Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2000-2002 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 | #include "xfs_macros.h" |
| 35 | #include "xfs_types.h" |
| 36 | #include "xfs_inum.h" |
| 37 | #include "xfs_log.h" |
| 38 | #include "xfs_trans.h" |
| 39 | #include "xfs_sb.h" |
| 40 | #include "xfs_dir.h" |
| 41 | #include "xfs_dir2.h" |
| 42 | #include "xfs_dmapi.h" |
| 43 | #include "xfs_mount.h" |
| 44 | #include "xfs_bmap_btree.h" |
| 45 | #include "xfs_attr_sf.h" |
| 46 | #include "xfs_dir_sf.h" |
| 47 | #include "xfs_dir2_sf.h" |
| 48 | #include "xfs_dinode.h" |
| 49 | #include "xfs_inode_item.h" |
| 50 | #include "xfs_inode.h" |
| 51 | #include "xfs_bmap.h" |
| 52 | #include "xfs_error.h" |
| 53 | #include "xfs_quota.h" |
| 54 | #include "xfs_rw.h" |
| 55 | #include "xfs_itable.h" |
| 56 | #include "xfs_utils.h" |
| 57 | |
| 58 | /* |
| 59 | * xfs_get_dir_entry is used to get a reference to an inode given |
| 60 | * its parent directory inode and the name of the file. It does |
| 61 | * not lock the child inode, and it unlocks the directory before |
| 62 | * returning. The directory's generation number is returned for |
| 63 | * use by a later call to xfs_lock_dir_and_entry. |
| 64 | */ |
| 65 | int |
| 66 | xfs_get_dir_entry( |
| 67 | vname_t *dentry, |
| 68 | xfs_inode_t **ipp) |
| 69 | { |
| 70 | vnode_t *vp; |
| 71 | bhv_desc_t *bdp; |
| 72 | |
| 73 | vp = VNAME_TO_VNODE(dentry); |
| 74 | bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(vp), &xfs_vnodeops); |
| 75 | if (!bdp) { |
| 76 | *ipp = NULL; |
| 77 | return XFS_ERROR(ENOENT); |
| 78 | } |
| 79 | VN_HOLD(vp); |
| 80 | *ipp = XFS_BHVTOI(bdp); |
| 81 | return 0; |
| 82 | } |
| 83 | |
| 84 | int |
| 85 | xfs_dir_lookup_int( |
| 86 | bhv_desc_t *dir_bdp, |
| 87 | uint lock_mode, |
| 88 | vname_t *dentry, |
| 89 | xfs_ino_t *inum, |
| 90 | xfs_inode_t **ipp) |
| 91 | { |
| 92 | vnode_t *dir_vp; |
| 93 | xfs_inode_t *dp; |
| 94 | int error; |
| 95 | |
| 96 | dir_vp = BHV_TO_VNODE(dir_bdp); |
| 97 | vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address); |
| 98 | |
| 99 | dp = XFS_BHVTOI(dir_bdp); |
| 100 | |
| 101 | error = XFS_DIR_LOOKUP(dp->i_mount, NULL, dp, |
| 102 | VNAME(dentry), VNAMELEN(dentry), inum); |
| 103 | if (!error) { |
| 104 | /* |
| 105 | * Unlock the directory. We do this because we can't |
| 106 | * hold the directory lock while doing the vn_get() |
| 107 | * in xfs_iget(). Doing so could cause us to hold |
| 108 | * a lock while waiting for the inode to finish |
| 109 | * being inactive while it's waiting for a log |
| 110 | * reservation in the inactive routine. |
| 111 | */ |
| 112 | xfs_iunlock(dp, lock_mode); |
| 113 | error = xfs_iget(dp->i_mount, NULL, *inum, 0, 0, ipp, 0); |
| 114 | xfs_ilock(dp, lock_mode); |
| 115 | |
| 116 | if (error) { |
| 117 | *ipp = NULL; |
| 118 | } else if ((*ipp)->i_d.di_mode == 0) { |
| 119 | /* |
| 120 | * The inode has been freed. Something is |
| 121 | * wrong so just get out of here. |
| 122 | */ |
| 123 | xfs_iunlock(dp, lock_mode); |
| 124 | xfs_iput_new(*ipp, 0); |
| 125 | *ipp = NULL; |
| 126 | xfs_ilock(dp, lock_mode); |
| 127 | error = XFS_ERROR(ENOENT); |
| 128 | } |
| 129 | } |
| 130 | return error; |
| 131 | } |
| 132 | |
| 133 | /* |
| 134 | * Allocates a new inode from disk and return a pointer to the |
| 135 | * incore copy. This routine will internally commit the current |
| 136 | * transaction and allocate a new one if the Space Manager needed |
| 137 | * to do an allocation to replenish the inode free-list. |
| 138 | * |
| 139 | * This routine is designed to be called from xfs_create and |
| 140 | * xfs_create_dir. |
| 141 | * |
| 142 | */ |
| 143 | int |
| 144 | xfs_dir_ialloc( |
| 145 | xfs_trans_t **tpp, /* input: current transaction; |
| 146 | output: may be a new transaction. */ |
| 147 | xfs_inode_t *dp, /* directory within whose allocate |
| 148 | the inode. */ |
| 149 | mode_t mode, |
Nathan Scott | 31b084a | 2005-05-05 13:25:00 -0700 | [diff] [blame] | 150 | xfs_nlink_t nlink, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | xfs_dev_t rdev, |
| 152 | cred_t *credp, |
| 153 | prid_t prid, /* project id */ |
| 154 | int okalloc, /* ok to allocate new space */ |
| 155 | xfs_inode_t **ipp, /* pointer to inode; it will be |
| 156 | locked. */ |
| 157 | int *committed) |
| 158 | |
| 159 | { |
| 160 | xfs_trans_t *tp; |
| 161 | xfs_trans_t *ntp; |
| 162 | xfs_inode_t *ip; |
| 163 | xfs_buf_t *ialloc_context = NULL; |
| 164 | boolean_t call_again = B_FALSE; |
| 165 | int code; |
| 166 | uint log_res; |
| 167 | uint log_count; |
| 168 | void *dqinfo; |
| 169 | uint tflags; |
| 170 | |
| 171 | tp = *tpp; |
| 172 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
| 173 | |
| 174 | /* |
| 175 | * xfs_ialloc will return a pointer to an incore inode if |
| 176 | * the Space Manager has an available inode on the free |
| 177 | * list. Otherwise, it will do an allocation and replenish |
| 178 | * the freelist. Since we can only do one allocation per |
| 179 | * transaction without deadlocks, we will need to commit the |
| 180 | * current transaction and start a new one. We will then |
| 181 | * need to call xfs_ialloc again to get the inode. |
| 182 | * |
| 183 | * If xfs_ialloc did an allocation to replenish the freelist, |
| 184 | * it returns the bp containing the head of the freelist as |
| 185 | * ialloc_context. We will hold a lock on it across the |
| 186 | * transaction commit so that no other process can steal |
| 187 | * the inode(s) that we've just allocated. |
| 188 | */ |
| 189 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid, okalloc, |
| 190 | &ialloc_context, &call_again, &ip); |
| 191 | |
| 192 | /* |
| 193 | * Return an error if we were unable to allocate a new inode. |
| 194 | * This should only happen if we run out of space on disk or |
| 195 | * encounter a disk error. |
| 196 | */ |
| 197 | if (code) { |
| 198 | *ipp = NULL; |
| 199 | return code; |
| 200 | } |
| 201 | if (!call_again && (ip == NULL)) { |
| 202 | *ipp = NULL; |
| 203 | return XFS_ERROR(ENOSPC); |
| 204 | } |
| 205 | |
| 206 | /* |
| 207 | * If call_again is set, then we were unable to get an |
| 208 | * inode in one operation. We need to commit the current |
| 209 | * transaction and call xfs_ialloc() again. It is guaranteed |
| 210 | * to succeed the second time. |
| 211 | */ |
| 212 | if (call_again) { |
| 213 | |
| 214 | /* |
| 215 | * Normally, xfs_trans_commit releases all the locks. |
| 216 | * We call bhold to hang on to the ialloc_context across |
| 217 | * the commit. Holding this buffer prevents any other |
| 218 | * processes from doing any allocations in this |
| 219 | * allocation group. |
| 220 | */ |
| 221 | xfs_trans_bhold(tp, ialloc_context); |
| 222 | /* |
| 223 | * Save the log reservation so we can use |
| 224 | * them in the next transaction. |
| 225 | */ |
| 226 | log_res = xfs_trans_get_log_res(tp); |
| 227 | log_count = xfs_trans_get_log_count(tp); |
| 228 | |
| 229 | /* |
| 230 | * We want the quota changes to be associated with the next |
| 231 | * transaction, NOT this one. So, detach the dqinfo from this |
| 232 | * and attach it to the next transaction. |
| 233 | */ |
| 234 | dqinfo = NULL; |
| 235 | tflags = 0; |
| 236 | if (tp->t_dqinfo) { |
| 237 | dqinfo = (void *)tp->t_dqinfo; |
| 238 | tp->t_dqinfo = NULL; |
| 239 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; |
| 240 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); |
| 241 | } |
| 242 | |
| 243 | ntp = xfs_trans_dup(tp); |
| 244 | code = xfs_trans_commit(tp, 0, NULL); |
| 245 | tp = ntp; |
| 246 | if (committed != NULL) { |
| 247 | *committed = 1; |
| 248 | } |
| 249 | /* |
| 250 | * If we get an error during the commit processing, |
| 251 | * release the buffer that is still held and return |
| 252 | * to the caller. |
| 253 | */ |
| 254 | if (code) { |
| 255 | xfs_buf_relse(ialloc_context); |
| 256 | if (dqinfo) { |
| 257 | tp->t_dqinfo = dqinfo; |
| 258 | XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp); |
| 259 | } |
| 260 | *tpp = ntp; |
| 261 | *ipp = NULL; |
| 262 | return code; |
| 263 | } |
| 264 | code = xfs_trans_reserve(tp, 0, log_res, 0, |
| 265 | XFS_TRANS_PERM_LOG_RES, log_count); |
| 266 | /* |
| 267 | * Re-attach the quota info that we detached from prev trx. |
| 268 | */ |
| 269 | if (dqinfo) { |
| 270 | tp->t_dqinfo = dqinfo; |
| 271 | tp->t_flags |= tflags; |
| 272 | } |
| 273 | |
| 274 | if (code) { |
| 275 | xfs_buf_relse(ialloc_context); |
| 276 | *tpp = ntp; |
| 277 | *ipp = NULL; |
| 278 | return code; |
| 279 | } |
| 280 | xfs_trans_bjoin(tp, ialloc_context); |
| 281 | |
| 282 | /* |
| 283 | * Call ialloc again. Since we've locked out all |
| 284 | * other allocations in this allocation group, |
| 285 | * this call should always succeed. |
| 286 | */ |
| 287 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, credp, prid, |
| 288 | okalloc, &ialloc_context, &call_again, &ip); |
| 289 | |
| 290 | /* |
| 291 | * If we get an error at this point, return to the caller |
| 292 | * so that the current transaction can be aborted. |
| 293 | */ |
| 294 | if (code) { |
| 295 | *tpp = tp; |
| 296 | *ipp = NULL; |
| 297 | return code; |
| 298 | } |
| 299 | ASSERT ((!call_again) && (ip != NULL)); |
| 300 | |
| 301 | } else { |
| 302 | if (committed != NULL) { |
| 303 | *committed = 0; |
| 304 | } |
| 305 | } |
| 306 | |
| 307 | *ipp = ip; |
| 308 | *tpp = tp; |
| 309 | |
| 310 | return 0; |
| 311 | } |
| 312 | |
| 313 | /* |
| 314 | * Decrement the link count on an inode & log the change. |
| 315 | * If this causes the link count to go to zero, initiate the |
| 316 | * logging activity required to truncate a file. |
| 317 | */ |
| 318 | int /* error */ |
| 319 | xfs_droplink( |
| 320 | xfs_trans_t *tp, |
| 321 | xfs_inode_t *ip) |
| 322 | { |
| 323 | int error; |
| 324 | |
| 325 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); |
| 326 | |
| 327 | ASSERT (ip->i_d.di_nlink > 0); |
| 328 | ip->i_d.di_nlink--; |
| 329 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 330 | |
| 331 | error = 0; |
| 332 | if (ip->i_d.di_nlink == 0) { |
| 333 | /* |
| 334 | * We're dropping the last link to this file. |
| 335 | * Move the on-disk inode to the AGI unlinked list. |
| 336 | * From xfs_inactive() we will pull the inode from |
| 337 | * the list and free it. |
| 338 | */ |
| 339 | error = xfs_iunlink(tp, ip); |
| 340 | } |
| 341 | return error; |
| 342 | } |
| 343 | |
| 344 | /* |
| 345 | * This gets called when the inode's version needs to be changed from 1 to 2. |
| 346 | * Currently this happens when the nlink field overflows the old 16-bit value |
| 347 | * or when chproj is called to change the project for the first time. |
| 348 | * As a side effect the superblock version will also get rev'd |
| 349 | * to contain the NLINK bit. |
| 350 | */ |
| 351 | void |
| 352 | xfs_bump_ino_vers2( |
| 353 | xfs_trans_t *tp, |
| 354 | xfs_inode_t *ip) |
| 355 | { |
| 356 | xfs_mount_t *mp; |
| 357 | unsigned long s; |
| 358 | |
| 359 | ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE)); |
| 360 | ASSERT(ip->i_d.di_version == XFS_DINODE_VERSION_1); |
| 361 | |
| 362 | ip->i_d.di_version = XFS_DINODE_VERSION_2; |
| 363 | ip->i_d.di_onlink = 0; |
| 364 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
| 365 | mp = tp->t_mountp; |
| 366 | if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) { |
| 367 | s = XFS_SB_LOCK(mp); |
| 368 | if (!XFS_SB_VERSION_HASNLINK(&mp->m_sb)) { |
| 369 | XFS_SB_VERSION_ADDNLINK(&mp->m_sb); |
| 370 | XFS_SB_UNLOCK(mp, s); |
| 371 | xfs_mod_sb(tp, XFS_SB_VERSIONNUM); |
| 372 | } else { |
| 373 | XFS_SB_UNLOCK(mp, s); |
| 374 | } |
| 375 | } |
| 376 | /* Caller must log the inode */ |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * Increment the link count on an inode & log the change. |
| 381 | */ |
| 382 | int |
| 383 | xfs_bumplink( |
| 384 | xfs_trans_t *tp, |
| 385 | xfs_inode_t *ip) |
| 386 | { |
| 387 | if (ip->i_d.di_nlink >= XFS_MAXLINK) |
| 388 | return XFS_ERROR(EMLINK); |
| 389 | xfs_ichgtime(ip, XFS_ICHGTIME_CHG); |
| 390 | |
| 391 | ASSERT(ip->i_d.di_nlink > 0); |
| 392 | ip->i_d.di_nlink++; |
| 393 | if ((ip->i_d.di_version == XFS_DINODE_VERSION_1) && |
| 394 | (ip->i_d.di_nlink > XFS_MAXLINK_1)) { |
| 395 | /* |
| 396 | * The inode has increased its number of links beyond |
| 397 | * what can fit in an old format inode. It now needs |
| 398 | * to be converted to a version 2 inode with a 32 bit |
| 399 | * link count. If this is the first inode in the file |
| 400 | * system to do this, then we need to bump the superblock |
| 401 | * version number as well. |
| 402 | */ |
| 403 | xfs_bump_ino_vers2(tp, ip); |
| 404 | } |
| 405 | |
| 406 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 407 | return 0; |
| 408 | } |
| 409 | |
| 410 | /* |
| 411 | * Try to truncate the given file to 0 length. Currently called |
| 412 | * only out of xfs_remove when it has to truncate a file to free |
| 413 | * up space for the remove to proceed. |
| 414 | */ |
| 415 | int |
| 416 | xfs_truncate_file( |
| 417 | xfs_mount_t *mp, |
| 418 | xfs_inode_t *ip) |
| 419 | { |
| 420 | xfs_trans_t *tp; |
| 421 | int error; |
| 422 | |
| 423 | #ifdef QUOTADEBUG |
| 424 | /* |
| 425 | * This is called to truncate the quotainodes too. |
| 426 | */ |
| 427 | if (XFS_IS_UQUOTA_ON(mp)) { |
| 428 | if (ip->i_ino != mp->m_sb.sb_uquotino) |
| 429 | ASSERT(ip->i_udquot); |
| 430 | } |
Nathan Scott | c8ad20f | 2005-06-21 15:38:48 +1000 | [diff] [blame] | 431 | if (XFS_IS_OQUOTA_ON(mp)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | if (ip->i_ino != mp->m_sb.sb_gquotino) |
| 433 | ASSERT(ip->i_gdquot); |
| 434 | } |
| 435 | #endif |
| 436 | /* |
| 437 | * Make the call to xfs_itruncate_start before starting the |
| 438 | * transaction, because we cannot make the call while we're |
| 439 | * in a transaction. |
| 440 | */ |
| 441 | xfs_ilock(ip, XFS_IOLOCK_EXCL); |
| 442 | xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, (xfs_fsize_t)0); |
| 443 | |
| 444 | tp = xfs_trans_alloc(mp, XFS_TRANS_TRUNCATE_FILE); |
| 445 | if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, |
| 446 | XFS_TRANS_PERM_LOG_RES, |
| 447 | XFS_ITRUNCATE_LOG_COUNT))) { |
| 448 | xfs_trans_cancel(tp, 0); |
| 449 | xfs_iunlock(ip, XFS_IOLOCK_EXCL); |
| 450 | return error; |
| 451 | } |
| 452 | |
| 453 | /* |
| 454 | * Follow the normal truncate locking protocol. Since we |
| 455 | * hold the inode in the transaction, we know that it's number |
| 456 | * of references will stay constant. |
| 457 | */ |
| 458 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 459 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
| 460 | xfs_trans_ihold(tp, ip); |
| 461 | /* |
| 462 | * Signal a sync xaction. The only case where that isn't |
| 463 | * the case is if we're truncating an already unlinked file |
| 464 | * on a wsync fs. In that case, we know the blocks can't |
| 465 | * reappear in the file because the links to file are |
| 466 | * permanently toast. Currently, we're always going to |
| 467 | * want a sync transaction because this code is being |
| 468 | * called from places where nlink is guaranteed to be 1 |
| 469 | * but I'm leaving the tests in to protect against future |
| 470 | * changes -- rcc. |
| 471 | */ |
| 472 | error = xfs_itruncate_finish(&tp, ip, (xfs_fsize_t)0, |
| 473 | XFS_DATA_FORK, |
| 474 | ((ip->i_d.di_nlink != 0 || |
| 475 | !(mp->m_flags & XFS_MOUNT_WSYNC)) |
| 476 | ? 1 : 0)); |
| 477 | if (error) { |
| 478 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | |
| 479 | XFS_TRANS_ABORT); |
| 480 | } else { |
| 481 | xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 482 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, |
| 483 | NULL); |
| 484 | } |
| 485 | xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL); |
| 486 | |
| 487 | return error; |
| 488 | } |