 3990 offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1); 3991 break; 3992 } 3993 3994 /* Skip to requested offset and skip empty entries */ 3995 if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) { 3996 ushort_t this_reclen = 3997 DIRENT64_RECLEN(idp->d_namlen); 3998 /* Buffer too small for any entries */ 3999 if (!outcount && this_reclen > bufsize) { 4000 fbrelse(fbp, S_OTHER); 4001 error = EINVAL; 4002 goto update_inode; 4003 } 4004 /* If would overrun the buffer, quit */ 4005 if (outcount + this_reclen > bufsize) { 4006 break; 4007 } 4008 /* Take this entry */ 4009 odp->d_ino = (ino64_t)idp->d_ino; 4010 odp->d_reclen = (ushort_t)this_reclen; 4011 odp->d_off = (offset_t)(offset + idp->d_reclen); 4012 4013 /* use strncpy(9f) to zero out uninitialized bytes */ 4014 4015 ASSERT(strlen(idp->d_name) + 1 <= 4016 DIRENT64_NAMELEN(this_reclen)); 4017 (void) strncpy(odp->d_name, idp->d_name, 4018 DIRENT64_NAMELEN(this_reclen)); 4019 outcount += odp->d_reclen; 4020 odp = (struct dirent64 *) 4021 ((intptr_t)odp + odp->d_reclen); 4022 ASSERT(outcount <= bufsize); 4023 } 4024 if (idp->d_reclen) { 4025 incount += idp->d_reclen; 4026 offset += idp->d_reclen; 4027 idp = (struct direct *)((intptr_t)idp + idp->d_reclen); 4028 } else { 4029 offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1); 4030 break; 4031 } 4032 } 4033 /* Release the chunk */ 4034 fbrelse(fbp, S_OTHER); 4035 4036 /* Read whole block, but got no entries, read another if not eof */ 4037 4038 /* 4039 * Large Files: casting i_size to int here is not a problem 4040 * because directory sizes are always less than MAXOFF32_T. 4041 * See assertion above. 4042 */ 4043 4044 if (offset < (int)ip->i_size && !outcount) 4045 goto nextblk; 4046 4047 /* Copy out the entry data */ 4048 if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) { 4049 iovp->iov_base += outcount; 4050 iovp->iov_len -= outcount; 4051 uiop->uio_resid -= outcount; 4052 uiop->uio_offset = offset; 4053 } else if ((error = uiomove(outbuf, (long)outcount, UIO_READ, 4054 uiop)) == 0) 4055 uiop->uio_offset = offset; 4056 update_inode: 4057 ITIMES(ip); 4058 if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) 4059 kmem_free(outbuf, bufsize); 4060 4061 if (eofp && error == 0) 4062 *eofp = (uiop->uio_offset >= (int)ip->i_size); 4063 unlock: 4064 if (ulp) { 4065 ufs_lockfs_end(ulp); 4066 } 4067 out: 4068 return (error); 4069 } 4070 4071 /*ARGSUSED*/ 4072 static int 4073 ufs_symlink( 4074 struct vnode *dvp, /* ptr to parent dir vnode */ 4075 char *linkname, /* name of symbolic link */ 4076 struct vattr *vap, /* attributes */ 4077 char *target, /* target path */ 4078 struct cred *cr, /* user credentials */ 4079 caller_context_t *ct, 4080 int flags) 4081 { 4082 struct inode *ip, *dip = VTOI(dvp); 4083 struct ufsvfs *ufsvfsp = dip->i_ufsvfs; 4084 struct ulockfs *ulp; 4085 int error; 4086 int issync; 4087 int trans_size; 4088 int residual; 4089 int ioflag; 4090 int retry = 1; 4091 4092 /* 4093 * No symlinks in attrdirs at this time 4094 */ 4095 if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR) 4096 return (EINVAL); 4097 4098 again: 4099 ip = (struct inode *)NULL; 4100 vap->va_type = VLNK; 4101 vap->va_rdev = 0; 4102 4103 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK); 4104 if (error) 4105 goto out; 4106 4107 if (ulp) 4108 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK, 4109 trans_size = (int)TOP_SYMLINK_SIZE(dip)); 4110 4111 /* 4112 * We must create the inode before the directory entry, to avoid 4113 * racing with readlink(). ufs_dirmakeinode requires that we 4114 * hold the quota lock as reader, and directory locks as writer. 4115 */ 4116 4117 rw_enter(&dip->i_rwlock, RW_WRITER); 4118 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 4119 rw_enter(&dip->i_contents, RW_WRITER); 4120 4121 /* 4122 * Suppress any out of inodes messages if we will retry on 4123 * ENOSP 4124 */ 4125 if (retry) 4126 dip->i_flag |= IQUIET; 4127 4128 error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr); 4129 4130 dip->i_flag &= ~IQUIET; 4131 4132 rw_exit(&dip->i_contents); 4133 rw_exit(&ufsvfsp->vfs_dqrwlock); 4134 rw_exit(&dip->i_rwlock); 4135 4136 if (error) 4137 goto unlock; 4138 4139 /* 4140 * OK. The inode has been created. Write out the data of the 4141 * symbolic link. Since symbolic links are metadata, and should 4142 * remain consistent across a system crash, we need to force the 4143 * data out synchronously. 4144 * 4145 * (This is a change from the semantics in earlier releases, which 4146 * only created symbolic links synchronously if the semi-documented 4147 * 'syncdir' option was set, or if we were being invoked by the NFS 4148 * server, which requires symbolic links to be created synchronously.) 4149 * 4150 * We need to pass in a pointer for the residual length; otherwise 4151 * ufs_rdwri() will always return EIO if it can't write the data, 4152 * even if the error was really ENOSPC or EDQUOT. 4153 */ 4154 4155 ioflag = FWRITE | FDSYNC; 4156 residual = 0; 4157 4158 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 4159 rw_enter(&ip->i_contents, RW_WRITER); 4160 4161 /* 4162 * Suppress file system full messages if we will retry 4163 */ 4164 if (retry) 4165 ip->i_flag |= IQUIET; 4166 4167 error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target), 4168 (offset_t)0, UIO_SYSSPACE, &residual, cr); 4169 4170 ip->i_flag &= ~IQUIET; 4171 4172 if (error) { 4173 rw_exit(&ip->i_contents); 4174 rw_exit(&ufsvfsp->vfs_dqrwlock); 4175 goto remove; 4176 } 4177 4178 /* 4179 * If the link's data is small enough, we can cache it in the inode. 4180 * This is a "fast symbolic link". We don't use the first direct 4181 * block because that's actually used to point at the symbolic link's 4182 * contents on disk; but we know that none of the other direct or 4183 * indirect blocks can be used because symbolic links are restricted 4184 * to be smaller than a file system block. 4185 */ 4186 4187 ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip))); 4188 4189 if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) { 4190 if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) { 4191 ip->i_flag |= IFASTSYMLNK; 4192 } else { 4193 int i; 4194 /* error, clear garbage left behind */ 4195 for (i = 1; i < NDADDR; i++) 4196 ip->i_db[i] = 0; 4197 for (i = 0; i < NIADDR; i++) 4198 ip->i_ib[i] = 0; 4199 } 4200 } 4201 4202 rw_exit(&ip->i_contents); 4203 rw_exit(&ufsvfsp->vfs_dqrwlock); 4204 4205 /* 4206 * OK. We've successfully created the symbolic link. All that 4207 * remains is to insert it into the appropriate directory. 4208 */ 4209 4210 rw_enter(&dip->i_rwlock, RW_WRITER); 4211 error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr, NULL); 4212 rw_exit(&dip->i_rwlock); 4213 4214 /* 4215 * Fall through into remove-on-error code. We're either done, or we 4216 * need to remove the inode (if we couldn't insert it). 4217 */ 4218 4219 remove: 4220 if (error && (ip != NULL)) { 4221 rw_enter(&ip->i_contents, RW_WRITER); 4222 ip->i_nlink--; 4223 ip->i_flag |= ICHG; 4224 ip->i_seq++; 4225 ufs_setreclaim(ip); 4226 rw_exit(&ip->i_contents); 4227 } 4228 4229 unlock: 4230 if (ip != NULL) 4231 VN_RELE(ITOV(ip)); 4232 4233 if (ulp) { 4234 int terr = 0; 4235 4236 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK, 4237 trans_size); 4238 ufs_lockfs_end(ulp); 4239 if (error == 0) 4240 error = terr; 4241 } 4242 4243 /* 4244 * We may have failed due to lack of an inode or of a block to 4245 * store the target in. Try flushing the delete queue to free 4246 * logically-available things up and try again. 4247 */ 4248 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) { 4249 ufs_delete_drain_wait(ufsvfsp, 1); 4250 retry = 0; 4251 goto again; 4252 } 4253 4254 out: 4255 return (error); 4256 } 4257 4258 /* 4259 * Ufs specific routine used to do ufs io. 4260 */ 4261 int 4262 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base, 4263 ssize_t len, offset_t offset, enum uio_seg seg, int *aresid, 4264 struct cred *cr) 4265 { 4266 struct uio auio; 4267 struct iovec aiov; 4268 int error; 4269 4270 ASSERT(RW_LOCK_HELD(&ip->i_contents)); 4271 4272 bzero((caddr_t)&auio, sizeof (uio_t)); 4273 bzero((caddr_t)&aiov, sizeof (iovec_t)); 4274 4275 aiov.iov_base = base; 4276 aiov.iov_len = len; 4277 auio.uio_iov = &aiov; 4278 auio.uio_iovcnt = 1; 4279 auio.uio_loffset = offset; 4280 auio.uio_segflg = (short)seg; 4281 auio.uio_resid = len; 4282 4283 if (rw == UIO_WRITE) { 4284 auio.uio_fmode = FWRITE; 4285 auio.uio_extflg = UIO_COPY_DEFAULT; 4286 auio.uio_llimit = curproc->p_fsz_ctl; 4287 error = wrip(ip, &auio, ioflag, cr); 4288 } else { 4289 auio.uio_fmode = FREAD; 4290 auio.uio_extflg = UIO_COPY_CACHED; 4291 auio.uio_llimit = MAXOFFSET_T; 4292 error = rdip(ip, &auio, ioflag, cr); 4293 } 4294 4295 if (aresid) { 4296 *aresid = auio.uio_resid; 4297 } else if (auio.uio_resid) { 4298 error = EIO; 4299 } 4300 return (error); 4301 } 4302 4303 /*ARGSUSED*/ 4304 static int 4305 ufs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct) 4306 { 4307 struct ufid *ufid; 4308 struct inode *ip = VTOI(vp); 4309 4310 if (ip->i_ufsvfs == NULL) 4311 return (EIO); 4312 4313 if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) { 4314 fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t); 4315 return (ENOSPC); 4316 } 4317 4318 ufid = (struct ufid *)fidp; 4319 bzero((char *)ufid, sizeof (struct ufid)); 4320 ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t); 4321 ufid->ufid_ino = ip->i_number; 4322 ufid->ufid_gen = ip->i_gen; 4323 4324 return (0); 4325 } 4326 4327 /* ARGSUSED2 */ 4328 static int 4329 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp) 4330 { 4331 struct inode *ip = VTOI(vp); 4332 struct ufsvfs *ufsvfsp; 4333 int forcedirectio; 4334 4335 /* 4336 * Read case is easy. 4337 */ 4338 if (!write_lock) { 4339 rw_enter(&ip->i_rwlock, RW_READER); 4340 return (V_WRITELOCK_FALSE); 4341 } 4342 4343 /* 4344 * Caller has requested a writer lock, but that inhibits any 4345 * concurrency in the VOPs that follow. Acquire the lock shared 4346 * and defer exclusive access until it is known to be needed in 4347 * other VOP handlers. Some cases can be determined here. 4348 */ 4349 4350 /* 4351 * If directio is not set, there is no chance of concurrency, 4352 * so just acquire the lock exclusive. Beware of a forced 4353 * unmount before looking at the mount option. 4354 */ 4355 ufsvfsp = ip->i_ufsvfs; 4356 forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0; 4357 if (!(ip->i_flag & IDIRECTIO || forcedirectio) || 4358 !ufs_allow_shared_writes) { 4359 rw_enter(&ip->i_rwlock, RW_WRITER); 4360 return (V_WRITELOCK_TRUE); 4361 } 4362 4363 /* 4364 * Mandatory locking forces acquiring i_rwlock exclusive. 4365 */ 4366 if (MANDLOCK(vp, ip->i_mode)) { 4367 rw_enter(&ip->i_rwlock, RW_WRITER); 4368 return (V_WRITELOCK_TRUE); 4369 } 4370 4371 /* 4372 * Acquire the lock shared in case a concurrent write follows. 4373 * Mandatory locking could have become enabled before the lock 4374 * was acquired. Re-check and upgrade if needed. 4375 */ 4376 rw_enter(&ip->i_rwlock, RW_READER); 4377 if (MANDLOCK(vp, ip->i_mode)) { 4378 rw_exit(&ip->i_rwlock); 4379 rw_enter(&ip->i_rwlock, RW_WRITER); 4380 return (V_WRITELOCK_TRUE); 4381 } 4382 return (V_WRITELOCK_FALSE); 4383 } 4384 4385 /*ARGSUSED*/ 4386 static void 4387 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp) 4388 { 4389 struct inode *ip = VTOI(vp); 4390 4391 rw_exit(&ip->i_rwlock); 4392 } 4393 4394 /* ARGSUSED */ 4395 static int 4396 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp, 4397 caller_context_t *ct) 4398 { 4399 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4400 } 4401 4402 /* ARGSUSED */ 4403 static int 4404 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag, 4405 offset_t offset, struct flk_callback *flk_cbp, struct cred *cr, 4406 caller_context_t *ct) 4407 { 4408 struct inode *ip = VTOI(vp); 4409 4410 if (ip->i_ufsvfs == NULL) 4411 return (EIO); 4412 4413 /* 4414 * If file is being mapped, disallow frlock. 4415 * XXX I am not holding tlock while checking i_mapcnt because the 4416 * current locking strategy drops all locks before calling fs_frlock. 4417 * So, mapcnt could change before we enter fs_frlock making is 4418 * meaningless to have held tlock in the first place. 4419 */ 4420 if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode)) 4421 return (EAGAIN); 4422 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4423 } 4424 4425 /* ARGSUSED */ 4426 static int 4427 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag, 4428 offset_t offset, cred_t *cr, caller_context_t *ct) 4429 { 4430 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs; 4431 struct ulockfs *ulp; 4432 int error; 4433 4434 if ((error = convoff(vp, bfp, 0, offset)) == 0) { 4435 if (cmd == F_FREESP) { 4436 error = ufs_lockfs_begin(ufsvfsp, &ulp, 4437 ULOCKFS_SPACE_MASK); 4438 if (error) 4439 return (error); 4440 error = ufs_freesp(vp, bfp, flag, cr); 4441 } else if (cmd == F_ALLOCSP) { 4442 error = ufs_lockfs_begin(ufsvfsp, &ulp, 4443 ULOCKFS_FALLOCATE_MASK); 4444 if (error) 4445 return (error); 4446 error = ufs_allocsp(vp, bfp, cr); 4447 } else 4448 return (EINVAL); /* Command not handled here */ 4449 4450 if (ulp) 4451 ufs_lockfs_end(ulp); 4452 4453 } 4454 return (error); 4455 } 4456 4457 /* 4458 * Used to determine if read ahead should be done. Also used to 4459 * to determine when write back occurs. 4460 */ 4461 #define CLUSTSZ(ip) ((ip)->i_ufsvfs->vfs_ioclustsz) 4462 4463 /* 4464 * A faster version of ufs_getpage. 4465 * 4466 * We optimize by inlining the pvn_getpages iterator, eliminating 4467 * calls to bmap_read if file doesn't have UFS holes, and avoiding 4468 * the overhead of page_exists(). 4469 * 4470 * When files has UFS_HOLES and ufs_getpage is called with S_READ, 4471 * we set *protp to PROT_READ to avoid calling bmap_read. This approach 4472 * victimizes performance when a file with UFS holes is faulted 4473 * first in the S_READ mode, and then in the S_WRITE mode. We will get 4474 * two MMU faults in this case. 4475 * 4476 * XXX - the inode fields which control the sequential mode are not 4477 * protected by any mutex. The read ahead will act wild if 4478 * multiple processes will access the file concurrently and 4479 * some of them in sequential mode. One particulary bad case 4480 * is if another thread will change the value of i_nextrio between 4481 * the time this thread tests the i_nextrio value and then reads it 4482 * again to use it as the offset for the read ahead. 4483 */ 4484 /*ARGSUSED*/ 4485 static int 4486 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp, 4487 page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr, 4488 enum seg_rw rw, struct cred *cr, caller_context_t *ct) 4489 { 4490 u_offset_t uoff = (u_offset_t)off; /* type conversion */ 4491 u_offset_t pgoff; 4492 u_offset_t eoff; 4493 struct inode *ip = VTOI(vp); 4494 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 4495 struct fs *fs; 4496 struct ulockfs *ulp; 4497 page_t **pl; 4498 caddr_t pgaddr; 4499 krw_t rwtype; 4500 int err; 4501 int has_holes; 4502 int beyond_eof; 4503 int seqmode; 4504 int pgsize = PAGESIZE; 4505 int dolock; 4506 int do_qlock; 4507 int trans_size; 4508 4509 ASSERT((uoff & PAGEOFFSET) == 0); 4510 4511 if (protp) 4512 *protp = PROT_ALL; 4513 4514 /* 4515 * Obey the lockfs protocol 4516 */ 4517 err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg, 4518 rw == S_READ || rw == S_EXEC, protp); 4519 if (err) 4520 goto out; 4521 4522 fs = ufsvfsp->vfs_fs; 4523 4524 if (ulp && (rw == S_CREATE || rw == S_WRITE) && 4525 !(vp->v_flag & VISSWAP)) { 4526 /* 4527 * Try to start a transaction, will return if blocking is 4528 * expected to occur and the address space is not the 4529 * kernel address space. 4530 */ 4531 trans_size = TOP_GETPAGE_SIZE(ip); 4532 if (seg->s_as != &kas) { 4533 TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, 4534 trans_size, err) 4535 if (err == EWOULDBLOCK) { 4536 /* 4537 * Use EDEADLK here because the VM code 4538 * can normally never see this error. 4539 */ 4540 err = EDEADLK; 4541 ufs_lockfs_end(ulp); 4542 goto out; 4543 } 4544 } else { 4545 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size); 4546 } 4547 } 4548 4549 if (vp->v_flag & VNOMAP) { 4550 err = ENOSYS; 4551 goto unlock; 4552 } 4553 4554 seqmode = ip->i_nextr == uoff && rw != S_CREATE; 4555 4556 rwtype = RW_READER; /* start as a reader */ 4557 dolock = (rw_owner(&ip->i_contents) != curthread); 4558 /* 4559 * If this thread owns the lock, i.e., this thread grabbed it 4560 * as writer somewhere above, then we don't need to grab the 4561 * lock as reader in this routine. 4562 */ 4563 do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread); 4564 4565 retrylock: 4566 if (dolock) { 4567 /* 4568 * Grab the quota lock if we need to call 4569 * bmap_write() below (with i_contents as writer). 4570 */ 4571 if (do_qlock && rwtype == RW_WRITER) 4572 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER); 4573 rw_enter(&ip->i_contents, rwtype); 4574 } 4575 4576 /* 4577 * We may be getting called as a side effect of a bmap using 4578 * fbread() when the blocks might be being allocated and the 4579 * size has not yet been up'ed. In this case we want to be 4580 * able to return zero pages if we get back UFS_HOLE from 4581 * calling bmap for a non write case here. We also might have 4582 * to read some frags from the disk into a page if we are 4583 * extending the number of frags for a given lbn in bmap(). 4584 * Large Files: The read of i_size here is atomic because 4585 * i_contents is held here. If dolock is zero, the lock 4586 * is held in bmap routines. 4587 */ 4588 beyond_eof = uoff + len > 4589 P2ROUNDUP_TYPED(ip->i_size, PAGESIZE, u_offset_t); 4590 if (beyond_eof && seg != segkmap) { 4591 if (dolock) { 4592 rw_exit(&ip->i_contents); 4593 if (do_qlock && rwtype == RW_WRITER) 4594 rw_exit(&ufsvfsp->vfs_dqrwlock); 4595 } 4596 err = EFAULT; 4597 goto unlock; 4598 } 4599 4600 /* 4601 * Must hold i_contents lock throughout the call to pvn_getpages 4602 * since locked pages are returned from each call to ufs_getapage. 4603 * Must *not* return locked pages and then try for contents lock 4604 * due to lock ordering requirements (inode > page) 4605 */ 4606 4607 has_holes = bmap_has_holes(ip); 4608 4609 if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) { 4610 int blk_size; 4611 u_offset_t offset; 4612 4613 /* 4614 * We must acquire the RW_WRITER lock in order to 4615 * call bmap_write(). 4616 */ 4617 if (dolock && rwtype == RW_READER) { 4618 rwtype = RW_WRITER; 4619 4620 /* 4621 * Grab the quota lock before 4622 * upgrading i_contents, but if we can't grab it 4623 * don't wait here due to lock order: 4624 * vfs_dqrwlock > i_contents. 4625 */ 4626 if (do_qlock && 4627 rw_tryenter(&ufsvfsp->vfs_dqrwlock, RW_READER) 4628 == 0) { 4629 rw_exit(&ip->i_contents); 4630 goto retrylock; 4631 } 4632 if (!rw_tryupgrade(&ip->i_contents)) { 4633 rw_exit(&ip->i_contents); 4634 if (do_qlock) 4635 rw_exit(&ufsvfsp->vfs_dqrwlock); 4636 goto retrylock; 4637 } 4638 } 4639 4640 /* 4641 * May be allocating disk blocks for holes here as 4642 * a result of mmap faults. write(2) does the bmap_write 4643 * in rdip/wrip, not here. We are not dealing with frags 4644 * in this case. 4645 */ 4646 /* 4647 * Large Files: We cast fs_bmask field to offset_t 4648 * just as we do for MAXBMASK because uoff is a 64-bit 4649 * data type. fs_bmask will still be a 32-bit type 4650 * as we cannot change any ondisk data structures. 4651 */ 4652 4653 offset = uoff & (offset_t)fs->fs_bmask; 4654 while (offset < uoff + len) { 4655 blk_size = (int)blksize(fs, ip, lblkno(fs, offset)); 4656 err = bmap_write(ip, offset, blk_size, 4657 BI_NORMAL, NULL, cr); 4658 if (ip->i_flag & (ICHG|IUPD)) 4659 ip->i_seq++; 4660 if (err) 4661 goto update_inode; 4662 offset += blk_size; /* XXX - make this contig */ 4663 } 4664 } 4665 4666 /* 4667 * Can be a reader from now on. 4668 */ 4669 if (dolock && rwtype == RW_WRITER) { 4670 rw_downgrade(&ip->i_contents); 4671 /* 4672 * We can release vfs_dqrwlock early so do it, but make 4673 * sure we don't try to release it again at the bottom. 4674 */ 4675 if (do_qlock) { 4676 rw_exit(&ufsvfsp->vfs_dqrwlock); 4677 do_qlock = 0; 4678 } 4679 } 4680 4681 /* 4682 * We remove PROT_WRITE in cases when the file has UFS holes 4683 * because we don't want to call bmap_read() to check each 4684 * page if it is backed with a disk block. 4685 */ 4686 if (protp && has_holes && rw != S_WRITE && rw != S_CREATE) 4687 *protp &= ~PROT_WRITE; 4688 4689 err = 0; 4690 4691 /* 4692 * The loop looks up pages in the range [off, off + len). 4693 * For each page, we first check if we should initiate an asynchronous 4694 * read ahead before we call page_lookup (we may sleep in page_lookup 4695 * for a previously initiated disk read). 4696 */ 4697 eoff = (uoff + len); 4698 for (pgoff = uoff, pgaddr = addr, pl = plarr; 4699 pgoff < eoff; /* empty */) { 4700 page_t *pp; 4701 u_offset_t nextrio; 4702 se_t se; 4703 int retval; 4704 4705 se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED); 4706 4707 /* Handle async getpage (faultahead) */ 4708 if (plarr == NULL) { 4709 ip->i_nextrio = pgoff; 4710 (void) ufs_getpage_ra(vp, pgoff, seg, pgaddr); 4711 pgoff += pgsize; 4712 pgaddr += pgsize; 4713 continue; 4714 } 4715 /* 4716 * Check if we should initiate read ahead of next cluster. 4717 * We call page_exists only when we need to confirm that 4718 * we have the current page before we initiate the read ahead. 4719 */ 4720 nextrio = ip->i_nextrio; 4721 if (seqmode && 4722 pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio && 4723 nextrio < ip->i_size && page_exists(vp, pgoff)) { 4724 retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr); 4725 /* 4726 * We always read ahead the next cluster of data 4727 * starting from i_nextrio. If the page (vp,nextrio) 4728 * is actually in core at this point, the routine 4729 * ufs_getpage_ra() will stop pre-fetching data 4730 * until we read that page in a synchronized manner 4731 * through ufs_getpage_miss(). So, we should increase 4732 * i_nextrio if the page (vp, nextrio) exists. 4733 */ 4734 if ((retval == 0) && page_exists(vp, nextrio)) { 4735 ip->i_nextrio = nextrio + pgsize; 4736 } 4737 } 4738 4739 if ((pp = page_lookup(vp, pgoff, se)) != NULL) { 4740 /* 4741 * We found the page in the page cache. 4742 */ 4743 *pl++ = pp; 4744 pgoff += pgsize; 4745 pgaddr += pgsize; 4746 len -= pgsize; 4747 plsz -= pgsize; 4748 } else { 4749 /* 4750 * We have to create the page, or read it from disk. 4751 */ 4752 if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr, 4753 pl, plsz, rw, seqmode)) 4754 goto error; 4755 4756 while (*pl != NULL) { 4757 pl++; 4758 pgoff += pgsize; 4759 pgaddr += pgsize; 4760 len -= pgsize; 4761 plsz -= pgsize; 4762 } 4763 } 4764 } 4765 4766 /* 4767 * Return pages up to plsz if they are in the page cache. 4768 * We cannot return pages if there is a chance that they are 4769 * backed with a UFS hole and rw is S_WRITE or S_CREATE. 4770 */ 4771 if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) { 4772 4773 ASSERT((protp == NULL) || 4774 !(has_holes && (*protp & PROT_WRITE))); 4775 4776 eoff = pgoff + plsz; 4777 while (pgoff < eoff) { 4778 page_t *pp; 4779 4780 if ((pp = page_lookup_nowait(vp, pgoff, 4781 SE_SHARED)) == NULL) 4782 break; 4783 4784 *pl++ = pp; 4785 pgoff += pgsize; 4786 plsz -= pgsize; 4787 } 4788 } 4789 4790 if (plarr) 4791 *pl = NULL; /* Terminate page list */ 4792 ip->i_nextr = pgoff; 4793 4794 error: 4795 if (err && plarr) { 4796 /* 4797 * Release any pages we have locked. 4798 */ 4799 while (pl > &plarr[0]) 4800 page_unlock(*--pl); 4801 4802 plarr[0] = NULL; 4803 } 4804 4805 update_inode: 4806 /* 4807 * If the inode is not already marked for IACC (in rdip() for read) 4808 * and the inode is not marked for no access time update (in wrip() 4809 * for write) then update the inode access time and mod time now. 4810 */ 4811 if ((ip->i_flag & (IACC | INOACC)) == 0) { 4812 if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) { 4813 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && 4814 (fs->fs_ronly == 0) && 4815 (!ufsvfsp->vfs_noatime)) { 4816 mutex_enter(&ip->i_tlock); 4817 ip->i_flag |= IACC; 4818 ITIMES_NOLOCK(ip); 4819 mutex_exit(&ip->i_tlock); 4820 } 4821 } 4822 } 4823 4824 if (dolock) { 4825 rw_exit(&ip->i_contents); 4826 if (do_qlock && rwtype == RW_WRITER) 4827 rw_exit(&ufsvfsp->vfs_dqrwlock); 4828 } 4829 4830 unlock: 4831 if (ulp) { 4832 if ((rw == S_CREATE || rw == S_WRITE) && 4833 !(vp->v_flag & VISSWAP)) { 4834 TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size); 4835 } 4836 ufs_lockfs_end(ulp); 4837 } 4838 out: 4839 return (err); 4840 } 4841 4842 /* 4843 * ufs_getpage_miss is called when ufs_getpage missed the page in the page 4844 * cache. The page is either read from the disk, or it's created. 4845 * A page is created (without disk read) if rw == S_CREATE, or if 4846 * the page is not backed with a real disk block (UFS hole). 4847 */ 4848 /* ARGSUSED */ 4849 static int 4850 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg, 4851 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq) 4852 { 4853 struct inode *ip = VTOI(vp); 4854 page_t *pp; 4855 daddr_t bn; 4856 size_t io_len; 4857 int crpage = 0; 4858 int err; 4859 int contig; 4860 int bsize = ip->i_fs->fs_bsize; 4861 4862 /* 4863 * Figure out whether the page can be created, or must be 4864 * must be read from the disk. 4865 */ 4866 if (rw == S_CREATE) 4867 crpage = 1; 4868 else { 4869 contig = 0; 4870 if (err = bmap_read(ip, off, &bn, &contig)) 4871 return (err); 4872 4873 crpage = (bn == UFS_HOLE); 4874 4875 /* 4876 * If its also a fallocated block that hasn't been written to 4877 * yet, we will treat it just like a UFS_HOLE and create 4878 * a zero page for it 4879 */ 4880 if (ISFALLOCBLK(ip, bn)) 4881 crpage = 1; 4882 } 4883 4884 if (crpage) { 4885 if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg, 4886 addr)) == NULL) { 4887 return (ufs_fault(vp, 4888 "ufs_getpage_miss: page_create == NULL")); 4889 } 4890 4891 if (rw != S_CREATE) 4892 pagezero(pp, 0, PAGESIZE); 4893 4894 io_len = PAGESIZE; 4895 } else { 4896 u_offset_t io_off; 4897 uint_t xlen; 4898 struct buf *bp; 4899 ufsvfs_t *ufsvfsp = ip->i_ufsvfs; 4900 4901 /* 4902 * If access is not in sequential order, we read from disk 4903 * in bsize units. 4904 * 4905 * We limit the size of the transfer to bsize if we are reading 4906 * from the beginning of the file. Note in this situation we 4907 * will hedge our bets and initiate an async read ahead of 4908 * the second block. 4909 */ 4910 if (!seq || off == 0) 4911 contig = MIN(contig, bsize); 4912 4913 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4914 &io_len, off, contig, 0); 4915 4916 /* 4917 * Some other thread has entered the page. 4918 * ufs_getpage will retry page_lookup. 4919 */ 4920 if (pp == NULL) { 4921 pl[0] = NULL; 4922 return (0); 4923 } 4924 4925 /* 4926 * Zero part of the page which we are not 4927 * going to read from the disk. 4928 */ 4929 xlen = io_len & PAGEOFFSET; 4930 if (xlen != 0) 4931 pagezero(pp->p_prev, xlen, PAGESIZE - xlen); 4932 4933 bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ); 4934 bp->b_edev = ip->i_dev; 4935 bp->b_dev = cmpdev(ip->i_dev); 4936 bp->b_blkno = bn; 4937 bp->b_un.b_addr = (caddr_t)0; 4938 bp->b_file = ip->i_vnode; 4939 bp->b_offset = off; 4940 4941 if (ufsvfsp->vfs_log) { 4942 lufs_read_strategy(ufsvfsp->vfs_log, bp); 4943 } else if (ufsvfsp->vfs_snapshot) { 4944 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp); 4945 } else { 4946 ufsvfsp->vfs_iotstamp = lbolt; 4947 ub.ub_getpages.value.ul++; 4948 (void) bdev_strategy(bp); 4949 lwp_stat_update(LWP_STAT_INBLK, 1); 4950 } 4951 4952 ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK); 4953 4954 /* 4955 * If the file access is sequential, initiate read ahead 4956 * of the next cluster. 4957 */ 4958 if (seq && ip->i_nextrio < ip->i_size) 4959 (void) ufs_getpage_ra(vp, off, seg, addr); 4960 err = biowait(bp); 4961 pageio_done(bp); 4962 4963 if (err) { 4964 pvn_read_done(pp, B_ERROR); 4965 return (err); 4966 } 4967 } 4968 4969 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4970 return (0); 4971 } 4972 4973 /* 4974 * Read ahead a cluster from the disk. Returns the length in bytes. 4975 */ 4976 static int 4977 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr) 4978 { 4979 struct inode *ip = VTOI(vp); 4980 page_t *pp; 4981 u_offset_t io_off = ip->i_nextrio; 4982 ufsvfs_t *ufsvfsp; 4983 caddr_t addr2 = addr + (io_off - off); 4984 struct buf *bp; 4985 daddr_t bn; 4986 size_t io_len; 4987 int err; 4988 int contig; 4989 int xlen; 4990 int bsize = ip->i_fs->fs_bsize; 4991 4992 /* 4993 * If the directio advisory is in effect on this file, 4994 * then do not do buffered read ahead. Read ahead makes 4995 * it more difficult on threads using directio as they 4996 * will be forced to flush the pages from this vnode. 4997 */ 4998 if ((ufsvfsp = ip->i_ufsvfs) == NULL) 4999 return (0); 5000 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) 5001 return (0); 5002 5003 /* 5004 * Is this test needed? 5005 */ 5006 if (addr2 >= seg->s_base + seg->s_size) 5007 return (0); 5008 5009 contig = 0; 5010 err = bmap_read(ip, io_off, &bn, &contig); 5011 /* 5012 * If its a UFS_HOLE or a fallocated block, do not perform 5013 * any read ahead's since there probably is nothing to read ahead 5014 */ 5015 if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn)) 5016 return (0); 5017 5018 /* 5019 * Limit the transfer size to bsize if this is the 2nd block. 5020 */ 5021 if (io_off == (u_offset_t)bsize) 5022 contig = MIN(contig, bsize); 5023 5024 if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off, 5025 &io_len, io_off, contig, 1)) == NULL) 5026 return (0); 5027 5028 /* 5029 * Zero part of page which we are not going to read from disk 5030 */ 5031 if ((xlen = (io_len & PAGEOFFSET)) > 0) 5032 pagezero(pp->p_prev, xlen, PAGESIZE - xlen); 5033 5034 ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK; 5035 5036 bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC); 5037 bp->b_edev = ip->i_dev; 5038 bp->b_dev = cmpdev(ip->i_dev); 5039 bp->b_blkno = bn; 5040 bp->b_un.b_addr = (caddr_t)0; 5041 bp->b_file = ip->i_vnode; 5042 bp->b_offset = off; 5043 5044 if (ufsvfsp->vfs_log) { 5045 lufs_read_strategy(ufsvfsp->vfs_log, bp); 5046 } else if (ufsvfsp->vfs_snapshot) { 5047 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp); 5048 } else { 5049 ufsvfsp->vfs_iotstamp = lbolt; 5050 ub.ub_getras.value.ul++; 5051 (void) bdev_strategy(bp); 5052 lwp_stat_update(LWP_STAT_INBLK, 1); 5053 } 5054 5055 return (io_len); 5056 } 5057 5058 int ufs_delay = 1; 5059 /* 5060 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC} 5061 * 5062 * LMXXX - the inode really ought to contain a pointer to one of these 5063 * async args. Stuff gunk in there and just hand the whole mess off. 5064 * This would replace i_delaylen, i_delayoff. 5065 */ 5066 /*ARGSUSED*/ 5067 static int 5068 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags, 5069 struct cred *cr, caller_context_t *ct) 5070 { 5071 struct inode *ip = VTOI(vp); 5072 int err = 0; 5073 5074 if (vp->v_count == 0) { 5075 return (ufs_fault(vp, "ufs_putpage: bad v_count == 0")); 5076 } 5077 5078 /* 5079 * XXX - Why should this check be made here? 5080 */ 5081 if (vp->v_flag & VNOMAP) { 5082 err = ENOSYS; 5083 goto errout; 5084 } 5085 5086 if (ip->i_ufsvfs == NULL) { 5087 err = EIO; 5088 goto errout; 5089 } 5090 5091 if (flags & B_ASYNC) { 5092 if (ufs_delay && len && 5093 (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) { 5094 mutex_enter(&ip->i_tlock); 5095 /* 5096 * If nobody stalled, start a new cluster. 5097 */ 5098 if (ip->i_delaylen == 0) { 5099 ip->i_delayoff = off; 5100 ip->i_delaylen = len; 5101 mutex_exit(&ip->i_tlock); 5102 goto errout; 5103 } 5104 /* 5105 * If we have a full cluster or they are not contig, 5106 * then push last cluster and start over. 5107 */ 5108 if (ip->i_delaylen >= CLUSTSZ(ip) || 5109 ip->i_delayoff + ip->i_delaylen != off) { 5110 u_offset_t doff; 5111 size_t dlen; 5112 5113 doff = ip->i_delayoff; 5114 dlen = ip->i_delaylen; 5115 ip->i_delayoff = off; 5116 ip->i_delaylen = len; 5117 mutex_exit(&ip->i_tlock); 5118 err = ufs_putpages(vp, doff, dlen, 5119 flags, cr); 5120 /* LMXXX - flags are new val, not old */ 5121 goto errout; 5122 } 5123 /* 5124 * There is something there, it's not full, and 5125 * it is contig. 5126 */ 5127 ip->i_delaylen += len; 5128 mutex_exit(&ip->i_tlock); 5129 goto errout; 5130 } 5131 /* 5132 * Must have weird flags or we are not clustering. 5133 */ 5134 } 5135 5136 err = ufs_putpages(vp, off, len, flags, cr); 5137 5138 errout: 5139 return (err); 5140 } 5141 5142 /* 5143 * If len == 0, do from off to EOF. 5144 * 5145 * The normal cases should be len == 0 & off == 0 (entire vp list), 5146 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE 5147 * (from pageout). 5148 */ 5149 /*ARGSUSED*/ 5150 static int 5151 ufs_putpages( 5152 struct vnode *vp, 5153 offset_t off, 5154 size_t len, 5155 int flags, 5156 struct cred *cr) 5157 { 5158 u_offset_t io_off; 5159 u_offset_t eoff; 5160 struct inode *ip = VTOI(vp); 5161 page_t *pp; 5162 size_t io_len; 5163 int err = 0; 5164 int dolock; 5165 5166 if (vp->v_count == 0) 5167 return (ufs_fault(vp, "ufs_putpages: v_count == 0")); 5168 /* 5169 * Acquire the readers/write inode lock before locking 5170 * any pages in this inode. 5171 * The inode lock is held during i/o. 5172 */ 5173 if (len == 0) { 5174 mutex_enter(&ip->i_tlock); 5175 ip->i_delayoff = ip->i_delaylen = 0; 5176 mutex_exit(&ip->i_tlock); 5177 } 5178 dolock = (rw_owner(&ip->i_contents) != curthread); 5179 if (dolock) { 5180 /* 5181 * Must synchronize this thread and any possible thread 5182 * operating in the window of vulnerability in wrip(). 5183 * It is dangerous to allow both a thread doing a putpage 5184 * and a thread writing, so serialize them. The exception 5185 * is when the thread in wrip() does something which causes 5186 * a putpage operation. Then, the thread must be allowed 5187 * to continue. It may encounter a bmap_read problem in 5188 * ufs_putapage, but that is handled in ufs_putapage. 5189 * Allow async writers to proceed, we don't want to block 5190 * the pageout daemon. 5191 */ 5192 if (ip->i_writer == curthread) 5193 rw_enter(&ip->i_contents, RW_READER); 5194 else { 5195 for (;;) { 5196 rw_enter(&ip->i_contents, RW_READER); 5197 mutex_enter(&ip->i_tlock); 5198 /* 5199 * If there is no thread in the critical 5200 * section of wrip(), then proceed. 5201 * Otherwise, wait until there isn't one. 5202 */ 5203 if (ip->i_writer == NULL) { 5204 mutex_exit(&ip->i_tlock); 5205 break; 5206 } 5207 rw_exit(&ip->i_contents); 5208 /* 5209 * Bounce async writers when we have a writer 5210 * working on this file so we don't deadlock 5211 * the pageout daemon. 5212 */ 5213 if (flags & B_ASYNC) { 5214 mutex_exit(&ip->i_tlock); 5215 return (0); 5216 } 5217 cv_wait(&ip->i_wrcv, &ip->i_tlock); 5218 mutex_exit(&ip->i_tlock); 5219 } 5220 } 5221 } 5222 5223 if (!vn_has_cached_data(vp)) { 5224 if (dolock) 5225 rw_exit(&ip->i_contents); 5226 return (0); 5227 } 5228 5229 if (len == 0) { 5230 /* 5231 * Search the entire vp list for pages >= off. 5232 */ 5233 err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage, 5234 flags, cr); 5235 } else { 5236 /* 5237 * Loop over all offsets in the range looking for 5238 * pages to deal with. 5239 */ 5240 if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0) 5241 eoff = MIN(off + len, eoff); 5242 else 5243 eoff = off + len; 5244 5245 for (io_off = off; io_off < eoff; io_off += io_len) { 5246 /* 5247 * If we are not invalidating, synchronously 5248 * freeing or writing pages, use the routine 5249 * page_lookup_nowait() to prevent reclaiming 5250 * them from the free list. 5251 */ 5252 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 5253 pp = page_lookup(vp, io_off, 5254 (flags & (B_INVAL | B_FREE)) ? 5255 SE_EXCL : SE_SHARED); 5256 } else { 5257 pp = page_lookup_nowait(vp, io_off, 5258 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 5259 } 5260 5261 if (pp == NULL || pvn_getdirty(pp, flags) == 0) 5262 io_len = PAGESIZE; 5263 else { 5264 u_offset_t *io_offp = &io_off; 5265 5266 err = ufs_putapage(vp, pp, io_offp, &io_len, 5267 flags, cr); 5268 if (err != 0) 5269 break; 5270 /* 5271 * "io_off" and "io_len" are returned as 5272 * the range of pages we actually wrote. 5273 * This allows us to skip ahead more quickly 5274 * since several pages may've been dealt 5275 * with by this iteration of the loop. 5276 */ 5277 } 5278 } 5279 } 5280 if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) { 5281 /* 5282 * We have just sync'ed back all the pages on 5283 * the inode, turn off the IMODTIME flag. 5284 */ 5285 mutex_enter(&ip->i_tlock); 5286 ip->i_flag &= ~IMODTIME; 5287 mutex_exit(&ip->i_tlock); 5288 } 5289 if (dolock) 5290 rw_exit(&ip->i_contents); 5291 return (err); 5292 } 5293 5294 static void 5295 ufs_iodone(buf_t *bp) 5296 { 5297 struct inode *ip; 5298 5299 ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ)); 5300 5301 bp->b_iodone = NULL; 5302 5303 ip = VTOI(bp->b_pages->p_vnode); 5304 5305 mutex_enter(&ip->i_tlock); 5306 if (ip->i_writes >= ufs_LW) { 5307 if ((ip->i_writes -= bp->b_bcount) <= ufs_LW) 5308 if (ufs_WRITES) 5309 cv_broadcast(&ip->i_wrcv); /* wake all up */ 5310 } else { 5311 ip->i_writes -= bp->b_bcount; 5312 } 5313 5314 mutex_exit(&ip->i_tlock); 5315 iodone(bp); 5316 } 5317 5318 /* 5319 * Write out a single page, possibly klustering adjacent 5320 * dirty pages. The inode lock must be held. 5321 * 5322 * LMXXX - bsize < pagesize not done. 5323 */ 5324 /*ARGSUSED*/ 5325 int 5326 ufs_putapage( 5327 struct vnode *vp, 5328 page_t *pp, 5329 u_offset_t *offp, 5330 size_t *lenp, /* return values */ 5331 int flags, 5332 struct cred *cr) 5333 { 5334 u_offset_t io_off; 5335 u_offset_t off; 5336 struct inode *ip = VTOI(vp); 5337 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 5338 struct fs *fs; 5339 struct buf *bp; 5340 size_t io_len; 5341 daddr_t bn; 5342 int err; 5343 int contig; 5344 int dotrans; 5345 5346 ASSERT(RW_LOCK_HELD(&ip->i_contents)); 5347 5348 if (ufsvfsp == NULL) { 5349 err = EIO; 5350 goto out_trace; 5351 } 5352 5353 fs = ip->i_fs; 5354 ASSERT(fs->fs_ronly == 0); 5355 5356 /* 5357 * If the modified time on the inode has not already been 5358 * set elsewhere (e.g. for write/setattr) we set the time now. 5359 * This gives us approximate modified times for mmap'ed files 5360 * which are modified via stores in the user address space. 5361 */ 5362 if ((ip->i_flag & IMODTIME) == 0) { 5363 mutex_enter(&ip->i_tlock); 5364 ip->i_flag |= IUPD; 5365 ip->i_seq++; 5366 ITIMES_NOLOCK(ip); 5367 mutex_exit(&ip->i_tlock); 5368 } 5369 5370 /* 5371 * Align the request to a block boundry (for old file systems), 5372 * and go ask bmap() how contiguous things are for this file. 5373 */ 5374 off = pp->p_offset & (offset_t)fs->fs_bmask; /* block align it */ 5375 contig = 0; 5376 err = bmap_read(ip, off, &bn, &contig); 5377 if (err) 5378 goto out; 5379 if (bn == UFS_HOLE) { /* putpage never allocates */ 5380 /* 5381 * logging device is in error mode; simply return EIO 5382 */ 5383 if (TRANS_ISERROR(ufsvfsp)) { 5384 err = EIO; 5385 goto out; 5386 } 5387 /* 5388 * Oops, the thread in the window in wrip() did some 5389 * sort of operation which caused a putpage in the bad 5390 * range. In this case, just return an error which will 5391 * cause the software modified bit on the page to set 5392 * and the page will get written out again later. 5393 */ 5394 if (ip->i_writer == curthread) { 5395 err = EIO; 5396 goto out; 5397 } 5398 /* 5399 * If the pager is trying to push a page in the bad range 5400 * just tell him to try again later when things are better. 5401 */ 5402 if (flags & B_ASYNC) { 5403 err = EAGAIN; 5404 goto out; 5405 } 5406 err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE"); 5407 goto out; 5408 } 5409 5410 /* 5411 * If it is an fallocate'd block, reverse the negativity since 5412 * we are now writing to it 5413 */ 5414 if (ISFALLOCBLK(ip, bn)) { 5415 err = bmap_set_bn(vp, off, dbtofsb(fs, -bn)); 5416 if (err) 5417 goto out; 5418 5419 bn = -bn; 5420 } 5421 5422 /* 5423 * Take the length (of contiguous bytes) passed back from bmap() 5424 * and _try_ and get a set of pages covering that extent. 5425 */ 5426 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags); 5427 5428 /* 5429 * May have run out of memory and not clustered backwards. 5430 * off p_offset 5431 * [ pp - 1 ][ pp ] 5432 * [ block ] 5433 * We told bmap off, so we have to adjust the bn accordingly. 5434 */ 5435 if (io_off > off) { 5436 bn += btod(io_off - off); 5437 contig -= (io_off - off); 5438 } 5439 5440 /* 5441 * bmap was carefull to tell us the right size so use that. 5442 * There might be unallocated frags at the end. 5443 * LMXXX - bzero the end of the page? We must be writing after EOF. 5444 */ 5445 if (io_len > contig) { 5446 ASSERT(io_len - contig < fs->fs_bsize); 5447 io_len -= (io_len - contig); 5448 } 5449 5450 /* 5451 * Handle the case where we are writing the last page after EOF. 5452 * 5453 * XXX - just a patch for i-mt3. 5454 */ 5455 if (io_len == 0) { 5456 ASSERT(pp->p_offset >= 5457 (u_offset_t)(roundup(ip->i_size, PAGESIZE))); 5458 io_len = PAGESIZE; 5459 } 5460 5461 bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags); 5462 5463 ULOCKFS_SET_MOD(ITOUL(ip)); 5464 5465 bp->b_edev = ip->i_dev; 5466 bp->b_dev = cmpdev(ip->i_dev); 5467 bp->b_blkno = bn; 5468 bp->b_un.b_addr = (caddr_t)0; 5469 bp->b_file = ip->i_vnode; 5470 5471 /* 5472 * File contents of shadow or quota inodes are metadata, and updates 5473 * to these need to be put into a logging transaction. All direct 5474 * callers in UFS do that, but fsflush can come here _before_ the 5475 * normal codepath. An example would be updating ACL information, for 5476 * which the normal codepath would be: 5477 * ufs_si_store() 5478 * ufs_rdwri() 5479 * wrip() 5480 * segmap_release() 5481 * VOP_PUTPAGE() 5482 * Here, fsflush can pick up the dirty page before segmap_release() 5483 * forces it out. If that happens, there's no transaction. 5484 * We therefore need to test whether a transaction exists, and if not 5485 * create one - for fsflush. 5486 */ 5487 dotrans = 5488 (((ip->i_mode & IFMT) == IFSHAD || ufsvfsp->vfs_qinod == ip) && 5489 ((curthread->t_flag & T_DONTBLOCK) == 0) && 5490 (TRANS_ISTRANS(ufsvfsp))); 5491 5492 if (dotrans) { 5493 curthread->t_flag |= T_DONTBLOCK; 5494 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip)); 5495 } 5496 if (TRANS_ISTRANS(ufsvfsp)) { 5497 if ((ip->i_mode & IFMT) == IFSHAD) { 5498 TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD); 5499 } else if (ufsvfsp->vfs_qinod == ip) { 5500 TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR, 5501 0, 0); 5502 } 5503 } 5504 if (dotrans) { 5505 TRANS_END_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip)); 5506 curthread->t_flag &= ~T_DONTBLOCK; 5507 } 5508 5509 /* write throttle */ 5510 5511 ASSERT(bp->b_iodone == NULL); 5512 bp->b_iodone = (int (*)())ufs_iodone; 5513 mutex_enter(&ip->i_tlock); 5514 ip->i_writes += bp->b_bcount; 5515 mutex_exit(&ip->i_tlock); 5516 5517 if (bp->b_flags & B_ASYNC) { 5518 if (ufsvfsp->vfs_log) { 5519 lufs_write_strategy(ufsvfsp->vfs_log, bp); 5520 } else if (ufsvfsp->vfs_snapshot) { 5521 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp); 5522 } else { 5523 ufsvfsp->vfs_iotstamp = lbolt; 5524 ub.ub_putasyncs.value.ul++; 5525 (void) bdev_strategy(bp); 5526 lwp_stat_update(LWP_STAT_OUBLK, 1); 5527 } 5528 } else { 5529 if (ufsvfsp->vfs_log) { 5530 lufs_write_strategy(ufsvfsp->vfs_log, bp); 5531 } else if (ufsvfsp->vfs_snapshot) { 5532 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp); 5533 } else { 5534 ufsvfsp->vfs_iotstamp = lbolt; 5535 ub.ub_putsyncs.value.ul++; 5536 (void) bdev_strategy(bp); 5537 lwp_stat_update(LWP_STAT_OUBLK, 1); 5538 } 5539 err = biowait(bp); 5540 pageio_done(bp); 5541 pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags); 5542 } 5543 5544 pp = NULL; 5545 5546 out: 5547 if (err != 0 && pp != NULL) 5548 pvn_write_done(pp, B_ERROR | B_WRITE | flags); 5549 5550 if (offp) 5551 *offp = io_off; 5552 if (lenp) 5553 *lenp = io_len; 5554 out_trace: 5555 return (err); 5556 } 5557 5558 uint64_t ufs_map_alock_retry_cnt; 5559 uint64_t ufs_map_lockfs_retry_cnt; 5560 5561 /* ARGSUSED */ 5562 static int 5563 ufs_map(struct vnode *vp, 5564 offset_t off, 5565 struct as *as, 5566 caddr_t *addrp, 5567 size_t len, 5568 uchar_t prot, 5569 uchar_t maxprot, 5570 uint_t flags, 5571 struct cred *cr, 5572 caller_context_t *ct) 5573 { 5574 struct segvn_crargs vn_a; 5575 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs; 5576 struct ulockfs *ulp; 5577 int error, sig; 5578 k_sigset_t smask; 5579 caddr_t hint = *addrp; 5580 5581 if (vp->v_flag & VNOMAP) { 5582 error = ENOSYS; 5583 goto out; 5584 } 5585 5586 if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) { 5587 error = ENXIO; 5588 goto out; 5589 } 5590 5591 if (vp->v_type != VREG) { 5592 error = ENODEV; 5593 goto out; 5594 } 5595 5596 retry_map: 5597 *addrp = hint; 5598 /* 5599 * If file is being locked, disallow mapping. 5600 */ 5601 if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) { 5602 error = EAGAIN; 5603 goto out; 5604 } 5605 5606 as_rangelock(as); 5607 /* 5608 * Note that if we are retrying (because ufs_lockfs_trybegin failed in 5609 * the previous attempt), some other thread could have grabbed 5610 * the same VA range if MAP_FIXED is set. In that case, choose_addr 5611 * would unmap the valid VA range, that is ok. 5612 */ 5613 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 5614 if (error != 0) { 5615 as_rangeunlock(as); 5616 goto out; 5617 } 5618 5619 /* 5620 * a_lock has to be acquired before entering the lockfs protocol 5621 * because that is the order in which pagefault works. Also we cannot 5622 * block on a_lock here because this waiting writer will prevent 5623 * further readers like ufs_read from progressing and could cause 5624 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is 5625 * pending. 5626 */ 5627 while (!AS_LOCK_TRYENTER(as, &as->a_lock, RW_WRITER)) { 5628 ufs_map_alock_retry_cnt++; 5629 delay(RETRY_LOCK_DELAY); 5630 } 5631 5632 /* 5633 * We can't hold as->a_lock and wait for lockfs to succeed because 5634 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin() 5635 * instead. 5636 */ 5637 if (error = ufs_lockfs_trybegin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK)) { 5638 /* 5639 * ufs_lockfs_trybegin() did not succeed. It is safer to give up 5640 * as->a_lock and wait for ulp->ul_fs_lock status to change. 5641 */ 5642 ufs_map_lockfs_retry_cnt++; 5643 AS_LOCK_EXIT(as, &as->a_lock); 5644 as_rangeunlock(as); 5645 if (error == EIO) 5646 goto out; 5647 5648 mutex_enter(&ulp->ul_lock); 5649 while (ulp->ul_fs_lock & ULOCKFS_MAP_MASK) { 5650 if (ULOCKFS_IS_SLOCK(ulp) || ufsvfsp->vfs_nointr) { 5651 cv_wait(&ulp->ul_cv, &ulp->ul_lock); 5652 } else { 5653 sigintr(&smask, 1); 5654 sig = cv_wait_sig(&ulp->ul_cv, &ulp->ul_lock); 5655 sigunintr(&smask); 5656 if (((ulp->ul_fs_lock & ULOCKFS_MAP_MASK) && 5657 !sig) || ufsvfsp->vfs_dontblock) { 5658 mutex_exit(&ulp->ul_lock); 5659 return (EINTR); 5660 } 5661 } 5662 } 5663 mutex_exit(&ulp->ul_lock); 5664 goto retry_map; 5665 } 5666 5667 vn_a.vp = vp; 5668 vn_a.offset = (u_offset_t)off; 5669 vn_a.type = flags & MAP_TYPE; 5670 vn_a.prot = prot; 5671 vn_a.maxprot = maxprot; 5672 vn_a.cred = cr; 5673 vn_a.amp = NULL; 5674 vn_a.flags = flags & ~MAP_TYPE; 5675 vn_a.szc = 0; 5676 vn_a.lgrp_mem_policy_flags = 0; 5677 5678 error = as_map_locked(as, *addrp, len, segvn_create, &vn_a); 5679 if (ulp) 5680 ufs_lockfs_end(ulp); 5681 as_rangeunlock(as); 5682 out: 5683 return (error); 5684 } 5685 5686 /* ARGSUSED */ 5687 static int 5688 ufs_addmap(struct vnode *vp, 5689 offset_t off, 5690 struct as *as, 5691 caddr_t addr, 5692 size_t len, 5693 uchar_t prot, 5694 uchar_t maxprot, 5695 uint_t flags, 5696 struct cred *cr, 5697 caller_context_t *ct) 5698 { 5699 struct inode *ip = VTOI(vp); 5700 5701 if (vp->v_flag & VNOMAP) { 5702 return (ENOSYS); 5703 } 5704 5705 mutex_enter(&ip->i_tlock); 5706 ip->i_mapcnt += btopr(len); 5707 mutex_exit(&ip->i_tlock); 5708 return (0); 5709 } 5710 5711 /*ARGSUSED*/ 5712 static int 5713 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr, 5714 size_t len, uint_t prot, uint_t maxprot, uint_t flags, 5715 struct cred *cr, caller_context_t *ct) 5716 { 5717 struct inode *ip = VTOI(vp); 5718 5719 if (vp->v_flag & VNOMAP) { 5720 return (ENOSYS); 5721 } 5722 5723 mutex_enter(&ip->i_tlock); 5724 ip->i_mapcnt -= btopr(len); /* Count released mappings */ 5725 ASSERT(ip->i_mapcnt >= 0); 5726 mutex_exit(&ip->i_tlock); 5727 return (0); 5728 } 5729 /* 5730 * Return the answer requested to poll() for non-device files 5731 */ 5732 struct pollhead ufs_pollhd; 5733 5734 /* ARGSUSED */ 5735 int 5736 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp, 5737 caller_context_t *ct) 5738 { 5739 struct ufsvfs *ufsvfsp; 5740 5741 *revp = 0; 5742 ufsvfsp = VTOI(vp)->i_ufsvfs; 5743 5744 if (!ufsvfsp) { 5745 *revp = POLLHUP; 5746 goto out; 5747 } 5748 5749 if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) || 5750 ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) { 5751 *revp |= POLLERR; 5752 5753 } else { 5754 if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly && 5755 !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs)) 5756 *revp |= POLLOUT; 5757 5758 if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly && 5759 !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs)) 5760 *revp |= POLLWRBAND; 5761 5762 if (ev & POLLIN) 5763 *revp |= POLLIN; 5764 5765 if (ev & POLLRDNORM) 5766 *revp |= POLLRDNORM; 5767 5768 if (ev & POLLRDBAND) 5769 *revp |= POLLRDBAND; 5770 } 5771 5772 if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP))) 5773 *revp |= POLLPRI; 5774 out: 5775 *phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL; 5776 5777 return (0); 5778 } 5779 5780 /* ARGSUSED */ 5781 static int 5782 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr, 5783 caller_context_t *ct) 5784 { 5785 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs; 5786 struct ulockfs *ulp = NULL; 5787 struct inode *sip = NULL; 5788 int error; 5789 struct inode *ip = VTOI(vp); 5790 int issync; 5791 5792 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK); 5793 if (error) 5794 return (error); 5795 5796 switch (cmd) { 5797 /* 5798 * Have to handle _PC_NAME_MAX here, because the normal way 5799 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()] 5800 * results in a lock ordering reversal between 5801 * ufs_lockfs_{begin,end}() and 5802 * ufs_thread_{suspend,continue}(). 5803 * 5804 * Keep in sync with ufs_statvfs(). 5805 */ 5806 case _PC_NAME_MAX: 5807 *valp = MAXNAMLEN; 5808 break; 5809 5810 case _PC_FILESIZEBITS: 5811 if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES) 5812 *valp = UFS_FILESIZE_BITS; 5813 else 5814 *valp = 32; 5815 break; 5816 5817 case _PC_XATTR_EXISTS: 5818 if (vp->v_vfsp->vfs_flag & VFS_XATTR) { 5819 5820 error = 5821 ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR, cr); 5822 if (error == 0 && sip != NULL) { 5823 /* Start transaction */ 5824 if (ulp) { 5825 TRANS_BEGIN_CSYNC(ufsvfsp, issync, 5826 TOP_RMDIR, TOP_RMDIR_SIZE); 5827 } 5828 /* 5829 * Is directory empty 5830 */ 5831 rw_enter(&sip->i_rwlock, RW_WRITER); 5832 rw_enter(&sip->i_contents, RW_WRITER); 5833 if (ufs_xattrdirempty(sip, 5834 sip->i_number, CRED())) { 5835 rw_enter(&ip->i_contents, RW_WRITER); 5836 ufs_unhook_shadow(ip, sip); 5837 rw_exit(&ip->i_contents); 5838 5839 *valp = 0; 5840 5841 } else 5842 *valp = 1; 5843 rw_exit(&sip->i_contents); 5844 rw_exit(&sip->i_rwlock); 5845 if (ulp) { 5846 TRANS_END_CSYNC(ufsvfsp, error, issync, 5847 TOP_RMDIR, TOP_RMDIR_SIZE); 5848 } 5849 VN_RELE(ITOV(sip)); 5850 } else if (error == ENOENT) { 5851 *valp = 0; 5852 error = 0; 5853 } 5854 } else { 5855 error = fs_pathconf(vp, cmd, valp, cr, ct); 5856 } 5857 break; 5858 5859 case _PC_ACL_ENABLED: 5860 *valp = _ACL_ACLENT_ENABLED; 5861 break; 5862 5863 case _PC_MIN_HOLE_SIZE: 5864 *valp = (ulong_t)ip->i_fs->fs_bsize; 5865 break; 5866 5867 case _PC_SATTR_ENABLED: 5868 case _PC_SATTR_EXISTS: 5869 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) && 5870 (vp->v_type == VREG || vp->v_type == VDIR); 5871 break; 5872 5873 default: 5874 error = fs_pathconf(vp, cmd, valp, cr, ct); 5875 } 5876 5877 if (ulp != NULL) { 5878 ufs_lockfs_end(ulp); 5879 } 5880 return (error); 5881 } 5882 5883 int ufs_pageio_writes, ufs_pageio_reads; 5884 5885 /*ARGSUSED*/ 5886 static int 5887 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len, 5888 int flags, struct cred *cr, caller_context_t *ct) 5889 { 5890 struct inode *ip = VTOI(vp); 5891 struct ufsvfs *ufsvfsp; 5892 page_t *npp = NULL, *opp = NULL, *cpp = pp; 5893 struct buf *bp; 5894 daddr_t bn; 5895 size_t done_len = 0, cur_len = 0; 5896 int err = 0; 5897 int contig = 0; 5898 int dolock; 5899 int vmpss = 0; 5900 struct ulockfs *ulp; 5901 5902 if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp && 5903 vp->v_mpssdata != NULL) { 5904 vmpss = 1; 5905 } 5906 5907 dolock = (rw_owner(&ip->i_contents) != curthread); 5908 /* 5909 * We need a better check. Ideally, we would use another 5910 * vnodeops so that hlocked and forcibly unmounted file 5911 * systems would return EIO where appropriate and w/o the 5912 * need for these checks. 5913 */ 5914 if ((ufsvfsp = ip->i_ufsvfs) == NULL) 5915 return (EIO); 5916 5917 /* 5918 * For vmpss (pp can be NULL) case respect the quiesce protocol. 5919 * ul_lock must be taken before locking pages so we can't use it here 5920 * if pp is non NULL because segvn already locked pages 5921 * SE_EXCL. Instead we rely on the fact that a forced umount or 5922 * applying a filesystem lock via ufs_fiolfs() will block in the 5923 * implicit call to ufs_flush() until we unlock the pages after the 5924 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend 5925 * above 0 until they are done. We have to be careful not to increment 5926 * ul_vnops_cnt here after forceful unmount hlocks the file system. 5927 * 5928 * If pp is NULL use ul_lock to make sure we don't increment 5929 * ul_vnops_cnt after forceful unmount hlocks the file system. 5930 */ 5931 if (vmpss || pp == NULL) { 5932 ulp = &ufsvfsp->vfs_ulockfs; 5933 if (pp == NULL) 5934 mutex_enter(&ulp->ul_lock); 5935 if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) { 5936 if (pp == NULL) { 5937 mutex_exit(&ulp->ul_lock); 5938 } 5939 return (vmpss ? EIO : EINVAL); 5940 } 5941 atomic_add_long(&ulp->ul_vnops_cnt, 1); 5942 if (pp == NULL) 5943 mutex_exit(&ulp->ul_lock); 5944 if (ufs_quiesce_pend) { 5945 if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1)) 5946 cv_broadcast(&ulp->ul_cv); 5947 return (vmpss ? EIO : EINVAL); 5948 } 5949 } 5950 5951 if (dolock) { 5952 /* 5953 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to 5954 * handle a fault against a segment that maps vnode pages with 5955 * large mappings. Segvn creates pages and holds them locked 5956 * SE_EXCL during VOP_PAGEIO() call. In this case we have to 5957 * use rw_tryenter() to avoid a potential deadlock since in 5958 * lock order i_contents needs to be taken first. 5959 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails. 5960 */ 5961 if (!vmpss) { 5962 rw_enter(&ip->i_contents, RW_READER); 5963 } else if (!rw_tryenter(&ip->i_contents, RW_READER)) { 5964 if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1)) 5965 cv_broadcast(&ulp->ul_cv); 5966 return (EDEADLK); 5967 } 5968 } 5969 5970 /* 5971 * Return an error to segvn because the pagefault request is beyond 5972 * PAGESIZE rounded EOF. 5973 */ 5974 if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) { 5975 if (dolock) 5976 rw_exit(&ip->i_contents); 5977 if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1)) 5978 cv_broadcast(&ulp->ul_cv); 5979 return (EFAULT); 5980 } 5981 5982 if (pp == NULL) { 5983 if (bmap_has_holes(ip)) { 5984 err = ENOSYS; 5985 } else { 5986 err = EINVAL; 5987 } 5988 if (dolock) 5989 rw_exit(&ip->i_contents); 5990 if (!atomic_add_long_nv(&ulp->ul_vnops_cnt, -1)) 5991 cv_broadcast(&ulp->ul_cv); 5992 return (err); 5993 } 5994 5995 /* 5996 * Break the io request into chunks, one for each contiguous 5997 * stretch of disk blocks in the target file. 5998 */ 5999 while (done_len < io_len) { 6000 ASSERT(cpp); 6001 contig = 0; 6002 if (err = bmap_read(ip, (u_offset_t)(io_off + done_len), 6003 &bn, &contig)) 6004 break; 6005 6006 if (bn == UFS_HOLE) { /* No holey swapfiles */ 6007 if (vmpss) { 6008 err = EFAULT; 6009 break; 6010 } 6011 err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE"); 6012 break; 6013 } 6014 6015 cur_len = MIN(io_len - done_len, contig); 6016 /* 6017 * Zero out a page beyond EOF, when the last block of 6018 * a file is a UFS fragment so that ufs_pageio() can be used 6019 * instead of ufs_getpage() to handle faults against 6020 * segvn segments that use large pages. 6021 */ 6022 page_list_break(&cpp, &npp, btopr(cur_len)); 6023 if ((flags & B_READ) && (cur_len & PAGEOFFSET)) { 6024 size_t xlen = cur_len & PAGEOFFSET; 6025 pagezero(cpp->p_prev, xlen, PAGESIZE - xlen); 6026 } 6027 6028 bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags); 6029 ASSERT(bp != NULL); 6030 6031 bp->b_edev = ip->i_dev; 6032 bp->b_dev = cmpdev(ip->i_dev); 6033 bp->b_blkno = bn; 6034 bp->b_un.b_addr = (caddr_t)0; 6035 bp->b_file = ip->i_vnode; 6036 6037 ufsvfsp->vfs_iotstamp = lbolt; 6038 ub.ub_pageios.value.ul++; 6039 if (ufsvfsp->vfs_snapshot) 6040 fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp); 6041 else 6042 (void) bdev_strategy(bp); 6043 6044 if (flags & B_READ) 6045 ufs_pageio_reads++; 6046 else 6047 ufs_pageio_writes++; 6048 if (flags & B_READ) 6049 lwp_stat_update(LWP_STAT_INBLK, 1); 6050 else 6051 lwp_stat_update(LWP_STAT_OUBLK, 1); 6052 /* 6053 * If the request is not B_ASYNC, wait for i/o to complete 6054 * and re-assemble the page list to return to the caller. 6055 * If it is B_ASYNC we leave the page list in pieces and 6056 * cleanup() will dispose of them. 6057 */ 6058 if ((flags & B_ASYNC) == 0) { 6059 err = biowait(bp); 6060 pageio_done(bp); 6061 if (err) 6062 break; 6063 page_list_concat(&opp, &cpp); 6064 } 6065 cpp = npp; 6066 npp = NULL; 6067 if (flags & B_READ) 6068 cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t); 6069 done_len += cur_len; 6070 } 6071 ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len)); 6072 if (err) { 6073 if (flags & B_ASYNC) { 6074 /* Cleanup unprocessed parts of list */ 6075 page_list_concat(&cpp, &npp); 6076 if (flags & B_READ) 6077 pvn_read_done(cpp, B_ERROR); 6078 else 6079 pvn_write_done(cpp, B_ERROR); 6080 } else { 6081 /* Re-assemble list and let caller clean up */ 6082 page_list_concat(&opp, &cpp); 6083 page_list_concat(&opp, &npp); 6084 } 6085 } 6086 6087 if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) && 6088 ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) { 6089 mutex_enter(&ip->i_tlock); 6090 ip->i_flag |= IACC; 6091 ITIMES_NOLOCK(ip); 6092 mutex_exit(&ip->i_tlock); 6093 } 6094 6095 if (dolock) 6096 rw_exit(&ip->i_contents); 6097 if (vmpss && !atomic_add_long_nv(&ulp->ul_vnops_cnt, -1)) 6098 cv_broadcast(&ulp->ul_cv); 6099 return (err); 6100 } 6101 6102 /* 6103 * Called when the kernel is in a frozen state to dump data 6104 * directly to the device. It uses a private dump data structure, 6105 * set up by dump_ctl, to locate the correct disk block to which to dump. 6106 */ 6107 /*ARGSUSED*/ 6108 static int 6109 ufs_dump(vnode_t *vp, caddr_t addr, offset_t ldbn, offset_t dblks, 6110 caller_context_t *ct) 6111 { 6112 u_offset_t file_size; 6113 struct inode *ip = VTOI(vp); 6114 struct fs *fs = ip->i_fs; 6115 daddr_t dbn, lfsbn; 6116 int disk_blks = fs->fs_bsize >> DEV_BSHIFT; 6117 int error = 0; 6118 int ndbs, nfsbs; 6119 6120 /* 6121 * forced unmount case 6122 */ 6123 if (ip->i_ufsvfs == NULL) 6124 return (EIO); 6125 /* 6126 * Validate the inode that it has not been modified since 6127 * the dump structure is allocated. 6128 */ 6129 mutex_enter(&ip->i_tlock); 6130 if ((dump_info == NULL) || 6131 (dump_info->ip != ip) || 6132 (dump_info->time.tv_sec != ip->i_mtime.tv_sec) || 6133 (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) { 6134 mutex_exit(&ip->i_tlock); 6135 return (-1); 6136 } 6137 mutex_exit(&ip->i_tlock); 6138 6139 /* 6140 * See that the file has room for this write 6141 */ 6142 UFS_GET_ISIZE(&file_size, ip); 6143 6144 if (ldbtob(ldbn + dblks) > file_size) 6145 return (ENOSPC); 6146 6147 /* 6148 * Find the physical disk block numbers from the dump 6149 * private data structure directly and write out the data 6150 * in contiguous block lumps 6151 */ 6152 while (dblks > 0 && !error) { 6153 lfsbn = (daddr_t)lblkno(fs, ldbtob(ldbn)); 6154 dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks; 6155 nfsbs = 1; 6156 ndbs = disk_blks - ldbn % disk_blks; 6157 while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn + 6158 nfsbs]) == dbn + ndbs) { 6159 nfsbs++; 6160 ndbs += disk_blks; 6161 } 6162 if (ndbs > dblks) 6163 ndbs = dblks; 6164 error = bdev_dump(ip->i_dev, addr, dbn, ndbs); 6165 addr += ldbtob((offset_t)ndbs); 6166 dblks -= ndbs; 6167 ldbn += ndbs; 6168 } 6169 return (error); 6170 6171 } 6172 6173 /* 6174 * Prepare the file system before and after the dump operation. 6175 * 6176 * action = DUMP_ALLOC: 6177 * Preparation before dump, allocate dump private data structure 6178 * to hold all the direct and indirect block info for dump. 6179 * 6180 * action = DUMP_FREE: 6181 * Clean up after dump, deallocate the dump private data structure. 6182 * 6183 * action = DUMP_SCAN: 6184 * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space; 6185 * if found, the starting file-relative DEV_BSIZE lbn is written 6186 * to *bklp; that lbn is intended for use with VOP_DUMP() 6187 */ 6188 /*ARGSUSED*/ 6189 static int 6190 ufs_dumpctl(vnode_t *vp, int action, offset_t *blkp, caller_context_t *ct) 6191 { 6192 struct inode *ip = VTOI(vp); 6193 ufsvfs_t *ufsvfsp = ip->i_ufsvfs; 6194 struct fs *fs; 6195 daddr32_t *dblk, *storeblk; 6196 daddr32_t *nextblk, *endblk; 6197 struct buf *bp; 6198 int i, entry, entries; 6199 int n, ncontig; 6200 6201 /* 6202 * check for forced unmount 6203 */ 6204 if (ufsvfsp == NULL) 6205 return (EIO); 6206 6207 if (action == DUMP_ALLOC) { 6208 /* 6209 * alloc and record dump_info 6210 */ 6211 if (dump_info != NULL) 6212 return (EINVAL); 6213 6214 ASSERT(vp->v_type == VREG); 6215 fs = ufsvfsp->vfs_fs; 6216 6217 rw_enter(&ip->i_contents, RW_READER); 6218 6219 if (bmap_has_holes(ip)) { 6220 rw_exit(&ip->i_contents); 6221 return (EFAULT); 6222 } 6223 6224 /* 6225 * calculate and allocate space needed according to i_size 6226 */ 6227 entries = (int)lblkno(fs, blkroundup(fs, ip->i_size)); 6228 dump_info = kmem_alloc(sizeof (struct dump) + 6229 (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP); 6230 if (dump_info == NULL) { 6231 rw_exit(&ip->i_contents); 6232 return (ENOMEM); 6233 } 6234 6235 /* Start saving the info */ 6236 dump_info->fsbs = entries; 6237 dump_info->ip = ip; 6238 storeblk = &dump_info->dblk[0]; 6239 6240 /* Direct Blocks */ 6241 for (entry = 0; entry < NDADDR && entry < entries; entry++) 6242 *storeblk++ = ip->i_db[entry]; 6243 6244 /* Indirect Blocks */ 6245 for (i = 0; i < NIADDR; i++) { 6246 int error = 0; 6247 6248 bp = UFS_BREAD(ufsvfsp, 6249 ip->i_dev, fsbtodb(fs, ip->i_ib[i]), fs->fs_bsize); 6250 if (bp->b_flags & B_ERROR) 6251 error = EIO; 6252 else { 6253 dblk = bp->b_un.b_daddr; 6254 if ((storeblk = save_dblks(ip, ufsvfsp, 6255 storeblk, dblk, i, entries)) == NULL) 6256 error = EIO; 6257 } 6258 6259 brelse(bp); 6260 6261 if (error != 0) { 6262 kmem_free(dump_info, sizeof (struct dump) + 6263 (entries - 1) * sizeof (daddr32_t)); 6264 rw_exit(&ip->i_contents); 6265 dump_info = NULL; 6266 return (error); 6267 } 6268 } 6269 /* and time stamp the information */ 6270 mutex_enter(&ip->i_tlock); 6271 dump_info->time = ip->i_mtime; 6272 mutex_exit(&ip->i_tlock); 6273 6274 rw_exit(&ip->i_contents); 6275 } else if (action == DUMP_FREE) { 6276 /* 6277 * free dump_info 6278 */ 6279 if (dump_info == NULL) 6280 return (EINVAL); 6281 entries = dump_info->fsbs - 1; 6282 kmem_free(dump_info, sizeof (struct dump) + 6283 entries * sizeof (daddr32_t)); 6284 dump_info = NULL; 6285 } else if (action == DUMP_SCAN) { 6286 /* 6287 * scan dump_info 6288 */ 6289 if (dump_info == NULL) 6290 return (EINVAL); 6291 6292 dblk = dump_info->dblk; 6293 nextblk = dblk + 1; 6294 endblk = dblk + dump_info->fsbs - 1; 6295 fs = ufsvfsp->vfs_fs; 6296 ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT); 6297 6298 /* 6299 * scan dblk[] entries; contig fs space is found when: 6300 * ((current blkno + frags per block) == next blkno) 6301 */ 6302 n = 0; 6303 while (n < ncontig && dblk < endblk) { 6304 if ((*dblk + fs->fs_frag) == *nextblk) 6305 n++; 6306 else 6307 n = 0; 6308 dblk++; 6309 nextblk++; 6310 } 6311 6312 /* 6313 * index is where size bytes of contig space begins; 6314 * conversion from index to the file's DEV_BSIZE lbn 6315 * is equivalent to: (index * fs_bsize) / DEV_BSIZE 6316 */ 6317 if (n == ncontig) { 6318 i = (dblk - dump_info->dblk) - ncontig; 6319 *blkp = i << (fs->fs_bshift - DEV_BSHIFT); 6320 } else 6321 return (EFAULT); 6322 } 6323 return (0); 6324 } 6325 6326 /* 6327 * Recursive helper function for ufs_dumpctl(). It follows the indirect file 6328 * system blocks until it reaches the the disk block addresses, which are 6329 * then stored into the given buffer, storeblk. 6330 */ 6331 static daddr32_t * 6332 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp, daddr32_t *storeblk, 6333 daddr32_t *dblk, int level, int entries) 6334 { 6335 struct fs *fs = ufsvfsp->vfs_fs; 6336 struct buf *bp; 6337 int i; 6338 6339 if (level == 0) { 6340 for (i = 0; i < NINDIR(fs); i++) { 6341 if (storeblk - dump_info->dblk >= entries) 6342 break; 6343 *storeblk++ = dblk[i]; 6344 } 6345 return (storeblk); 6346 } 6347 for (i = 0; i < NINDIR(fs); i++) { 6348 if (storeblk - dump_info->dblk >= entries) 6349 break; 6350 bp = UFS_BREAD(ufsvfsp, 6351 ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize); 6352 if (bp->b_flags & B_ERROR) { 6353 brelse(bp); 6354 return (NULL); 6355 } 6356 storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr, 6357 level - 1, entries); 6358 brelse(bp); 6359 6360 if (storeblk == NULL) 6361 return (NULL); 6362 } 6363 return (storeblk); 6364 } 6365 6366 /* ARGSUSED */ 6367 static int 6368 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, 6369 struct cred *cr, caller_context_t *ct) 6370 { 6371 struct inode *ip = VTOI(vp); 6372 struct ulockfs *ulp; 6373 struct ufsvfs *ufsvfsp = ip->i_ufsvfs; 6374 ulong_t vsa_mask = vsap->vsa_mask; 6375 int err = EINVAL; 6376 6377 vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT); 6378 6379 /* 6380 * Only grab locks if needed - they're not needed to check vsa_mask 6381 * or if the mask contains no acl flags. 6382 */ 6383 if (vsa_mask != 0) { 6384 if (err = ufs_lockfs_begin(ufsvfsp, &ulp, 6385 ULOCKFS_GETATTR_MASK)) 6386 return (err); 6387 6388 rw_enter(&ip->i_contents, RW_READER); 6389 err = ufs_acl_get(ip, vsap, flag, cr); 6390 rw_exit(&ip->i_contents); 6391 6392 if (ulp) 6393 ufs_lockfs_end(ulp); 6394 } 6395 return (err); 6396 } 6397 6398 /* ARGSUSED */ 6399 static int 6400 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr, 6401 caller_context_t *ct) 6402 { 6403 struct inode *ip = VTOI(vp); 6404 struct ulockfs *ulp = NULL; 6405 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs; 6406 ulong_t vsa_mask = vsap->vsa_mask; 6407 int err; 6408 int haverwlock = 1; 6409 int trans_size; 6410 int donetrans = 0; 6411 int retry = 1; 6412 6413 ASSERT(RW_LOCK_HELD(&ip->i_rwlock)); 6414 6415 /* Abort now if the request is either empty or invalid. */ 6416 vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT); 6417 if ((vsa_mask == 0) || 6418 ((vsap->vsa_aclentp == NULL) && 6419 (vsap->vsa_dfaclentp == NULL))) { 6420 err = EINVAL; 6421 goto out; 6422 } 6423 6424 /* 6425 * Following convention, if this is a directory then we acquire the 6426 * inode's i_rwlock after starting a UFS logging transaction; 6427 * otherwise, we acquire it beforehand. Since we were called (and 6428 * must therefore return) with the lock held, we will have to drop it, 6429 * and later reacquire it, if operating on a directory. 6430 */ 6431 if (vp->v_type == VDIR) { 6432 rw_exit(&ip->i_rwlock); 6433 haverwlock = 0; 6434 } else { 6435 /* Upgrade the lock if required. */ 6436 if (!rw_write_held(&ip->i_rwlock)) { 6437 rw_exit(&ip->i_rwlock); 6438 rw_enter(&ip->i_rwlock, RW_WRITER); 6439 } 6440 } 6441 6442 again: 6443 ASSERT(!(vp->v_type == VDIR && haverwlock)); 6444 if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) { 6445 ulp = NULL; 6446 retry = 0; 6447 goto out; 6448 } 6449 6450 /* 6451 * Check that the file system supports this operation. Note that 6452 * ufs_lockfs_begin() will have checked that the file system had 6453 * not been forcibly unmounted. 6454 */ 6455 if (ufsvfsp->vfs_fs->fs_ronly) { 6456 err = EROFS; 6457 goto out; 6458 } 6459 if (ufsvfsp->vfs_nosetsec) { 6460 err = ENOSYS; 6461 goto out; 6462 } 6463 6464 if (ulp) { 6465 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR, 6466 trans_size = TOP_SETSECATTR_SIZE(VTOI(vp))); 6467 donetrans = 1; 6468 } 6469 6470 if (vp->v_type == VDIR) { 6471 rw_enter(&ip->i_rwlock, RW_WRITER); 6472 haverwlock = 1; 6473 } 6474 6475 ASSERT(haverwlock); 6476 6477 /* Do the actual work. */ 6478 rw_enter(&ip->i_contents, RW_WRITER); 6479 /* 6480 * Suppress out of inodes messages if we will retry. 6481 */ 6482 if (retry) 6483 ip->i_flag |= IQUIET; 6484 err = ufs_acl_set(ip, vsap, flag, cr); 6485 ip->i_flag &= ~IQUIET; 6486 rw_exit(&ip->i_contents); 6487 6488 out: 6489 if (ulp) { 6490 if (donetrans) { 6491 /* 6492 * top_end_async() can eventually call 6493 * top_end_sync(), which can block. We must 6494 * therefore observe the lock-ordering protocol 6495 * here as well. 6496 */ 6497 if (vp->v_type == VDIR) { 6498 rw_exit(&ip->i_rwlock); 6499 haverwlock = 0; 6500 } 6501 TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size); 6502 } 6503 ufs_lockfs_end(ulp); 6504 } 6505 /* 6506 * If no inodes available, try scaring a logically- 6507 * free one out of the delete queue to someplace 6508 * that we can find it. 6509 */ 6510 if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) { 6511 ufs_delete_drain_wait(ufsvfsp, 1); 6512 retry = 0; 6513 if (vp->v_type == VDIR && haverwlock) { 6514 rw_exit(&ip->i_rwlock); 6515 haverwlock = 0; 6516 } 6517 goto again; 6518 } 6519 /* 6520 * If we need to reacquire the lock then it is safe to do so 6521 * as a reader. This is because ufs_rwunlock(), which will be 6522 * called by our caller after we return, does not differentiate 6523 * between shared and exclusive locks. 6524 */ 6525 if (!haverwlock) { 6526 ASSERT(vp->v_type == VDIR); 6527 rw_enter(&ip->i_rwlock, RW_READER); 6528 } 6529 6530 return (err); 6531 } 6532