puffs_vnops.c revision 1.197
1 /* $NetBSD: puffs_vnops.c,v 1.197 2014/11/04 09:10:37 manu Exp $ */ 2 3 /* 4 * Copyright (c) 2005, 2006, 2007 Antti Kantee. All Rights Reserved. 5 * 6 * Development of this software was supported by the 7 * Google Summer of Code program and the Ulla Tuominen Foundation. 8 * The Google SoC project was mentored by Bill Studenmund. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS 20 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 21 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 22 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 25 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: puffs_vnops.c,v 1.197 2014/11/04 09:10:37 manu Exp $"); 34 35 #include <sys/param.h> 36 #include <sys/buf.h> 37 #include <sys/lockf.h> 38 #include <sys/malloc.h> 39 #include <sys/mount.h> 40 #include <sys/namei.h> 41 #include <sys/vnode.h> 42 #include <sys/proc.h> 43 #include <sys/kernel.h> /* For hz, hardclock_ticks */ 44 45 #include <uvm/uvm.h> 46 47 #include <fs/puffs/puffs_msgif.h> 48 #include <fs/puffs/puffs_sys.h> 49 50 #include <miscfs/fifofs/fifo.h> 51 #include <miscfs/genfs/genfs.h> 52 #include <miscfs/specfs/specdev.h> 53 54 int puffs_vnop_lookup(void *); 55 int puffs_vnop_create(void *); 56 int puffs_vnop_access(void *); 57 int puffs_vnop_mknod(void *); 58 int puffs_vnop_open(void *); 59 int puffs_vnop_close(void *); 60 int puffs_vnop_getattr(void *); 61 int puffs_vnop_setattr(void *); 62 int puffs_vnop_reclaim(void *); 63 int puffs_vnop_readdir(void *); 64 int puffs_vnop_poll(void *); 65 int puffs_vnop_fsync(void *); 66 int puffs_vnop_seek(void *); 67 int puffs_vnop_remove(void *); 68 int puffs_vnop_mkdir(void *); 69 int puffs_vnop_rmdir(void *); 70 int puffs_vnop_link(void *); 71 int puffs_vnop_readlink(void *); 72 int puffs_vnop_symlink(void *); 73 int puffs_vnop_rename(void *); 74 int puffs_vnop_read(void *); 75 int puffs_vnop_write(void *); 76 int puffs_vnop_fallocate(void *); 77 int puffs_vnop_fdiscard(void *); 78 int puffs_vnop_fcntl(void *); 79 int puffs_vnop_ioctl(void *); 80 int puffs_vnop_inactive(void *); 81 int puffs_vnop_print(void *); 82 int puffs_vnop_pathconf(void *); 83 int puffs_vnop_advlock(void *); 84 int puffs_vnop_strategy(void *); 85 int puffs_vnop_bmap(void *); 86 int puffs_vnop_mmap(void *); 87 int puffs_vnop_getpages(void *); 88 int puffs_vnop_abortop(void *); 89 int puffs_vnop_getextattr(void *); 90 int puffs_vnop_setextattr(void *); 91 int puffs_vnop_listextattr(void *); 92 int puffs_vnop_deleteextattr(void *); 93 94 int puffs_vnop_spec_read(void *); 95 int puffs_vnop_spec_write(void *); 96 int puffs_vnop_fifo_read(void *); 97 int puffs_vnop_fifo_write(void *); 98 99 int puffs_vnop_checkop(void *); 100 101 #define puffs_vnop_lock genfs_lock 102 #define puffs_vnop_unlock genfs_unlock 103 #define puffs_vnop_islocked genfs_islocked 104 105 int (**puffs_vnodeop_p)(void *); 106 const struct vnodeopv_entry_desc puffs_vnodeop_entries[] = { 107 { &vop_default_desc, vn_default_error }, 108 { &vop_lookup_desc, puffs_vnop_lookup }, /* REAL lookup */ 109 { &vop_create_desc, puffs_vnop_checkop }, /* create */ 110 { &vop_mknod_desc, puffs_vnop_checkop }, /* mknod */ 111 { &vop_open_desc, puffs_vnop_open }, /* REAL open */ 112 { &vop_close_desc, puffs_vnop_checkop }, /* close */ 113 { &vop_access_desc, puffs_vnop_access }, /* REAL access */ 114 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */ 115 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */ 116 { &vop_read_desc, puffs_vnop_checkop }, /* read */ 117 { &vop_write_desc, puffs_vnop_checkop }, /* write */ 118 { &vop_fallocate_desc, puffs_vnop_fallocate }, /* fallocate */ 119 { &vop_fdiscard_desc, puffs_vnop_fdiscard }, /* fdiscard */ 120 { &vop_fsync_desc, puffs_vnop_fsync }, /* REAL fsync */ 121 { &vop_seek_desc, puffs_vnop_checkop }, /* seek */ 122 { &vop_remove_desc, puffs_vnop_checkop }, /* remove */ 123 { &vop_link_desc, puffs_vnop_checkop }, /* link */ 124 { &vop_rename_desc, puffs_vnop_checkop }, /* rename */ 125 { &vop_mkdir_desc, puffs_vnop_checkop }, /* mkdir */ 126 { &vop_rmdir_desc, puffs_vnop_checkop }, /* rmdir */ 127 { &vop_symlink_desc, puffs_vnop_checkop }, /* symlink */ 128 { &vop_readdir_desc, puffs_vnop_checkop }, /* readdir */ 129 { &vop_readlink_desc, puffs_vnop_checkop }, /* readlink */ 130 { &vop_getpages_desc, puffs_vnop_checkop }, /* getpages */ 131 { &vop_putpages_desc, genfs_putpages }, /* REAL putpages */ 132 { &vop_pathconf_desc, puffs_vnop_checkop }, /* pathconf */ 133 { &vop_advlock_desc, puffs_vnop_advlock }, /* advlock */ 134 { &vop_strategy_desc, puffs_vnop_strategy }, /* REAL strategy */ 135 { &vop_revoke_desc, genfs_revoke }, /* REAL revoke */ 136 { &vop_abortop_desc, puffs_vnop_abortop }, /* REAL abortop */ 137 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */ 138 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */ 139 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */ 140 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */ 141 { &vop_bmap_desc, puffs_vnop_bmap }, /* REAL bmap */ 142 { &vop_print_desc, puffs_vnop_print }, /* REAL print */ 143 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */ 144 { &vop_bwrite_desc, genfs_nullop }, /* REAL bwrite */ 145 { &vop_mmap_desc, puffs_vnop_mmap }, /* REAL mmap */ 146 { &vop_poll_desc, puffs_vnop_poll }, /* REAL poll */ 147 { &vop_getextattr_desc, puffs_vnop_getextattr }, /* getextattr */ 148 { &vop_setextattr_desc, puffs_vnop_setextattr }, /* setextattr */ 149 { &vop_listextattr_desc, puffs_vnop_listextattr }, /* listextattr */ 150 { &vop_deleteextattr_desc, puffs_vnop_deleteextattr },/* deleteextattr */ 151 #if 0 152 { &vop_openextattr_desc, puffs_vnop_checkop }, /* openextattr */ 153 { &vop_closeextattr_desc, puffs_vnop_checkop }, /* closeextattr */ 154 #endif 155 { &vop_kqfilter_desc, genfs_eopnotsupp }, /* kqfilter XXX */ 156 { NULL, NULL } 157 }; 158 const struct vnodeopv_desc puffs_vnodeop_opv_desc = 159 { &puffs_vnodeop_p, puffs_vnodeop_entries }; 160 161 162 int (**puffs_specop_p)(void *); 163 const struct vnodeopv_entry_desc puffs_specop_entries[] = { 164 { &vop_default_desc, vn_default_error }, 165 { &vop_lookup_desc, spec_lookup }, /* lookup, ENOTDIR */ 166 { &vop_create_desc, spec_create }, /* genfs_badop */ 167 { &vop_mknod_desc, spec_mknod }, /* genfs_badop */ 168 { &vop_open_desc, spec_open }, /* spec_open */ 169 { &vop_close_desc, spec_close }, /* spec_close */ 170 { &vop_access_desc, puffs_vnop_checkop }, /* access */ 171 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */ 172 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */ 173 { &vop_read_desc, puffs_vnop_spec_read }, /* update, read */ 174 { &vop_write_desc, puffs_vnop_spec_write }, /* update, write */ 175 { &vop_fallocate_desc, spec_fallocate }, /* fallocate */ 176 { &vop_fdiscard_desc, spec_fdiscard }, /* fdiscard */ 177 { &vop_ioctl_desc, spec_ioctl }, /* spec_ioctl */ 178 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */ 179 { &vop_poll_desc, spec_poll }, /* spec_poll */ 180 { &vop_kqfilter_desc, spec_kqfilter }, /* spec_kqfilter */ 181 { &vop_revoke_desc, spec_revoke }, /* genfs_revoke */ 182 { &vop_mmap_desc, spec_mmap }, /* spec_mmap */ 183 { &vop_fsync_desc, spec_fsync }, /* vflushbuf */ 184 { &vop_seek_desc, spec_seek }, /* genfs_nullop */ 185 { &vop_remove_desc, spec_remove }, /* genfs_badop */ 186 { &vop_link_desc, spec_link }, /* genfs_badop */ 187 { &vop_rename_desc, spec_rename }, /* genfs_badop */ 188 { &vop_mkdir_desc, spec_mkdir }, /* genfs_badop */ 189 { &vop_rmdir_desc, spec_rmdir }, /* genfs_badop */ 190 { &vop_symlink_desc, spec_symlink }, /* genfs_badop */ 191 { &vop_readdir_desc, spec_readdir }, /* genfs_badop */ 192 { &vop_readlink_desc, spec_readlink }, /* genfs_badop */ 193 { &vop_abortop_desc, spec_abortop }, /* genfs_badop */ 194 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */ 195 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */ 196 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */ 197 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */ 198 { &vop_bmap_desc, spec_bmap }, /* dummy */ 199 { &vop_strategy_desc, spec_strategy }, /* dev strategy */ 200 { &vop_print_desc, puffs_vnop_print }, /* REAL print */ 201 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */ 202 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 203 { &vop_advlock_desc, spec_advlock }, /* lf_advlock */ 204 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 205 { &vop_getpages_desc, spec_getpages }, /* genfs_getpages */ 206 { &vop_putpages_desc, spec_putpages }, /* genfs_putpages */ 207 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */ 208 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */ 209 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */ 210 { &vop_deleteextattr_desc, puffs_vnop_checkop },/* deleteextattr */ 211 #if 0 212 { &vop_openextattr_desc, _openextattr }, /* openextattr */ 213 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */ 214 #endif 215 { NULL, NULL } 216 }; 217 const struct vnodeopv_desc puffs_specop_opv_desc = 218 { &puffs_specop_p, puffs_specop_entries }; 219 220 221 int (**puffs_fifoop_p)(void *); 222 const struct vnodeopv_entry_desc puffs_fifoop_entries[] = { 223 { &vop_default_desc, vn_default_error }, 224 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup, ENOTDIR */ 225 { &vop_create_desc, vn_fifo_bypass }, /* genfs_badop */ 226 { &vop_mknod_desc, vn_fifo_bypass }, /* genfs_badop */ 227 { &vop_open_desc, vn_fifo_bypass }, /* open */ 228 { &vop_close_desc, vn_fifo_bypass }, /* close */ 229 { &vop_access_desc, puffs_vnop_checkop }, /* access */ 230 { &vop_getattr_desc, puffs_vnop_checkop }, /* getattr */ 231 { &vop_setattr_desc, puffs_vnop_checkop }, /* setattr */ 232 { &vop_read_desc, puffs_vnop_fifo_read }, /* read, update */ 233 { &vop_write_desc, puffs_vnop_fifo_write }, /* write, update */ 234 { &vop_fallocate_desc, vn_fifo_bypass }, /* fallocate */ 235 { &vop_fdiscard_desc, vn_fifo_bypass }, /* fdiscard */ 236 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 237 { &vop_fcntl_desc, genfs_fcntl }, /* dummy */ 238 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 239 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 240 { &vop_revoke_desc, vn_fifo_bypass }, /* genfs_revoke */ 241 { &vop_mmap_desc, vn_fifo_bypass }, /* genfs_badop */ 242 { &vop_fsync_desc, vn_fifo_bypass }, /* genfs_nullop*/ 243 { &vop_seek_desc, vn_fifo_bypass }, /* genfs_badop */ 244 { &vop_remove_desc, vn_fifo_bypass }, /* genfs_badop */ 245 { &vop_link_desc, vn_fifo_bypass }, /* genfs_badop */ 246 { &vop_rename_desc, vn_fifo_bypass }, /* genfs_badop */ 247 { &vop_mkdir_desc, vn_fifo_bypass }, /* genfs_badop */ 248 { &vop_rmdir_desc, vn_fifo_bypass }, /* genfs_badop */ 249 { &vop_symlink_desc, vn_fifo_bypass }, /* genfs_badop */ 250 { &vop_readdir_desc, vn_fifo_bypass }, /* genfs_badop */ 251 { &vop_readlink_desc, vn_fifo_bypass }, /* genfs_badop */ 252 { &vop_abortop_desc, vn_fifo_bypass }, /* genfs_badop */ 253 { &vop_inactive_desc, puffs_vnop_inactive }, /* REAL inactive */ 254 { &vop_reclaim_desc, puffs_vnop_reclaim }, /* REAL reclaim */ 255 { &vop_lock_desc, puffs_vnop_lock }, /* REAL lock */ 256 { &vop_unlock_desc, puffs_vnop_unlock }, /* REAL unlock */ 257 { &vop_bmap_desc, vn_fifo_bypass }, /* dummy */ 258 { &vop_strategy_desc, vn_fifo_bypass }, /* genfs_badop */ 259 { &vop_print_desc, puffs_vnop_print }, /* REAL print */ 260 { &vop_islocked_desc, puffs_vnop_islocked }, /* REAL islocked */ 261 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 262 { &vop_advlock_desc, vn_fifo_bypass }, /* genfs_einval */ 263 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 264 { &vop_putpages_desc, vn_fifo_bypass }, /* genfs_null_putpages*/ 265 #if 0 266 { &vop_openextattr_desc, _openextattr }, /* openextattr */ 267 { &vop_closeextattr_desc, _closeextattr }, /* closeextattr */ 268 #endif 269 { &vop_getextattr_desc, puffs_vnop_checkop }, /* getextattr */ 270 { &vop_setextattr_desc, puffs_vnop_checkop }, /* setextattr */ 271 { &vop_listextattr_desc, puffs_vnop_checkop }, /* listextattr */ 272 { &vop_deleteextattr_desc, puffs_vnop_checkop }, /* deleteextattr */ 273 { NULL, NULL } 274 }; 275 const struct vnodeopv_desc puffs_fifoop_opv_desc = 276 { &puffs_fifoop_p, puffs_fifoop_entries }; 277 278 279 /* "real" vnode operations */ 280 int (**puffs_msgop_p)(void *); 281 const struct vnodeopv_entry_desc puffs_msgop_entries[] = { 282 { &vop_default_desc, vn_default_error }, 283 { &vop_create_desc, puffs_vnop_create }, /* create */ 284 { &vop_mknod_desc, puffs_vnop_mknod }, /* mknod */ 285 { &vop_open_desc, puffs_vnop_open }, /* open */ 286 { &vop_close_desc, puffs_vnop_close }, /* close */ 287 { &vop_access_desc, puffs_vnop_access }, /* access */ 288 { &vop_getattr_desc, puffs_vnop_getattr }, /* getattr */ 289 { &vop_setattr_desc, puffs_vnop_setattr }, /* setattr */ 290 { &vop_read_desc, puffs_vnop_read }, /* read */ 291 { &vop_write_desc, puffs_vnop_write }, /* write */ 292 { &vop_seek_desc, puffs_vnop_seek }, /* seek */ 293 { &vop_remove_desc, puffs_vnop_remove }, /* remove */ 294 { &vop_link_desc, puffs_vnop_link }, /* link */ 295 { &vop_rename_desc, puffs_vnop_rename }, /* rename */ 296 { &vop_mkdir_desc, puffs_vnop_mkdir }, /* mkdir */ 297 { &vop_rmdir_desc, puffs_vnop_rmdir }, /* rmdir */ 298 { &vop_symlink_desc, puffs_vnop_symlink }, /* symlink */ 299 { &vop_readdir_desc, puffs_vnop_readdir }, /* readdir */ 300 { &vop_readlink_desc, puffs_vnop_readlink }, /* readlink */ 301 { &vop_print_desc, puffs_vnop_print }, /* print */ 302 { &vop_islocked_desc, puffs_vnop_islocked }, /* islocked */ 303 { &vop_pathconf_desc, puffs_vnop_pathconf }, /* pathconf */ 304 { &vop_getpages_desc, puffs_vnop_getpages }, /* getpages */ 305 { NULL, NULL } 306 }; 307 const struct vnodeopv_desc puffs_msgop_opv_desc = 308 { &puffs_msgop_p, puffs_msgop_entries }; 309 310 /* 311 * for dosetattr / update_va 312 */ 313 #define SETATTR_CHSIZE 0x01 314 #define SETATTR_ASYNC 0x02 315 316 #define ERROUT(err) \ 317 do { \ 318 error = err; \ 319 goto out; \ 320 } while (/*CONSTCOND*/0) 321 322 /* 323 * This is a generic vnode operation handler. It checks if the necessary 324 * operations for the called vnode operation are implemented by userspace 325 * and either returns a dummy return value or proceeds to call the real 326 * vnode operation from puffs_msgop_v. 327 * 328 * XXX: this should described elsewhere and autogenerated, the complexity 329 * of the vnode operations vectors and their interrelationships is also 330 * getting a bit out of hand. Another problem is that we need this same 331 * information in the fs server code, so keeping the two in sync manually 332 * is not a viable (long term) plan. 333 */ 334 335 /* not supported, handle locking protocol */ 336 #define CHECKOP_NOTSUPP(op) \ 337 case VOP_##op##_DESCOFFSET: \ 338 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \ 339 return genfs_eopnotsupp(v); \ 340 break 341 342 /* always succeed, no locking */ 343 #define CHECKOP_SUCCESS(op) \ 344 case VOP_##op##_DESCOFFSET: \ 345 if (pmp->pmp_vnopmask[PUFFS_VN_##op] == 0) \ 346 return 0; \ 347 break 348 349 int 350 puffs_vnop_checkop(void *v) 351 { 352 struct vop_generic_args /* { 353 struct vnodeop_desc *a_desc; 354 spooky mystery contents; 355 } */ *ap = v; 356 struct vnodeop_desc *desc = ap->a_desc; 357 struct puffs_mount *pmp; 358 struct vnode *vp; 359 int offset, rv; 360 361 offset = ap->a_desc->vdesc_vp_offsets[0]; 362 #ifdef DIAGNOSTIC 363 if (offset == VDESC_NO_OFFSET) 364 panic("puffs_checkop: no vnode, why did you call me?"); 365 #endif 366 vp = *VOPARG_OFFSETTO(struct vnode **, offset, ap); 367 pmp = MPTOPUFFSMP(vp->v_mount); 368 369 DPRINTF_VERBOSE(("checkop call %s (%d), vp %p\n", 370 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp)); 371 372 if (!ALLOPS(pmp)) { 373 switch (desc->vdesc_offset) { 374 CHECKOP_NOTSUPP(CREATE); 375 CHECKOP_NOTSUPP(MKNOD); 376 CHECKOP_NOTSUPP(GETATTR); 377 CHECKOP_NOTSUPP(SETATTR); 378 CHECKOP_NOTSUPP(READ); 379 CHECKOP_NOTSUPP(WRITE); 380 CHECKOP_NOTSUPP(FCNTL); 381 CHECKOP_NOTSUPP(IOCTL); 382 CHECKOP_NOTSUPP(REMOVE); 383 CHECKOP_NOTSUPP(LINK); 384 CHECKOP_NOTSUPP(RENAME); 385 CHECKOP_NOTSUPP(MKDIR); 386 CHECKOP_NOTSUPP(RMDIR); 387 CHECKOP_NOTSUPP(SYMLINK); 388 CHECKOP_NOTSUPP(READDIR); 389 CHECKOP_NOTSUPP(READLINK); 390 CHECKOP_NOTSUPP(PRINT); 391 CHECKOP_NOTSUPP(PATHCONF); 392 CHECKOP_NOTSUPP(GETEXTATTR); 393 CHECKOP_NOTSUPP(SETEXTATTR); 394 CHECKOP_NOTSUPP(LISTEXTATTR); 395 CHECKOP_NOTSUPP(DELETEEXTATTR); 396 397 CHECKOP_SUCCESS(ACCESS); 398 CHECKOP_SUCCESS(CLOSE); 399 CHECKOP_SUCCESS(SEEK); 400 401 case VOP_GETPAGES_DESCOFFSET: 402 if (!EXISTSOP(pmp, READ)) 403 return genfs_eopnotsupp(v); 404 break; 405 406 default: 407 panic("puffs_checkop: unhandled vnop %d", 408 desc->vdesc_offset); 409 } 410 } 411 412 rv = VOCALL(puffs_msgop_p, ap->a_desc->vdesc_offset, v); 413 414 DPRINTF_VERBOSE(("checkop return %s (%d), vp %p: %d\n", 415 ap->a_desc->vdesc_name, ap->a_desc->vdesc_offset, vp, rv)); 416 417 return rv; 418 } 419 420 static int callremove(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t, 421 struct componentname *); 422 static int callrmdir(struct puffs_mount *, puffs_cookie_t, puffs_cookie_t, 423 struct componentname *); 424 static void callinactive(struct puffs_mount *, puffs_cookie_t, int); 425 static void callreclaim(struct puffs_mount *, puffs_cookie_t, int); 426 static int flushvncache(struct vnode *, off_t, off_t, bool); 427 static void update_va(struct vnode *, struct vattr *, struct vattr *, 428 struct timespec *, struct timespec *, int); 429 static void update_parent(struct vnode *, struct vnode *); 430 431 432 #define PUFFS_ABORT_LOOKUP 1 433 #define PUFFS_ABORT_CREATE 2 434 #define PUFFS_ABORT_MKNOD 3 435 #define PUFFS_ABORT_MKDIR 4 436 #define PUFFS_ABORT_SYMLINK 5 437 438 /* 439 * Press the pani^Wabort button! Kernel resource allocation failed. 440 */ 441 static void 442 puffs_abortbutton(struct puffs_mount *pmp, int what, 443 puffs_cookie_t dck, puffs_cookie_t ck, struct componentname *cnp) 444 { 445 446 switch (what) { 447 case PUFFS_ABORT_CREATE: 448 case PUFFS_ABORT_MKNOD: 449 case PUFFS_ABORT_SYMLINK: 450 callremove(pmp, dck, ck, cnp); 451 break; 452 case PUFFS_ABORT_MKDIR: 453 callrmdir(pmp, dck, ck, cnp); 454 break; 455 } 456 457 callinactive(pmp, ck, 0); 458 callreclaim(pmp, ck, 1); 459 } 460 461 /* 462 * Begin vnode operations. 463 * 464 * A word from the keymaster about locks: generally we don't want 465 * to use the vnode locks at all: it creates an ugly dependency between 466 * the userlandia file server and the kernel. But we'll play along with 467 * the kernel vnode locks for now. However, even currently we attempt 468 * to release locks as early as possible. This is possible for some 469 * operations which a) don't need a locked vnode after the userspace op 470 * and b) return with the vnode unlocked. Theoretically we could 471 * unlock-do op-lock for others and order the graph in userspace, but I 472 * don't want to think of the consequences for the time being. 473 */ 474 475 #define TTL_TO_TIMEOUT(ts) \ 476 (hardclock_ticks + (ts->tv_sec * hz) + (ts->tv_nsec * hz / 1000000000)) 477 #define TTL_VALID(ts) \ 478 ((ts != NULL) && !((ts->tv_sec == 0) && (ts->tv_nsec == 0))) 479 #define TIMED_OUT(expire) \ 480 ((int)((unsigned int)hardclock_ticks - (unsigned int)expire) > 0) 481 int 482 puffs_vnop_lookup(void *v) 483 { 484 struct vop_lookup_v2_args /* { 485 const struct vnodeop_desc *a_desc; 486 struct vnode *a_dvp; 487 struct vnode **a_vpp; 488 struct componentname *a_cnp; 489 } */ *ap = v; 490 PUFFS_MSG_VARS(vn, lookup); 491 struct puffs_mount *pmp; 492 struct componentname *cnp; 493 struct vnode *vp, *dvp, *cvp; 494 struct puffs_node *dpn, *cpn; 495 int isdot; 496 int error; 497 498 pmp = MPTOPUFFSMP(ap->a_dvp->v_mount); 499 cnp = ap->a_cnp; 500 dvp = ap->a_dvp; 501 cvp = NULL; 502 cpn = NULL; 503 *ap->a_vpp = NULL; 504 505 /* r/o fs? we check create later to handle EEXIST */ 506 if ((cnp->cn_flags & ISLASTCN) 507 && (dvp->v_mount->mnt_flag & MNT_RDONLY) 508 && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) 509 return EROFS; 510 511 isdot = cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.'; 512 513 DPRINTF(("puffs_lookup: \"%s\", parent vnode %p, op: %x\n", 514 cnp->cn_nameptr, dvp, cnp->cn_nameiop)); 515 516 /* 517 * If dotdot cache is enabled, add reference to .. and return. 518 */ 519 if (PUFFS_USE_DOTDOTCACHE(pmp) && (cnp->cn_flags & ISDOTDOT)) { 520 vp = VPTOPP(ap->a_dvp)->pn_parent; 521 vref(vp); 522 523 *ap->a_vpp = vp; 524 return 0; 525 } 526 527 /* 528 * Check if someone fed it into the cache 529 */ 530 if (!isdot && PUFFS_USE_NAMECACHE(pmp)) { 531 int found, iswhiteout; 532 533 found = cache_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen, 534 cnp->cn_nameiop, cnp->cn_flags, 535 &iswhiteout, ap->a_vpp); 536 if (iswhiteout) { 537 cnp->cn_flags |= ISWHITEOUT; 538 } 539 540 if (found && *ap->a_vpp != NULLVP && PUFFS_USE_FS_TTL(pmp)) { 541 cvp = *ap->a_vpp; 542 cpn = VPTOPP(cvp); 543 544 if (TIMED_OUT(cpn->pn_cn_timeout)) { 545 cache_purge(cvp); 546 /* 547 * cached vnode (cvp) is still referenced 548 * so that we can reuse it upon a new 549 * successful lookup. 550 */ 551 *ap->a_vpp = NULL; 552 found = 0; 553 } 554 } 555 556 /* 557 * Do not use negative caching, since the filesystem 558 * provides no TTL for it. 559 */ 560 if (found && *ap->a_vpp == NULLVP && PUFFS_USE_FS_TTL(pmp)) 561 found = 0; 562 563 if (found) { 564 return *ap->a_vpp == NULLVP ? ENOENT : 0; 565 } 566 567 /* 568 * This is what would have been left in ERROR before 569 * the rearrangement of cache_lookup(). What with all 570 * the macros, I am not sure if this is a dead value 571 * below or not. 572 */ 573 error = -1; 574 } 575 576 if (isdot) { 577 /* deal with rename lookup semantics */ 578 if (cnp->cn_nameiop == RENAME && (cnp->cn_flags & ISLASTCN)) 579 return EISDIR; 580 581 vp = ap->a_dvp; 582 vref(vp); 583 *ap->a_vpp = vp; 584 return 0; 585 } 586 587 if (cvp != NULL) { 588 if (vn_lock(cvp, LK_EXCLUSIVE) != 0) { 589 vrele(cvp); 590 cvp = NULL; 591 } else 592 mutex_enter(&cpn->pn_sizemtx); 593 } 594 595 PUFFS_MSG_ALLOC(vn, lookup); 596 puffs_makecn(&lookup_msg->pvnr_cn, &lookup_msg->pvnr_cn_cred, 597 cnp, PUFFS_USE_FULLPNBUF(pmp)); 598 599 if (cnp->cn_flags & ISDOTDOT) 600 VOP_UNLOCK(dvp); 601 602 puffs_msg_setinfo(park_lookup, PUFFSOP_VN, 603 PUFFS_VN_LOOKUP, VPTOPNC(dvp)); 604 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_lookup, dvp->v_data, NULL, error); 605 DPRINTF(("puffs_lookup: return of the userspace, part %d\n", error)); 606 607 /* 608 * In case of error, there is no new vnode to play with, so be 609 * happy with the NULL value given to vpp in the beginning. 610 * Also, check if this really was an error or the target was not 611 * present. Either treat it as a non-error for CREATE/RENAME or 612 * enter the component into the negative name cache (if desired). 613 */ 614 if (error) { 615 error = checkerr(pmp, error, __func__); 616 if (error == ENOENT) { 617 /* don't allow to create files on r/o fs */ 618 if ((dvp->v_mount->mnt_flag & MNT_RDONLY) 619 && cnp->cn_nameiop == CREATE) { 620 error = EROFS; 621 622 /* adjust values if we are creating */ 623 } else if ((cnp->cn_flags & ISLASTCN) 624 && (cnp->cn_nameiop == CREATE 625 || cnp->cn_nameiop == RENAME)) { 626 error = EJUSTRETURN; 627 628 /* save negative cache entry */ 629 } else { 630 if (PUFFS_USE_NAMECACHE(pmp) && 631 !PUFFS_USE_FS_TTL(pmp)) 632 cache_enter(dvp, NULL, cnp->cn_nameptr, 633 cnp->cn_namelen, cnp->cn_flags); 634 } 635 } 636 goto out; 637 } 638 639 /* 640 * Check that we don't get our parent node back, that would cause 641 * a pretty obvious deadlock. 642 */ 643 dpn = dvp->v_data; 644 if (lookup_msg->pvnr_newnode == dpn->pn_cookie) { 645 puffs_senderr(pmp, PUFFS_ERR_LOOKUP, EINVAL, 646 "lookup produced parent cookie", lookup_msg->pvnr_newnode); 647 error = EPROTO; 648 goto out; 649 } 650 651 /* 652 * Check if we looked up the cached vnode 653 */ 654 vp = NULL; 655 if (cvp && (VPTOPP(cvp)->pn_cookie == lookup_msg->pvnr_newnode)) { 656 int grace; 657 658 /* 659 * Bump grace time of this node so that it does not get 660 * reclaimed too fast. We try to increase a bit more the 661 * lifetime of busiest * nodes - with some limits. 662 */ 663 grace = 10 * puffs_sopreq_expire_timeout; 664 cpn->pn_cn_grace = hardclock_ticks + grace; 665 vp = cvp; 666 } 667 668 /* 669 * No cached vnode available, or the cached vnode does not 670 * match the userland cookie anymore: is the node known? 671 */ 672 if (vp == NULL) { 673 error = puffs_getvnode(dvp->v_mount, 674 lookup_msg->pvnr_newnode, lookup_msg->pvnr_vtype, 675 lookup_msg->pvnr_size, lookup_msg->pvnr_rdev, &vp); 676 if (error) { 677 puffs_abortbutton(pmp, PUFFS_ABORT_LOOKUP, 678 VPTOPNC(dvp), lookup_msg->pvnr_newnode, 679 ap->a_cnp); 680 goto out; 681 } 682 683 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 684 } 685 686 /* 687 * Update cache and TTL 688 */ 689 if (PUFFS_USE_FS_TTL(pmp)) { 690 struct timespec *va_ttl = &lookup_msg->pvnr_va_ttl; 691 struct timespec *cn_ttl = &lookup_msg->pvnr_cn_ttl; 692 update_va(vp, NULL, &lookup_msg->pvnr_va, 693 va_ttl, cn_ttl, SETATTR_CHSIZE); 694 } 695 696 KASSERT(lookup_msg->pvnr_newnode == VPTOPP(vp)->pn_cookie); 697 *ap->a_vpp = vp; 698 699 if (PUFFS_USE_NAMECACHE(pmp)) 700 cache_enter(dvp, vp, cnp->cn_nameptr, cnp->cn_namelen, 701 cnp->cn_flags); 702 703 /* XXX */ 704 if ((lookup_msg->pvnr_cn.pkcn_flags & REQUIREDIR) == 0) 705 cnp->cn_flags &= ~REQUIREDIR; 706 if (lookup_msg->pvnr_cn.pkcn_consume) 707 cnp->cn_consume = MIN(lookup_msg->pvnr_cn.pkcn_consume, 708 strlen(cnp->cn_nameptr) - cnp->cn_namelen); 709 710 VPTOPP(vp)->pn_nlookup++; 711 712 if (PUFFS_USE_DOTDOTCACHE(pmp) && 713 (VPTOPP(vp)->pn_parent != dvp)) 714 update_parent(vp, dvp); 715 716 out: 717 if (cvp != NULL) { 718 mutex_exit(&cpn->pn_sizemtx); 719 720 if (error || (cvp != vp)) 721 vput(cvp); 722 } 723 if (error == 0) 724 VOP_UNLOCK(*ap->a_vpp); 725 726 if (cnp->cn_flags & ISDOTDOT) 727 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 728 729 DPRINTF(("puffs_lookup: returning %d %p\n", error, *ap->a_vpp)); 730 PUFFS_MSG_RELEASE(lookup); 731 return error; 732 } 733 734 #define REFPN_AND_UNLOCKVP(a, b) \ 735 do { \ 736 mutex_enter(&b->pn_mtx); \ 737 puffs_referencenode(b); \ 738 mutex_exit(&b->pn_mtx); \ 739 VOP_UNLOCK(a); \ 740 } while (/*CONSTCOND*/0) 741 742 #define REFPN(b) \ 743 do { \ 744 mutex_enter(&b->pn_mtx); \ 745 puffs_referencenode(b); \ 746 mutex_exit(&b->pn_mtx); \ 747 } while (/*CONSTCOND*/0) 748 749 #define RELEPN_AND_VP(a, b) \ 750 do { \ 751 puffs_releasenode(b); \ 752 vrele(a); \ 753 } while (/*CONSTCOND*/0) 754 755 int 756 puffs_vnop_create(void *v) 757 { 758 struct vop_create_v3_args /* { 759 const struct vnodeop_desc *a_desc; 760 struct vnode *a_dvp; 761 struct vnode **a_vpp; 762 struct componentname *a_cnp; 763 struct vattr *a_vap; 764 } */ *ap = v; 765 PUFFS_MSG_VARS(vn, create); 766 struct vnode *dvp = ap->a_dvp; 767 struct puffs_node *dpn = VPTOPP(dvp); 768 struct componentname *cnp = ap->a_cnp; 769 struct mount *mp = dvp->v_mount; 770 struct puffs_mount *pmp = MPTOPUFFSMP(mp); 771 int error; 772 773 DPRINTF(("puffs_create: dvp %p, cnp: %s\n", 774 dvp, ap->a_cnp->cn_nameptr)); 775 776 PUFFS_MSG_ALLOC(vn, create); 777 puffs_makecn(&create_msg->pvnr_cn, &create_msg->pvnr_cn_cred, 778 cnp, PUFFS_USE_FULLPNBUF(pmp)); 779 create_msg->pvnr_va = *ap->a_vap; 780 puffs_msg_setinfo(park_create, PUFFSOP_VN, 781 PUFFS_VN_CREATE, VPTOPNC(dvp)); 782 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_create, dvp->v_data, NULL, error); 783 784 error = checkerr(pmp, error, __func__); 785 if (error) 786 goto out; 787 788 error = puffs_newnode(mp, dvp, ap->a_vpp, 789 create_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 0); 790 if (error) { 791 puffs_abortbutton(pmp, PUFFS_ABORT_CREATE, dpn->pn_cookie, 792 create_msg->pvnr_newnode, cnp); 793 goto out; 794 } 795 796 if (PUFFS_USE_FS_TTL(pmp)) { 797 struct timespec *va_ttl = &create_msg->pvnr_va_ttl; 798 struct timespec *cn_ttl = &create_msg->pvnr_cn_ttl; 799 struct vattr *rvap = &create_msg->pvnr_va; 800 801 update_va(*ap->a_vpp, NULL, rvap, 802 va_ttl, cn_ttl, SETATTR_CHSIZE); 803 } 804 805 VPTOPP(*ap->a_vpp)->pn_nlookup++; 806 807 if (PUFFS_USE_DOTDOTCACHE(pmp) && 808 (VPTOPP(*ap->a_vpp)->pn_parent != dvp)) 809 update_parent(*ap->a_vpp, dvp); 810 811 out: 812 DPRINTF(("puffs_create: return %d\n", error)); 813 PUFFS_MSG_RELEASE(create); 814 return error; 815 } 816 817 int 818 puffs_vnop_mknod(void *v) 819 { 820 struct vop_mknod_v3_args /* { 821 const struct vnodeop_desc *a_desc; 822 struct vnode *a_dvp; 823 struct vnode **a_vpp; 824 struct componentname *a_cnp; 825 struct vattr *a_vap; 826 } */ *ap = v; 827 PUFFS_MSG_VARS(vn, mknod); 828 struct vnode *dvp = ap->a_dvp; 829 struct puffs_node *dpn = VPTOPP(dvp); 830 struct componentname *cnp = ap->a_cnp; 831 struct mount *mp = dvp->v_mount; 832 struct puffs_mount *pmp = MPTOPUFFSMP(mp); 833 int error; 834 835 PUFFS_MSG_ALLOC(vn, mknod); 836 puffs_makecn(&mknod_msg->pvnr_cn, &mknod_msg->pvnr_cn_cred, 837 cnp, PUFFS_USE_FULLPNBUF(pmp)); 838 mknod_msg->pvnr_va = *ap->a_vap; 839 puffs_msg_setinfo(park_mknod, PUFFSOP_VN, 840 PUFFS_VN_MKNOD, VPTOPNC(dvp)); 841 842 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mknod, dvp->v_data, NULL, error); 843 844 error = checkerr(pmp, error, __func__); 845 if (error) 846 goto out; 847 848 error = puffs_newnode(mp, dvp, ap->a_vpp, 849 mknod_msg->pvnr_newnode, cnp, ap->a_vap->va_type, 850 ap->a_vap->va_rdev); 851 if (error) { 852 puffs_abortbutton(pmp, PUFFS_ABORT_MKNOD, dpn->pn_cookie, 853 mknod_msg->pvnr_newnode, cnp); 854 goto out; 855 } 856 857 if (PUFFS_USE_FS_TTL(pmp)) { 858 struct timespec *va_ttl = &mknod_msg->pvnr_va_ttl; 859 struct timespec *cn_ttl = &mknod_msg->pvnr_cn_ttl; 860 struct vattr *rvap = &mknod_msg->pvnr_va; 861 862 update_va(*ap->a_vpp, NULL, rvap, 863 va_ttl, cn_ttl, SETATTR_CHSIZE); 864 } 865 866 VPTOPP(*ap->a_vpp)->pn_nlookup++; 867 868 if (PUFFS_USE_DOTDOTCACHE(pmp) && 869 (VPTOPP(*ap->a_vpp)->pn_parent != dvp)) 870 update_parent(*ap->a_vpp, dvp); 871 872 out: 873 PUFFS_MSG_RELEASE(mknod); 874 return error; 875 } 876 877 int 878 puffs_vnop_open(void *v) 879 { 880 struct vop_open_args /* { 881 const struct vnodeop_desc *a_desc; 882 struct vnode *a_vp; 883 int a_mode; 884 kauth_cred_t a_cred; 885 } */ *ap = v; 886 PUFFS_MSG_VARS(vn, open); 887 struct vnode *vp = ap->a_vp; 888 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 889 struct puffs_node *pn = VPTOPP(vp); 890 int mode = ap->a_mode; 891 int error; 892 893 DPRINTF(("puffs_open: vp %p, mode 0x%x\n", vp, mode)); 894 895 if (vp->v_type == VREG && mode & FWRITE && !EXISTSOP(pmp, WRITE)) 896 ERROUT(EROFS); 897 898 if (!EXISTSOP(pmp, OPEN)) 899 ERROUT(0); 900 901 PUFFS_MSG_ALLOC(vn, open); 902 open_msg->pvnr_mode = mode; 903 puffs_credcvt(&open_msg->pvnr_cred, ap->a_cred); 904 puffs_msg_setinfo(park_open, PUFFSOP_VN, 905 PUFFS_VN_OPEN, VPTOPNC(vp)); 906 907 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_open, vp->v_data, NULL, error); 908 error = checkerr(pmp, error, __func__); 909 910 if (open_msg->pvnr_oflags & PUFFS_OPEN_IO_DIRECT) { 911 if (mode & FREAD) 912 pn->pn_stat |= PNODE_RDIRECT; 913 if (mode & FWRITE) 914 pn->pn_stat |= PNODE_WDIRECT; 915 } 916 out: 917 DPRINTF(("puffs_open: returning %d\n", error)); 918 PUFFS_MSG_RELEASE(open); 919 return error; 920 } 921 922 int 923 puffs_vnop_close(void *v) 924 { 925 struct vop_close_args /* { 926 const struct vnodeop_desc *a_desc; 927 struct vnode *a_vp; 928 int a_fflag; 929 kauth_cred_t a_cred; 930 } */ *ap = v; 931 PUFFS_MSG_VARS(vn, close); 932 struct vnode *vp = ap->a_vp; 933 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 934 935 PUFFS_MSG_ALLOC(vn, close); 936 puffs_msg_setfaf(park_close); 937 close_msg->pvnr_fflag = ap->a_fflag; 938 puffs_credcvt(&close_msg->pvnr_cred, ap->a_cred); 939 puffs_msg_setinfo(park_close, PUFFSOP_VN, 940 PUFFS_VN_CLOSE, VPTOPNC(vp)); 941 942 puffs_msg_enqueue(pmp, park_close); 943 PUFFS_MSG_RELEASE(close); 944 return 0; 945 } 946 947 int 948 puffs_vnop_access(void *v) 949 { 950 struct vop_access_args /* { 951 const struct vnodeop_desc *a_desc; 952 struct vnode *a_vp; 953 int a_mode; 954 kauth_cred_t a_cred; 955 } */ *ap = v; 956 PUFFS_MSG_VARS(vn, access); 957 struct vnode *vp = ap->a_vp; 958 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 959 int mode = ap->a_mode; 960 int error; 961 962 if (mode & VWRITE) { 963 switch (vp->v_type) { 964 case VDIR: 965 case VLNK: 966 case VREG: 967 if ((vp->v_mount->mnt_flag & MNT_RDONLY) 968 || !EXISTSOP(pmp, WRITE)) 969 return EROFS; 970 break; 971 default: 972 break; 973 } 974 } 975 976 if (!EXISTSOP(pmp, ACCESS)) 977 return 0; 978 979 PUFFS_MSG_ALLOC(vn, access); 980 access_msg->pvnr_mode = ap->a_mode; 981 puffs_credcvt(&access_msg->pvnr_cred, ap->a_cred); 982 puffs_msg_setinfo(park_access, PUFFSOP_VN, 983 PUFFS_VN_ACCESS, VPTOPNC(vp)); 984 985 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_access, vp->v_data, NULL, error); 986 error = checkerr(pmp, error, __func__); 987 PUFFS_MSG_RELEASE(access); 988 989 return error; 990 } 991 992 static void 993 update_va(struct vnode *vp, struct vattr *vap, struct vattr *rvap, 994 struct timespec *va_ttl, struct timespec *cn_ttl, int flags) 995 { 996 struct puffs_node *pn = VPTOPP(vp); 997 998 if (TTL_VALID(cn_ttl)) { 999 pn->pn_cn_timeout = TTL_TO_TIMEOUT(cn_ttl); 1000 pn->pn_cn_grace = MAX(pn->pn_cn_timeout, pn->pn_cn_grace); 1001 } 1002 1003 /* 1004 * Don't listen to the file server regarding special device 1005 * size info, the file server doesn't know anything about them. 1006 */ 1007 if (vp->v_type == VBLK || vp->v_type == VCHR) 1008 rvap->va_size = vp->v_size; 1009 1010 /* Ditto for blocksize (ufs comment: this doesn't belong here) */ 1011 if (vp->v_type == VBLK) 1012 rvap->va_blocksize = BLKDEV_IOSIZE; 1013 else if (vp->v_type == VCHR) 1014 rvap->va_blocksize = MAXBSIZE; 1015 1016 if (vap != NULL) { 1017 (void) memcpy(vap, rvap, sizeof(struct vattr)); 1018 vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0]; 1019 1020 if (pn->pn_stat & PNODE_METACACHE_ATIME) 1021 vap->va_atime = pn->pn_mc_atime; 1022 if (pn->pn_stat & PNODE_METACACHE_CTIME) 1023 vap->va_ctime = pn->pn_mc_ctime; 1024 if (pn->pn_stat & PNODE_METACACHE_MTIME) 1025 vap->va_mtime = pn->pn_mc_mtime; 1026 if (pn->pn_stat & PNODE_METACACHE_SIZE) 1027 vap->va_size = pn->pn_mc_size; 1028 } 1029 1030 if (!(pn->pn_stat & PNODE_METACACHE_SIZE) && (flags & SETATTR_CHSIZE)) { 1031 if (rvap->va_size != VNOVAL 1032 && vp->v_type != VBLK && vp->v_type != VCHR) { 1033 uvm_vnp_setsize(vp, rvap->va_size); 1034 pn->pn_serversize = rvap->va_size; 1035 } 1036 } 1037 1038 if ((va_ttl != NULL) && TTL_VALID(va_ttl)) { 1039 if (pn->pn_va_cache == NULL) 1040 pn->pn_va_cache = pool_get(&puffs_vapool, PR_WAITOK); 1041 1042 (void)memcpy(pn->pn_va_cache, rvap, sizeof(*rvap)); 1043 1044 pn->pn_va_timeout = TTL_TO_TIMEOUT(va_ttl); 1045 } 1046 } 1047 1048 static void 1049 update_parent(struct vnode *vp, struct vnode *dvp) 1050 { 1051 struct puffs_node *pn = VPTOPP(vp); 1052 1053 if (pn->pn_parent != NULL) { 1054 KASSERT(pn->pn_parent != dvp); 1055 vrele(pn->pn_parent); 1056 } 1057 1058 vref(dvp); 1059 pn->pn_parent = dvp; 1060 } 1061 1062 int 1063 puffs_vnop_getattr(void *v) 1064 { 1065 struct vop_getattr_args /* { 1066 const struct vnodeop_desc *a_desc; 1067 struct vnode *a_vp; 1068 struct vattr *a_vap; 1069 kauth_cred_t a_cred; 1070 } */ *ap = v; 1071 PUFFS_MSG_VARS(vn, getattr); 1072 struct vnode *vp = ap->a_vp; 1073 struct mount *mp = vp->v_mount; 1074 struct puffs_mount *pmp = MPTOPUFFSMP(mp); 1075 struct vattr *vap, *rvap; 1076 struct puffs_node *pn = VPTOPP(vp); 1077 struct timespec *va_ttl = NULL; 1078 int error = 0; 1079 1080 /* 1081 * A lock is required so that we do not race with 1082 * setattr, write and fsync when changing vp->v_size. 1083 * This is critical, since setting a stall smaler value 1084 * triggers a file truncate in uvm_vnp_setsize(), which 1085 * most of the time means data corruption (a chunk of 1086 * data is replaced by zeroes). This can be removed if 1087 * we decide one day that VOP_GETATTR must operate on 1088 * a locked vnode. 1089 * 1090 * XXX Should be useless now that VOP_GETATTR has been 1091 * fixed to always require a shared lock at least. 1092 */ 1093 mutex_enter(&pn->pn_sizemtx); 1094 1095 REFPN(pn); 1096 vap = ap->a_vap; 1097 1098 if (PUFFS_USE_FS_TTL(pmp)) { 1099 if (!TIMED_OUT(pn->pn_va_timeout)) { 1100 update_va(vp, vap, pn->pn_va_cache, 1101 NULL, NULL, SETATTR_CHSIZE); 1102 goto out2; 1103 } 1104 } 1105 1106 PUFFS_MSG_ALLOC(vn, getattr); 1107 vattr_null(&getattr_msg->pvnr_va); 1108 puffs_credcvt(&getattr_msg->pvnr_cred, ap->a_cred); 1109 puffs_msg_setinfo(park_getattr, PUFFSOP_VN, 1110 PUFFS_VN_GETATTR, VPTOPNC(vp)); 1111 1112 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getattr, vp->v_data, NULL, error); 1113 error = checkerr(pmp, error, __func__); 1114 if (error) 1115 goto out; 1116 1117 rvap = &getattr_msg->pvnr_va; 1118 1119 if (PUFFS_USE_FS_TTL(pmp)) 1120 va_ttl = &getattr_msg->pvnr_va_ttl; 1121 1122 update_va(vp, vap, rvap, va_ttl, NULL, SETATTR_CHSIZE); 1123 1124 out: 1125 PUFFS_MSG_RELEASE(getattr); 1126 1127 out2: 1128 puffs_releasenode(pn); 1129 1130 mutex_exit(&pn->pn_sizemtx); 1131 1132 return error; 1133 } 1134 1135 static void 1136 zerofill_lastpage(struct vnode *vp, voff_t off) 1137 { 1138 char zbuf[PAGE_SIZE]; 1139 struct iovec iov; 1140 struct uio uio; 1141 vsize_t len; 1142 int error; 1143 1144 if (trunc_page(off) == off) 1145 return; 1146 1147 if (vp->v_writecount == 0) 1148 return; 1149 1150 len = round_page(off) - off; 1151 memset(zbuf, 0, len); 1152 1153 iov.iov_base = zbuf; 1154 iov.iov_len = len; 1155 UIO_SETUP_SYSSPACE(&uio); 1156 uio.uio_iov = &iov; 1157 uio.uio_iovcnt = 1; 1158 uio.uio_offset = off; 1159 uio.uio_resid = len; 1160 uio.uio_rw = UIO_WRITE; 1161 1162 error = ubc_uiomove(&vp->v_uobj, &uio, len, 1163 UVM_ADV_SEQUENTIAL, UBC_WRITE|UBC_UNMAP_FLAG(vp)); 1164 if (error) { 1165 DPRINTF(("zero-fill 0x%" PRIxVSIZE "@0x%" PRIx64 1166 " failed: error = %d\n", len, off, error)); 1167 } 1168 1169 return; 1170 } 1171 1172 static int 1173 dosetattr(struct vnode *vp, struct vattr *vap, kauth_cred_t cred, int flags) 1174 { 1175 PUFFS_MSG_VARS(vn, setattr); 1176 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 1177 struct puffs_node *pn = vp->v_data; 1178 vsize_t oldsize = vp->v_size; 1179 int error = 0; 1180 1181 KASSERT(!(flags & SETATTR_CHSIZE) || mutex_owned(&pn->pn_sizemtx)); 1182 1183 if ((vp->v_mount->mnt_flag & MNT_RDONLY) && 1184 (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL 1185 || vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL 1186 || vap->va_mode != (mode_t)VNOVAL)) 1187 return EROFS; 1188 1189 if ((vp->v_mount->mnt_flag & MNT_RDONLY) 1190 && vp->v_type == VREG && vap->va_size != VNOVAL) 1191 return EROFS; 1192 1193 /* 1194 * Flush metacache first. If we are called with some explicit 1195 * parameters, treat them as information overriding metacache 1196 * information. 1197 */ 1198 if (pn->pn_stat & PNODE_METACACHE_MASK) { 1199 if ((pn->pn_stat & PNODE_METACACHE_ATIME) 1200 && vap->va_atime.tv_sec == VNOVAL) 1201 vap->va_atime = pn->pn_mc_atime; 1202 if ((pn->pn_stat & PNODE_METACACHE_CTIME) 1203 && vap->va_ctime.tv_sec == VNOVAL) 1204 vap->va_ctime = pn->pn_mc_ctime; 1205 if ((pn->pn_stat & PNODE_METACACHE_MTIME) 1206 && vap->va_mtime.tv_sec == VNOVAL) 1207 vap->va_mtime = pn->pn_mc_mtime; 1208 if ((pn->pn_stat & PNODE_METACACHE_SIZE) 1209 && vap->va_size == VNOVAL) 1210 vap->va_size = pn->pn_mc_size; 1211 1212 pn->pn_stat &= ~PNODE_METACACHE_MASK; 1213 } 1214 1215 /* 1216 * Flush attribute cache so that another thread do 1217 * not get a stale value during the operation. 1218 */ 1219 if (PUFFS_USE_FS_TTL(pmp)) 1220 pn->pn_va_timeout = 0; 1221 1222 PUFFS_MSG_ALLOC(vn, setattr); 1223 (void)memcpy(&setattr_msg->pvnr_va, vap, sizeof(struct vattr)); 1224 puffs_credcvt(&setattr_msg->pvnr_cred, cred); 1225 puffs_msg_setinfo(park_setattr, PUFFSOP_VN, 1226 PUFFS_VN_SETATTR, VPTOPNC(vp)); 1227 if (flags & SETATTR_ASYNC) 1228 puffs_msg_setfaf(park_setattr); 1229 1230 puffs_msg_enqueue(pmp, park_setattr); 1231 if ((flags & SETATTR_ASYNC) == 0) 1232 error = puffs_msg_wait2(pmp, park_setattr, vp->v_data, NULL); 1233 1234 if ((error == 0) && PUFFS_USE_FS_TTL(pmp)) { 1235 struct timespec *va_ttl = &setattr_msg->pvnr_va_ttl; 1236 struct vattr *rvap = &setattr_msg->pvnr_va; 1237 1238 update_va(vp, NULL, rvap, va_ttl, NULL, flags); 1239 } 1240 1241 PUFFS_MSG_RELEASE(setattr); 1242 if ((flags & SETATTR_ASYNC) == 0) { 1243 error = checkerr(pmp, error, __func__); 1244 if (error) 1245 return error; 1246 } else { 1247 error = 0; 1248 } 1249 1250 if (vap->va_size != VNOVAL) { 1251 /* 1252 * If we truncated the file, make sure the data beyond 1253 * EOF in last page does not remain in cache, otherwise 1254 * if the file is later truncated to a larger size (creating 1255 * a hole), that area will not return zeroes as it 1256 * should. 1257 */ 1258 if ((flags & SETATTR_CHSIZE) && PUFFS_USE_PAGECACHE(pmp) && 1259 (vap->va_size < oldsize)) 1260 zerofill_lastpage(vp, vap->va_size); 1261 1262 pn->pn_serversize = vap->va_size; 1263 if (flags & SETATTR_CHSIZE) 1264 uvm_vnp_setsize(vp, vap->va_size); 1265 } 1266 1267 return 0; 1268 } 1269 1270 int 1271 puffs_vnop_setattr(void *v) 1272 { 1273 struct vop_getattr_args /* { 1274 const struct vnodeop_desc *a_desc; 1275 struct vnode *a_vp; 1276 struct vattr *a_vap; 1277 kauth_cred_t a_cred; 1278 } */ *ap = v; 1279 struct puffs_node *pn = ap->a_vp->v_data; 1280 int error; 1281 1282 mutex_enter(&pn->pn_sizemtx); 1283 error = dosetattr(ap->a_vp, ap->a_vap, ap->a_cred, SETATTR_CHSIZE); 1284 mutex_exit(&pn->pn_sizemtx); 1285 1286 return error; 1287 } 1288 1289 static __inline int 1290 doinact(struct puffs_mount *pmp, int iaflag) 1291 { 1292 1293 if (EXISTSOP(pmp, INACTIVE)) 1294 if (pmp->pmp_flags & PUFFS_KFLAG_IAONDEMAND) 1295 if (iaflag || ALLOPS(pmp)) 1296 return 1; 1297 else 1298 return 0; 1299 else 1300 return 1; 1301 else 1302 return 0; 1303 } 1304 1305 static void 1306 callinactive(struct puffs_mount *pmp, puffs_cookie_t ck, int iaflag) 1307 { 1308 PUFFS_MSG_VARS(vn, inactive); 1309 1310 if (doinact(pmp, iaflag)) { 1311 PUFFS_MSG_ALLOC(vn, inactive); 1312 puffs_msg_setinfo(park_inactive, PUFFSOP_VN, 1313 PUFFS_VN_INACTIVE, ck); 1314 PUFFS_MSG_ENQUEUEWAIT_NOERROR(pmp, park_inactive); 1315 PUFFS_MSG_RELEASE(inactive); 1316 } 1317 } 1318 1319 /* XXX: callinactive can't setback */ 1320 int 1321 puffs_vnop_inactive(void *v) 1322 { 1323 struct vop_inactive_args /* { 1324 const struct vnodeop_desc *a_desc; 1325 struct vnode *a_vp; 1326 } */ *ap = v; 1327 PUFFS_MSG_VARS(vn, inactive); 1328 struct vnode *vp = ap->a_vp; 1329 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 1330 struct puffs_node *pnode; 1331 bool recycle = false; 1332 1333 /* 1334 * When puffs_cookie2vnode() misses an entry, vcache_get() 1335 * creates a new node (puffs_vfsop_loadvnode being called to 1336 * initialize the PUFFS part), then it discovers it is VNON, 1337 * and tries to vrele() it. This leads us there, while the 1338 * cookie was stall and the node likely already reclaimed. 1339 */ 1340 if (vp->v_type == VNON) { 1341 VOP_UNLOCK(vp); 1342 return 0; 1343 } 1344 1345 pnode = vp->v_data; 1346 mutex_enter(&pnode->pn_sizemtx); 1347 1348 if (doinact(pmp, pnode->pn_stat & PNODE_DOINACT)) { 1349 flushvncache(vp, 0, 0, false); 1350 PUFFS_MSG_ALLOC(vn, inactive); 1351 puffs_msg_setinfo(park_inactive, PUFFSOP_VN, 1352 PUFFS_VN_INACTIVE, VPTOPNC(vp)); 1353 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_inactive, vp->v_data, 1354 NULL); 1355 PUFFS_MSG_RELEASE(inactive); 1356 } 1357 pnode->pn_stat &= ~PNODE_DOINACT; 1358 1359 /* 1360 * file server thinks it's gone? then don't be afraid care, 1361 * node's life was already all it would ever be 1362 */ 1363 if (pnode->pn_stat & PNODE_NOREFS) { 1364 pnode->pn_stat |= PNODE_DYING; 1365 recycle = true; 1366 } 1367 1368 /* 1369 * Handle node TTL. 1370 * If grace has already timed out, make it reclaimed. 1371 * Otherwise, we queue its expiration by sop thread, so 1372 * that it does not remain for ages in the freelist, 1373 * holding memory in userspace, while we will have 1374 * to look it up again anyway. 1375 */ 1376 if (PUFFS_USE_FS_TTL(pmp) && !(vp->v_vflag & VV_ROOT) && !recycle) { 1377 bool incache = !TIMED_OUT(pnode->pn_cn_timeout); 1378 bool ingrace = !TIMED_OUT(pnode->pn_cn_grace); 1379 bool reclaimqueued = pnode->pn_stat & PNODE_SOPEXP; 1380 1381 if (!incache && !ingrace && !reclaimqueued) { 1382 pnode->pn_stat |= PNODE_DYING; 1383 recycle = true; 1384 } 1385 1386 if (!recycle && !reclaimqueued) { 1387 struct puffs_sopreq *psopr; 1388 int at = MAX(pnode->pn_cn_grace, pnode->pn_cn_timeout); 1389 1390 KASSERT(curlwp != uvm.pagedaemon_lwp); 1391 psopr = kmem_alloc(sizeof(*psopr), KM_SLEEP); 1392 psopr->psopr_ck = VPTOPNC(pnode->pn_vp); 1393 psopr->psopr_sopreq = PUFFS_SOPREQ_EXPIRE; 1394 psopr->psopr_at = at; 1395 1396 mutex_enter(&pmp->pmp_sopmtx); 1397 1398 /* 1399 * If thread has disapeared, just give up. The 1400 * fs is being unmounted and the node will be 1401 * be reclaimed anyway. 1402 * 1403 * Otherwise, we queue the request but do not 1404 * immediatly signal the thread, as the node 1405 * has not been expired yet. 1406 */ 1407 if (pmp->pmp_sopthrcount == 0) { 1408 kmem_free(psopr, sizeof(*psopr)); 1409 } else { 1410 TAILQ_INSERT_TAIL(&pmp->pmp_sopnodereqs, 1411 psopr, psopr_entries); 1412 pnode->pn_stat |= PNODE_SOPEXP; 1413 } 1414 1415 mutex_exit(&pmp->pmp_sopmtx); 1416 } 1417 } 1418 1419 *ap->a_recycle = recycle; 1420 1421 mutex_exit(&pnode->pn_sizemtx); 1422 VOP_UNLOCK(vp); 1423 1424 return 0; 1425 } 1426 1427 static void 1428 callreclaim(struct puffs_mount *pmp, puffs_cookie_t ck, int nlookup) 1429 { 1430 PUFFS_MSG_VARS(vn, reclaim); 1431 1432 if (!EXISTSOP(pmp, RECLAIM)) 1433 return; 1434 1435 PUFFS_MSG_ALLOC(vn, reclaim); 1436 reclaim_msg->pvnr_nlookup = nlookup; 1437 puffs_msg_setfaf(park_reclaim); 1438 puffs_msg_setinfo(park_reclaim, PUFFSOP_VN, PUFFS_VN_RECLAIM, ck); 1439 1440 puffs_msg_enqueue(pmp, park_reclaim); 1441 PUFFS_MSG_RELEASE(reclaim); 1442 return; 1443 } 1444 1445 /* 1446 * always FAF, we don't really care if the server wants to fail to 1447 * reclaim the node or not 1448 */ 1449 int 1450 puffs_vnop_reclaim(void *v) 1451 { 1452 struct vop_reclaim_args /* { 1453 const struct vnodeop_desc *a_desc; 1454 struct vnode *a_vp; 1455 } */ *ap = v; 1456 struct vnode *vp = ap->a_vp; 1457 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 1458 bool notifyserver = true; 1459 1460 /* 1461 * first things first: check if someone is trying to reclaim the 1462 * root vnode. do not allow that to travel to userspace. 1463 * Note that we don't need to take the lock similarly to 1464 * puffs_root(), since there is only one of us. 1465 */ 1466 if (vp->v_vflag & VV_ROOT) { 1467 mutex_enter(&pmp->pmp_lock); 1468 KASSERT(pmp->pmp_root != NULL); 1469 pmp->pmp_root = NULL; 1470 mutex_exit(&pmp->pmp_lock); 1471 notifyserver = false; 1472 } 1473 1474 /* 1475 * purge info from kernel before issueing FAF, since we 1476 * don't really know when we'll get around to it after 1477 * that and someone might race us into node creation 1478 */ 1479 mutex_enter(&pmp->pmp_lock); 1480 if (PUFFS_USE_NAMECACHE(pmp)) 1481 cache_purge(vp); 1482 mutex_exit(&pmp->pmp_lock); 1483 1484 if (notifyserver) { 1485 int nlookup = VPTOPP(vp)->pn_nlookup; 1486 1487 callreclaim(MPTOPUFFSMP(vp->v_mount), VPTOPNC(vp), nlookup); 1488 } 1489 1490 if (PUFFS_USE_DOTDOTCACHE(pmp)) { 1491 if (__predict_true(VPTOPP(vp)->pn_parent != NULL)) 1492 vrele(VPTOPP(vp)->pn_parent); 1493 else 1494 KASSERT(vp->v_type == VNON || (vp->v_vflag & VV_ROOT)); 1495 } 1496 1497 puffs_putvnode(vp); 1498 1499 return 0; 1500 } 1501 1502 #define CSIZE sizeof(**ap->a_cookies) 1503 int 1504 puffs_vnop_readdir(void *v) 1505 { 1506 struct vop_readdir_args /* { 1507 const struct vnodeop_desc *a_desc; 1508 struct vnode *a_vp; 1509 struct uio *a_uio; 1510 kauth_cred_t a_cred; 1511 int *a_eofflag; 1512 off_t **a_cookies; 1513 int *a_ncookies; 1514 } */ *ap = v; 1515 PUFFS_MSG_VARS(vn, readdir); 1516 struct vnode *vp = ap->a_vp; 1517 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 1518 size_t argsize, tomove, cookiemem, cookiesmax; 1519 struct uio *uio = ap->a_uio; 1520 size_t howmuch, resid; 1521 int error; 1522 1523 /* 1524 * ok, so we need: resid + cookiemem = maxreq 1525 * => resid + cookiesize * (resid/minsize) = maxreq 1526 * => resid + cookiesize/minsize * resid = maxreq 1527 * => (cookiesize/minsize + 1) * resid = maxreq 1528 * => resid = maxreq / (cookiesize/minsize + 1) 1529 * 1530 * Since cookiesize <= minsize and we're not very big on floats, 1531 * we approximate that to be 1. Therefore: 1532 * 1533 * resid = maxreq / 2; 1534 * 1535 * Well, at least we didn't have to use differential equations 1536 * or the Gram-Schmidt process. 1537 * 1538 * (yes, I'm very afraid of this) 1539 */ 1540 KASSERT(CSIZE <= _DIRENT_MINSIZE((struct dirent *)0)); 1541 1542 if (ap->a_cookies) { 1543 KASSERT(ap->a_ncookies != NULL); 1544 if (pmp->pmp_args.pa_fhsize == 0) 1545 return EOPNOTSUPP; 1546 resid = PUFFS_TOMOVE(uio->uio_resid, pmp) / 2; 1547 cookiesmax = resid/_DIRENT_MINSIZE((struct dirent *)0); 1548 cookiemem = ALIGN(cookiesmax*CSIZE); /* play safe */ 1549 } else { 1550 resid = PUFFS_TOMOVE(uio->uio_resid, pmp); 1551 cookiesmax = 0; 1552 cookiemem = 0; 1553 } 1554 1555 argsize = sizeof(struct puffs_vnmsg_readdir); 1556 tomove = resid + cookiemem; 1557 puffs_msgmem_alloc(argsize + tomove, &park_readdir, 1558 (void *)&readdir_msg, 1); 1559 1560 puffs_credcvt(&readdir_msg->pvnr_cred, ap->a_cred); 1561 readdir_msg->pvnr_offset = uio->uio_offset; 1562 readdir_msg->pvnr_resid = resid; 1563 readdir_msg->pvnr_ncookies = cookiesmax; 1564 readdir_msg->pvnr_eofflag = 0; 1565 readdir_msg->pvnr_dentoff = cookiemem; 1566 puffs_msg_setinfo(park_readdir, PUFFSOP_VN, 1567 PUFFS_VN_READDIR, VPTOPNC(vp)); 1568 puffs_msg_setdelta(park_readdir, tomove); 1569 1570 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readdir, vp->v_data, NULL, error); 1571 error = checkerr(pmp, error, __func__); 1572 if (error) 1573 goto out; 1574 1575 /* userspace is cheating? */ 1576 if (readdir_msg->pvnr_resid > resid) { 1577 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG, 1578 "resid grew", VPTOPNC(vp)); 1579 ERROUT(EPROTO); 1580 } 1581 if (readdir_msg->pvnr_ncookies > cookiesmax) { 1582 puffs_senderr(pmp, PUFFS_ERR_READDIR, E2BIG, 1583 "too many cookies", VPTOPNC(vp)); 1584 ERROUT(EPROTO); 1585 } 1586 1587 /* check eof */ 1588 if (readdir_msg->pvnr_eofflag) 1589 *ap->a_eofflag = 1; 1590 1591 /* bouncy-wouncy with the directory data */ 1592 howmuch = resid - readdir_msg->pvnr_resid; 1593 1594 /* force eof if no data was returned (getcwd() needs this) */ 1595 if (howmuch == 0) { 1596 *ap->a_eofflag = 1; 1597 goto out; 1598 } 1599 1600 error = uiomove(readdir_msg->pvnr_data + cookiemem, howmuch, uio); 1601 if (error) 1602 goto out; 1603 1604 /* provide cookies to caller if so desired */ 1605 if (ap->a_cookies) { 1606 KASSERT(curlwp != uvm.pagedaemon_lwp); 1607 *ap->a_cookies = malloc(readdir_msg->pvnr_ncookies*CSIZE, 1608 M_TEMP, M_WAITOK); 1609 *ap->a_ncookies = readdir_msg->pvnr_ncookies; 1610 memcpy(*ap->a_cookies, readdir_msg->pvnr_data, 1611 *ap->a_ncookies*CSIZE); 1612 } 1613 1614 /* next readdir starts here */ 1615 uio->uio_offset = readdir_msg->pvnr_offset; 1616 1617 out: 1618 puffs_msgmem_release(park_readdir); 1619 return error; 1620 } 1621 #undef CSIZE 1622 1623 /* 1624 * poll works by consuming the bitmask in pn_revents. If there are 1625 * events available, poll returns immediately. If not, it issues a 1626 * poll to userspace, selrecords itself and returns with no available 1627 * events. When the file server returns, it executes puffs_parkdone_poll(), 1628 * where available events are added to the bitmask. selnotify() is 1629 * then also executed by that function causing us to enter here again 1630 * and hopefully find the missing bits (unless someone got them first, 1631 * in which case it starts all over again). 1632 */ 1633 int 1634 puffs_vnop_poll(void *v) 1635 { 1636 struct vop_poll_args /* { 1637 const struct vnodeop_desc *a_desc; 1638 struct vnode *a_vp; 1639 int a_events; 1640 } */ *ap = v; 1641 PUFFS_MSG_VARS(vn, poll); 1642 struct vnode *vp = ap->a_vp; 1643 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 1644 struct puffs_node *pn = vp->v_data; 1645 int events; 1646 1647 if (EXISTSOP(pmp, POLL)) { 1648 mutex_enter(&pn->pn_mtx); 1649 events = pn->pn_revents & ap->a_events; 1650 if (events & ap->a_events) { 1651 pn->pn_revents &= ~ap->a_events; 1652 mutex_exit(&pn->pn_mtx); 1653 1654 return events; 1655 } else { 1656 puffs_referencenode(pn); 1657 mutex_exit(&pn->pn_mtx); 1658 1659 PUFFS_MSG_ALLOC(vn, poll); 1660 poll_msg->pvnr_events = ap->a_events; 1661 puffs_msg_setinfo(park_poll, PUFFSOP_VN, 1662 PUFFS_VN_POLL, VPTOPNC(vp)); 1663 puffs_msg_setcall(park_poll, puffs_parkdone_poll, pn); 1664 selrecord(curlwp, &pn->pn_sel); 1665 1666 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_poll, 1667 vp->v_data, NULL); 1668 PUFFS_MSG_RELEASE(poll); 1669 1670 return 0; 1671 } 1672 } else { 1673 return genfs_poll(v); 1674 } 1675 } 1676 1677 static int 1678 flushvncache(struct vnode *vp, off_t offlo, off_t offhi, bool wait) 1679 { 1680 struct puffs_node *pn = VPTOPP(vp); 1681 struct vattr va; 1682 int pflags, error; 1683 1684 /* flush out information from our metacache, see vop_setattr */ 1685 if (pn->pn_stat & PNODE_METACACHE_MASK 1686 && (pn->pn_stat & PNODE_DYING) == 0) { 1687 vattr_null(&va); 1688 error = dosetattr(vp, &va, FSCRED, 1689 SETATTR_CHSIZE | (wait ? 0 : SETATTR_ASYNC)); 1690 if (error) 1691 return error; 1692 } 1693 1694 /* 1695 * flush pages to avoid being overly dirty 1696 */ 1697 pflags = PGO_CLEANIT; 1698 if (wait) 1699 pflags |= PGO_SYNCIO; 1700 1701 mutex_enter(vp->v_interlock); 1702 return VOP_PUTPAGES(vp, trunc_page(offlo), round_page(offhi), pflags); 1703 } 1704 1705 int 1706 puffs_vnop_fsync(void *v) 1707 { 1708 struct vop_fsync_args /* { 1709 const struct vnodeop_desc *a_desc; 1710 struct vnode *a_vp; 1711 kauth_cred_t a_cred; 1712 int a_flags; 1713 off_t a_offlo; 1714 off_t a_offhi; 1715 } */ *ap = v; 1716 PUFFS_MSG_VARS(vn, fsync); 1717 struct vnode *vp; 1718 struct puffs_node *pn; 1719 struct puffs_mount *pmp; 1720 int error, dofaf; 1721 1722 vp = ap->a_vp; 1723 KASSERT(vp != NULL); 1724 pn = VPTOPP(vp); 1725 KASSERT(pn != NULL); 1726 pmp = MPTOPUFFSMP(vp->v_mount); 1727 if (ap->a_flags & FSYNC_WAIT) { 1728 mutex_enter(&pn->pn_sizemtx); 1729 } else { 1730 if (mutex_tryenter(&pn->pn_sizemtx) == 0) 1731 return EDEADLK; 1732 } 1733 1734 error = flushvncache(vp, ap->a_offlo, ap->a_offhi, 1735 (ap->a_flags & FSYNC_WAIT) == FSYNC_WAIT); 1736 if (error) 1737 goto out; 1738 1739 /* 1740 * HELLO! We exit already here if the user server does not 1741 * support fsync OR if we should call fsync for a node which 1742 * has references neither in the kernel or the fs server. 1743 * Otherwise we continue to issue fsync() forward. 1744 */ 1745 error = 0; 1746 if (!EXISTSOP(pmp, FSYNC) || (pn->pn_stat & PNODE_DYING)) 1747 goto out; 1748 1749 dofaf = (ap->a_flags & FSYNC_WAIT) == 0 || ap->a_flags == FSYNC_LAZY; 1750 /* 1751 * We abuse VXLOCK to mean "vnode is going to die", so we issue 1752 * only FAFs for those. Otherwise there's a danger of deadlock, 1753 * since the execution context here might be the user server 1754 * doing some operation on another fs, which in turn caused a 1755 * vnode to be reclaimed from the freelist for this fs. 1756 */ 1757 if (dofaf == 0) { 1758 mutex_enter(vp->v_interlock); 1759 if (vdead_check(vp, VDEAD_NOWAIT) != 0) 1760 dofaf = 1; 1761 mutex_exit(vp->v_interlock); 1762 } 1763 1764 PUFFS_MSG_ALLOC(vn, fsync); 1765 if (dofaf) 1766 puffs_msg_setfaf(park_fsync); 1767 1768 puffs_credcvt(&fsync_msg->pvnr_cred, ap->a_cred); 1769 fsync_msg->pvnr_flags = ap->a_flags; 1770 fsync_msg->pvnr_offlo = ap->a_offlo; 1771 fsync_msg->pvnr_offhi = ap->a_offhi; 1772 puffs_msg_setinfo(park_fsync, PUFFSOP_VN, 1773 PUFFS_VN_FSYNC, VPTOPNC(vp)); 1774 1775 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fsync, vp->v_data, NULL, error); 1776 PUFFS_MSG_RELEASE(fsync); 1777 1778 error = checkerr(pmp, error, __func__); 1779 1780 out: 1781 mutex_exit(&pn->pn_sizemtx); 1782 return error; 1783 } 1784 1785 int 1786 puffs_vnop_seek(void *v) 1787 { 1788 struct vop_seek_args /* { 1789 const struct vnodeop_desc *a_desc; 1790 struct vnode *a_vp; 1791 off_t a_oldoff; 1792 off_t a_newoff; 1793 kauth_cred_t a_cred; 1794 } */ *ap = v; 1795 PUFFS_MSG_VARS(vn, seek); 1796 struct vnode *vp = ap->a_vp; 1797 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 1798 int error; 1799 1800 PUFFS_MSG_ALLOC(vn, seek); 1801 seek_msg->pvnr_oldoff = ap->a_oldoff; 1802 seek_msg->pvnr_newoff = ap->a_newoff; 1803 puffs_credcvt(&seek_msg->pvnr_cred, ap->a_cred); 1804 puffs_msg_setinfo(park_seek, PUFFSOP_VN, 1805 PUFFS_VN_SEEK, VPTOPNC(vp)); 1806 1807 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_seek, vp->v_data, NULL, error); 1808 PUFFS_MSG_RELEASE(seek); 1809 return checkerr(pmp, error, __func__); 1810 } 1811 1812 static int 1813 callremove(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck, 1814 struct componentname *cnp) 1815 { 1816 PUFFS_MSG_VARS(vn, remove); 1817 int error; 1818 1819 PUFFS_MSG_ALLOC(vn, remove); 1820 remove_msg->pvnr_cookie_targ = ck; 1821 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred, 1822 cnp, PUFFS_USE_FULLPNBUF(pmp)); 1823 puffs_msg_setinfo(park_remove, PUFFSOP_VN, PUFFS_VN_REMOVE, dck); 1824 1825 PUFFS_MSG_ENQUEUEWAIT(pmp, park_remove, error); 1826 PUFFS_MSG_RELEASE(remove); 1827 1828 return checkerr(pmp, error, __func__); 1829 } 1830 1831 /* 1832 * XXX: can't use callremove now because can't catch setbacks with 1833 * it due to lack of a pnode argument. 1834 */ 1835 int 1836 puffs_vnop_remove(void *v) 1837 { 1838 struct vop_remove_args /* { 1839 const struct vnodeop_desc *a_desc; 1840 struct vnode *a_dvp; 1841 struct vnode *a_vp; 1842 struct componentname *a_cnp; 1843 } */ *ap = v; 1844 PUFFS_MSG_VARS(vn, remove); 1845 struct vnode *dvp = ap->a_dvp; 1846 struct vnode *vp = ap->a_vp; 1847 struct puffs_node *dpn = VPTOPP(dvp); 1848 struct puffs_node *pn = VPTOPP(vp); 1849 struct componentname *cnp = ap->a_cnp; 1850 struct mount *mp = dvp->v_mount; 1851 struct puffs_mount *pmp = MPTOPUFFSMP(mp); 1852 int error; 1853 1854 PUFFS_MSG_ALLOC(vn, remove); 1855 remove_msg->pvnr_cookie_targ = VPTOPNC(vp); 1856 puffs_makecn(&remove_msg->pvnr_cn, &remove_msg->pvnr_cn_cred, 1857 cnp, PUFFS_USE_FULLPNBUF(pmp)); 1858 puffs_msg_setinfo(park_remove, PUFFSOP_VN, 1859 PUFFS_VN_REMOVE, VPTOPNC(dvp)); 1860 1861 puffs_msg_enqueue(pmp, park_remove); 1862 REFPN_AND_UNLOCKVP(dvp, dpn); 1863 if (dvp == vp) 1864 REFPN(pn); 1865 else 1866 REFPN_AND_UNLOCKVP(vp, pn); 1867 error = puffs_msg_wait2(pmp, park_remove, dpn, pn); 1868 1869 PUFFS_MSG_RELEASE(remove); 1870 1871 puffs_updatenode(VPTOPP(dvp), PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0); 1872 1873 RELEPN_AND_VP(dvp, dpn); 1874 RELEPN_AND_VP(vp, pn); 1875 1876 error = checkerr(pmp, error, __func__); 1877 return error; 1878 } 1879 1880 int 1881 puffs_vnop_mkdir(void *v) 1882 { 1883 struct vop_mkdir_v3_args /* { 1884 const struct vnodeop_desc *a_desc; 1885 struct vnode *a_dvp; 1886 struct vnode **a_vpp; 1887 struct componentname *a_cnp; 1888 struct vattr *a_vap; 1889 } */ *ap = v; 1890 PUFFS_MSG_VARS(vn, mkdir); 1891 struct vnode *dvp = ap->a_dvp; 1892 struct puffs_node *dpn = VPTOPP(dvp); 1893 struct componentname *cnp = ap->a_cnp; 1894 struct mount *mp = dvp->v_mount; 1895 struct puffs_mount *pmp = MPTOPUFFSMP(mp); 1896 int error; 1897 1898 PUFFS_MSG_ALLOC(vn, mkdir); 1899 puffs_makecn(&mkdir_msg->pvnr_cn, &mkdir_msg->pvnr_cn_cred, 1900 cnp, PUFFS_USE_FULLPNBUF(pmp)); 1901 mkdir_msg->pvnr_va = *ap->a_vap; 1902 puffs_msg_setinfo(park_mkdir, PUFFSOP_VN, 1903 PUFFS_VN_MKDIR, VPTOPNC(dvp)); 1904 1905 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mkdir, dvp->v_data, NULL, error); 1906 1907 error = checkerr(pmp, error, __func__); 1908 if (error) 1909 goto out; 1910 1911 error = puffs_newnode(mp, dvp, ap->a_vpp, 1912 mkdir_msg->pvnr_newnode, cnp, VDIR, 0); 1913 if (error) { 1914 puffs_abortbutton(pmp, PUFFS_ABORT_MKDIR, dpn->pn_cookie, 1915 mkdir_msg->pvnr_newnode, cnp); 1916 goto out; 1917 } 1918 1919 if (PUFFS_USE_FS_TTL(pmp)) { 1920 struct timespec *va_ttl = &mkdir_msg->pvnr_va_ttl; 1921 struct timespec *cn_ttl = &mkdir_msg->pvnr_cn_ttl; 1922 struct vattr *rvap = &mkdir_msg->pvnr_va; 1923 1924 update_va(*ap->a_vpp, NULL, rvap, 1925 va_ttl, cn_ttl, SETATTR_CHSIZE); 1926 } 1927 1928 VPTOPP(*ap->a_vpp)->pn_nlookup++; 1929 1930 if (PUFFS_USE_DOTDOTCACHE(pmp) && 1931 (VPTOPP(*ap->a_vpp)->pn_parent != dvp)) 1932 update_parent(*ap->a_vpp, dvp); 1933 1934 out: 1935 PUFFS_MSG_RELEASE(mkdir); 1936 return error; 1937 } 1938 1939 static int 1940 callrmdir(struct puffs_mount *pmp, puffs_cookie_t dck, puffs_cookie_t ck, 1941 struct componentname *cnp) 1942 { 1943 PUFFS_MSG_VARS(vn, rmdir); 1944 int error; 1945 1946 PUFFS_MSG_ALLOC(vn, rmdir); 1947 rmdir_msg->pvnr_cookie_targ = ck; 1948 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred, 1949 cnp, PUFFS_USE_FULLPNBUF(pmp)); 1950 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, PUFFS_VN_RMDIR, dck); 1951 1952 PUFFS_MSG_ENQUEUEWAIT(pmp, park_rmdir, error); 1953 PUFFS_MSG_RELEASE(rmdir); 1954 1955 return checkerr(pmp, error, __func__); 1956 } 1957 1958 int 1959 puffs_vnop_rmdir(void *v) 1960 { 1961 struct vop_rmdir_args /* { 1962 const struct vnodeop_desc *a_desc; 1963 struct vnode *a_dvp; 1964 struct vnode *a_vp; 1965 struct componentname *a_cnp; 1966 } */ *ap = v; 1967 PUFFS_MSG_VARS(vn, rmdir); 1968 struct vnode *dvp = ap->a_dvp; 1969 struct vnode *vp = ap->a_vp; 1970 struct puffs_node *dpn = VPTOPP(dvp); 1971 struct puffs_node *pn = VPTOPP(vp); 1972 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount); 1973 struct componentname *cnp = ap->a_cnp; 1974 int error; 1975 1976 PUFFS_MSG_ALLOC(vn, rmdir); 1977 rmdir_msg->pvnr_cookie_targ = VPTOPNC(vp); 1978 puffs_makecn(&rmdir_msg->pvnr_cn, &rmdir_msg->pvnr_cn_cred, 1979 cnp, PUFFS_USE_FULLPNBUF(pmp)); 1980 puffs_msg_setinfo(park_rmdir, PUFFSOP_VN, 1981 PUFFS_VN_RMDIR, VPTOPNC(dvp)); 1982 1983 puffs_msg_enqueue(pmp, park_rmdir); 1984 REFPN_AND_UNLOCKVP(dvp, dpn); 1985 REFPN_AND_UNLOCKVP(vp, pn); 1986 error = puffs_msg_wait2(pmp, park_rmdir, dpn, pn); 1987 1988 PUFFS_MSG_RELEASE(rmdir); 1989 1990 puffs_updatenode(VPTOPP(dvp), PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0); 1991 1992 /* XXX: some call cache_purge() *for both vnodes* here, investigate */ 1993 RELEPN_AND_VP(dvp, dpn); 1994 RELEPN_AND_VP(vp, pn); 1995 1996 return error; 1997 } 1998 1999 int 2000 puffs_vnop_link(void *v) 2001 { 2002 struct vop_link_args /* { 2003 const struct vnodeop_desc *a_desc; 2004 struct vnode *a_dvp; 2005 struct vnode *a_vp; 2006 struct componentname *a_cnp; 2007 } */ *ap = v; 2008 PUFFS_MSG_VARS(vn, link); 2009 struct vnode *dvp = ap->a_dvp; 2010 struct vnode *vp = ap->a_vp; 2011 struct puffs_node *dpn = VPTOPP(dvp); 2012 struct puffs_node *pn = VPTOPP(vp); 2013 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount); 2014 struct componentname *cnp = ap->a_cnp; 2015 int error; 2016 2017 PUFFS_MSG_ALLOC(vn, link); 2018 link_msg->pvnr_cookie_targ = VPTOPNC(vp); 2019 puffs_makecn(&link_msg->pvnr_cn, &link_msg->pvnr_cn_cred, 2020 cnp, PUFFS_USE_FULLPNBUF(pmp)); 2021 puffs_msg_setinfo(park_link, PUFFSOP_VN, 2022 PUFFS_VN_LINK, VPTOPNC(dvp)); 2023 2024 puffs_msg_enqueue(pmp, park_link); 2025 REFPN_AND_UNLOCKVP(dvp, dpn); 2026 REFPN(pn); 2027 error = puffs_msg_wait2(pmp, park_link, dpn, pn); 2028 2029 PUFFS_MSG_RELEASE(link); 2030 2031 error = checkerr(pmp, error, __func__); 2032 2033 /* 2034 * XXX: stay in touch with the cache. I don't like this, but 2035 * don't have a better solution either. See also puffs_rename(). 2036 */ 2037 if (error == 0) { 2038 puffs_updatenode(pn, PUFFS_UPDATECTIME, 0); 2039 puffs_updatenode(VPTOPP(dvp), 2040 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0); 2041 } 2042 2043 RELEPN_AND_VP(dvp, dpn); 2044 puffs_releasenode(pn); 2045 2046 return error; 2047 } 2048 2049 int 2050 puffs_vnop_symlink(void *v) 2051 { 2052 struct vop_symlink_v3_args /* { 2053 const struct vnodeop_desc *a_desc; 2054 struct vnode *a_dvp; 2055 struct vnode **a_vpp; 2056 struct componentname *a_cnp; 2057 struct vattr *a_vap; 2058 char *a_target; 2059 } */ *ap = v; 2060 PUFFS_MSG_VARS(vn, symlink); 2061 struct vnode *dvp = ap->a_dvp; 2062 struct puffs_node *dpn = VPTOPP(dvp); 2063 struct mount *mp = dvp->v_mount; 2064 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount); 2065 struct componentname *cnp = ap->a_cnp; 2066 int error; 2067 2068 *ap->a_vpp = NULL; 2069 2070 PUFFS_MSG_ALLOC(vn, symlink); 2071 puffs_makecn(&symlink_msg->pvnr_cn, &symlink_msg->pvnr_cn_cred, 2072 cnp, PUFFS_USE_FULLPNBUF(pmp)); 2073 symlink_msg->pvnr_va = *ap->a_vap; 2074 (void)strlcpy(symlink_msg->pvnr_link, ap->a_target, 2075 sizeof(symlink_msg->pvnr_link)); 2076 puffs_msg_setinfo(park_symlink, PUFFSOP_VN, 2077 PUFFS_VN_SYMLINK, VPTOPNC(dvp)); 2078 2079 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_symlink, dvp->v_data, NULL, error); 2080 2081 error = checkerr(pmp, error, __func__); 2082 if (error) 2083 goto out; 2084 2085 error = puffs_newnode(mp, dvp, ap->a_vpp, 2086 symlink_msg->pvnr_newnode, cnp, VLNK, 0); 2087 if (error) { 2088 puffs_abortbutton(pmp, PUFFS_ABORT_SYMLINK, dpn->pn_cookie, 2089 symlink_msg->pvnr_newnode, cnp); 2090 goto out; 2091 } 2092 2093 if (PUFFS_USE_FS_TTL(pmp)) { 2094 struct timespec *va_ttl = &symlink_msg->pvnr_va_ttl; 2095 struct timespec *cn_ttl = &symlink_msg->pvnr_cn_ttl; 2096 struct vattr *rvap = &symlink_msg->pvnr_va; 2097 2098 update_va(*ap->a_vpp, NULL, rvap, 2099 va_ttl, cn_ttl, SETATTR_CHSIZE); 2100 } 2101 2102 VPTOPP(*ap->a_vpp)->pn_nlookup++; 2103 2104 if (PUFFS_USE_DOTDOTCACHE(pmp) && 2105 (VPTOPP(*ap->a_vpp)->pn_parent != dvp)) 2106 update_parent(*ap->a_vpp, dvp); 2107 2108 out: 2109 PUFFS_MSG_RELEASE(symlink); 2110 2111 return error; 2112 } 2113 2114 int 2115 puffs_vnop_readlink(void *v) 2116 { 2117 struct vop_readlink_args /* { 2118 const struct vnodeop_desc *a_desc; 2119 struct vnode *a_vp; 2120 struct uio *a_uio; 2121 kauth_cred_t a_cred; 2122 } */ *ap = v; 2123 PUFFS_MSG_VARS(vn, readlink); 2124 struct vnode *vp = ap->a_vp; 2125 struct puffs_mount *pmp = MPTOPUFFSMP(ap->a_vp->v_mount); 2126 size_t linklen; 2127 int error; 2128 2129 PUFFS_MSG_ALLOC(vn, readlink); 2130 puffs_credcvt(&readlink_msg->pvnr_cred, ap->a_cred); 2131 linklen = sizeof(readlink_msg->pvnr_link); 2132 readlink_msg->pvnr_linklen = linklen; 2133 puffs_msg_setinfo(park_readlink, PUFFSOP_VN, 2134 PUFFS_VN_READLINK, VPTOPNC(vp)); 2135 2136 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_readlink, vp->v_data, NULL, error); 2137 error = checkerr(pmp, error, __func__); 2138 if (error) 2139 goto out; 2140 2141 /* bad bad user file server */ 2142 if (readlink_msg->pvnr_linklen > linklen) { 2143 puffs_senderr(pmp, PUFFS_ERR_READLINK, E2BIG, 2144 "linklen too big", VPTOPNC(ap->a_vp)); 2145 error = EPROTO; 2146 goto out; 2147 } 2148 2149 error = uiomove(&readlink_msg->pvnr_link, readlink_msg->pvnr_linklen, 2150 ap->a_uio); 2151 out: 2152 PUFFS_MSG_RELEASE(readlink); 2153 return error; 2154 } 2155 2156 int 2157 puffs_vnop_rename(void *v) 2158 { 2159 struct vop_rename_args /* { 2160 const struct vnodeop_desc *a_desc; 2161 struct vnode *a_fdvp; 2162 struct vnode *a_fvp; 2163 struct componentname *a_fcnp; 2164 struct vnode *a_tdvp; 2165 struct vnode *a_tvp; 2166 struct componentname *a_tcnp; 2167 } */ *ap = v; 2168 PUFFS_MSG_VARS(vn, rename); 2169 struct vnode *fdvp = ap->a_fdvp, *fvp = ap->a_fvp; 2170 struct vnode *tdvp = ap->a_tdvp, *tvp = ap->a_tvp; 2171 struct puffs_node *fpn = ap->a_fvp->v_data; 2172 struct puffs_mount *pmp = MPTOPUFFSMP(fdvp->v_mount); 2173 int error; 2174 bool doabort = true; 2175 2176 if ((fvp->v_mount != tdvp->v_mount) || 2177 (tvp && (fvp->v_mount != tvp->v_mount))) { 2178 ERROUT(EXDEV); 2179 } 2180 2181 PUFFS_MSG_ALLOC(vn, rename); 2182 rename_msg->pvnr_cookie_src = VPTOPNC(fvp); 2183 rename_msg->pvnr_cookie_targdir = VPTOPNC(tdvp); 2184 if (tvp) 2185 rename_msg->pvnr_cookie_targ = VPTOPNC(tvp); 2186 else 2187 rename_msg->pvnr_cookie_targ = NULL; 2188 puffs_makecn(&rename_msg->pvnr_cn_src, &rename_msg->pvnr_cn_src_cred, 2189 ap->a_fcnp, PUFFS_USE_FULLPNBUF(pmp)); 2190 puffs_makecn(&rename_msg->pvnr_cn_targ, &rename_msg->pvnr_cn_targ_cred, 2191 ap->a_tcnp, PUFFS_USE_FULLPNBUF(pmp)); 2192 puffs_msg_setinfo(park_rename, PUFFSOP_VN, 2193 PUFFS_VN_RENAME, VPTOPNC(fdvp)); 2194 2195 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rename, fdvp->v_data, NULL, error); 2196 doabort = false; 2197 PUFFS_MSG_RELEASE(rename); 2198 error = checkerr(pmp, error, __func__); 2199 2200 /* 2201 * XXX: stay in touch with the cache. I don't like this, but 2202 * don't have a better solution either. See also puffs_link(). 2203 */ 2204 if (error == 0) { 2205 puffs_updatenode(fpn, PUFFS_UPDATECTIME, 0); 2206 puffs_updatenode(VPTOPP(fdvp), 2207 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 0); 2208 if (fdvp != tdvp) 2209 puffs_updatenode(VPTOPP(tdvp), 2210 PUFFS_UPDATECTIME|PUFFS_UPDATEMTIME, 2211 0); 2212 2213 if (PUFFS_USE_DOTDOTCACHE(pmp) && 2214 (VPTOPP(fvp)->pn_parent != tdvp)) 2215 update_parent(fvp, tdvp); 2216 } 2217 2218 2219 out: 2220 if (doabort) 2221 VOP_ABORTOP(tdvp, ap->a_tcnp); 2222 if (tvp != NULL) 2223 vput(tvp); 2224 if (tdvp == tvp) 2225 vrele(tdvp); 2226 else 2227 vput(tdvp); 2228 2229 if (doabort) 2230 VOP_ABORTOP(fdvp, ap->a_fcnp); 2231 vrele(fdvp); 2232 vrele(fvp); 2233 2234 return error; 2235 } 2236 2237 #define RWARGS(cont, iofl, move, offset, creds) \ 2238 (cont)->pvnr_ioflag = (iofl); \ 2239 (cont)->pvnr_resid = (move); \ 2240 (cont)->pvnr_offset = (offset); \ 2241 puffs_credcvt(&(cont)->pvnr_cred, creds) 2242 2243 int 2244 puffs_vnop_read(void *v) 2245 { 2246 struct vop_read_args /* { 2247 const struct vnodeop_desc *a_desc; 2248 struct vnode *a_vp; 2249 struct uio *a_uio; 2250 int a_ioflag; 2251 kauth_cred_t a_cred; 2252 } */ *ap = v; 2253 PUFFS_MSG_VARS(vn, read); 2254 struct vnode *vp = ap->a_vp; 2255 struct puffs_node *pn = VPTOPP(vp); 2256 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2257 struct uio *uio = ap->a_uio; 2258 size_t tomove, argsize; 2259 vsize_t bytelen; 2260 int error; 2261 2262 read_msg = NULL; 2263 error = 0; 2264 2265 /* std sanity */ 2266 if (uio->uio_resid == 0) 2267 return 0; 2268 if (uio->uio_offset < 0) 2269 return EFBIG; 2270 2271 if (vp->v_type == VREG && 2272 PUFFS_USE_PAGECACHE(pmp) && 2273 !(pn->pn_stat & PNODE_RDIRECT)) { 2274 const int advice = IO_ADV_DECODE(ap->a_ioflag); 2275 2276 while (uio->uio_resid > 0) { 2277 if (vp->v_size <= uio->uio_offset) { 2278 break; 2279 } 2280 bytelen = MIN(uio->uio_resid, 2281 vp->v_size - uio->uio_offset); 2282 if (bytelen == 0) 2283 break; 2284 2285 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice, 2286 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp)); 2287 if (error) 2288 break; 2289 } 2290 2291 if ((vp->v_mount->mnt_flag & MNT_NOATIME) == 0) 2292 puffs_updatenode(VPTOPP(vp), PUFFS_UPDATEATIME, 0); 2293 } else { 2294 /* 2295 * in case it's not a regular file or we're operating 2296 * uncached, do read in the old-fashioned style, 2297 * i.e. explicit read operations 2298 */ 2299 2300 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp); 2301 argsize = sizeof(struct puffs_vnmsg_read); 2302 puffs_msgmem_alloc(argsize + tomove, &park_read, 2303 (void *)&read_msg, 1); 2304 2305 error = 0; 2306 while (uio->uio_resid > 0) { 2307 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp); 2308 memset(read_msg, 0, argsize); /* XXX: touser KASSERT */ 2309 RWARGS(read_msg, ap->a_ioflag, tomove, 2310 uio->uio_offset, ap->a_cred); 2311 puffs_msg_setinfo(park_read, PUFFSOP_VN, 2312 PUFFS_VN_READ, VPTOPNC(vp)); 2313 puffs_msg_setdelta(park_read, tomove); 2314 2315 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_read, vp->v_data, 2316 NULL, error); 2317 error = checkerr(pmp, error, __func__); 2318 if (error) 2319 break; 2320 2321 if (read_msg->pvnr_resid > tomove) { 2322 puffs_senderr(pmp, PUFFS_ERR_READ, 2323 E2BIG, "resid grew", VPTOPNC(ap->a_vp)); 2324 error = EPROTO; 2325 break; 2326 } 2327 2328 error = uiomove(read_msg->pvnr_data, 2329 tomove - read_msg->pvnr_resid, uio); 2330 2331 /* 2332 * in case the file is out of juice, resid from 2333 * userspace is != 0. and the error-case is 2334 * quite obvious 2335 */ 2336 if (error || read_msg->pvnr_resid) 2337 break; 2338 } 2339 2340 puffs_msgmem_release(park_read); 2341 } 2342 2343 return error; 2344 } 2345 2346 /* 2347 * XXX: in case of a failure, this leaves uio in a bad state. 2348 * We could theoretically copy the uio and iovecs and "replay" 2349 * them the right amount after the userspace trip, but don't 2350 * bother for now. 2351 */ 2352 int 2353 puffs_vnop_write(void *v) 2354 { 2355 struct vop_write_args /* { 2356 const struct vnodeop_desc *a_desc; 2357 struct vnode *a_vp; 2358 struct uio *a_uio; 2359 int a_ioflag; 2360 kauth_cred_t a_cred; 2361 } */ *ap = v; 2362 PUFFS_MSG_VARS(vn, write); 2363 struct vnode *vp = ap->a_vp; 2364 struct puffs_node *pn = VPTOPP(vp); 2365 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2366 struct uio *uio = ap->a_uio; 2367 size_t tomove, argsize; 2368 off_t oldoff, newoff, origoff; 2369 vsize_t bytelen; 2370 int error, uflags; 2371 int ubcflags; 2372 2373 error = uflags = 0; 2374 write_msg = NULL; 2375 2376 /* std sanity */ 2377 if (uio->uio_resid == 0) 2378 return 0; 2379 if (uio->uio_offset < 0) 2380 return EFBIG; 2381 2382 mutex_enter(&pn->pn_sizemtx); 2383 2384 if (vp->v_type == VREG && 2385 PUFFS_USE_PAGECACHE(pmp) && 2386 !(pn->pn_stat & PNODE_WDIRECT)) { 2387 ubcflags = UBC_WRITE | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp); 2388 2389 /* 2390 * userspace *should* be allowed to control this, 2391 * but with UBC it's a bit unclear how to handle it 2392 */ 2393 if (ap->a_ioflag & IO_APPEND) 2394 uio->uio_offset = vp->v_size; 2395 2396 origoff = uio->uio_offset; 2397 while (uio->uio_resid > 0) { 2398 oldoff = uio->uio_offset; 2399 bytelen = uio->uio_resid; 2400 2401 newoff = oldoff + bytelen; 2402 if (vp->v_size < newoff) { 2403 uvm_vnp_setwritesize(vp, newoff); 2404 } 2405 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, 2406 UVM_ADV_RANDOM, ubcflags); 2407 2408 /* 2409 * In case of a ubc_uiomove() error, 2410 * opt to not extend the file at all and 2411 * return an error. Otherwise, if we attempt 2412 * to clear the memory we couldn't fault to, 2413 * we might generate a kernel page fault. 2414 */ 2415 if (vp->v_size < newoff) { 2416 if (error == 0) { 2417 uflags |= PUFFS_UPDATESIZE; 2418 uvm_vnp_setsize(vp, newoff); 2419 } else { 2420 uvm_vnp_setwritesize(vp, vp->v_size); 2421 } 2422 } 2423 if (error) 2424 break; 2425 2426 /* 2427 * If we're writing large files, flush to file server 2428 * every 64k. Otherwise we can very easily exhaust 2429 * kernel and user memory, as the file server cannot 2430 * really keep up with our writing speed. 2431 * 2432 * Note: this does *NOT* honor MNT_ASYNC, because 2433 * that gives userland too much say in the kernel. 2434 */ 2435 if (oldoff >> 16 != uio->uio_offset >> 16) { 2436 mutex_enter(vp->v_interlock); 2437 error = VOP_PUTPAGES(vp, oldoff & ~0xffff, 2438 uio->uio_offset & ~0xffff, 2439 PGO_CLEANIT | PGO_SYNCIO); 2440 if (error) 2441 break; 2442 } 2443 } 2444 2445 /* synchronous I/O? */ 2446 if (error == 0 && ap->a_ioflag & IO_SYNC) { 2447 mutex_enter(vp->v_interlock); 2448 error = VOP_PUTPAGES(vp, trunc_page(origoff), 2449 round_page(uio->uio_offset), 2450 PGO_CLEANIT | PGO_SYNCIO); 2451 2452 /* write through page cache? */ 2453 } else if (error == 0 && pmp->pmp_flags & PUFFS_KFLAG_WTCACHE) { 2454 mutex_enter(vp->v_interlock); 2455 error = VOP_PUTPAGES(vp, trunc_page(origoff), 2456 round_page(uio->uio_offset), PGO_CLEANIT); 2457 } 2458 } else { 2459 /* tomove is non-increasing */ 2460 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp); 2461 argsize = sizeof(struct puffs_vnmsg_write) + tomove; 2462 puffs_msgmem_alloc(argsize, &park_write, (void *)&write_msg,1); 2463 2464 while (uio->uio_resid > 0) { 2465 /* move data to buffer */ 2466 tomove = PUFFS_TOMOVE(uio->uio_resid, pmp); 2467 memset(write_msg, 0, argsize); /* XXX: touser KASSERT */ 2468 RWARGS(write_msg, ap->a_ioflag, tomove, 2469 uio->uio_offset, ap->a_cred); 2470 error = uiomove(write_msg->pvnr_data, tomove, uio); 2471 if (error) 2472 break; 2473 2474 /* move buffer to userspace */ 2475 puffs_msg_setinfo(park_write, PUFFSOP_VN, 2476 PUFFS_VN_WRITE, VPTOPNC(vp)); 2477 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_write, vp->v_data, 2478 NULL, error); 2479 error = checkerr(pmp, error, __func__); 2480 if (error) 2481 break; 2482 2483 if (write_msg->pvnr_resid > tomove) { 2484 puffs_senderr(pmp, PUFFS_ERR_WRITE, 2485 E2BIG, "resid grew", VPTOPNC(ap->a_vp)); 2486 error = EPROTO; 2487 break; 2488 } 2489 2490 /* adjust file size */ 2491 if (vp->v_size < uio->uio_offset) { 2492 uflags |= PUFFS_UPDATESIZE; 2493 uvm_vnp_setsize(vp, uio->uio_offset); 2494 } 2495 2496 /* didn't move everything? bad userspace. bail */ 2497 if (write_msg->pvnr_resid != 0) { 2498 error = EIO; 2499 break; 2500 } 2501 } 2502 puffs_msgmem_release(park_write); 2503 } 2504 2505 if (vp->v_mount->mnt_flag & MNT_RELATIME) 2506 uflags |= PUFFS_UPDATEATIME; 2507 uflags |= PUFFS_UPDATECTIME; 2508 uflags |= PUFFS_UPDATEMTIME; 2509 puffs_updatenode(VPTOPP(vp), uflags, vp->v_size); 2510 2511 mutex_exit(&pn->pn_sizemtx); 2512 return error; 2513 } 2514 2515 int 2516 puffs_vnop_fallocate(void *v) 2517 { 2518 struct vop_fallocate_args /* { 2519 const struct vnodeop_desc *a_desc; 2520 struct vnode *a_vp; 2521 off_t a_pos; 2522 off_t a_len; 2523 } */ *ap = v; 2524 struct vnode *vp = ap->a_vp; 2525 struct puffs_node *pn = VPTOPP(vp); 2526 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2527 PUFFS_MSG_VARS(vn, fallocate); 2528 int error; 2529 2530 mutex_enter(&pn->pn_sizemtx); 2531 2532 PUFFS_MSG_ALLOC(vn, fallocate); 2533 fallocate_msg->pvnr_off = ap->a_pos; 2534 fallocate_msg->pvnr_len = ap->a_len; 2535 puffs_msg_setinfo(park_fallocate, PUFFSOP_VN, 2536 PUFFS_VN_FALLOCATE, VPTOPNC(vp)); 2537 2538 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fallocate, vp->v_data, NULL, error); 2539 error = checkerr(pmp, error, __func__); 2540 PUFFS_MSG_RELEASE(fallocate); 2541 2542 switch (error) { 2543 case 0: 2544 break; 2545 case EAGAIN: 2546 error = EIO; 2547 /* FALLTHROUGH */ 2548 default: 2549 goto out; 2550 } 2551 2552 if (ap->a_pos + ap->a_len > vp->v_size) { 2553 uvm_vnp_setsize(vp, ap->a_pos + ap->a_len); 2554 puffs_updatenode(pn, PUFFS_UPDATESIZE, vp->v_size); 2555 } 2556 out: 2557 mutex_exit(&pn->pn_sizemtx); 2558 2559 return error; 2560 } 2561 2562 int 2563 puffs_vnop_fdiscard(void *v) 2564 { 2565 struct vop_fdiscard_args /* { 2566 const struct vnodeop_desc *a_desc; 2567 struct vnode *a_vp; 2568 off_t a_pos; 2569 off_t a_len; 2570 } */ *ap = v; 2571 struct vnode *vp = ap->a_vp; 2572 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2573 PUFFS_MSG_VARS(vn, fdiscard); 2574 int error; 2575 2576 PUFFS_MSG_ALLOC(vn, fdiscard); 2577 fdiscard_msg->pvnr_off = ap->a_pos; 2578 fdiscard_msg->pvnr_len = ap->a_len; 2579 puffs_msg_setinfo(park_fdiscard, PUFFSOP_VN, 2580 PUFFS_VN_FALLOCATE, VPTOPNC(vp)); 2581 2582 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_fdiscard, vp->v_data, NULL, error); 2583 error = checkerr(pmp, error, __func__); 2584 PUFFS_MSG_RELEASE(fdiscard); 2585 2586 return error; 2587 } 2588 2589 int 2590 puffs_vnop_print(void *v) 2591 { 2592 struct vop_print_args /* { 2593 struct vnode *a_vp; 2594 } */ *ap = v; 2595 PUFFS_MSG_VARS(vn, print); 2596 struct vnode *vp = ap->a_vp; 2597 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2598 struct puffs_node *pn = vp->v_data; 2599 2600 /* kernel portion */ 2601 printf("tag VT_PUFFS, vnode %p, puffs node: %p,\n" 2602 "\tuserspace cookie: %p", vp, pn, pn->pn_cookie); 2603 if (vp->v_type == VFIFO) 2604 VOCALL(fifo_vnodeop_p, VOFFSET(vop_print), v); 2605 printf("\n"); 2606 2607 /* userspace portion */ 2608 if (EXISTSOP(pmp, PRINT)) { 2609 PUFFS_MSG_ALLOC(vn, print); 2610 puffs_msg_setinfo(park_print, PUFFSOP_VN, 2611 PUFFS_VN_PRINT, VPTOPNC(vp)); 2612 PUFFS_MSG_ENQUEUEWAIT2_NOERROR(pmp, park_print, vp->v_data, 2613 NULL); 2614 PUFFS_MSG_RELEASE(print); 2615 } 2616 2617 return 0; 2618 } 2619 2620 int 2621 puffs_vnop_pathconf(void *v) 2622 { 2623 struct vop_pathconf_args /* { 2624 const struct vnodeop_desc *a_desc; 2625 struct vnode *a_vp; 2626 int a_name; 2627 register_t *a_retval; 2628 } */ *ap = v; 2629 PUFFS_MSG_VARS(vn, pathconf); 2630 struct vnode *vp = ap->a_vp; 2631 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2632 int error; 2633 2634 PUFFS_MSG_ALLOC(vn, pathconf); 2635 pathconf_msg->pvnr_name = ap->a_name; 2636 puffs_msg_setinfo(park_pathconf, PUFFSOP_VN, 2637 PUFFS_VN_PATHCONF, VPTOPNC(vp)); 2638 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_pathconf, vp->v_data, NULL, error); 2639 error = checkerr(pmp, error, __func__); 2640 if (!error) 2641 *ap->a_retval = pathconf_msg->pvnr_retval; 2642 PUFFS_MSG_RELEASE(pathconf); 2643 2644 return error; 2645 } 2646 2647 int 2648 puffs_vnop_advlock(void *v) 2649 { 2650 struct vop_advlock_args /* { 2651 const struct vnodeop_desc *a_desc; 2652 struct vnode *a_vp; 2653 void *a_id; 2654 int a_op; 2655 struct flock *a_fl; 2656 int a_flags; 2657 } */ *ap = v; 2658 PUFFS_MSG_VARS(vn, advlock); 2659 struct vnode *vp = ap->a_vp; 2660 struct puffs_node *pn = VPTOPP(vp); 2661 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2662 int error; 2663 2664 if (!EXISTSOP(pmp, ADVLOCK)) 2665 return lf_advlock(ap, &pn->pn_lockf, vp->v_size); 2666 2667 PUFFS_MSG_ALLOC(vn, advlock); 2668 (void)memcpy(&advlock_msg->pvnr_fl, ap->a_fl, 2669 sizeof(advlock_msg->pvnr_fl)); 2670 advlock_msg->pvnr_id = ap->a_id; 2671 advlock_msg->pvnr_op = ap->a_op; 2672 advlock_msg->pvnr_flags = ap->a_flags; 2673 puffs_msg_setinfo(park_advlock, PUFFSOP_VN, 2674 PUFFS_VN_ADVLOCK, VPTOPNC(vp)); 2675 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_advlock, vp->v_data, NULL, error); 2676 error = checkerr(pmp, error, __func__); 2677 PUFFS_MSG_RELEASE(advlock); 2678 2679 return error; 2680 } 2681 2682 int 2683 puffs_vnop_abortop(void *v) 2684 { 2685 struct vop_abortop_args /* { 2686 struct vnode *a_dvp; 2687 struct componentname *a_cnp; 2688 }; */ *ap = v; 2689 PUFFS_MSG_VARS(vn, abortop); 2690 struct vnode *dvp = ap->a_dvp; 2691 struct puffs_mount *pmp = MPTOPUFFSMP(dvp->v_mount); 2692 struct componentname *cnp = ap->a_cnp; 2693 2694 if (EXISTSOP(pmp, ABORTOP)) { 2695 PUFFS_MSG_ALLOC(vn, abortop); 2696 puffs_makecn(&abortop_msg->pvnr_cn, &abortop_msg->pvnr_cn_cred, 2697 cnp, PUFFS_USE_FULLPNBUF(pmp)); 2698 puffs_msg_setfaf(park_abortop); 2699 puffs_msg_setinfo(park_abortop, PUFFSOP_VN, 2700 PUFFS_VN_ABORTOP, VPTOPNC(dvp)); 2701 2702 puffs_msg_enqueue(pmp, park_abortop); 2703 PUFFS_MSG_RELEASE(abortop); 2704 } 2705 2706 return genfs_abortop(v); 2707 } 2708 2709 #define BIOASYNC(bp) (bp->b_flags & B_ASYNC) 2710 2711 /* 2712 * This maps itself to PUFFS_VN_READ/WRITE for data transfer. 2713 */ 2714 int 2715 puffs_vnop_strategy(void *v) 2716 { 2717 struct vop_strategy_args /* { 2718 const struct vnodeop_desc *a_desc; 2719 struct vnode *a_vp; 2720 struct buf *a_bp; 2721 } */ *ap = v; 2722 PUFFS_MSG_VARS(vn, rw); 2723 struct vnode *vp = ap->a_vp; 2724 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2725 struct puffs_node *pn; 2726 struct buf *bp; 2727 size_t argsize; 2728 size_t tomove, moved; 2729 int error, dofaf, cansleep, dobiodone; 2730 2731 pmp = MPTOPUFFSMP(vp->v_mount); 2732 bp = ap->a_bp; 2733 error = 0; 2734 dofaf = 0; 2735 cansleep = 0; 2736 pn = VPTOPP(vp); 2737 park_rw = NULL; /* explicit */ 2738 dobiodone = 1; 2739 2740 if ((BUF_ISREAD(bp) && !EXISTSOP(pmp, READ)) 2741 || (BUF_ISWRITE(bp) && !EXISTSOP(pmp, WRITE))) 2742 ERROUT(EOPNOTSUPP); 2743 2744 /* 2745 * Short-circuit optimization: don't flush buffer in between 2746 * VOP_INACTIVE and VOP_RECLAIM in case the node has no references. 2747 */ 2748 if (pn->pn_stat & PNODE_DYING) { 2749 KASSERT(BUF_ISWRITE(bp)); 2750 bp->b_resid = 0; 2751 goto out; 2752 } 2753 2754 #ifdef DIAGNOSTIC 2755 if (bp->b_bcount > pmp->pmp_msg_maxsize - PUFFS_MSGSTRUCT_MAX) 2756 panic("puffs_strategy: wildly inappropriate buf bcount %d", 2757 bp->b_bcount); 2758 #endif 2759 2760 /* 2761 * See explanation for the necessity of a FAF in puffs_fsync. 2762 * 2763 * Also, do FAF in case we're suspending. 2764 * See puffs_vfsops.c:pageflush() 2765 */ 2766 if (BUF_ISWRITE(bp)) { 2767 mutex_enter(vp->v_interlock); 2768 if (vdead_check(vp, VDEAD_NOWAIT) != 0) 2769 dofaf = 1; 2770 if (pn->pn_stat & PNODE_FAF) 2771 dofaf = 1; 2772 mutex_exit(vp->v_interlock); 2773 } 2774 2775 cansleep = (curlwp == uvm.pagedaemon_lwp || dofaf) ? 0 : 1; 2776 2777 KASSERT(curlwp != uvm.pagedaemon_lwp || dofaf || BIOASYNC(bp)); 2778 2779 /* allocate transport structure */ 2780 tomove = PUFFS_TOMOVE(bp->b_bcount, pmp); 2781 argsize = sizeof(struct puffs_vnmsg_rw); 2782 error = puffs_msgmem_alloc(argsize + tomove, &park_rw, 2783 (void *)&rw_msg, cansleep); 2784 if (error) 2785 goto out; 2786 RWARGS(rw_msg, 0, tomove, bp->b_blkno << DEV_BSHIFT, FSCRED); 2787 2788 /* 2x2 cases: read/write, faf/nofaf */ 2789 if (BUF_ISREAD(bp)) { 2790 puffs_msg_setinfo(park_rw, PUFFSOP_VN, 2791 PUFFS_VN_READ, VPTOPNC(vp)); 2792 puffs_msg_setdelta(park_rw, tomove); 2793 if (BIOASYNC(bp)) { 2794 puffs_msg_setcall(park_rw, 2795 puffs_parkdone_asyncbioread, bp); 2796 puffs_msg_enqueue(pmp, park_rw); 2797 dobiodone = 0; 2798 } else { 2799 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, 2800 NULL, error); 2801 error = checkerr(pmp, error, __func__); 2802 if (error) 2803 goto out; 2804 2805 if (rw_msg->pvnr_resid > tomove) { 2806 puffs_senderr(pmp, PUFFS_ERR_READ, 2807 E2BIG, "resid grew", VPTOPNC(vp)); 2808 ERROUT(EPROTO); 2809 } 2810 2811 moved = tomove - rw_msg->pvnr_resid; 2812 2813 (void)memcpy(bp->b_data, rw_msg->pvnr_data, moved); 2814 bp->b_resid = bp->b_bcount - moved; 2815 } 2816 } else { 2817 puffs_msg_setinfo(park_rw, PUFFSOP_VN, 2818 PUFFS_VN_WRITE, VPTOPNC(vp)); 2819 /* 2820 * make pages read-only before we write them if we want 2821 * write caching info 2822 */ 2823 if (PUFFS_WCACHEINFO(pmp)) { 2824 struct uvm_object *uobj = &vp->v_uobj; 2825 int npages = (bp->b_bcount + PAGE_SIZE-1) >> PAGE_SHIFT; 2826 struct vm_page *vmp; 2827 int i; 2828 2829 for (i = 0; i < npages; i++) { 2830 vmp= uvm_pageratop((vaddr_t)bp->b_data 2831 + (i << PAGE_SHIFT)); 2832 DPRINTF(("puffs_strategy: write-protecting " 2833 "vp %p page %p, offset %" PRId64"\n", 2834 vp, vmp, vmp->offset)); 2835 mutex_enter(uobj->vmobjlock); 2836 vmp->flags |= PG_RDONLY; 2837 pmap_page_protect(vmp, VM_PROT_READ); 2838 mutex_exit(uobj->vmobjlock); 2839 } 2840 } 2841 2842 (void)memcpy(&rw_msg->pvnr_data, bp->b_data, tomove); 2843 if (dofaf) { 2844 puffs_msg_setfaf(park_rw); 2845 } else if (BIOASYNC(bp)) { 2846 puffs_msg_setcall(park_rw, 2847 puffs_parkdone_asyncbiowrite, bp); 2848 dobiodone = 0; 2849 } 2850 2851 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_rw, vp->v_data, NULL, error); 2852 2853 if (dobiodone == 0) 2854 goto out; 2855 2856 error = checkerr(pmp, error, __func__); 2857 if (error) 2858 goto out; 2859 2860 if (rw_msg->pvnr_resid > tomove) { 2861 puffs_senderr(pmp, PUFFS_ERR_WRITE, 2862 E2BIG, "resid grew", VPTOPNC(vp)); 2863 ERROUT(EPROTO); 2864 } 2865 2866 /* 2867 * FAF moved everything. Frankly, we don't 2868 * really have a choice. 2869 */ 2870 if (dofaf && error == 0) 2871 moved = tomove; 2872 else 2873 moved = tomove - rw_msg->pvnr_resid; 2874 2875 bp->b_resid = bp->b_bcount - moved; 2876 if (bp->b_resid != 0) { 2877 ERROUT(EIO); 2878 } 2879 } 2880 2881 out: 2882 if (park_rw) 2883 puffs_msgmem_release(park_rw); 2884 2885 if (error) 2886 bp->b_error = error; 2887 2888 if (error || dobiodone) 2889 biodone(bp); 2890 2891 return error; 2892 } 2893 2894 int 2895 puffs_vnop_mmap(void *v) 2896 { 2897 struct vop_mmap_args /* { 2898 const struct vnodeop_desc *a_desc; 2899 struct vnode *a_vp; 2900 vm_prot_t a_prot; 2901 kauth_cred_t a_cred; 2902 } */ *ap = v; 2903 PUFFS_MSG_VARS(vn, mmap); 2904 struct vnode *vp = ap->a_vp; 2905 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 2906 int error; 2907 2908 if (!PUFFS_USE_PAGECACHE(pmp)) 2909 return genfs_eopnotsupp(v); 2910 2911 if (EXISTSOP(pmp, MMAP)) { 2912 PUFFS_MSG_ALLOC(vn, mmap); 2913 mmap_msg->pvnr_prot = ap->a_prot; 2914 puffs_credcvt(&mmap_msg->pvnr_cred, ap->a_cred); 2915 puffs_msg_setinfo(park_mmap, PUFFSOP_VN, 2916 PUFFS_VN_MMAP, VPTOPNC(vp)); 2917 2918 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_mmap, vp->v_data, NULL, error); 2919 error = checkerr(pmp, error, __func__); 2920 PUFFS_MSG_RELEASE(mmap); 2921 } else { 2922 error = genfs_mmap(v); 2923 } 2924 2925 return error; 2926 } 2927 2928 2929 /* 2930 * The rest don't get a free trip to userspace and back, they 2931 * have to stay within the kernel. 2932 */ 2933 2934 /* 2935 * bmap doesn't really make any sense for puffs, so just 1:1 map it. 2936 * well, maybe somehow, somewhere, some day .... 2937 */ 2938 int 2939 puffs_vnop_bmap(void *v) 2940 { 2941 struct vop_bmap_args /* { 2942 const struct vnodeop_desc *a_desc; 2943 struct vnode *a_vp; 2944 daddr_t a_bn; 2945 struct vnode **a_vpp; 2946 daddr_t *a_bnp; 2947 int *a_runp; 2948 } */ *ap = v; 2949 struct puffs_mount *pmp; 2950 2951 pmp = MPTOPUFFSMP(ap->a_vp->v_mount); 2952 2953 if (ap->a_vpp) 2954 *ap->a_vpp = ap->a_vp; 2955 if (ap->a_bnp) 2956 *ap->a_bnp = ap->a_bn; 2957 if (ap->a_runp) 2958 *ap->a_runp 2959 = (PUFFS_TOMOVE(pmp->pmp_msg_maxsize, pmp)>>DEV_BSHIFT) - 1; 2960 2961 return 0; 2962 } 2963 2964 /* 2965 * Handle getpages faults in puffs. We let genfs_getpages() do most 2966 * of the dirty work, but we come in this route to do accounting tasks. 2967 * If the user server has specified functions for cache notifications 2968 * about reads and/or writes, we record which type of operation we got, 2969 * for which page range, and proceed to issue a FAF notification to the 2970 * server about it. 2971 */ 2972 int 2973 puffs_vnop_getpages(void *v) 2974 { 2975 struct vop_getpages_args /* { 2976 const struct vnodeop_desc *a_desc; 2977 struct vnode *a_vp; 2978 voff_t a_offset; 2979 struct vm_page **a_m; 2980 int *a_count; 2981 int a_centeridx; 2982 vm_prot_t a_access_type; 2983 int a_advice; 2984 int a_flags; 2985 } */ *ap = v; 2986 struct puffs_mount *pmp; 2987 struct puffs_node *pn; 2988 struct vnode *vp; 2989 struct vm_page **pgs; 2990 struct puffs_cacheinfo *pcinfo = NULL; 2991 struct puffs_cacherun *pcrun; 2992 void *parkmem = NULL; 2993 size_t runsizes; 2994 int i, npages, si, streakon; 2995 int error, locked, write; 2996 2997 pmp = MPTOPUFFSMP(ap->a_vp->v_mount); 2998 npages = *ap->a_count; 2999 pgs = ap->a_m; 3000 vp = ap->a_vp; 3001 pn = vp->v_data; 3002 locked = (ap->a_flags & PGO_LOCKED) != 0; 3003 write = (ap->a_access_type & VM_PROT_WRITE) != 0; 3004 3005 /* ccg xnaht - gets Wuninitialized wrong */ 3006 pcrun = NULL; 3007 runsizes = 0; 3008 3009 /* 3010 * Check that we aren't trying to fault in pages which our file 3011 * server doesn't know about. This happens if we extend a file by 3012 * skipping some pages and later try to fault in pages which 3013 * are between pn_serversize and vp_size. This check optimizes 3014 * away the common case where a file is being extended. 3015 */ 3016 if (ap->a_offset >= pn->pn_serversize && ap->a_offset < vp->v_size) { 3017 struct vattr va; 3018 3019 /* try again later when we can block */ 3020 if (locked) 3021 ERROUT(EBUSY); 3022 3023 mutex_exit(vp->v_interlock); 3024 vattr_null(&va); 3025 va.va_size = vp->v_size; 3026 error = dosetattr(vp, &va, FSCRED, 0); 3027 if (error) 3028 ERROUT(error); 3029 mutex_enter(vp->v_interlock); 3030 } 3031 3032 if (write && PUFFS_WCACHEINFO(pmp)) { 3033 #ifdef notnowjohn 3034 /* allocate worst-case memory */ 3035 runsizes = ((npages / 2) + 1) * sizeof(struct puffs_cacherun); 3036 KASSERT(curlwp != uvm.pagedaemon_lwp || locked); 3037 pcinfo = kmem_zalloc(sizeof(struct puffs_cacheinfo) + runsize, 3038 locked ? KM_NOSLEEP : KM_SLEEP); 3039 3040 /* 3041 * can't block if we're locked and can't mess up caching 3042 * information for fs server. so come back later, please 3043 */ 3044 if (pcinfo == NULL) 3045 ERROUT(ENOMEM); 3046 3047 parkmem = puffs_park_alloc(locked == 0); 3048 if (parkmem == NULL) 3049 ERROUT(ENOMEM); 3050 3051 pcrun = pcinfo->pcache_runs; 3052 #else 3053 (void)parkmem; 3054 #endif 3055 } 3056 3057 error = genfs_getpages(v); 3058 if (error) 3059 goto out; 3060 3061 if (PUFFS_WCACHEINFO(pmp) == 0) 3062 goto out; 3063 3064 /* 3065 * Let's see whose fault it was and inform the user server of 3066 * possibly read/written pages. Map pages from read faults 3067 * strictly read-only, since otherwise we might miss info on 3068 * when the page is actually write-faulted to. 3069 */ 3070 if (!locked) 3071 mutex_enter(vp->v_uobj.vmobjlock); 3072 for (i = 0, si = 0, streakon = 0; i < npages; i++) { 3073 if (pgs[i] == NULL || pgs[i] == PGO_DONTCARE) { 3074 if (streakon && write) { 3075 streakon = 0; 3076 pcrun[si].pcache_runend 3077 = trunc_page(pgs[i]->offset) + PAGE_MASK; 3078 si++; 3079 } 3080 continue; 3081 } 3082 if (streakon == 0 && write) { 3083 streakon = 1; 3084 pcrun[si].pcache_runstart = pgs[i]->offset; 3085 } 3086 3087 if (!write) 3088 pgs[i]->flags |= PG_RDONLY; 3089 } 3090 /* was the last page part of our streak? */ 3091 if (streakon) { 3092 pcrun[si].pcache_runend 3093 = trunc_page(pgs[i-1]->offset) + PAGE_MASK; 3094 si++; 3095 } 3096 if (!locked) 3097 mutex_exit(vp->v_uobj.vmobjlock); 3098 3099 KASSERT(si <= (npages / 2) + 1); 3100 3101 #ifdef notnowjohn 3102 /* send results to userspace */ 3103 if (write) 3104 puffs_cacheop(pmp, parkmem, pcinfo, 3105 sizeof(struct puffs_cacheinfo) + runsizes, VPTOPNC(vp)); 3106 #endif 3107 3108 out: 3109 if (error) { 3110 if (pcinfo != NULL) 3111 kmem_free(pcinfo, 3112 sizeof(struct puffs_cacheinfo) + runsizes); 3113 #ifdef notnowjohn 3114 if (parkmem != NULL) 3115 puffs_park_release(parkmem, 1); 3116 #endif 3117 } 3118 3119 return error; 3120 } 3121 3122 /* 3123 * Extended attribute support. 3124 */ 3125 3126 int 3127 puffs_vnop_getextattr(void *v) 3128 { 3129 struct vop_getextattr_args /* 3130 struct vnode *a_vp; 3131 int a_attrnamespace; 3132 const char *a_name; 3133 struct uio *a_uio; 3134 size_t *a_size; 3135 kauth_cred_t a_cred; 3136 }; */ *ap = v; 3137 PUFFS_MSG_VARS(vn, getextattr); 3138 struct vnode *vp = ap->a_vp; 3139 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 3140 int attrnamespace = ap->a_attrnamespace; 3141 const char *name = ap->a_name; 3142 struct uio *uio = ap->a_uio; 3143 size_t *sizep = ap->a_size; 3144 size_t tomove, resid; 3145 int error; 3146 3147 if (uio) 3148 resid = uio->uio_resid; 3149 else 3150 resid = 0; 3151 3152 tomove = PUFFS_TOMOVE(resid, pmp); 3153 if (tomove != resid) { 3154 error = E2BIG; 3155 goto out; 3156 } 3157 3158 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_getextattr) + tomove, 3159 &park_getextattr, (void *)&getextattr_msg, 1); 3160 3161 getextattr_msg->pvnr_attrnamespace = attrnamespace; 3162 strlcpy(getextattr_msg->pvnr_attrname, name, 3163 sizeof(getextattr_msg->pvnr_attrname)); 3164 puffs_credcvt(&getextattr_msg->pvnr_cred, ap->a_cred); 3165 if (sizep) 3166 getextattr_msg->pvnr_datasize = 1; 3167 getextattr_msg->pvnr_resid = tomove; 3168 3169 puffs_msg_setinfo(park_getextattr, 3170 PUFFSOP_VN, PUFFS_VN_GETEXTATTR, VPTOPNC(vp)); 3171 puffs_msg_setdelta(park_getextattr, tomove); 3172 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_getextattr, vp->v_data, NULL, error); 3173 3174 error = checkerr(pmp, error, __func__); 3175 if (error) 3176 goto out; 3177 3178 resid = getextattr_msg->pvnr_resid; 3179 if (resid > tomove) { 3180 puffs_senderr(pmp, PUFFS_ERR_GETEXTATTR, E2BIG, 3181 "resid grew", VPTOPNC(vp)); 3182 error = EPROTO; 3183 goto out; 3184 } 3185 3186 if (sizep) 3187 *sizep = getextattr_msg->pvnr_datasize; 3188 if (uio) 3189 error = uiomove(getextattr_msg->pvnr_data, tomove - resid, uio); 3190 3191 out: 3192 PUFFS_MSG_RELEASE(getextattr); 3193 return error; 3194 } 3195 3196 int 3197 puffs_vnop_setextattr(void *v) 3198 { 3199 struct vop_setextattr_args /* { 3200 struct vnode *a_vp; 3201 int a_attrnamespace; 3202 const char *a_name; 3203 struct uio *a_uio; 3204 kauth_cred_t a_cred; 3205 }; */ *ap = v; 3206 PUFFS_MSG_VARS(vn, setextattr); 3207 struct vnode *vp = ap->a_vp; 3208 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 3209 int attrnamespace = ap->a_attrnamespace; 3210 const char *name = ap->a_name; 3211 struct uio *uio = ap->a_uio; 3212 size_t tomove, resid; 3213 int error; 3214 3215 if (uio) 3216 resid = uio->uio_resid; 3217 else 3218 resid = 0; 3219 3220 tomove = PUFFS_TOMOVE(resid, pmp); 3221 if (tomove != resid) { 3222 error = E2BIG; 3223 goto out; 3224 } 3225 3226 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_setextattr) + tomove, 3227 &park_setextattr, (void *)&setextattr_msg, 1); 3228 3229 setextattr_msg->pvnr_attrnamespace = attrnamespace; 3230 strlcpy(setextattr_msg->pvnr_attrname, name, 3231 sizeof(setextattr_msg->pvnr_attrname)); 3232 puffs_credcvt(&setextattr_msg->pvnr_cred, ap->a_cred); 3233 setextattr_msg->pvnr_resid = tomove; 3234 3235 if (uio) { 3236 error = uiomove(setextattr_msg->pvnr_data, tomove, uio); 3237 if (error) 3238 goto out; 3239 } 3240 3241 puffs_msg_setinfo(park_setextattr, 3242 PUFFSOP_VN, PUFFS_VN_SETEXTATTR, VPTOPNC(vp)); 3243 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_setextattr, vp->v_data, NULL, error); 3244 3245 error = checkerr(pmp, error, __func__); 3246 if (error) 3247 goto out; 3248 3249 if (setextattr_msg->pvnr_resid != 0) 3250 error = EIO; 3251 3252 out: 3253 PUFFS_MSG_RELEASE(setextattr); 3254 3255 return error; 3256 } 3257 3258 int 3259 puffs_vnop_listextattr(void *v) 3260 { 3261 struct vop_listextattr_args /* { 3262 struct vnode *a_vp; 3263 int a_attrnamespace; 3264 struct uio *a_uio; 3265 size_t *a_size; 3266 int a_flag, 3267 kauth_cred_t a_cred; 3268 }; */ *ap = v; 3269 PUFFS_MSG_VARS(vn, listextattr); 3270 struct vnode *vp = ap->a_vp; 3271 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 3272 int attrnamespace = ap->a_attrnamespace; 3273 struct uio *uio = ap->a_uio; 3274 size_t *sizep = ap->a_size; 3275 int flag = ap->a_flag; 3276 size_t tomove, resid; 3277 int error; 3278 3279 if (uio) 3280 resid = uio->uio_resid; 3281 else 3282 resid = 0; 3283 3284 tomove = PUFFS_TOMOVE(resid, pmp); 3285 if (tomove != resid) { 3286 error = E2BIG; 3287 goto out; 3288 } 3289 3290 puffs_msgmem_alloc(sizeof(struct puffs_vnmsg_listextattr) + tomove, 3291 &park_listextattr, (void *)&listextattr_msg, 1); 3292 3293 listextattr_msg->pvnr_attrnamespace = attrnamespace; 3294 listextattr_msg->pvnr_flag = flag; 3295 puffs_credcvt(&listextattr_msg->pvnr_cred, ap->a_cred); 3296 listextattr_msg->pvnr_resid = tomove; 3297 if (sizep) 3298 listextattr_msg->pvnr_datasize = 1; 3299 3300 puffs_msg_setinfo(park_listextattr, 3301 PUFFSOP_VN, PUFFS_VN_LISTEXTATTR, VPTOPNC(vp)); 3302 puffs_msg_setdelta(park_listextattr, tomove); 3303 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_listextattr, vp->v_data, NULL, error); 3304 3305 error = checkerr(pmp, error, __func__); 3306 if (error) 3307 goto out; 3308 3309 resid = listextattr_msg->pvnr_resid; 3310 if (resid > tomove) { 3311 puffs_senderr(pmp, PUFFS_ERR_LISTEXTATTR, E2BIG, 3312 "resid grew", VPTOPNC(vp)); 3313 error = EPROTO; 3314 goto out; 3315 } 3316 3317 if (sizep) 3318 *sizep = listextattr_msg->pvnr_datasize; 3319 if (uio) 3320 error = uiomove(listextattr_msg->pvnr_data, tomove-resid, uio); 3321 3322 out: 3323 PUFFS_MSG_RELEASE(listextattr); 3324 return error; 3325 } 3326 3327 int 3328 puffs_vnop_deleteextattr(void *v) 3329 { 3330 struct vop_deleteextattr_args /* { 3331 struct vnode *a_vp; 3332 int a_attrnamespace; 3333 const char *a_name; 3334 kauth_cred_t a_cred; 3335 }; */ *ap = v; 3336 PUFFS_MSG_VARS(vn, deleteextattr); 3337 struct vnode *vp = ap->a_vp; 3338 struct puffs_mount *pmp = MPTOPUFFSMP(vp->v_mount); 3339 int attrnamespace = ap->a_attrnamespace; 3340 const char *name = ap->a_name; 3341 int error; 3342 3343 PUFFS_MSG_ALLOC(vn, deleteextattr); 3344 deleteextattr_msg->pvnr_attrnamespace = attrnamespace; 3345 strlcpy(deleteextattr_msg->pvnr_attrname, name, 3346 sizeof(deleteextattr_msg->pvnr_attrname)); 3347 puffs_credcvt(&deleteextattr_msg->pvnr_cred, ap->a_cred); 3348 3349 puffs_msg_setinfo(park_deleteextattr, 3350 PUFFSOP_VN, PUFFS_VN_DELETEEXTATTR, VPTOPNC(vp)); 3351 PUFFS_MSG_ENQUEUEWAIT2(pmp, park_deleteextattr, 3352 vp->v_data, NULL, error); 3353 3354 error = checkerr(pmp, error, __func__); 3355 3356 PUFFS_MSG_RELEASE(deleteextattr); 3357 return error; 3358 } 3359 3360 /* 3361 * spec & fifo. These call the miscfs spec and fifo vectors, but issue 3362 * FAF update information for the puffs node first. 3363 */ 3364 int 3365 puffs_vnop_spec_read(void *v) 3366 { 3367 struct vop_read_args /* { 3368 const struct vnodeop_desc *a_desc; 3369 struct vnode *a_vp; 3370 struct uio *a_uio; 3371 int a_ioflag; 3372 kauth_cred_t a_cred; 3373 } */ *ap = v; 3374 3375 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0); 3376 return VOCALL(spec_vnodeop_p, VOFFSET(vop_read), v); 3377 } 3378 3379 int 3380 puffs_vnop_spec_write(void *v) 3381 { 3382 struct vop_write_args /* { 3383 const struct vnodeop_desc *a_desc; 3384 struct vnode *a_vp; 3385 struct uio *a_uio; 3386 int a_ioflag; 3387 kauth_cred_t a_cred; 3388 } */ *ap = v; 3389 3390 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0); 3391 return VOCALL(spec_vnodeop_p, VOFFSET(vop_write), v); 3392 } 3393 3394 int 3395 puffs_vnop_fifo_read(void *v) 3396 { 3397 struct vop_read_args /* { 3398 const struct vnodeop_desc *a_desc; 3399 struct vnode *a_vp; 3400 struct uio *a_uio; 3401 int a_ioflag; 3402 kauth_cred_t a_cred; 3403 } */ *ap = v; 3404 3405 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEATIME, 0); 3406 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_read), v); 3407 } 3408 3409 int 3410 puffs_vnop_fifo_write(void *v) 3411 { 3412 struct vop_write_args /* { 3413 const struct vnodeop_desc *a_desc; 3414 struct vnode *a_vp; 3415 struct uio *a_uio; 3416 int a_ioflag; 3417 kauth_cred_t a_cred; 3418 } */ *ap = v; 3419 3420 puffs_updatenode(VPTOPP(ap->a_vp), PUFFS_UPDATEMTIME, 0); 3421 return VOCALL(fifo_vnodeop_p, VOFFSET(vop_write), v); 3422 } 3423
Indexes created Tue Sep 30 06:09:59 GMT 2025