Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or modify it |
| 5 | * under the terms of version 2 of the GNU General Public License as |
| 6 | * published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it would be useful, but |
| 9 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. |
| 11 | * |
| 12 | * Further, this software is distributed without any warranty that it is |
| 13 | * free of the rightful claim of any third person regarding infringement |
| 14 | * or the like. Any license provided herein, whether implied or |
| 15 | * otherwise, applies only to this software file. Patent licenses, if |
| 16 | * any, provided herein do not apply to combinations of this program with |
| 17 | * other software, or any other product whatsoever. |
| 18 | * |
| 19 | * You should have received a copy of the GNU General Public License along |
| 20 | * with this program; if not, write the Free Software Foundation, Inc., 59 |
| 21 | * Temple Place - Suite 330, Boston MA 02111-1307, USA. |
| 22 | * |
| 23 | * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy, |
| 24 | * Mountain View, CA 94043, or: |
| 25 | * |
| 26 | * http://www.sgi.com |
| 27 | * |
| 28 | * For further information regarding this notice, see: |
| 29 | * |
| 30 | * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/ |
| 31 | */ |
| 32 | |
| 33 | #include "xfs.h" |
| 34 | |
| 35 | #include "xfs_macros.h" |
| 36 | #include "xfs_types.h" |
| 37 | #include "xfs_inum.h" |
| 38 | #include "xfs_log.h" |
| 39 | #include "xfs_trans.h" |
| 40 | #include "xfs_sb.h" |
| 41 | #include "xfs_ag.h" |
| 42 | #include "xfs_dir.h" |
| 43 | #include "xfs_dir2.h" |
| 44 | #include "xfs_dmapi.h" |
| 45 | #include "xfs_mount.h" |
| 46 | #include "xfs_alloc_btree.h" |
| 47 | #include "xfs_bmap_btree.h" |
| 48 | #include "xfs_ialloc_btree.h" |
| 49 | #include "xfs_btree.h" |
| 50 | #include "xfs_ialloc.h" |
| 51 | #include "xfs_attr_sf.h" |
| 52 | #include "xfs_dir_sf.h" |
| 53 | #include "xfs_dir2_sf.h" |
| 54 | #include "xfs_dinode.h" |
| 55 | #include "xfs_inode.h" |
| 56 | #include "xfs_quota.h" |
| 57 | #include "xfs_utils.h" |
| 58 | #include "xfs_bit.h" |
| 59 | |
| 60 | /* |
| 61 | * Initialize the inode hash table for the newly mounted file system. |
| 62 | * Choose an initial table size based on user specified value, else |
| 63 | * use a simple algorithm using the maximum number of inodes as an |
| 64 | * indicator for table size, and clamp it between one and some large |
| 65 | * number of pages. |
| 66 | */ |
| 67 | void |
| 68 | xfs_ihash_init(xfs_mount_t *mp) |
| 69 | { |
| 70 | __uint64_t icount; |
| 71 | uint i, flags = KM_SLEEP | KM_MAYFAIL; |
| 72 | |
| 73 | if (!mp->m_ihsize) { |
| 74 | icount = mp->m_maxicount ? mp->m_maxicount : |
| 75 | (mp->m_sb.sb_dblocks << mp->m_sb.sb_inopblog); |
| 76 | mp->m_ihsize = 1 << max_t(uint, 8, |
| 77 | (xfs_highbit64(icount) + 1) / 2); |
| 78 | mp->m_ihsize = min_t(uint, mp->m_ihsize, |
| 79 | (64 * NBPP) / sizeof(xfs_ihash_t)); |
| 80 | } |
| 81 | |
| 82 | while (!(mp->m_ihash = (xfs_ihash_t *)kmem_zalloc(mp->m_ihsize * |
| 83 | sizeof(xfs_ihash_t), flags))) { |
| 84 | if ((mp->m_ihsize >>= 1) <= NBPP) |
| 85 | flags = KM_SLEEP; |
| 86 | } |
| 87 | for (i = 0; i < mp->m_ihsize; i++) { |
| 88 | rwlock_init(&(mp->m_ihash[i].ih_lock)); |
| 89 | } |
| 90 | } |
| 91 | |
| 92 | /* |
| 93 | * Free up structures allocated by xfs_ihash_init, at unmount time. |
| 94 | */ |
| 95 | void |
| 96 | xfs_ihash_free(xfs_mount_t *mp) |
| 97 | { |
| 98 | kmem_free(mp->m_ihash, mp->m_ihsize*sizeof(xfs_ihash_t)); |
| 99 | mp->m_ihash = NULL; |
| 100 | } |
| 101 | |
| 102 | /* |
| 103 | * Initialize the inode cluster hash table for the newly mounted file system. |
| 104 | * Its size is derived from the ihash table size. |
| 105 | */ |
| 106 | void |
| 107 | xfs_chash_init(xfs_mount_t *mp) |
| 108 | { |
| 109 | uint i; |
| 110 | |
| 111 | mp->m_chsize = max_t(uint, 1, mp->m_ihsize / |
| 112 | (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)); |
| 113 | mp->m_chsize = min_t(uint, mp->m_chsize, mp->m_ihsize); |
| 114 | mp->m_chash = (xfs_chash_t *)kmem_zalloc(mp->m_chsize |
| 115 | * sizeof(xfs_chash_t), |
| 116 | KM_SLEEP); |
| 117 | for (i = 0; i < mp->m_chsize; i++) { |
| 118 | spinlock_init(&mp->m_chash[i].ch_lock,"xfshash"); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | /* |
| 123 | * Free up structures allocated by xfs_chash_init, at unmount time. |
| 124 | */ |
| 125 | void |
| 126 | xfs_chash_free(xfs_mount_t *mp) |
| 127 | { |
| 128 | int i; |
| 129 | |
| 130 | for (i = 0; i < mp->m_chsize; i++) { |
| 131 | spinlock_destroy(&mp->m_chash[i].ch_lock); |
| 132 | } |
| 133 | |
| 134 | kmem_free(mp->m_chash, mp->m_chsize*sizeof(xfs_chash_t)); |
| 135 | mp->m_chash = NULL; |
| 136 | } |
| 137 | |
| 138 | /* |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 139 | * Try to move an inode to the front of its hash list if possible |
| 140 | * (and if its not there already). Called right after obtaining |
| 141 | * the list version number and then dropping the read_lock on the |
| 142 | * hash list in question (which is done right after looking up the |
| 143 | * inode in question...). |
| 144 | */ |
| 145 | STATIC void |
| 146 | xfs_ihash_promote( |
| 147 | xfs_ihash_t *ih, |
| 148 | xfs_inode_t *ip, |
| 149 | ulong version) |
| 150 | { |
| 151 | xfs_inode_t *iq; |
| 152 | |
| 153 | if ((ip->i_prevp != &ih->ih_next) && write_trylock(&ih->ih_lock)) { |
| 154 | if (likely(version == ih->ih_version)) { |
| 155 | /* remove from list */ |
| 156 | if ((iq = ip->i_next)) { |
| 157 | iq->i_prevp = ip->i_prevp; |
| 158 | } |
| 159 | *ip->i_prevp = iq; |
| 160 | |
| 161 | /* insert at list head */ |
| 162 | iq = ih->ih_next; |
| 163 | iq->i_prevp = &ip->i_next; |
| 164 | ip->i_next = iq; |
| 165 | ip->i_prevp = &ih->ih_next; |
| 166 | ih->ih_next = ip; |
| 167 | } |
| 168 | write_unlock(&ih->ih_lock); |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | * Look up an inode by number in the given file system. |
| 174 | * The inode is looked up in the hash table for the file system |
| 175 | * represented by the mount point parameter mp. Each bucket of |
| 176 | * the hash table is guarded by an individual semaphore. |
| 177 | * |
| 178 | * If the inode is found in the hash table, its corresponding vnode |
| 179 | * is obtained with a call to vn_get(). This call takes care of |
| 180 | * coordination with the reclamation of the inode and vnode. Note |
| 181 | * that the vmap structure is filled in while holding the hash lock. |
| 182 | * This gives us the state of the inode/vnode when we found it and |
| 183 | * is used for coordination in vn_get(). |
| 184 | * |
| 185 | * If it is not in core, read it in from the file system's device and |
| 186 | * add the inode into the hash table. |
| 187 | * |
| 188 | * The inode is locked according to the value of the lock_flags parameter. |
| 189 | * This flag parameter indicates how and if the inode's IO lock and inode lock |
| 190 | * should be taken. |
| 191 | * |
| 192 | * mp -- the mount point structure for the current file system. It points |
| 193 | * to the inode hash table. |
| 194 | * tp -- a pointer to the current transaction if there is one. This is |
| 195 | * simply passed through to the xfs_iread() call. |
| 196 | * ino -- the number of the inode desired. This is the unique identifier |
| 197 | * within the file system for the inode being requested. |
| 198 | * lock_flags -- flags indicating how to lock the inode. See the comment |
| 199 | * for xfs_ilock() for a list of valid values. |
| 200 | * bno -- the block number starting the buffer containing the inode, |
| 201 | * if known (as by bulkstat), else 0. |
| 202 | */ |
| 203 | STATIC int |
| 204 | xfs_iget_core( |
| 205 | vnode_t *vp, |
| 206 | xfs_mount_t *mp, |
| 207 | xfs_trans_t *tp, |
| 208 | xfs_ino_t ino, |
| 209 | uint flags, |
| 210 | uint lock_flags, |
| 211 | xfs_inode_t **ipp, |
| 212 | xfs_daddr_t bno) |
| 213 | { |
| 214 | xfs_ihash_t *ih; |
| 215 | xfs_inode_t *ip; |
| 216 | xfs_inode_t *iq; |
| 217 | vnode_t *inode_vp; |
| 218 | ulong version; |
| 219 | int error; |
| 220 | /* REFERENCED */ |
| 221 | xfs_chash_t *ch; |
| 222 | xfs_chashlist_t *chl, *chlnew; |
| 223 | SPLDECL(s); |
| 224 | |
| 225 | |
| 226 | ih = XFS_IHASH(mp, ino); |
| 227 | |
| 228 | again: |
| 229 | read_lock(&ih->ih_lock); |
| 230 | |
| 231 | for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { |
| 232 | if (ip->i_ino == ino) { |
| 233 | /* |
| 234 | * If INEW is set this inode is being set up |
| 235 | * we need to pause and try again. |
| 236 | */ |
| 237 | if (ip->i_flags & XFS_INEW) { |
| 238 | read_unlock(&ih->ih_lock); |
| 239 | delay(1); |
| 240 | XFS_STATS_INC(xs_ig_frecycle); |
| 241 | |
| 242 | goto again; |
| 243 | } |
| 244 | |
| 245 | inode_vp = XFS_ITOV_NULL(ip); |
| 246 | if (inode_vp == NULL) { |
| 247 | /* |
| 248 | * If IRECLAIM is set this inode is |
| 249 | * on its way out of the system, |
| 250 | * we need to pause and try again. |
| 251 | */ |
| 252 | if (ip->i_flags & XFS_IRECLAIM) { |
| 253 | read_unlock(&ih->ih_lock); |
| 254 | delay(1); |
| 255 | XFS_STATS_INC(xs_ig_frecycle); |
| 256 | |
| 257 | goto again; |
| 258 | } |
| 259 | |
| 260 | vn_trace_exit(vp, "xfs_iget.alloc", |
| 261 | (inst_t *)__return_address); |
| 262 | |
| 263 | XFS_STATS_INC(xs_ig_found); |
| 264 | |
| 265 | ip->i_flags &= ~XFS_IRECLAIMABLE; |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 266 | version = ih->ih_version; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | read_unlock(&ih->ih_lock); |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 268 | xfs_ihash_promote(ih, ip, version); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | |
| 270 | XFS_MOUNT_ILOCK(mp); |
| 271 | list_del_init(&ip->i_reclaim); |
| 272 | XFS_MOUNT_IUNLOCK(mp); |
| 273 | |
| 274 | goto finish_inode; |
| 275 | |
| 276 | } else if (vp != inode_vp) { |
| 277 | struct inode *inode = LINVFS_GET_IP(inode_vp); |
| 278 | |
| 279 | /* The inode is being torn down, pause and |
| 280 | * try again. |
| 281 | */ |
| 282 | if (inode->i_state & (I_FREEING | I_CLEAR)) { |
| 283 | read_unlock(&ih->ih_lock); |
| 284 | delay(1); |
| 285 | XFS_STATS_INC(xs_ig_frecycle); |
| 286 | |
| 287 | goto again; |
| 288 | } |
| 289 | /* Chances are the other vnode (the one in the inode) is being torn |
| 290 | * down right now, and we landed on top of it. Question is, what do |
| 291 | * we do? Unhook the old inode and hook up the new one? |
| 292 | */ |
| 293 | cmn_err(CE_PANIC, |
| 294 | "xfs_iget_core: ambiguous vns: vp/0x%p, invp/0x%p", |
| 295 | inode_vp, vp); |
| 296 | } |
| 297 | |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 298 | /* |
| 299 | * Inode cache hit: if ip is not at the front of |
| 300 | * its hash chain, move it there now. |
| 301 | * Do this with the lock held for update, but |
| 302 | * do statistics after releasing the lock. |
| 303 | */ |
| 304 | version = ih->ih_version; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | read_unlock(&ih->ih_lock); |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 306 | xfs_ihash_promote(ih, ip, version); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | XFS_STATS_INC(xs_ig_found); |
| 308 | |
| 309 | finish_inode: |
| 310 | if (ip->i_d.di_mode == 0) { |
| 311 | if (!(flags & IGET_CREATE)) |
| 312 | return ENOENT; |
| 313 | xfs_iocore_inode_reinit(ip); |
| 314 | } |
| 315 | |
| 316 | if (lock_flags != 0) |
| 317 | xfs_ilock(ip, lock_flags); |
| 318 | |
| 319 | ip->i_flags &= ~XFS_ISTALE; |
| 320 | |
| 321 | vn_trace_exit(vp, "xfs_iget.found", |
| 322 | (inst_t *)__return_address); |
| 323 | goto return_ip; |
| 324 | } |
| 325 | } |
| 326 | |
| 327 | /* |
| 328 | * Inode cache miss: save the hash chain version stamp and unlock |
| 329 | * the chain, so we don't deadlock in vn_alloc. |
| 330 | */ |
| 331 | XFS_STATS_INC(xs_ig_missed); |
| 332 | |
| 333 | version = ih->ih_version; |
| 334 | |
| 335 | read_unlock(&ih->ih_lock); |
| 336 | |
| 337 | /* |
| 338 | * Read the disk inode attributes into a new inode structure and get |
| 339 | * a new vnode for it. This should also initialize i_ino and i_mount. |
| 340 | */ |
| 341 | error = xfs_iread(mp, tp, ino, &ip, bno); |
| 342 | if (error) { |
| 343 | return error; |
| 344 | } |
| 345 | |
| 346 | vn_trace_exit(vp, "xfs_iget.alloc", (inst_t *)__return_address); |
| 347 | |
| 348 | xfs_inode_lock_init(ip, vp); |
| 349 | xfs_iocore_inode_init(ip); |
| 350 | |
| 351 | if (lock_flags != 0) { |
| 352 | xfs_ilock(ip, lock_flags); |
| 353 | } |
| 354 | |
| 355 | if ((ip->i_d.di_mode == 0) && !(flags & IGET_CREATE)) { |
| 356 | xfs_idestroy(ip); |
| 357 | return ENOENT; |
| 358 | } |
| 359 | |
| 360 | /* |
| 361 | * Put ip on its hash chain, unless someone else hashed a duplicate |
| 362 | * after we released the hash lock. |
| 363 | */ |
| 364 | write_lock(&ih->ih_lock); |
| 365 | |
| 366 | if (ih->ih_version != version) { |
| 367 | for (iq = ih->ih_next; iq != NULL; iq = iq->i_next) { |
| 368 | if (iq->i_ino == ino) { |
| 369 | write_unlock(&ih->ih_lock); |
| 370 | xfs_idestroy(ip); |
| 371 | |
| 372 | XFS_STATS_INC(xs_ig_dup); |
| 373 | goto again; |
| 374 | } |
| 375 | } |
| 376 | } |
| 377 | |
| 378 | /* |
| 379 | * These values _must_ be set before releasing ihlock! |
| 380 | */ |
| 381 | ip->i_hash = ih; |
| 382 | if ((iq = ih->ih_next)) { |
| 383 | iq->i_prevp = &ip->i_next; |
| 384 | } |
| 385 | ip->i_next = iq; |
| 386 | ip->i_prevp = &ih->ih_next; |
| 387 | ih->ih_next = ip; |
| 388 | ip->i_udquot = ip->i_gdquot = NULL; |
| 389 | ih->ih_version++; |
| 390 | ip->i_flags |= XFS_INEW; |
| 391 | |
| 392 | write_unlock(&ih->ih_lock); |
| 393 | |
| 394 | /* |
| 395 | * put ip on its cluster's hash chain |
| 396 | */ |
| 397 | ASSERT(ip->i_chash == NULL && ip->i_cprev == NULL && |
| 398 | ip->i_cnext == NULL); |
| 399 | |
| 400 | chlnew = NULL; |
| 401 | ch = XFS_CHASH(mp, ip->i_blkno); |
| 402 | chlredo: |
| 403 | s = mutex_spinlock(&ch->ch_lock); |
| 404 | for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) { |
| 405 | if (chl->chl_blkno == ip->i_blkno) { |
| 406 | |
| 407 | /* insert this inode into the doubly-linked list |
| 408 | * where chl points */ |
| 409 | if ((iq = chl->chl_ip)) { |
| 410 | ip->i_cprev = iq->i_cprev; |
| 411 | iq->i_cprev->i_cnext = ip; |
| 412 | iq->i_cprev = ip; |
| 413 | ip->i_cnext = iq; |
| 414 | } else { |
| 415 | ip->i_cnext = ip; |
| 416 | ip->i_cprev = ip; |
| 417 | } |
| 418 | chl->chl_ip = ip; |
| 419 | ip->i_chash = chl; |
| 420 | break; |
| 421 | } |
| 422 | } |
| 423 | |
| 424 | /* no hash list found for this block; add a new hash list */ |
| 425 | if (chl == NULL) { |
| 426 | if (chlnew == NULL) { |
| 427 | mutex_spinunlock(&ch->ch_lock, s); |
| 428 | ASSERT(xfs_chashlist_zone != NULL); |
| 429 | chlnew = (xfs_chashlist_t *) |
| 430 | kmem_zone_alloc(xfs_chashlist_zone, |
| 431 | KM_SLEEP); |
| 432 | ASSERT(chlnew != NULL); |
| 433 | goto chlredo; |
| 434 | } else { |
| 435 | ip->i_cnext = ip; |
| 436 | ip->i_cprev = ip; |
| 437 | ip->i_chash = chlnew; |
| 438 | chlnew->chl_ip = ip; |
| 439 | chlnew->chl_blkno = ip->i_blkno; |
| 440 | chlnew->chl_next = ch->ch_list; |
| 441 | ch->ch_list = chlnew; |
| 442 | chlnew = NULL; |
| 443 | } |
| 444 | } else { |
| 445 | if (chlnew != NULL) { |
| 446 | kmem_zone_free(xfs_chashlist_zone, chlnew); |
| 447 | } |
| 448 | } |
| 449 | |
| 450 | mutex_spinunlock(&ch->ch_lock, s); |
| 451 | |
| 452 | |
| 453 | /* |
| 454 | * Link ip to its mount and thread it on the mount's inode list. |
| 455 | */ |
| 456 | XFS_MOUNT_ILOCK(mp); |
| 457 | if ((iq = mp->m_inodes)) { |
| 458 | ASSERT(iq->i_mprev->i_mnext == iq); |
| 459 | ip->i_mprev = iq->i_mprev; |
| 460 | iq->i_mprev->i_mnext = ip; |
| 461 | iq->i_mprev = ip; |
| 462 | ip->i_mnext = iq; |
| 463 | } else { |
| 464 | ip->i_mnext = ip; |
| 465 | ip->i_mprev = ip; |
| 466 | } |
| 467 | mp->m_inodes = ip; |
| 468 | |
| 469 | XFS_MOUNT_IUNLOCK(mp); |
| 470 | |
| 471 | return_ip: |
| 472 | ASSERT(ip->i_df.if_ext_max == |
| 473 | XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t)); |
| 474 | |
| 475 | ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) == |
| 476 | ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0)); |
| 477 | |
| 478 | *ipp = ip; |
| 479 | |
| 480 | /* |
| 481 | * If we have a real type for an on-disk inode, we can set ops(&unlock) |
| 482 | * now. If it's a new inode being created, xfs_ialloc will handle it. |
| 483 | */ |
| 484 | VFS_INIT_VNODE(XFS_MTOVFS(mp), vp, XFS_ITOBHV(ip), 1); |
| 485 | |
| 486 | return 0; |
| 487 | } |
| 488 | |
| 489 | |
| 490 | /* |
| 491 | * The 'normal' internal xfs_iget, if needed it will |
| 492 | * 'allocate', or 'get', the vnode. |
| 493 | */ |
| 494 | int |
| 495 | xfs_iget( |
| 496 | xfs_mount_t *mp, |
| 497 | xfs_trans_t *tp, |
| 498 | xfs_ino_t ino, |
| 499 | uint flags, |
| 500 | uint lock_flags, |
| 501 | xfs_inode_t **ipp, |
| 502 | xfs_daddr_t bno) |
| 503 | { |
| 504 | struct inode *inode; |
| 505 | vnode_t *vp = NULL; |
| 506 | int error; |
| 507 | |
| 508 | retry: |
| 509 | XFS_STATS_INC(xs_ig_attempts); |
| 510 | |
| 511 | if ((inode = iget_locked(XFS_MTOVFS(mp)->vfs_super, ino))) { |
| 512 | bhv_desc_t *bdp; |
| 513 | xfs_inode_t *ip; |
| 514 | int newnode; |
| 515 | |
| 516 | vp = LINVFS_GET_VP(inode); |
| 517 | if (inode->i_state & I_NEW) { |
| 518 | inode_allocate: |
| 519 | vn_initialize(inode); |
| 520 | error = xfs_iget_core(vp, mp, tp, ino, flags, |
| 521 | lock_flags, ipp, bno); |
| 522 | if (error) { |
| 523 | vn_mark_bad(vp); |
| 524 | if (inode->i_state & I_NEW) |
| 525 | unlock_new_inode(inode); |
| 526 | iput(inode); |
| 527 | } |
| 528 | } else { |
| 529 | /* These are true if the inode is in inactive or |
| 530 | * reclaim. The linux inode is about to go away, |
| 531 | * wait for that path to finish, and try again. |
| 532 | */ |
| 533 | if (vp->v_flag & (VINACT | VRECLM)) { |
| 534 | vn_wait(vp); |
| 535 | iput(inode); |
| 536 | goto retry; |
| 537 | } |
| 538 | |
| 539 | if (is_bad_inode(inode)) { |
| 540 | iput(inode); |
| 541 | return EIO; |
| 542 | } |
| 543 | |
| 544 | bdp = vn_bhv_lookup(VN_BHV_HEAD(vp), &xfs_vnodeops); |
| 545 | if (bdp == NULL) { |
| 546 | XFS_STATS_INC(xs_ig_dup); |
| 547 | goto inode_allocate; |
| 548 | } |
| 549 | ip = XFS_BHVTOI(bdp); |
| 550 | if (lock_flags != 0) |
| 551 | xfs_ilock(ip, lock_flags); |
| 552 | newnode = (ip->i_d.di_mode == 0); |
| 553 | if (newnode) |
| 554 | xfs_iocore_inode_reinit(ip); |
| 555 | XFS_STATS_INC(xs_ig_found); |
| 556 | *ipp = ip; |
| 557 | error = 0; |
| 558 | } |
| 559 | } else |
| 560 | error = ENOMEM; /* If we got no inode we are out of memory */ |
| 561 | |
| 562 | return error; |
| 563 | } |
| 564 | |
| 565 | /* |
| 566 | * Do the setup for the various locks within the incore inode. |
| 567 | */ |
| 568 | void |
| 569 | xfs_inode_lock_init( |
| 570 | xfs_inode_t *ip, |
| 571 | vnode_t *vp) |
| 572 | { |
| 573 | mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER, |
| 574 | "xfsino", (long)vp->v_number); |
| 575 | mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", vp->v_number); |
| 576 | init_waitqueue_head(&ip->i_ipin_wait); |
| 577 | atomic_set(&ip->i_pincount, 0); |
| 578 | init_sema(&ip->i_flock, 1, "xfsfino", vp->v_number); |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * Look for the inode corresponding to the given ino in the hash table. |
| 583 | * If it is there and its i_transp pointer matches tp, return it. |
| 584 | * Otherwise, return NULL. |
| 585 | */ |
| 586 | xfs_inode_t * |
| 587 | xfs_inode_incore(xfs_mount_t *mp, |
| 588 | xfs_ino_t ino, |
| 589 | xfs_trans_t *tp) |
| 590 | { |
| 591 | xfs_ihash_t *ih; |
| 592 | xfs_inode_t *ip; |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 593 | ulong version; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 594 | |
| 595 | ih = XFS_IHASH(mp, ino); |
| 596 | read_lock(&ih->ih_lock); |
| 597 | for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) { |
| 598 | if (ip->i_ino == ino) { |
| 599 | /* |
| 600 | * If we find it and tp matches, return it. |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 601 | * Also move it to the front of the hash list |
| 602 | * if we find it and it is not already there. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | * Otherwise break from the loop and return |
| 604 | * NULL. |
| 605 | */ |
| 606 | if (ip->i_transp == tp) { |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 607 | version = ih->ih_version; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 608 | read_unlock(&ih->ih_lock); |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 609 | xfs_ihash_promote(ih, ip, version); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 610 | return (ip); |
| 611 | } |
| 612 | break; |
| 613 | } |
| 614 | } |
| 615 | read_unlock(&ih->ih_lock); |
| 616 | return (NULL); |
| 617 | } |
| 618 | |
| 619 | /* |
| 620 | * Decrement reference count of an inode structure and unlock it. |
| 621 | * |
| 622 | * ip -- the inode being released |
| 623 | * lock_flags -- this parameter indicates the inode's locks to be |
| 624 | * to be released. See the comment on xfs_iunlock() for a list |
| 625 | * of valid values. |
| 626 | */ |
| 627 | void |
| 628 | xfs_iput(xfs_inode_t *ip, |
| 629 | uint lock_flags) |
| 630 | { |
| 631 | vnode_t *vp = XFS_ITOV(ip); |
| 632 | |
| 633 | vn_trace_entry(vp, "xfs_iput", (inst_t *)__return_address); |
| 634 | |
| 635 | xfs_iunlock(ip, lock_flags); |
| 636 | |
| 637 | VN_RELE(vp); |
| 638 | } |
| 639 | |
| 640 | /* |
| 641 | * Special iput for brand-new inodes that are still locked |
| 642 | */ |
| 643 | void |
| 644 | xfs_iput_new(xfs_inode_t *ip, |
| 645 | uint lock_flags) |
| 646 | { |
| 647 | vnode_t *vp = XFS_ITOV(ip); |
| 648 | struct inode *inode = LINVFS_GET_IP(vp); |
| 649 | |
| 650 | vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address); |
| 651 | |
| 652 | if ((ip->i_d.di_mode == 0)) { |
| 653 | ASSERT(!(ip->i_flags & XFS_IRECLAIMABLE)); |
| 654 | vn_mark_bad(vp); |
| 655 | } |
| 656 | if (inode->i_state & I_NEW) |
| 657 | unlock_new_inode(inode); |
| 658 | if (lock_flags) |
| 659 | xfs_iunlock(ip, lock_flags); |
| 660 | VN_RELE(vp); |
| 661 | } |
| 662 | |
| 663 | |
| 664 | /* |
| 665 | * This routine embodies the part of the reclaim code that pulls |
| 666 | * the inode from the inode hash table and the mount structure's |
| 667 | * inode list. |
| 668 | * This should only be called from xfs_reclaim(). |
| 669 | */ |
| 670 | void |
| 671 | xfs_ireclaim(xfs_inode_t *ip) |
| 672 | { |
| 673 | vnode_t *vp; |
| 674 | |
| 675 | /* |
| 676 | * Remove from old hash list and mount list. |
| 677 | */ |
| 678 | XFS_STATS_INC(xs_ig_reclaims); |
| 679 | |
| 680 | xfs_iextract(ip); |
| 681 | |
| 682 | /* |
| 683 | * Here we do a spurious inode lock in order to coordinate with |
| 684 | * xfs_sync(). This is because xfs_sync() references the inodes |
| 685 | * in the mount list without taking references on the corresponding |
| 686 | * vnodes. We make that OK here by ensuring that we wait until |
| 687 | * the inode is unlocked in xfs_sync() before we go ahead and |
| 688 | * free it. We get both the regular lock and the io lock because |
| 689 | * the xfs_sync() code may need to drop the regular one but will |
| 690 | * still hold the io lock. |
| 691 | */ |
| 692 | xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
| 693 | |
| 694 | /* |
| 695 | * Release dquots (and their references) if any. An inode may escape |
| 696 | * xfs_inactive and get here via vn_alloc->vn_reclaim path. |
| 697 | */ |
| 698 | XFS_QM_DQDETACH(ip->i_mount, ip); |
| 699 | |
| 700 | /* |
| 701 | * Pull our behavior descriptor from the vnode chain. |
| 702 | */ |
| 703 | vp = XFS_ITOV_NULL(ip); |
| 704 | if (vp) { |
| 705 | vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip)); |
| 706 | } |
| 707 | |
| 708 | /* |
| 709 | * Free all memory associated with the inode. |
| 710 | */ |
| 711 | xfs_idestroy(ip); |
| 712 | } |
| 713 | |
| 714 | /* |
| 715 | * This routine removes an about-to-be-destroyed inode from |
| 716 | * all of the lists in which it is located with the exception |
| 717 | * of the behavior chain. |
| 718 | */ |
| 719 | void |
| 720 | xfs_iextract( |
| 721 | xfs_inode_t *ip) |
| 722 | { |
| 723 | xfs_ihash_t *ih; |
| 724 | xfs_inode_t *iq; |
| 725 | xfs_mount_t *mp; |
| 726 | xfs_chash_t *ch; |
| 727 | xfs_chashlist_t *chl, *chm; |
| 728 | SPLDECL(s); |
| 729 | |
| 730 | ih = ip->i_hash; |
| 731 | write_lock(&ih->ih_lock); |
| 732 | if ((iq = ip->i_next)) { |
| 733 | iq->i_prevp = ip->i_prevp; |
| 734 | } |
| 735 | *ip->i_prevp = iq; |
Nathan Scott | 71bce25 | 2005-05-05 13:23:27 -0700 | [diff] [blame^] | 736 | ih->ih_version++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | write_unlock(&ih->ih_lock); |
| 738 | |
| 739 | /* |
| 740 | * Remove from cluster hash list |
| 741 | * 1) delete the chashlist if this is the last inode on the chashlist |
| 742 | * 2) unchain from list of inodes |
| 743 | * 3) point chashlist->chl_ip to 'chl_next' if to this inode. |
| 744 | */ |
| 745 | mp = ip->i_mount; |
| 746 | ch = XFS_CHASH(mp, ip->i_blkno); |
| 747 | s = mutex_spinlock(&ch->ch_lock); |
| 748 | |
| 749 | if (ip->i_cnext == ip) { |
| 750 | /* Last inode on chashlist */ |
| 751 | ASSERT(ip->i_cnext == ip && ip->i_cprev == ip); |
| 752 | ASSERT(ip->i_chash != NULL); |
| 753 | chm=NULL; |
| 754 | for (chl = ch->ch_list; chl != NULL; chl = chl->chl_next) { |
| 755 | if (chl->chl_blkno == ip->i_blkno) { |
| 756 | if (chm == NULL) { |
| 757 | /* first item on the list */ |
| 758 | ch->ch_list = chl->chl_next; |
| 759 | } else { |
| 760 | chm->chl_next = chl->chl_next; |
| 761 | } |
| 762 | kmem_zone_free(xfs_chashlist_zone, chl); |
| 763 | break; |
| 764 | } else { |
| 765 | ASSERT(chl->chl_ip != ip); |
| 766 | chm = chl; |
| 767 | } |
| 768 | } |
| 769 | ASSERT_ALWAYS(chl != NULL); |
| 770 | } else { |
| 771 | /* delete one inode from a non-empty list */ |
| 772 | iq = ip->i_cnext; |
| 773 | iq->i_cprev = ip->i_cprev; |
| 774 | ip->i_cprev->i_cnext = iq; |
| 775 | if (ip->i_chash->chl_ip == ip) { |
| 776 | ip->i_chash->chl_ip = iq; |
| 777 | } |
| 778 | ip->i_chash = __return_address; |
| 779 | ip->i_cprev = __return_address; |
| 780 | ip->i_cnext = __return_address; |
| 781 | } |
| 782 | mutex_spinunlock(&ch->ch_lock, s); |
| 783 | |
| 784 | /* |
| 785 | * Remove from mount's inode list. |
| 786 | */ |
| 787 | XFS_MOUNT_ILOCK(mp); |
| 788 | ASSERT((ip->i_mnext != NULL) && (ip->i_mprev != NULL)); |
| 789 | iq = ip->i_mnext; |
| 790 | iq->i_mprev = ip->i_mprev; |
| 791 | ip->i_mprev->i_mnext = iq; |
| 792 | |
| 793 | /* |
| 794 | * Fix up the head pointer if it points to the inode being deleted. |
| 795 | */ |
| 796 | if (mp->m_inodes == ip) { |
| 797 | if (ip == iq) { |
| 798 | mp->m_inodes = NULL; |
| 799 | } else { |
| 800 | mp->m_inodes = iq; |
| 801 | } |
| 802 | } |
| 803 | |
| 804 | /* Deal with the deleted inodes list */ |
| 805 | list_del_init(&ip->i_reclaim); |
| 806 | |
| 807 | mp->m_ireclaims++; |
| 808 | XFS_MOUNT_IUNLOCK(mp); |
| 809 | } |
| 810 | |
| 811 | /* |
| 812 | * This is a wrapper routine around the xfs_ilock() routine |
| 813 | * used to centralize some grungy code. It is used in places |
| 814 | * that wish to lock the inode solely for reading the extents. |
| 815 | * The reason these places can't just call xfs_ilock(SHARED) |
| 816 | * is that the inode lock also guards to bringing in of the |
| 817 | * extents from disk for a file in b-tree format. If the inode |
| 818 | * is in b-tree format, then we need to lock the inode exclusively |
| 819 | * until the extents are read in. Locking it exclusively all |
| 820 | * the time would limit our parallelism unnecessarily, though. |
| 821 | * What we do instead is check to see if the extents have been |
| 822 | * read in yet, and only lock the inode exclusively if they |
| 823 | * have not. |
| 824 | * |
| 825 | * The function returns a value which should be given to the |
| 826 | * corresponding xfs_iunlock_map_shared(). This value is |
| 827 | * the mode in which the lock was actually taken. |
| 828 | */ |
| 829 | uint |
| 830 | xfs_ilock_map_shared( |
| 831 | xfs_inode_t *ip) |
| 832 | { |
| 833 | uint lock_mode; |
| 834 | |
| 835 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && |
| 836 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { |
| 837 | lock_mode = XFS_ILOCK_EXCL; |
| 838 | } else { |
| 839 | lock_mode = XFS_ILOCK_SHARED; |
| 840 | } |
| 841 | |
| 842 | xfs_ilock(ip, lock_mode); |
| 843 | |
| 844 | return lock_mode; |
| 845 | } |
| 846 | |
| 847 | /* |
| 848 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). |
| 849 | * All it does is call xfs_iunlock() with the given lock_mode. |
| 850 | */ |
| 851 | void |
| 852 | xfs_iunlock_map_shared( |
| 853 | xfs_inode_t *ip, |
| 854 | unsigned int lock_mode) |
| 855 | { |
| 856 | xfs_iunlock(ip, lock_mode); |
| 857 | } |
| 858 | |
| 859 | /* |
| 860 | * The xfs inode contains 2 locks: a multi-reader lock called the |
| 861 | * i_iolock and a multi-reader lock called the i_lock. This routine |
| 862 | * allows either or both of the locks to be obtained. |
| 863 | * |
| 864 | * The 2 locks should always be ordered so that the IO lock is |
| 865 | * obtained first in order to prevent deadlock. |
| 866 | * |
| 867 | * ip -- the inode being locked |
| 868 | * lock_flags -- this parameter indicates the inode's locks |
| 869 | * to be locked. It can be: |
| 870 | * XFS_IOLOCK_SHARED, |
| 871 | * XFS_IOLOCK_EXCL, |
| 872 | * XFS_ILOCK_SHARED, |
| 873 | * XFS_ILOCK_EXCL, |
| 874 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, |
| 875 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, |
| 876 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, |
| 877 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
| 878 | */ |
| 879 | void |
| 880 | xfs_ilock(xfs_inode_t *ip, |
| 881 | uint lock_flags) |
| 882 | { |
| 883 | /* |
| 884 | * You can't set both SHARED and EXCL for the same lock, |
| 885 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 886 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 887 | */ |
| 888 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 889 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 890 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 891 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
| 892 | ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0); |
| 893 | |
| 894 | if (lock_flags & XFS_IOLOCK_EXCL) { |
| 895 | mrupdate(&ip->i_iolock); |
| 896 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
| 897 | mraccess(&ip->i_iolock); |
| 898 | } |
| 899 | if (lock_flags & XFS_ILOCK_EXCL) { |
| 900 | mrupdate(&ip->i_lock); |
| 901 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
| 902 | mraccess(&ip->i_lock); |
| 903 | } |
| 904 | xfs_ilock_trace(ip, 1, lock_flags, (inst_t *)__return_address); |
| 905 | } |
| 906 | |
| 907 | /* |
| 908 | * This is just like xfs_ilock(), except that the caller |
| 909 | * is guaranteed not to sleep. It returns 1 if it gets |
| 910 | * the requested locks and 0 otherwise. If the IO lock is |
| 911 | * obtained but the inode lock cannot be, then the IO lock |
| 912 | * is dropped before returning. |
| 913 | * |
| 914 | * ip -- the inode being locked |
| 915 | * lock_flags -- this parameter indicates the inode's locks to be |
| 916 | * to be locked. See the comment for xfs_ilock() for a list |
| 917 | * of valid values. |
| 918 | * |
| 919 | */ |
| 920 | int |
| 921 | xfs_ilock_nowait(xfs_inode_t *ip, |
| 922 | uint lock_flags) |
| 923 | { |
| 924 | int iolocked; |
| 925 | int ilocked; |
| 926 | |
| 927 | /* |
| 928 | * You can't set both SHARED and EXCL for the same lock, |
| 929 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 930 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 931 | */ |
| 932 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 933 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 934 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 935 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
| 936 | ASSERT((lock_flags & ~XFS_LOCK_MASK) == 0); |
| 937 | |
| 938 | iolocked = 0; |
| 939 | if (lock_flags & XFS_IOLOCK_EXCL) { |
| 940 | iolocked = mrtryupdate(&ip->i_iolock); |
| 941 | if (!iolocked) { |
| 942 | return 0; |
| 943 | } |
| 944 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
| 945 | iolocked = mrtryaccess(&ip->i_iolock); |
| 946 | if (!iolocked) { |
| 947 | return 0; |
| 948 | } |
| 949 | } |
| 950 | if (lock_flags & XFS_ILOCK_EXCL) { |
| 951 | ilocked = mrtryupdate(&ip->i_lock); |
| 952 | if (!ilocked) { |
| 953 | if (iolocked) { |
| 954 | mrunlock(&ip->i_iolock); |
| 955 | } |
| 956 | return 0; |
| 957 | } |
| 958 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
| 959 | ilocked = mrtryaccess(&ip->i_lock); |
| 960 | if (!ilocked) { |
| 961 | if (iolocked) { |
| 962 | mrunlock(&ip->i_iolock); |
| 963 | } |
| 964 | return 0; |
| 965 | } |
| 966 | } |
| 967 | xfs_ilock_trace(ip, 2, lock_flags, (inst_t *)__return_address); |
| 968 | return 1; |
| 969 | } |
| 970 | |
| 971 | /* |
| 972 | * xfs_iunlock() is used to drop the inode locks acquired with |
| 973 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass |
| 974 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so |
| 975 | * that we know which locks to drop. |
| 976 | * |
| 977 | * ip -- the inode being unlocked |
| 978 | * lock_flags -- this parameter indicates the inode's locks to be |
| 979 | * to be unlocked. See the comment for xfs_ilock() for a list |
| 980 | * of valid values for this parameter. |
| 981 | * |
| 982 | */ |
| 983 | void |
| 984 | xfs_iunlock(xfs_inode_t *ip, |
| 985 | uint lock_flags) |
| 986 | { |
| 987 | /* |
| 988 | * You can't set both SHARED and EXCL for the same lock, |
| 989 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 990 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 991 | */ |
| 992 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 993 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 994 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 995 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
| 996 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_IUNLOCK_NONOTIFY)) == 0); |
| 997 | ASSERT(lock_flags != 0); |
| 998 | |
| 999 | if (lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) { |
| 1000 | ASSERT(!(lock_flags & XFS_IOLOCK_SHARED) || |
| 1001 | (ismrlocked(&ip->i_iolock, MR_ACCESS))); |
| 1002 | ASSERT(!(lock_flags & XFS_IOLOCK_EXCL) || |
| 1003 | (ismrlocked(&ip->i_iolock, MR_UPDATE))); |
| 1004 | mrunlock(&ip->i_iolock); |
| 1005 | } |
| 1006 | |
| 1007 | if (lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) { |
| 1008 | ASSERT(!(lock_flags & XFS_ILOCK_SHARED) || |
| 1009 | (ismrlocked(&ip->i_lock, MR_ACCESS))); |
| 1010 | ASSERT(!(lock_flags & XFS_ILOCK_EXCL) || |
| 1011 | (ismrlocked(&ip->i_lock, MR_UPDATE))); |
| 1012 | mrunlock(&ip->i_lock); |
| 1013 | |
| 1014 | /* |
| 1015 | * Let the AIL know that this item has been unlocked in case |
| 1016 | * it is in the AIL and anyone is waiting on it. Don't do |
| 1017 | * this if the caller has asked us not to. |
| 1018 | */ |
| 1019 | if (!(lock_flags & XFS_IUNLOCK_NONOTIFY) && |
| 1020 | ip->i_itemp != NULL) { |
| 1021 | xfs_trans_unlocked_item(ip->i_mount, |
| 1022 | (xfs_log_item_t*)(ip->i_itemp)); |
| 1023 | } |
| 1024 | } |
| 1025 | xfs_ilock_trace(ip, 3, lock_flags, (inst_t *)__return_address); |
| 1026 | } |
| 1027 | |
| 1028 | /* |
| 1029 | * give up write locks. the i/o lock cannot be held nested |
| 1030 | * if it is being demoted. |
| 1031 | */ |
| 1032 | void |
| 1033 | xfs_ilock_demote(xfs_inode_t *ip, |
| 1034 | uint lock_flags) |
| 1035 | { |
| 1036 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); |
| 1037 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); |
| 1038 | |
| 1039 | if (lock_flags & XFS_ILOCK_EXCL) { |
| 1040 | ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE)); |
| 1041 | mrdemote(&ip->i_lock); |
| 1042 | } |
| 1043 | if (lock_flags & XFS_IOLOCK_EXCL) { |
| 1044 | ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE)); |
| 1045 | mrdemote(&ip->i_iolock); |
| 1046 | } |
| 1047 | } |
| 1048 | |
| 1049 | /* |
| 1050 | * The following three routines simply manage the i_flock |
| 1051 | * semaphore embedded in the inode. This semaphore synchronizes |
| 1052 | * processes attempting to flush the in-core inode back to disk. |
| 1053 | */ |
| 1054 | void |
| 1055 | xfs_iflock(xfs_inode_t *ip) |
| 1056 | { |
| 1057 | psema(&(ip->i_flock), PINOD|PLTWAIT); |
| 1058 | } |
| 1059 | |
| 1060 | int |
| 1061 | xfs_iflock_nowait(xfs_inode_t *ip) |
| 1062 | { |
| 1063 | return (cpsema(&(ip->i_flock))); |
| 1064 | } |
| 1065 | |
| 1066 | void |
| 1067 | xfs_ifunlock(xfs_inode_t *ip) |
| 1068 | { |
| 1069 | ASSERT(valusema(&(ip->i_flock)) <= 0); |
| 1070 | vsema(&(ip->i_flock)); |
| 1071 | } |