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