1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. 4 * 5 * This copyrighted material is made available to anyone wishing to use, 6 * modify, copy, or redistribute it subject to the terms and conditions 7 * of the GNU General Public License v.2. 8 */ 9 10 #include <linux/sched.h> 11 #include <linux/slab.h> 12 #include <linux/spinlock.h> 13 #include <linux/completion.h> 14 #include <linux/buffer_head.h> 15 #include <linux/posix_acl.h> 16 #include <linux/sort.h> 17 #include <linux/gfs2_ondisk.h> 18 #include <asm/semaphore.h> 19 20 #include "gfs2.h" 21 #include "lm_interface.h" 22 #include "incore.h" 23 #include "acl.h" 24 #include "bmap.h" 25 #include "dir.h" 26 #include "eattr.h" 27 #include "glock.h" 28 #include "glops.h" 29 #include "inode.h" 30 #include "log.h" 31 #include "meta_io.h" 32 #include "ops_address.h" 33 #include "ops_file.h" 34 #include "ops_inode.h" 35 #include "quota.h" 36 #include "rgrp.h" 37 #include "trans.h" 38 #include "unlinked.h" 39 #include "util.h" 40 41 /** 42 * inode_attr_in - Copy attributes from the dinode into the VFS inode 43 * @ip: The GFS2 inode (with embedded disk inode data) 44 * @inode: The Linux VFS inode 45 * 46 */ 47 48 static void inode_attr_in(struct gfs2_inode *ip, struct inode *inode) 49 { 50 inode->i_ino = ip->i_num.no_formal_ino; 51 52 switch (ip->i_di.di_mode & S_IFMT) { 53 case S_IFBLK: 54 case S_IFCHR: 55 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor); 56 break; 57 default: 58 inode->i_rdev = 0; 59 break; 60 }; 61 62 inode->i_mode = ip->i_di.di_mode; 63 inode->i_nlink = ip->i_di.di_nlink; 64 inode->i_uid = ip->i_di.di_uid; 65 inode->i_gid = ip->i_di.di_gid; 66 i_size_write(inode, ip->i_di.di_size); 67 inode->i_atime.tv_sec = ip->i_di.di_atime; 68 inode->i_mtime.tv_sec = ip->i_di.di_mtime; 69 inode->i_ctime.tv_sec = ip->i_di.di_ctime; 70 inode->i_atime.tv_nsec = 0; 71 inode->i_mtime.tv_nsec = 0; 72 inode->i_ctime.tv_nsec = 0; 73 inode->i_blksize = PAGE_SIZE; 74 inode->i_blocks = ip->i_di.di_blocks << 75 (ip->i_sbd->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT); 76 77 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE) 78 inode->i_flags |= S_IMMUTABLE; 79 else 80 inode->i_flags &= ~S_IMMUTABLE; 81 82 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY) 83 inode->i_flags |= S_APPEND; 84 else 85 inode->i_flags &= ~S_APPEND; 86 } 87 88 /** 89 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode 90 * @ip: The GFS2 inode (with embedded disk inode data) 91 * 92 */ 93 94 void gfs2_inode_attr_in(struct gfs2_inode *ip) 95 { 96 struct inode *inode; 97 98 inode = gfs2_ip2v_lookup(ip); 99 if (inode) { 100 inode_attr_in(ip, inode); 101 iput(inode); 102 } 103 } 104 105 /** 106 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode 107 * @ip: The GFS2 inode 108 * 109 * Only copy out the attributes that we want the VFS layer 110 * to be able to modify. 111 */ 112 113 void gfs2_inode_attr_out(struct gfs2_inode *ip) 114 { 115 struct inode *inode = ip->i_vnode; 116 117 gfs2_assert_withdraw(ip->i_sbd, 118 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT)); 119 ip->i_di.di_mode = inode->i_mode; 120 ip->i_di.di_uid = inode->i_uid; 121 ip->i_di.di_gid = inode->i_gid; 122 ip->i_di.di_atime = inode->i_atime.tv_sec; 123 ip->i_di.di_mtime = inode->i_mtime.tv_sec; 124 ip->i_di.di_ctime = inode->i_ctime.tv_sec; 125 } 126 127 /** 128 * gfs2_ip2v_lookup - Get the struct inode for a struct gfs2_inode 129 * @ip: the struct gfs2_inode to get the struct inode for 130 * 131 * Returns: A VFS inode, or NULL if none 132 */ 133 134 struct inode *gfs2_ip2v_lookup(struct gfs2_inode *ip) 135 { 136 struct inode *inode = NULL; 137 138 gfs2_assert_warn(ip->i_sbd, test_bit(GIF_MIN_INIT, &ip->i_flags)); 139 140 spin_lock(&ip->i_spin); 141 if (ip->i_vnode) 142 inode = igrab(ip->i_vnode); 143 spin_unlock(&ip->i_spin); 144 145 return inode; 146 } 147 148 /** 149 * gfs2_ip2v - Get/Create a struct inode for a struct gfs2_inode 150 * @ip: the struct gfs2_inode to get the struct inode for 151 * 152 * Returns: A VFS inode, or NULL if no mem 153 */ 154 155 struct inode *gfs2_ip2v(struct gfs2_inode *ip) 156 { 157 struct inode *inode, *tmp; 158 159 inode = gfs2_ip2v_lookup(ip); 160 if (inode) 161 return inode; 162 163 tmp = new_inode(ip->i_sbd->sd_vfs); 164 if (!tmp) 165 return NULL; 166 167 inode_attr_in(ip, tmp); 168 169 if (S_ISREG(ip->i_di.di_mode)) { 170 tmp->i_op = &gfs2_file_iops; 171 tmp->i_fop = &gfs2_file_fops; 172 tmp->i_mapping->a_ops = &gfs2_file_aops; 173 } else if (S_ISDIR(ip->i_di.di_mode)) { 174 tmp->i_op = &gfs2_dir_iops; 175 tmp->i_fop = &gfs2_dir_fops; 176 } else if (S_ISLNK(ip->i_di.di_mode)) { 177 tmp->i_op = &gfs2_symlink_iops; 178 } else { 179 tmp->i_op = &gfs2_dev_iops; 180 init_special_inode(tmp, tmp->i_mode, tmp->i_rdev); 181 } 182 183 tmp->u.generic_ip = NULL; 184 185 for (;;) { 186 spin_lock(&ip->i_spin); 187 if (!ip->i_vnode) 188 break; 189 inode = igrab(ip->i_vnode); 190 spin_unlock(&ip->i_spin); 191 192 if (inode) { 193 iput(tmp); 194 return inode; 195 } 196 yield(); 197 } 198 199 inode = tmp; 200 201 gfs2_inode_hold(ip); 202 ip->i_vnode = inode; 203 inode->u.generic_ip = ip; 204 205 spin_unlock(&ip->i_spin); 206 207 insert_inode_hash(inode); 208 209 return inode; 210 } 211 212 static int iget_test(struct inode *inode, void *opaque) 213 { 214 struct gfs2_inode *ip = inode->u.generic_ip; 215 struct gfs2_inum *inum = (struct gfs2_inum *)opaque; 216 217 if (ip && ip->i_num.no_addr == inum->no_addr) 218 return 1; 219 220 return 0; 221 } 222 223 struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum) 224 { 225 return ilookup5(sb, (unsigned long)inum->no_formal_ino, 226 iget_test, inum); 227 } 228 229 void gfs2_inode_min_init(struct gfs2_inode *ip, unsigned int type) 230 { 231 if (!test_and_set_bit(GIF_MIN_INIT, &ip->i_flags)) { 232 ip->i_di.di_nlink = 1; 233 ip->i_di.di_mode = DT2IF(type); 234 } 235 } 236 237 /** 238 * gfs2_inode_refresh - Refresh the incore copy of the dinode 239 * @ip: The GFS2 inode 240 * 241 * Returns: errno 242 */ 243 244 int gfs2_inode_refresh(struct gfs2_inode *ip) 245 { 246 struct buffer_head *dibh; 247 int error; 248 249 error = gfs2_meta_inode_buffer(ip, &dibh); 250 if (error) 251 return error; 252 253 if (gfs2_metatype_check(ip->i_sbd, dibh, GFS2_METATYPE_DI)) { 254 brelse(dibh); 255 return -EIO; 256 } 257 258 gfs2_dinode_in(&ip->i_di, dibh->b_data); 259 set_bit(GIF_MIN_INIT, &ip->i_flags); 260 261 brelse(dibh); 262 263 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) { 264 if (gfs2_consist_inode(ip)) 265 gfs2_dinode_print(&ip->i_di); 266 return -EIO; 267 } 268 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino) 269 return -ESTALE; 270 271 ip->i_vn = ip->i_gl->gl_vn; 272 273 return 0; 274 } 275 276 /** 277 * inode_create - create a struct gfs2_inode 278 * @i_gl: The glock covering the inode 279 * @inum: The inode number 280 * @io_gl: the iopen glock to acquire/hold (using holder in new gfs2_inode) 281 * @io_state: the state the iopen glock should be acquired in 282 * @ipp: pointer to put the returned inode in 283 * 284 * Returns: errno 285 */ 286 287 static int inode_create(struct gfs2_glock *i_gl, const struct gfs2_inum *inum, 288 struct gfs2_glock *io_gl, unsigned int io_state, 289 struct gfs2_inode **ipp) 290 { 291 struct gfs2_sbd *sdp = i_gl->gl_sbd; 292 struct gfs2_inode *ip; 293 int error = 0; 294 295 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL); 296 if (!ip) 297 return -ENOMEM; 298 memset(ip, 0, sizeof(struct gfs2_inode)); 299 300 ip->i_num = *inum; 301 302 atomic_set(&ip->i_count, 1); 303 304 ip->i_vn = i_gl->gl_vn - 1; 305 306 ip->i_gl = i_gl; 307 ip->i_sbd = sdp; 308 309 spin_lock_init(&ip->i_spin); 310 init_rwsem(&ip->i_rw_mutex); 311 312 ip->i_greedy = gfs2_tune_get(sdp, gt_greedy_default); 313 314 error = gfs2_glock_nq_init(io_gl, 315 io_state, GL_LOCAL_EXCL | GL_EXACT, 316 &ip->i_iopen_gh); 317 if (error) 318 goto fail; 319 ip->i_iopen_gh.gh_owner = NULL; 320 321 spin_lock(&io_gl->gl_spin); 322 gfs2_glock_hold(i_gl); 323 io_gl->gl_object = i_gl; 324 spin_unlock(&io_gl->gl_spin); 325 326 gfs2_glock_hold(i_gl); 327 i_gl->gl_object = ip; 328 329 atomic_inc(&sdp->sd_inode_count); 330 331 *ipp = ip; 332 333 return 0; 334 335 fail: 336 gfs2_meta_cache_flush(ip); 337 kmem_cache_free(gfs2_inode_cachep, ip); 338 *ipp = NULL; 339 340 return error; 341 } 342 343 /** 344 * gfs2_inode_get - Create or get a reference on an inode 345 * @i_gl: The glock covering the inode 346 * @inum: The inode number 347 * @create: 348 * @ipp: pointer to put the returned inode in 349 * 350 * Returns: errno 351 */ 352 353 int gfs2_inode_get(struct gfs2_glock *i_gl, const struct gfs2_inum *inum, 354 int create, struct gfs2_inode **ipp) 355 { 356 struct gfs2_sbd *sdp = i_gl->gl_sbd; 357 struct gfs2_glock *io_gl; 358 int error = 0; 359 360 gfs2_glmutex_lock(i_gl); 361 362 *ipp = i_gl->gl_object; 363 if (*ipp) { 364 error = -ESTALE; 365 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino) 366 goto out; 367 atomic_inc(&(*ipp)->i_count); 368 error = 0; 369 goto out; 370 } 371 372 if (!create) 373 goto out; 374 375 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, 376 CREATE, &io_gl); 377 if (!error) { 378 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp); 379 gfs2_glock_put(io_gl); 380 } 381 382 out: 383 gfs2_glmutex_unlock(i_gl); 384 385 return error; 386 } 387 388 void gfs2_inode_hold(struct gfs2_inode *ip) 389 { 390 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0); 391 atomic_inc(&ip->i_count); 392 } 393 394 void gfs2_inode_put(struct gfs2_inode *ip) 395 { 396 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0); 397 atomic_dec(&ip->i_count); 398 } 399 400 void gfs2_inode_destroy(struct gfs2_inode *ip) 401 { 402 struct gfs2_sbd *sdp = ip->i_sbd; 403 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl; 404 struct gfs2_glock *i_gl = ip->i_gl; 405 406 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count)); 407 gfs2_assert(sdp, io_gl->gl_object == i_gl); 408 409 spin_lock(&io_gl->gl_spin); 410 io_gl->gl_object = NULL; 411 gfs2_glock_put(i_gl); 412 spin_unlock(&io_gl->gl_spin); 413 414 gfs2_glock_dq_uninit(&ip->i_iopen_gh); 415 416 gfs2_meta_cache_flush(ip); 417 kmem_cache_free(gfs2_inode_cachep, ip); 418 419 i_gl->gl_object = NULL; 420 gfs2_glock_put(i_gl); 421 422 atomic_dec(&sdp->sd_inode_count); 423 } 424 425 static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul) 426 { 427 struct gfs2_sbd *sdp = ip->i_sbd; 428 struct gfs2_alloc *al; 429 struct gfs2_rgrpd *rgd; 430 int error; 431 432 if (ip->i_di.di_blocks != 1) { 433 if (gfs2_consist_inode(ip)) 434 gfs2_dinode_print(&ip->i_di); 435 return -EIO; 436 } 437 438 al = gfs2_alloc_get(ip); 439 440 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 441 if (error) 442 goto out; 443 444 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 445 if (error) 446 goto out_qs; 447 448 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 449 if (!rgd) { 450 gfs2_consist_inode(ip); 451 error = -EIO; 452 goto out_rindex_relse; 453 } 454 455 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 456 &al->al_rgd_gh); 457 if (error) 458 goto out_rindex_relse; 459 460 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED + 461 RES_STATFS + RES_QUOTA, 1); 462 if (error) 463 goto out_rg_gunlock; 464 465 gfs2_trans_add_gl(ip->i_gl); 466 467 gfs2_free_di(rgd, ip); 468 469 error = gfs2_unlinked_ondisk_rm(sdp, ul); 470 471 gfs2_trans_end(sdp); 472 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 473 474 out_rg_gunlock: 475 gfs2_glock_dq_uninit(&al->al_rgd_gh); 476 477 out_rindex_relse: 478 gfs2_glock_dq_uninit(&al->al_ri_gh); 479 480 out_qs: 481 gfs2_quota_unhold(ip); 482 483 out: 484 gfs2_alloc_put(ip); 485 486 return error; 487 } 488 489 /** 490 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode) 491 * @sdp: the filesystem 492 * @inum: the inode number to deallocate 493 * @io_gh: a holder for the iopen glock for this inode 494 * 495 * Returns: errno 496 */ 497 498 static int inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul, 499 struct gfs2_holder *io_gh) 500 { 501 struct gfs2_inode *ip; 502 struct gfs2_holder i_gh; 503 int error; 504 505 error = gfs2_glock_nq_num(sdp, 506 ul->ul_ut.ut_inum.no_addr, &gfs2_inode_glops, 507 LM_ST_EXCLUSIVE, 0, &i_gh); 508 if (error) 509 return error; 510 511 /* We reacquire the iopen lock here to avoid a race with the NFS server 512 calling gfs2_read_inode() with the inode number of a inode we're in 513 the process of deallocating. And we can't keep our hold on the lock 514 from inode_dealloc_init() for deadlock reasons. */ 515 516 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY, io_gh); 517 error = gfs2_glock_nq(io_gh); 518 switch (error) { 519 case 0: 520 break; 521 case GLR_TRYFAILED: 522 error = 1; 523 default: 524 goto out; 525 } 526 527 gfs2_assert_warn(sdp, !i_gh.gh_gl->gl_object); 528 error = inode_create(i_gh.gh_gl, &ul->ul_ut.ut_inum, io_gh->gh_gl, 529 LM_ST_EXCLUSIVE, &ip); 530 531 gfs2_glock_dq(io_gh); 532 533 if (error) 534 goto out; 535 536 error = gfs2_inode_refresh(ip); 537 if (error) 538 goto out_iput; 539 540 if (ip->i_di.di_nlink) { 541 if (gfs2_consist_inode(ip)) 542 gfs2_dinode_print(&ip->i_di); 543 error = -EIO; 544 goto out_iput; 545 } 546 547 if (S_ISDIR(ip->i_di.di_mode) && 548 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) { 549 error = gfs2_dir_exhash_dealloc(ip); 550 if (error) 551 goto out_iput; 552 } 553 554 if (ip->i_di.di_eattr) { 555 error = gfs2_ea_dealloc(ip); 556 if (error) 557 goto out_iput; 558 } 559 560 if (!gfs2_is_stuffed(ip)) { 561 error = gfs2_file_dealloc(ip); 562 if (error) 563 goto out_iput; 564 } 565 566 error = dinode_dealloc(ip, ul); 567 if (error) 568 goto out_iput; 569 570 out_iput: 571 gfs2_glmutex_lock(i_gh.gh_gl); 572 gfs2_inode_put(ip); 573 gfs2_inode_destroy(ip); 574 gfs2_glmutex_unlock(i_gh.gh_gl); 575 576 out: 577 gfs2_glock_dq_uninit(&i_gh); 578 579 return error; 580 } 581 582 /** 583 * try_inode_dealloc - Try to deallocate an inode and all its blocks 584 * @sdp: the filesystem 585 * 586 * Returns: 0 on success, -errno on error, 1 on busy (inode open) 587 */ 588 589 static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul) 590 { 591 struct gfs2_holder io_gh; 592 int error = 0; 593 594 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum); 595 596 error = gfs2_glock_nq_num(sdp, 597 ul->ul_ut.ut_inum.no_addr, &gfs2_iopen_glops, 598 LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB, &io_gh); 599 switch (error) { 600 case 0: 601 break; 602 case GLR_TRYFAILED: 603 return 1; 604 default: 605 return error; 606 } 607 608 gfs2_glock_dq(&io_gh); 609 error = inode_dealloc(sdp, ul, &io_gh); 610 gfs2_holder_uninit(&io_gh); 611 612 return error; 613 } 614 615 static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul) 616 { 617 struct gfs2_rgrpd *rgd; 618 struct gfs2_holder ri_gh, rgd_gh; 619 int error; 620 621 error = gfs2_rindex_hold(sdp, &ri_gh); 622 if (error) 623 return error; 624 625 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr); 626 if (!rgd) { 627 gfs2_consist(sdp); 628 error = -EIO; 629 goto out; 630 } 631 632 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh); 633 if (error) 634 goto out; 635 636 error = gfs2_trans_begin(sdp, 637 RES_RG_BIT + RES_UNLINKED + RES_STATFS, 638 0); 639 if (error) 640 goto out_gunlock; 641 642 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr); 643 gfs2_unlinked_ondisk_rm(sdp, ul); 644 645 gfs2_trans_end(sdp); 646 647 out_gunlock: 648 gfs2_glock_dq_uninit(&rgd_gh); 649 out: 650 gfs2_glock_dq_uninit(&ri_gh); 651 652 return error; 653 } 654 655 int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul) 656 { 657 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT) 658 return inode_dealloc_uninit(sdp, ul); 659 else 660 return try_inode_dealloc(sdp, ul); 661 } 662 663 /** 664 * gfs2_change_nlink - Change nlink count on inode 665 * @ip: The GFS2 inode 666 * @diff: The change in the nlink count required 667 * 668 * Returns: errno 669 */ 670 671 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 672 { 673 struct buffer_head *dibh; 674 uint32_t nlink; 675 int error; 676 677 nlink = ip->i_di.di_nlink + diff; 678 679 /* If we are reducing the nlink count, but the new value ends up being 680 bigger than the old one, we must have underflowed. */ 681 if (diff < 0 && nlink > ip->i_di.di_nlink) { 682 if (gfs2_consist_inode(ip)) 683 gfs2_dinode_print(&ip->i_di); 684 return -EIO; 685 } 686 687 error = gfs2_meta_inode_buffer(ip, &dibh); 688 if (error) 689 return error; 690 691 ip->i_di.di_nlink = nlink; 692 ip->i_di.di_ctime = get_seconds(); 693 694 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 695 gfs2_dinode_out(&ip->i_di, dibh->b_data); 696 brelse(dibh); 697 698 return 0; 699 } 700 701 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 702 { 703 struct qstr qstr; 704 qstr.name = name; 705 qstr.len = strlen(name); 706 qstr.hash = gfs2_disk_hash(qstr.name, qstr.len); 707 return gfs2_lookupi(dip, &qstr, 1, NULL); 708 } 709 710 711 /** 712 * gfs2_lookupi - Look up a filename in a directory and return its inode 713 * @d_gh: An initialized holder for the directory glock 714 * @name: The name of the inode to look for 715 * @is_root: If 1, ignore the caller's permissions 716 * @i_gh: An uninitialized holder for the new inode glock 717 * 718 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 719 * @is_root is true. 720 * 721 * Returns: errno 722 */ 723 724 struct inode *gfs2_lookupi(struct inode *dir, struct qstr *name, int is_root, 725 struct nameidata *nd) 726 727 { 728 struct super_block *sb = dir->i_sb; 729 struct gfs2_inode *ipp; 730 struct gfs2_inode *dip = dir->u.generic_ip; 731 struct gfs2_sbd *sdp = dip->i_sbd; 732 struct gfs2_holder d_gh; 733 struct gfs2_inum inum; 734 unsigned int type; 735 struct gfs2_glock *gl; 736 int error = 0; 737 struct inode *inode = NULL; 738 739 if (!name->len || name->len > GFS2_FNAMESIZE) 740 return ERR_PTR(-ENAMETOOLONG); 741 742 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 743 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 744 dir == sb->s_root->d_inode)) { 745 gfs2_inode_hold(dip); 746 ipp = dip; 747 goto done; 748 } 749 750 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 751 if (error) 752 return ERR_PTR(error); 753 754 if (!is_root) { 755 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL); 756 if (error) 757 goto out; 758 } 759 760 error = gfs2_dir_search(dir, name, &inum, &type); 761 if (error) 762 goto out; 763 764 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops, 765 CREATE, &gl); 766 if (error) 767 goto out; 768 769 error = gfs2_inode_get(gl, &inum, CREATE, &ipp); 770 if (!error) 771 gfs2_inode_min_init(ipp, type); 772 773 gfs2_glock_put(gl); 774 775 out: 776 gfs2_glock_dq_uninit(&d_gh); 777 done: 778 if (error == -ENOENT) 779 return NULL; 780 if (error == 0) { 781 inode = gfs2_ip2v(ipp); 782 gfs2_inode_put(ipp); 783 if (!inode) 784 return ERR_PTR(-ENOMEM); 785 return inode; 786 } 787 return ERR_PTR(error); 788 } 789 790 static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino) 791 { 792 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip; 793 struct buffer_head *bh; 794 struct gfs2_inum_range ir; 795 int error; 796 797 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 798 if (error) 799 return error; 800 mutex_lock(&sdp->sd_inum_mutex); 801 802 error = gfs2_meta_inode_buffer(ip, &bh); 803 if (error) { 804 mutex_unlock(&sdp->sd_inum_mutex); 805 gfs2_trans_end(sdp); 806 return error; 807 } 808 809 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 810 811 if (ir.ir_length) { 812 *formal_ino = ir.ir_start++; 813 ir.ir_length--; 814 gfs2_trans_add_bh(ip->i_gl, bh, 1); 815 gfs2_inum_range_out(&ir, 816 bh->b_data + sizeof(struct gfs2_dinode)); 817 brelse(bh); 818 mutex_unlock(&sdp->sd_inum_mutex); 819 gfs2_trans_end(sdp); 820 return 0; 821 } 822 823 brelse(bh); 824 825 mutex_unlock(&sdp->sd_inum_mutex); 826 gfs2_trans_end(sdp); 827 828 return 1; 829 } 830 831 static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino) 832 { 833 struct gfs2_inode *ip = sdp->sd_ir_inode->u.generic_ip; 834 struct gfs2_inode *m_ip = sdp->sd_inum_inode->u.generic_ip; 835 struct gfs2_holder gh; 836 struct buffer_head *bh; 837 struct gfs2_inum_range ir; 838 int error; 839 840 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 841 if (error) 842 return error; 843 844 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 845 if (error) 846 goto out; 847 mutex_lock(&sdp->sd_inum_mutex); 848 849 error = gfs2_meta_inode_buffer(ip, &bh); 850 if (error) 851 goto out_end_trans; 852 853 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 854 855 if (!ir.ir_length) { 856 struct buffer_head *m_bh; 857 uint64_t x, y; 858 859 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 860 if (error) 861 goto out_brelse; 862 863 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 864 x = y = be64_to_cpu(x); 865 ir.ir_start = x; 866 ir.ir_length = GFS2_INUM_QUANTUM; 867 x += GFS2_INUM_QUANTUM; 868 if (x < y) 869 gfs2_consist_inode(m_ip); 870 x = cpu_to_be64(x); 871 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 872 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x; 873 874 brelse(m_bh); 875 } 876 877 *formal_ino = ir.ir_start++; 878 ir.ir_length--; 879 880 gfs2_trans_add_bh(ip->i_gl, bh, 1); 881 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 882 883 out_brelse: 884 brelse(bh); 885 886 out_end_trans: 887 mutex_unlock(&sdp->sd_inum_mutex); 888 gfs2_trans_end(sdp); 889 890 out: 891 gfs2_glock_dq_uninit(&gh); 892 893 return error; 894 } 895 896 static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum) 897 { 898 int error; 899 900 error = pick_formal_ino_1(sdp, inum); 901 if (error <= 0) 902 return error; 903 904 error = pick_formal_ino_2(sdp, inum); 905 906 return error; 907 } 908 909 /** 910 * create_ok - OK to create a new on-disk inode here? 911 * @dip: Directory in which dinode is to be created 912 * @name: Name of new dinode 913 * @mode: 914 * 915 * Returns: errno 916 */ 917 918 static int create_ok(struct gfs2_inode *dip, struct qstr *name, 919 unsigned int mode) 920 { 921 int error; 922 923 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL); 924 if (error) 925 return error; 926 927 /* Don't create entries in an unlinked directory */ 928 if (!dip->i_di.di_nlink) 929 return -EPERM; 930 931 error = gfs2_dir_search(dip->i_vnode, name, NULL, NULL); 932 switch (error) { 933 case -ENOENT: 934 error = 0; 935 break; 936 case 0: 937 return -EEXIST; 938 default: 939 return error; 940 } 941 942 if (dip->i_di.di_entries == (uint32_t)-1) 943 return -EFBIG; 944 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1) 945 return -EMLINK; 946 947 return 0; 948 } 949 950 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 951 unsigned int *uid, unsigned int *gid) 952 { 953 if (dip->i_sbd->sd_args.ar_suiddir && 954 (dip->i_di.di_mode & S_ISUID) && 955 dip->i_di.di_uid) { 956 if (S_ISDIR(*mode)) 957 *mode |= S_ISUID; 958 else if (dip->i_di.di_uid != current->fsuid) 959 *mode &= ~07111; 960 *uid = dip->i_di.di_uid; 961 } else 962 *uid = current->fsuid; 963 964 if (dip->i_di.di_mode & S_ISGID) { 965 if (S_ISDIR(*mode)) 966 *mode |= S_ISGID; 967 *gid = dip->i_di.di_gid; 968 } else 969 *gid = current->fsgid; 970 } 971 972 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul) 973 { 974 struct gfs2_sbd *sdp = dip->i_sbd; 975 int error; 976 977 gfs2_alloc_get(dip); 978 979 dip->i_alloc.al_requested = RES_DINODE; 980 error = gfs2_inplace_reserve(dip); 981 if (error) 982 goto out; 983 984 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED + 985 RES_STATFS, 0); 986 if (error) 987 goto out_ipreserv; 988 989 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip); 990 991 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT; 992 error = gfs2_unlinked_ondisk_add(sdp, ul); 993 994 gfs2_trans_end(sdp); 995 996 out_ipreserv: 997 gfs2_inplace_release(dip); 998 999 out: 1000 gfs2_alloc_put(dip); 1001 1002 return error; 1003 } 1004 1005 /** 1006 * init_dinode - Fill in a new dinode structure 1007 * @dip: the directory this inode is being created in 1008 * @gl: The glock covering the new inode 1009 * @inum: the inode number 1010 * @mode: the file permissions 1011 * @uid: 1012 * @gid: 1013 * 1014 */ 1015 1016 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 1017 struct gfs2_inum *inum, unsigned int mode, 1018 unsigned int uid, unsigned int gid) 1019 { 1020 struct gfs2_sbd *sdp = dip->i_sbd; 1021 struct gfs2_dinode *di; 1022 struct buffer_head *dibh; 1023 1024 dibh = gfs2_meta_new(gl, inum->no_addr); 1025 gfs2_trans_add_bh(gl, dibh, 1); 1026 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 1027 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 1028 di = (struct gfs2_dinode *)dibh->b_data; 1029 1030 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 1031 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 1032 di->di_mode = cpu_to_be32(mode); 1033 di->di_uid = cpu_to_be32(uid); 1034 di->di_gid = cpu_to_be32(gid); 1035 di->di_nlink = cpu_to_be32(0); 1036 di->di_size = cpu_to_be64(0); 1037 di->di_blocks = cpu_to_be64(1); 1038 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 1039 di->di_major = di->di_minor = cpu_to_be32(0); 1040 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 1041 di->__pad[0] = di->__pad[1] = 0; 1042 di->di_flags = cpu_to_be32(0); 1043 1044 if (S_ISREG(mode)) { 1045 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 1046 gfs2_tune_get(sdp, gt_new_files_jdata)) 1047 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 1048 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 1049 gfs2_tune_get(sdp, gt_new_files_directio)) 1050 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 1051 } else if (S_ISDIR(mode)) { 1052 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 1053 GFS2_DIF_INHERIT_DIRECTIO); 1054 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 1055 GFS2_DIF_INHERIT_JDATA); 1056 } 1057 1058 di->__pad1 = 0; 1059 di->di_height = cpu_to_be32(0); 1060 di->__pad2 = 0; 1061 di->__pad3 = 0; 1062 di->di_depth = cpu_to_be16(0); 1063 di->di_entries = cpu_to_be32(0); 1064 memset(&di->__pad4, 0, sizeof(di->__pad4)); 1065 di->di_eattr = cpu_to_be64(0); 1066 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 1067 1068 brelse(dibh); 1069 } 1070 1071 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 1072 unsigned int mode, struct gfs2_unlinked *ul) 1073 { 1074 struct gfs2_sbd *sdp = dip->i_sbd; 1075 unsigned int uid, gid; 1076 int error; 1077 1078 munge_mode_uid_gid(dip, &mode, &uid, &gid); 1079 1080 gfs2_alloc_get(dip); 1081 1082 error = gfs2_quota_lock(dip, uid, gid); 1083 if (error) 1084 goto out; 1085 1086 error = gfs2_quota_check(dip, uid, gid); 1087 if (error) 1088 goto out_quota; 1089 1090 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED + 1091 RES_QUOTA, 0); 1092 if (error) 1093 goto out_quota; 1094 1095 ul->ul_ut.ut_flags = 0; 1096 error = gfs2_unlinked_ondisk_munge(sdp, ul); 1097 1098 init_dinode(dip, gl, &ul->ul_ut.ut_inum, 1099 mode, uid, gid); 1100 1101 gfs2_quota_change(dip, +1, uid, gid); 1102 1103 gfs2_trans_end(sdp); 1104 1105 out_quota: 1106 gfs2_quota_unlock(dip); 1107 1108 out: 1109 gfs2_alloc_put(dip); 1110 1111 return error; 1112 } 1113 1114 static int link_dinode(struct gfs2_inode *dip, struct qstr *name, 1115 struct gfs2_inode *ip, struct gfs2_unlinked *ul) 1116 { 1117 struct gfs2_sbd *sdp = dip->i_sbd; 1118 struct gfs2_alloc *al; 1119 int alloc_required; 1120 struct buffer_head *dibh; 1121 int error; 1122 1123 al = gfs2_alloc_get(dip); 1124 1125 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 1126 if (error) 1127 goto fail; 1128 1129 error = alloc_required = gfs2_diradd_alloc_required(dip->i_vnode, name); 1130 if (alloc_required < 0) 1131 goto fail; 1132 if (alloc_required) { 1133 error = gfs2_quota_check(dip, dip->i_di.di_uid, 1134 dip->i_di.di_gid); 1135 if (error) 1136 goto fail_quota_locks; 1137 1138 al->al_requested = sdp->sd_max_dirres; 1139 1140 error = gfs2_inplace_reserve(dip); 1141 if (error) 1142 goto fail_quota_locks; 1143 1144 error = gfs2_trans_begin(sdp, 1145 sdp->sd_max_dirres + 1146 al->al_rgd->rd_ri.ri_length + 1147 2 * RES_DINODE + RES_UNLINKED + 1148 RES_STATFS + RES_QUOTA, 0); 1149 if (error) 1150 goto fail_ipreserv; 1151 } else { 1152 error = gfs2_trans_begin(sdp, 1153 RES_LEAF + 1154 2 * RES_DINODE + 1155 RES_UNLINKED, 0); 1156 if (error) 1157 goto fail_quota_locks; 1158 } 1159 1160 error = gfs2_dir_add(dip->i_vnode, name, &ip->i_num, IF2DT(ip->i_di.di_mode)); 1161 if (error) 1162 goto fail_end_trans; 1163 1164 error = gfs2_meta_inode_buffer(ip, &dibh); 1165 if (error) 1166 goto fail_end_trans; 1167 ip->i_di.di_nlink = 1; 1168 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1169 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1170 brelse(dibh); 1171 1172 error = gfs2_unlinked_ondisk_rm(sdp, ul); 1173 if (error) 1174 goto fail_end_trans; 1175 1176 return 0; 1177 1178 fail_end_trans: 1179 gfs2_trans_end(sdp); 1180 1181 fail_ipreserv: 1182 if (dip->i_alloc.al_rgd) 1183 gfs2_inplace_release(dip); 1184 1185 fail_quota_locks: 1186 gfs2_quota_unlock(dip); 1187 1188 fail: 1189 gfs2_alloc_put(dip); 1190 1191 return error; 1192 } 1193 1194 /** 1195 * gfs2_createi - Create a new inode 1196 * @ghs: An array of two holders 1197 * @name: The name of the new file 1198 * @mode: the permissions on the new inode 1199 * 1200 * @ghs[0] is an initialized holder for the directory 1201 * @ghs[1] is the holder for the inode lock 1202 * 1203 * If the return value is not NULL, the glocks on both the directory and the new 1204 * file are held. A transaction has been started and an inplace reservation 1205 * is held, as well. 1206 * 1207 * Returns: An inode 1208 */ 1209 1210 struct inode *gfs2_createi(struct gfs2_holder *ghs, struct qstr *name, 1211 unsigned int mode) 1212 { 1213 struct inode *inode; 1214 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 1215 struct gfs2_sbd *sdp = dip->i_sbd; 1216 struct gfs2_unlinked *ul; 1217 struct gfs2_inode *ip; 1218 int error; 1219 1220 if (!name->len || name->len > GFS2_FNAMESIZE) 1221 return ERR_PTR(-ENAMETOOLONG); 1222 1223 error = gfs2_unlinked_get(sdp, &ul); 1224 if (error) 1225 return ERR_PTR(error); 1226 1227 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 1228 error = gfs2_glock_nq(ghs); 1229 if (error) 1230 goto fail; 1231 1232 error = create_ok(dip, name, mode); 1233 if (error) 1234 goto fail_gunlock; 1235 1236 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino); 1237 if (error) 1238 goto fail_gunlock; 1239 1240 error = alloc_dinode(dip, ul); 1241 if (error) 1242 goto fail_gunlock; 1243 1244 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) { 1245 gfs2_glock_dq(ghs); 1246 1247 error = gfs2_glock_nq_num(sdp, 1248 ul->ul_ut.ut_inum.no_addr, 1249 &gfs2_inode_glops, 1250 LM_ST_EXCLUSIVE, GL_SKIP, 1251 ghs + 1); 1252 if (error) { 1253 gfs2_unlinked_put(sdp, ul); 1254 return ERR_PTR(error); 1255 } 1256 1257 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 1258 error = gfs2_glock_nq(ghs); 1259 if (error) { 1260 gfs2_glock_dq_uninit(ghs + 1); 1261 gfs2_unlinked_put(sdp, ul); 1262 return ERR_PTR(error); 1263 } 1264 1265 error = create_ok(dip, name, mode); 1266 if (error) 1267 goto fail_gunlock2; 1268 } else { 1269 error = gfs2_glock_nq_num(sdp, 1270 ul->ul_ut.ut_inum.no_addr, 1271 &gfs2_inode_glops, 1272 LM_ST_EXCLUSIVE, GL_SKIP, 1273 ghs + 1); 1274 if (error) 1275 goto fail_gunlock; 1276 } 1277 1278 error = make_dinode(dip, ghs[1].gh_gl, mode, ul); 1279 if (error) 1280 goto fail_gunlock2; 1281 1282 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip); 1283 if (error) 1284 goto fail_gunlock2; 1285 1286 error = gfs2_inode_refresh(ip); 1287 if (error) 1288 goto fail_iput; 1289 1290 error = gfs2_acl_create(dip, ip); 1291 if (error) 1292 goto fail_iput; 1293 1294 error = link_dinode(dip, name, ip, ul); 1295 if (error) 1296 goto fail_iput; 1297 1298 gfs2_unlinked_put(sdp, ul); 1299 1300 inode = gfs2_ip2v(ip); 1301 gfs2_inode_put(ip); 1302 if (!inode) 1303 return ERR_PTR(-ENOMEM); 1304 return inode; 1305 1306 fail_iput: 1307 gfs2_inode_put(ip); 1308 1309 fail_gunlock2: 1310 gfs2_glock_dq_uninit(ghs + 1); 1311 1312 fail_gunlock: 1313 gfs2_glock_dq(ghs); 1314 1315 fail: 1316 gfs2_unlinked_put(sdp, ul); 1317 1318 return ERR_PTR(error); 1319 } 1320 1321 /** 1322 * gfs2_unlinki - Unlink a file 1323 * @dip: The inode of the directory 1324 * @name: The name of the file to be unlinked 1325 * @ip: The inode of the file to be removed 1326 * 1327 * Assumes Glocks on both dip and ip are held. 1328 * 1329 * Returns: errno 1330 */ 1331 1332 int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name, 1333 struct gfs2_inode *ip, struct gfs2_unlinked *ul) 1334 { 1335 struct gfs2_sbd *sdp = dip->i_sbd; 1336 int error; 1337 1338 error = gfs2_dir_del(dip, name); 1339 if (error) 1340 return error; 1341 1342 error = gfs2_change_nlink(ip, -1); 1343 if (error) 1344 return error; 1345 1346 /* If this inode is being unlinked from the directory structure, 1347 we need to mark that in the log so that it isn't lost during 1348 a crash. */ 1349 1350 if (!ip->i_di.di_nlink) { 1351 ul->ul_ut.ut_inum = ip->i_num; 1352 error = gfs2_unlinked_ondisk_add(sdp, ul); 1353 if (!error) 1354 set_bit(GLF_STICKY, &ip->i_gl->gl_flags); 1355 } 1356 1357 return error; 1358 } 1359 1360 /** 1361 * gfs2_rmdiri - Remove a directory 1362 * @dip: The parent directory of the directory to be removed 1363 * @name: The name of the directory to be removed 1364 * @ip: The GFS2 inode of the directory to be removed 1365 * 1366 * Assumes Glocks on dip and ip are held 1367 * 1368 * Returns: errno 1369 */ 1370 1371 int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name, 1372 struct gfs2_inode *ip, struct gfs2_unlinked *ul) 1373 { 1374 struct gfs2_sbd *sdp = dip->i_sbd; 1375 struct qstr dotname; 1376 int error; 1377 1378 if (ip->i_di.di_entries != 2) { 1379 if (gfs2_consist_inode(ip)) 1380 gfs2_dinode_print(&ip->i_di); 1381 return -EIO; 1382 } 1383 1384 error = gfs2_dir_del(dip, name); 1385 if (error) 1386 return error; 1387 1388 error = gfs2_change_nlink(dip, -1); 1389 if (error) 1390 return error; 1391 1392 dotname.len = 1; 1393 dotname.name = "."; 1394 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len); 1395 error = gfs2_dir_del(ip, &dotname); 1396 if (error) 1397 return error; 1398 1399 dotname.len = 2; 1400 dotname.name = ".."; 1401 dotname.hash = gfs2_disk_hash(dotname.name, dotname.len); 1402 error = gfs2_dir_del(ip, &dotname); 1403 if (error) 1404 return error; 1405 1406 error = gfs2_change_nlink(ip, -2); 1407 if (error) 1408 return error; 1409 1410 /* This inode is being unlinked from the directory structure and 1411 we need to mark that in the log so that it isn't lost during 1412 a crash. */ 1413 1414 ul->ul_ut.ut_inum = ip->i_num; 1415 error = gfs2_unlinked_ondisk_add(sdp, ul); 1416 if (!error) 1417 set_bit(GLF_STICKY, &ip->i_gl->gl_flags); 1418 1419 return error; 1420 } 1421 1422 /* 1423 * gfs2_unlink_ok - check to see that a inode is still in a directory 1424 * @dip: the directory 1425 * @name: the name of the file 1426 * @ip: the inode 1427 * 1428 * Assumes that the lock on (at least) @dip is held. 1429 * 1430 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 1431 */ 1432 1433 int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name, 1434 struct gfs2_inode *ip) 1435 { 1436 struct gfs2_inum inum; 1437 unsigned int type; 1438 int error; 1439 1440 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode)) 1441 return -EPERM; 1442 1443 if ((dip->i_di.di_mode & S_ISVTX) && 1444 dip->i_di.di_uid != current->fsuid && 1445 ip->i_di.di_uid != current->fsuid && 1446 !capable(CAP_FOWNER)) 1447 return -EPERM; 1448 1449 if (IS_APPEND(dip->i_vnode)) 1450 return -EPERM; 1451 1452 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL); 1453 if (error) 1454 return error; 1455 1456 error = gfs2_dir_search(dip->i_vnode, name, &inum, &type); 1457 if (error) 1458 return error; 1459 1460 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1461 return -ENOENT; 1462 1463 if (IF2DT(ip->i_di.di_mode) != type) { 1464 gfs2_consist_inode(dip); 1465 return -EIO; 1466 } 1467 1468 return 0; 1469 } 1470 1471 /* 1472 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1473 * @this: move this 1474 * @to: to here 1475 * 1476 * Follow @to back to the root and make sure we don't encounter @this 1477 * Assumes we already hold the rename lock. 1478 * 1479 * Returns: errno 1480 */ 1481 1482 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1483 { 1484 struct inode *dir = to->i_vnode; 1485 struct super_block *sb = dir->i_sb; 1486 struct inode *tmp; 1487 struct qstr dotdot; 1488 int error = 0; 1489 1490 memset(&dotdot, 0, sizeof(struct qstr)); 1491 dotdot.name = ".."; 1492 dotdot.len = 2; 1493 dotdot.hash = gfs2_disk_hash(dotdot.name, dotdot.len); 1494 1495 igrab(dir); 1496 1497 for (;;) { 1498 if (dir == this->i_vnode) { 1499 error = -EINVAL; 1500 break; 1501 } 1502 if (dir == sb->s_root->d_inode) { 1503 error = 0; 1504 break; 1505 } 1506 1507 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1508 if (IS_ERR(tmp)) { 1509 error = PTR_ERR(tmp); 1510 break; 1511 } 1512 1513 iput(dir); 1514 dir = tmp; 1515 } 1516 1517 iput(dir); 1518 1519 return error; 1520 } 1521 1522 /** 1523 * gfs2_readlinki - return the contents of a symlink 1524 * @ip: the symlink's inode 1525 * @buf: a pointer to the buffer to be filled 1526 * @len: a pointer to the length of @buf 1527 * 1528 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1529 * to be freed by the caller. 1530 * 1531 * Returns: errno 1532 */ 1533 1534 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1535 { 1536 struct gfs2_holder i_gh; 1537 struct buffer_head *dibh; 1538 unsigned int x; 1539 int error; 1540 1541 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1542 error = gfs2_glock_nq_atime(&i_gh); 1543 if (error) { 1544 gfs2_holder_uninit(&i_gh); 1545 return error; 1546 } 1547 1548 if (!ip->i_di.di_size) { 1549 gfs2_consist_inode(ip); 1550 error = -EIO; 1551 goto out; 1552 } 1553 1554 error = gfs2_meta_inode_buffer(ip, &dibh); 1555 if (error) 1556 goto out; 1557 1558 x = ip->i_di.di_size + 1; 1559 if (x > *len) { 1560 *buf = kmalloc(x, GFP_KERNEL); 1561 if (!*buf) { 1562 error = -ENOMEM; 1563 goto out_brelse; 1564 } 1565 } 1566 1567 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1568 *len = x; 1569 1570 out_brelse: 1571 brelse(dibh); 1572 1573 out: 1574 gfs2_glock_dq_uninit(&i_gh); 1575 1576 return error; 1577 } 1578 1579 /** 1580 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1581 * conditionally update the inode's atime 1582 * @gh: the holder to acquire 1583 * 1584 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1585 * Update if the difference between the current time and the inode's current 1586 * atime is greater than an interval specified at mount. 1587 * 1588 * Returns: errno 1589 */ 1590 1591 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1592 { 1593 struct gfs2_glock *gl = gh->gh_gl; 1594 struct gfs2_sbd *sdp = gl->gl_sbd; 1595 struct gfs2_inode *ip = gl->gl_object; 1596 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1597 unsigned int state; 1598 int flags; 1599 int error; 1600 1601 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1602 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1603 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1604 return -EINVAL; 1605 1606 state = gh->gh_state; 1607 flags = gh->gh_flags; 1608 1609 error = gfs2_glock_nq(gh); 1610 if (error) 1611 return error; 1612 1613 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1614 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1615 return 0; 1616 1617 curtime = get_seconds(); 1618 if (curtime - ip->i_di.di_atime >= quantum) { 1619 gfs2_glock_dq(gh); 1620 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 1621 gh->gh_flags & ~LM_FLAG_ANY, 1622 gh); 1623 error = gfs2_glock_nq(gh); 1624 if (error) 1625 return error; 1626 1627 /* Verify that atime hasn't been updated while we were 1628 trying to get exclusive lock. */ 1629 1630 curtime = get_seconds(); 1631 if (curtime - ip->i_di.di_atime >= quantum) { 1632 struct buffer_head *dibh; 1633 1634 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1635 if (error == -EROFS) 1636 return 0; 1637 if (error) 1638 goto fail; 1639 1640 error = gfs2_meta_inode_buffer(ip, &dibh); 1641 if (error) 1642 goto fail_end_trans; 1643 1644 ip->i_di.di_atime = curtime; 1645 1646 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1647 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1648 brelse(dibh); 1649 1650 gfs2_trans_end(sdp); 1651 } 1652 1653 /* If someone else has asked for the glock, 1654 unlock and let them have it. Then reacquire 1655 in the original state. */ 1656 if (gfs2_glock_is_blocking(gl)) { 1657 gfs2_glock_dq(gh); 1658 gfs2_holder_reinit(state, flags, gh); 1659 return gfs2_glock_nq(gh); 1660 } 1661 } 1662 1663 return 0; 1664 1665 fail_end_trans: 1666 gfs2_trans_end(sdp); 1667 1668 fail: 1669 gfs2_glock_dq(gh); 1670 1671 return error; 1672 } 1673 1674 /** 1675 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1676 * @arg_a: the first structure 1677 * @arg_b: the second structure 1678 * 1679 * Returns: 1 if A > B 1680 * -1 if A < B 1681 * 0 if A = B 1682 */ 1683 1684 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1685 { 1686 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a; 1687 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b; 1688 struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1689 struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1690 int ret = 0; 1691 1692 if (a->ln_number > b->ln_number) 1693 ret = 1; 1694 else if (a->ln_number < b->ln_number) 1695 ret = -1; 1696 else { 1697 if (gh_a->gh_state == LM_ST_SHARED && 1698 gh_b->gh_state == LM_ST_EXCLUSIVE) 1699 ret = 1; 1700 else if (gh_a->gh_state == LM_ST_SHARED && 1701 (gh_b->gh_flags & GL_ATIME)) 1702 ret = 1; 1703 } 1704 1705 return ret; 1706 } 1707 1708 /** 1709 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1710 * atime update 1711 * @num_gh: the number of structures 1712 * @ghs: an array of struct gfs2_holder structures 1713 * 1714 * Returns: 0 on success (all glocks acquired), 1715 * errno on failure (no glocks acquired) 1716 */ 1717 1718 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1719 { 1720 struct gfs2_holder **p; 1721 unsigned int x; 1722 int error = 0; 1723 1724 if (!num_gh) 1725 return 0; 1726 1727 if (num_gh == 1) { 1728 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1729 if (ghs->gh_flags & GL_ATIME) 1730 error = gfs2_glock_nq_atime(ghs); 1731 else 1732 error = gfs2_glock_nq(ghs); 1733 return error; 1734 } 1735 1736 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1737 if (!p) 1738 return -ENOMEM; 1739 1740 for (x = 0; x < num_gh; x++) 1741 p[x] = &ghs[x]; 1742 1743 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1744 1745 for (x = 0; x < num_gh; x++) { 1746 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1747 1748 if (p[x]->gh_flags & GL_ATIME) 1749 error = gfs2_glock_nq_atime(p[x]); 1750 else 1751 error = gfs2_glock_nq(p[x]); 1752 1753 if (error) { 1754 while (x--) 1755 gfs2_glock_dq(p[x]); 1756 break; 1757 } 1758 } 1759 1760 kfree(p); 1761 1762 return error; 1763 } 1764 1765 /** 1766 * gfs2_try_toss_vnode - See if we can toss a vnode from memory 1767 * @ip: the inode 1768 * 1769 * Returns: 1 if the vnode was tossed 1770 */ 1771 1772 void gfs2_try_toss_vnode(struct gfs2_inode *ip) 1773 { 1774 struct inode *inode; 1775 1776 inode = gfs2_ip2v_lookup(ip); 1777 if (!inode) 1778 return; 1779 1780 d_prune_aliases(inode); 1781 1782 if (S_ISDIR(ip->i_di.di_mode)) { 1783 struct list_head *head = &inode->i_dentry; 1784 struct dentry *d = NULL; 1785 1786 spin_lock(&dcache_lock); 1787 if (list_empty(head)) 1788 spin_unlock(&dcache_lock); 1789 else { 1790 d = list_entry(head->next, struct dentry, d_alias); 1791 dget_locked(d); 1792 spin_unlock(&dcache_lock); 1793 1794 if (have_submounts(d)) 1795 dput(d); 1796 else { 1797 shrink_dcache_parent(d); 1798 dput(d); 1799 d_prune_aliases(inode); 1800 } 1801 } 1802 } 1803 1804 inode->i_nlink = 0; 1805 iput(inode); 1806 } 1807 1808 1809 static int 1810 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1811 { 1812 struct buffer_head *dibh; 1813 int error; 1814 1815 error = gfs2_meta_inode_buffer(ip, &dibh); 1816 if (!error) { 1817 error = inode_setattr(ip->i_vnode, attr); 1818 gfs2_assert_warn(ip->i_sbd, !error); 1819 gfs2_inode_attr_out(ip); 1820 1821 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1822 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1823 brelse(dibh); 1824 } 1825 return error; 1826 } 1827 1828 /** 1829 * gfs2_setattr_simple - 1830 * @ip: 1831 * @attr: 1832 * 1833 * Called with a reference on the vnode. 1834 * 1835 * Returns: errno 1836 */ 1837 1838 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1839 { 1840 int error; 1841 1842 if (current->journal_info) 1843 return __gfs2_setattr_simple(ip, attr); 1844 1845 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0); 1846 if (error) 1847 return error; 1848 1849 error = __gfs2_setattr_simple(ip, attr); 1850 1851 gfs2_trans_end(ip->i_sbd); 1852 1853 return error; 1854 } 1855 1856 int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd) 1857 { 1858 return permission(inode, mask, nd); 1859 } 1860 1861