1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2006 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 version 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 <linux/crc32.h> 19 #include <linux/lm_interface.h> 20 21 #include "gfs2.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 "util.h" 39 40 /** 41 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode 42 * @ip: The GFS2 inode (with embedded disk inode data) 43 * @inode: The Linux VFS inode 44 * 45 */ 46 47 void gfs2_inode_attr_in(struct gfs2_inode *ip) 48 { 49 struct inode *inode = &ip->i_inode; 50 struct gfs2_dinode *di = &ip->i_di; 51 52 inode->i_ino = ip->i_num.no_addr; 53 54 switch (di->di_mode & S_IFMT) { 55 case S_IFBLK: 56 case S_IFCHR: 57 inode->i_rdev = MKDEV(di->di_major, di->di_minor); 58 break; 59 default: 60 inode->i_rdev = 0; 61 break; 62 }; 63 64 inode->i_mode = di->di_mode; 65 inode->i_nlink = di->di_nlink; 66 inode->i_uid = di->di_uid; 67 inode->i_gid = di->di_gid; 68 i_size_write(inode, di->di_size); 69 inode->i_atime.tv_sec = di->di_atime; 70 inode->i_mtime.tv_sec = di->di_mtime; 71 inode->i_ctime.tv_sec = di->di_ctime; 72 inode->i_atime.tv_nsec = 0; 73 inode->i_mtime.tv_nsec = 0; 74 inode->i_ctime.tv_nsec = 0; 75 inode->i_blocks = di->di_blocks << 76 (GFS2_SB(inode)->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT); 77 78 if (di->di_flags & GFS2_DIF_IMMUTABLE) 79 inode->i_flags |= S_IMMUTABLE; 80 else 81 inode->i_flags &= ~S_IMMUTABLE; 82 83 if (di->di_flags & GFS2_DIF_APPENDONLY) 84 inode->i_flags |= S_APPEND; 85 else 86 inode->i_flags &= ~S_APPEND; 87 } 88 89 /** 90 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode 91 * @ip: The GFS2 inode 92 * 93 * Only copy out the attributes that we want the VFS layer 94 * to be able to modify. 95 */ 96 97 void gfs2_inode_attr_out(struct gfs2_inode *ip) 98 { 99 struct inode *inode = &ip->i_inode; 100 struct gfs2_dinode *di = &ip->i_di; 101 gfs2_assert_withdraw(GFS2_SB(inode), 102 (di->di_mode & S_IFMT) == (inode->i_mode & S_IFMT)); 103 di->di_mode = inode->i_mode; 104 di->di_uid = inode->i_uid; 105 di->di_gid = inode->i_gid; 106 di->di_atime = inode->i_atime.tv_sec; 107 di->di_mtime = inode->i_mtime.tv_sec; 108 di->di_ctime = inode->i_ctime.tv_sec; 109 } 110 111 static int iget_test(struct inode *inode, void *opaque) 112 { 113 struct gfs2_inode *ip = GFS2_I(inode); 114 struct gfs2_inum *inum = opaque; 115 116 if (ip && ip->i_num.no_addr == inum->no_addr) 117 return 1; 118 119 return 0; 120 } 121 122 static int iget_set(struct inode *inode, void *opaque) 123 { 124 struct gfs2_inode *ip = GFS2_I(inode); 125 struct gfs2_inum *inum = opaque; 126 127 ip->i_num = *inum; 128 return 0; 129 } 130 131 struct inode *gfs2_ilookup(struct super_block *sb, struct gfs2_inum *inum) 132 { 133 return ilookup5(sb, (unsigned long)inum->no_formal_ino, 134 iget_test, inum); 135 } 136 137 static struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum) 138 { 139 return iget5_locked(sb, (unsigned long)inum->no_formal_ino, 140 iget_test, iget_set, inum); 141 } 142 143 /** 144 * gfs2_inode_lookup - Lookup an inode 145 * @sb: The super block 146 * @inum: The inode number 147 * @type: The type of the inode 148 * 149 * Returns: A VFS inode, or an error 150 */ 151 152 struct inode *gfs2_inode_lookup(struct super_block *sb, struct gfs2_inum *inum, unsigned int type) 153 { 154 struct inode *inode = gfs2_iget(sb, inum); 155 struct gfs2_inode *ip = GFS2_I(inode); 156 struct gfs2_glock *io_gl; 157 int error; 158 159 if (inode->i_state & I_NEW) { 160 struct gfs2_sbd *sdp = GFS2_SB(inode); 161 umode_t mode = DT2IF(type); 162 inode->i_private = ip; 163 inode->i_mode = mode; 164 165 if (S_ISREG(mode)) { 166 inode->i_op = &gfs2_file_iops; 167 inode->i_fop = &gfs2_file_fops; 168 inode->i_mapping->a_ops = &gfs2_file_aops; 169 } else if (S_ISDIR(mode)) { 170 inode->i_op = &gfs2_dir_iops; 171 inode->i_fop = &gfs2_dir_fops; 172 } else if (S_ISLNK(mode)) { 173 inode->i_op = &gfs2_symlink_iops; 174 } else { 175 inode->i_op = &gfs2_dev_iops; 176 } 177 178 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_inode_glops, CREATE, &ip->i_gl); 179 if (unlikely(error)) 180 goto fail; 181 ip->i_gl->gl_object = ip; 182 183 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops, CREATE, &io_gl); 184 if (unlikely(error)) 185 goto fail_put; 186 187 ip->i_vn = ip->i_gl->gl_vn - 1; 188 error = gfs2_glock_nq_init(io_gl, LM_ST_SHARED, GL_EXACT, &ip->i_iopen_gh); 189 if (unlikely(error)) 190 goto fail_iopen; 191 192 gfs2_glock_put(io_gl); 193 unlock_new_inode(inode); 194 } 195 196 return inode; 197 fail_iopen: 198 gfs2_glock_put(io_gl); 199 fail_put: 200 ip->i_gl->gl_object = NULL; 201 gfs2_glock_put(ip->i_gl); 202 fail: 203 iput(inode); 204 return ERR_PTR(error); 205 } 206 207 /** 208 * gfs2_inode_refresh - Refresh the incore copy of the dinode 209 * @ip: The GFS2 inode 210 * 211 * Returns: errno 212 */ 213 214 int gfs2_inode_refresh(struct gfs2_inode *ip) 215 { 216 struct buffer_head *dibh; 217 int error; 218 219 error = gfs2_meta_inode_buffer(ip, &dibh); 220 if (error) 221 return error; 222 223 if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), dibh, GFS2_METATYPE_DI)) { 224 brelse(dibh); 225 return -EIO; 226 } 227 228 gfs2_dinode_in(&ip->i_di, dibh->b_data); 229 230 brelse(dibh); 231 232 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) { 233 if (gfs2_consist_inode(ip)) 234 gfs2_dinode_print(&ip->i_di); 235 return -EIO; 236 } 237 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino) 238 return -ESTALE; 239 240 ip->i_vn = ip->i_gl->gl_vn; 241 242 return 0; 243 } 244 245 int gfs2_dinode_dealloc(struct gfs2_inode *ip) 246 { 247 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); 248 struct gfs2_alloc *al; 249 struct gfs2_rgrpd *rgd; 250 int error; 251 252 if (ip->i_di.di_blocks != 1) { 253 if (gfs2_consist_inode(ip)) 254 gfs2_dinode_print(&ip->i_di); 255 return -EIO; 256 } 257 258 al = gfs2_alloc_get(ip); 259 260 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 261 if (error) 262 goto out; 263 264 error = gfs2_rindex_hold(sdp, &al->al_ri_gh); 265 if (error) 266 goto out_qs; 267 268 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 269 if (!rgd) { 270 gfs2_consist_inode(ip); 271 error = -EIO; 272 goto out_rindex_relse; 273 } 274 275 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, 276 &al->al_rgd_gh); 277 if (error) 278 goto out_rindex_relse; 279 280 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA, 1); 281 if (error) 282 goto out_rg_gunlock; 283 284 gfs2_trans_add_gl(ip->i_gl); 285 286 gfs2_free_di(rgd, ip); 287 288 gfs2_trans_end(sdp); 289 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags); 290 291 out_rg_gunlock: 292 gfs2_glock_dq_uninit(&al->al_rgd_gh); 293 out_rindex_relse: 294 gfs2_glock_dq_uninit(&al->al_ri_gh); 295 out_qs: 296 gfs2_quota_unhold(ip); 297 out: 298 gfs2_alloc_put(ip); 299 return error; 300 } 301 302 /** 303 * gfs2_change_nlink - Change nlink count on inode 304 * @ip: The GFS2 inode 305 * @diff: The change in the nlink count required 306 * 307 * Returns: errno 308 */ 309 310 int gfs2_change_nlink(struct gfs2_inode *ip, int diff) 311 { 312 struct gfs2_sbd *sdp = ip->i_inode.i_sb->s_fs_info; 313 struct buffer_head *dibh; 314 u32 nlink; 315 int error; 316 317 BUG_ON(ip->i_di.di_nlink != ip->i_inode.i_nlink); 318 nlink = ip->i_di.di_nlink + diff; 319 320 /* If we are reducing the nlink count, but the new value ends up being 321 bigger than the old one, we must have underflowed. */ 322 if (diff < 0 && nlink > ip->i_di.di_nlink) { 323 if (gfs2_consist_inode(ip)) 324 gfs2_dinode_print(&ip->i_di); 325 return -EIO; 326 } 327 328 error = gfs2_meta_inode_buffer(ip, &dibh); 329 if (error) 330 return error; 331 332 ip->i_di.di_nlink = nlink; 333 ip->i_di.di_ctime = get_seconds(); 334 ip->i_inode.i_nlink = nlink; 335 336 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 337 gfs2_dinode_out(&ip->i_di, dibh->b_data); 338 brelse(dibh); 339 mark_inode_dirty(&ip->i_inode); 340 341 if (ip->i_di.di_nlink == 0) { 342 struct gfs2_rgrpd *rgd; 343 struct gfs2_holder ri_gh, rg_gh; 344 345 error = gfs2_rindex_hold(sdp, &ri_gh); 346 if (error) 347 goto out; 348 error = -EIO; 349 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr); 350 if (!rgd) 351 goto out_norgrp; 352 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rg_gh); 353 if (error) 354 goto out_norgrp; 355 356 gfs2_unlink_di(&ip->i_inode); /* mark inode unlinked */ 357 gfs2_glock_dq_uninit(&rg_gh); 358 out_norgrp: 359 gfs2_glock_dq_uninit(&ri_gh); 360 } 361 out: 362 return error; 363 } 364 365 struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) 366 { 367 struct qstr qstr; 368 gfs2_str2qstr(&qstr, name); 369 return gfs2_lookupi(dip, &qstr, 1, NULL); 370 } 371 372 373 /** 374 * gfs2_lookupi - Look up a filename in a directory and return its inode 375 * @d_gh: An initialized holder for the directory glock 376 * @name: The name of the inode to look for 377 * @is_root: If 1, ignore the caller's permissions 378 * @i_gh: An uninitialized holder for the new inode glock 379 * 380 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless 381 * @is_root is true. 382 * 383 * Returns: errno 384 */ 385 386 struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, 387 int is_root, struct nameidata *nd) 388 { 389 struct super_block *sb = dir->i_sb; 390 struct gfs2_inode *dip = GFS2_I(dir); 391 struct gfs2_holder d_gh; 392 struct gfs2_inum inum; 393 unsigned int type; 394 int error = 0; 395 struct inode *inode = NULL; 396 397 if (!name->len || name->len > GFS2_FNAMESIZE) 398 return ERR_PTR(-ENAMETOOLONG); 399 400 if ((name->len == 1 && memcmp(name->name, ".", 1) == 0) || 401 (name->len == 2 && memcmp(name->name, "..", 2) == 0 && 402 dir == sb->s_root->d_inode)) { 403 igrab(dir); 404 return dir; 405 } 406 407 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); 408 if (error) 409 return ERR_PTR(error); 410 411 if (!is_root) { 412 error = permission(dir, MAY_EXEC, NULL); 413 if (error) 414 goto out; 415 } 416 417 error = gfs2_dir_search(dir, name, &inum, &type); 418 if (error) 419 goto out; 420 421 inode = gfs2_inode_lookup(sb, &inum, type); 422 423 out: 424 gfs2_glock_dq_uninit(&d_gh); 425 if (error == -ENOENT) 426 return NULL; 427 return inode; 428 } 429 430 static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino) 431 { 432 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 433 struct buffer_head *bh; 434 struct gfs2_inum_range ir; 435 int error; 436 437 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 438 if (error) 439 return error; 440 mutex_lock(&sdp->sd_inum_mutex); 441 442 error = gfs2_meta_inode_buffer(ip, &bh); 443 if (error) { 444 mutex_unlock(&sdp->sd_inum_mutex); 445 gfs2_trans_end(sdp); 446 return error; 447 } 448 449 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 450 451 if (ir.ir_length) { 452 *formal_ino = ir.ir_start++; 453 ir.ir_length--; 454 gfs2_trans_add_bh(ip->i_gl, bh, 1); 455 gfs2_inum_range_out(&ir, 456 bh->b_data + sizeof(struct gfs2_dinode)); 457 brelse(bh); 458 mutex_unlock(&sdp->sd_inum_mutex); 459 gfs2_trans_end(sdp); 460 return 0; 461 } 462 463 brelse(bh); 464 465 mutex_unlock(&sdp->sd_inum_mutex); 466 gfs2_trans_end(sdp); 467 468 return 1; 469 } 470 471 static int pick_formal_ino_2(struct gfs2_sbd *sdp, u64 *formal_ino) 472 { 473 struct gfs2_inode *ip = GFS2_I(sdp->sd_ir_inode); 474 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_inum_inode); 475 struct gfs2_holder gh; 476 struct buffer_head *bh; 477 struct gfs2_inum_range ir; 478 int error; 479 480 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 481 if (error) 482 return error; 483 484 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 485 if (error) 486 goto out; 487 mutex_lock(&sdp->sd_inum_mutex); 488 489 error = gfs2_meta_inode_buffer(ip, &bh); 490 if (error) 491 goto out_end_trans; 492 493 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 494 495 if (!ir.ir_length) { 496 struct buffer_head *m_bh; 497 u64 x, y; 498 499 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 500 if (error) 501 goto out_brelse; 502 503 x = *(u64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)); 504 x = y = be64_to_cpu(x); 505 ir.ir_start = x; 506 ir.ir_length = GFS2_INUM_QUANTUM; 507 x += GFS2_INUM_QUANTUM; 508 if (x < y) 509 gfs2_consist_inode(m_ip); 510 x = cpu_to_be64(x); 511 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 512 *(u64 *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x; 513 514 brelse(m_bh); 515 } 516 517 *formal_ino = ir.ir_start++; 518 ir.ir_length--; 519 520 gfs2_trans_add_bh(ip->i_gl, bh, 1); 521 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode)); 522 523 out_brelse: 524 brelse(bh); 525 out_end_trans: 526 mutex_unlock(&sdp->sd_inum_mutex); 527 gfs2_trans_end(sdp); 528 out: 529 gfs2_glock_dq_uninit(&gh); 530 return error; 531 } 532 533 static int pick_formal_ino(struct gfs2_sbd *sdp, u64 *inum) 534 { 535 int error; 536 537 error = pick_formal_ino_1(sdp, inum); 538 if (error <= 0) 539 return error; 540 541 error = pick_formal_ino_2(sdp, inum); 542 543 return error; 544 } 545 546 /** 547 * create_ok - OK to create a new on-disk inode here? 548 * @dip: Directory in which dinode is to be created 549 * @name: Name of new dinode 550 * @mode: 551 * 552 * Returns: errno 553 */ 554 555 static int create_ok(struct gfs2_inode *dip, const struct qstr *name, 556 unsigned int mode) 557 { 558 int error; 559 560 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 561 if (error) 562 return error; 563 564 /* Don't create entries in an unlinked directory */ 565 if (!dip->i_di.di_nlink) 566 return -EPERM; 567 568 error = gfs2_dir_search(&dip->i_inode, name, NULL, NULL); 569 switch (error) { 570 case -ENOENT: 571 error = 0; 572 break; 573 case 0: 574 return -EEXIST; 575 default: 576 return error; 577 } 578 579 if (dip->i_di.di_entries == (u32)-1) 580 return -EFBIG; 581 if (S_ISDIR(mode) && dip->i_di.di_nlink == (u32)-1) 582 return -EMLINK; 583 584 return 0; 585 } 586 587 static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode, 588 unsigned int *uid, unsigned int *gid) 589 { 590 if (GFS2_SB(&dip->i_inode)->sd_args.ar_suiddir && 591 (dip->i_di.di_mode & S_ISUID) && dip->i_di.di_uid) { 592 if (S_ISDIR(*mode)) 593 *mode |= S_ISUID; 594 else if (dip->i_di.di_uid != current->fsuid) 595 *mode &= ~07111; 596 *uid = dip->i_di.di_uid; 597 } else 598 *uid = current->fsuid; 599 600 if (dip->i_di.di_mode & S_ISGID) { 601 if (S_ISDIR(*mode)) 602 *mode |= S_ISGID; 603 *gid = dip->i_di.di_gid; 604 } else 605 *gid = current->fsgid; 606 } 607 608 static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_inum *inum, 609 u64 *generation) 610 { 611 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 612 int error; 613 614 gfs2_alloc_get(dip); 615 616 dip->i_alloc.al_requested = RES_DINODE; 617 error = gfs2_inplace_reserve(dip); 618 if (error) 619 goto out; 620 621 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS, 0); 622 if (error) 623 goto out_ipreserv; 624 625 inum->no_addr = gfs2_alloc_di(dip, generation); 626 627 gfs2_trans_end(sdp); 628 629 out_ipreserv: 630 gfs2_inplace_release(dip); 631 out: 632 gfs2_alloc_put(dip); 633 return error; 634 } 635 636 /** 637 * init_dinode - Fill in a new dinode structure 638 * @dip: the directory this inode is being created in 639 * @gl: The glock covering the new inode 640 * @inum: the inode number 641 * @mode: the file permissions 642 * @uid: 643 * @gid: 644 * 645 */ 646 647 static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 648 const struct gfs2_inum *inum, unsigned int mode, 649 unsigned int uid, unsigned int gid, 650 const u64 *generation) 651 { 652 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 653 struct gfs2_dinode *di; 654 struct buffer_head *dibh; 655 656 dibh = gfs2_meta_new(gl, inum->no_addr); 657 gfs2_trans_add_bh(gl, dibh, 1); 658 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI); 659 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); 660 di = (struct gfs2_dinode *)dibh->b_data; 661 662 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino); 663 di->di_num.no_addr = cpu_to_be64(inum->no_addr); 664 di->di_mode = cpu_to_be32(mode); 665 di->di_uid = cpu_to_be32(uid); 666 di->di_gid = cpu_to_be32(gid); 667 di->di_nlink = cpu_to_be32(0); 668 di->di_size = cpu_to_be64(0); 669 di->di_blocks = cpu_to_be64(1); 670 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds()); 671 di->di_major = di->di_minor = cpu_to_be32(0); 672 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr); 673 di->di_generation = cpu_to_be64(*generation); 674 di->di_flags = cpu_to_be32(0); 675 676 if (S_ISREG(mode)) { 677 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) || 678 gfs2_tune_get(sdp, gt_new_files_jdata)) 679 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA); 680 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) || 681 gfs2_tune_get(sdp, gt_new_files_directio)) 682 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO); 683 } else if (S_ISDIR(mode)) { 684 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 685 GFS2_DIF_INHERIT_DIRECTIO); 686 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & 687 GFS2_DIF_INHERIT_JDATA); 688 } 689 690 di->__pad1 = 0; 691 di->di_payload_format = cpu_to_be32(0); 692 di->di_height = cpu_to_be32(0); 693 di->__pad2 = 0; 694 di->__pad3 = 0; 695 di->di_depth = cpu_to_be16(0); 696 di->di_entries = cpu_to_be32(0); 697 memset(&di->__pad4, 0, sizeof(di->__pad4)); 698 di->di_eattr = cpu_to_be64(0); 699 memset(&di->di_reserved, 0, sizeof(di->di_reserved)); 700 701 brelse(dibh); 702 } 703 704 static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl, 705 unsigned int mode, const struct gfs2_inum *inum, 706 const u64 *generation) 707 { 708 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 709 unsigned int uid, gid; 710 int error; 711 712 munge_mode_uid_gid(dip, &mode, &uid, &gid); 713 gfs2_alloc_get(dip); 714 715 error = gfs2_quota_lock(dip, uid, gid); 716 if (error) 717 goto out; 718 719 error = gfs2_quota_check(dip, uid, gid); 720 if (error) 721 goto out_quota; 722 723 error = gfs2_trans_begin(sdp, RES_DINODE + RES_QUOTA, 0); 724 if (error) 725 goto out_quota; 726 727 init_dinode(dip, gl, inum, mode, uid, gid, generation); 728 gfs2_quota_change(dip, +1, uid, gid); 729 gfs2_trans_end(sdp); 730 731 out_quota: 732 gfs2_quota_unlock(dip); 733 out: 734 gfs2_alloc_put(dip); 735 return error; 736 } 737 738 static int link_dinode(struct gfs2_inode *dip, const struct qstr *name, 739 struct gfs2_inode *ip) 740 { 741 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 742 struct gfs2_alloc *al; 743 int alloc_required; 744 struct buffer_head *dibh; 745 int error; 746 747 al = gfs2_alloc_get(dip); 748 749 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); 750 if (error) 751 goto fail; 752 753 error = alloc_required = gfs2_diradd_alloc_required(&dip->i_inode, name); 754 if (alloc_required < 0) 755 goto fail; 756 if (alloc_required) { 757 error = gfs2_quota_check(dip, dip->i_di.di_uid, 758 dip->i_di.di_gid); 759 if (error) 760 goto fail_quota_locks; 761 762 al->al_requested = sdp->sd_max_dirres; 763 764 error = gfs2_inplace_reserve(dip); 765 if (error) 766 goto fail_quota_locks; 767 768 error = gfs2_trans_begin(sdp, sdp->sd_max_dirres + 769 al->al_rgd->rd_ri.ri_length + 770 2 * RES_DINODE + 771 RES_STATFS + RES_QUOTA, 0); 772 if (error) 773 goto fail_ipreserv; 774 } else { 775 error = gfs2_trans_begin(sdp, RES_LEAF + 2 * RES_DINODE, 0); 776 if (error) 777 goto fail_quota_locks; 778 } 779 780 error = gfs2_dir_add(&dip->i_inode, name, &ip->i_num, IF2DT(ip->i_di.di_mode)); 781 if (error) 782 goto fail_end_trans; 783 784 error = gfs2_meta_inode_buffer(ip, &dibh); 785 if (error) 786 goto fail_end_trans; 787 ip->i_di.di_nlink = 1; 788 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 789 gfs2_dinode_out(&ip->i_di, dibh->b_data); 790 brelse(dibh); 791 return 0; 792 793 fail_end_trans: 794 gfs2_trans_end(sdp); 795 796 fail_ipreserv: 797 if (dip->i_alloc.al_rgd) 798 gfs2_inplace_release(dip); 799 800 fail_quota_locks: 801 gfs2_quota_unlock(dip); 802 803 fail: 804 gfs2_alloc_put(dip); 805 return error; 806 } 807 808 /** 809 * gfs2_createi - Create a new inode 810 * @ghs: An array of two holders 811 * @name: The name of the new file 812 * @mode: the permissions on the new inode 813 * 814 * @ghs[0] is an initialized holder for the directory 815 * @ghs[1] is the holder for the inode lock 816 * 817 * If the return value is not NULL, the glocks on both the directory and the new 818 * file are held. A transaction has been started and an inplace reservation 819 * is held, as well. 820 * 821 * Returns: An inode 822 */ 823 824 struct inode *gfs2_createi(struct gfs2_holder *ghs, const struct qstr *name, 825 unsigned int mode) 826 { 827 struct inode *inode; 828 struct gfs2_inode *dip = ghs->gh_gl->gl_object; 829 struct inode *dir = &dip->i_inode; 830 struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); 831 struct gfs2_inum inum; 832 int error; 833 u64 generation; 834 835 if (!name->len || name->len > GFS2_FNAMESIZE) 836 return ERR_PTR(-ENAMETOOLONG); 837 838 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 839 error = gfs2_glock_nq(ghs); 840 if (error) 841 goto fail; 842 843 error = create_ok(dip, name, mode); 844 if (error) 845 goto fail_gunlock; 846 847 error = pick_formal_ino(sdp, &inum.no_formal_ino); 848 if (error) 849 goto fail_gunlock; 850 851 error = alloc_dinode(dip, &inum, &generation); 852 if (error) 853 goto fail_gunlock; 854 855 if (inum.no_addr < dip->i_num.no_addr) { 856 gfs2_glock_dq(ghs); 857 858 error = gfs2_glock_nq_num(sdp, inum.no_addr, 859 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 860 GL_SKIP, ghs + 1); 861 if (error) { 862 return ERR_PTR(error); 863 } 864 865 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs); 866 error = gfs2_glock_nq(ghs); 867 if (error) { 868 gfs2_glock_dq_uninit(ghs + 1); 869 return ERR_PTR(error); 870 } 871 872 error = create_ok(dip, name, mode); 873 if (error) 874 goto fail_gunlock2; 875 } else { 876 error = gfs2_glock_nq_num(sdp, inum.no_addr, 877 &gfs2_inode_glops, LM_ST_EXCLUSIVE, 878 GL_SKIP, ghs + 1); 879 if (error) 880 goto fail_gunlock; 881 } 882 883 error = make_dinode(dip, ghs[1].gh_gl, mode, &inum, &generation); 884 if (error) 885 goto fail_gunlock2; 886 887 inode = gfs2_inode_lookup(dir->i_sb, &inum, IF2DT(mode)); 888 if (IS_ERR(inode)) 889 goto fail_gunlock2; 890 891 error = gfs2_inode_refresh(GFS2_I(inode)); 892 if (error) 893 goto fail_iput; 894 895 error = gfs2_acl_create(dip, GFS2_I(inode)); 896 if (error) 897 goto fail_iput; 898 899 error = link_dinode(dip, name, GFS2_I(inode)); 900 if (error) 901 goto fail_iput; 902 903 if (!inode) 904 return ERR_PTR(-ENOMEM); 905 return inode; 906 907 fail_iput: 908 iput(inode); 909 fail_gunlock2: 910 gfs2_glock_dq_uninit(ghs + 1); 911 fail_gunlock: 912 gfs2_glock_dq(ghs); 913 fail: 914 return ERR_PTR(error); 915 } 916 917 /** 918 * gfs2_rmdiri - Remove a directory 919 * @dip: The parent directory of the directory to be removed 920 * @name: The name of the directory to be removed 921 * @ip: The GFS2 inode of the directory to be removed 922 * 923 * Assumes Glocks on dip and ip are held 924 * 925 * Returns: errno 926 */ 927 928 int gfs2_rmdiri(struct gfs2_inode *dip, const struct qstr *name, 929 struct gfs2_inode *ip) 930 { 931 struct qstr dotname; 932 int error; 933 934 if (ip->i_di.di_entries != 2) { 935 if (gfs2_consist_inode(ip)) 936 gfs2_dinode_print(&ip->i_di); 937 return -EIO; 938 } 939 940 error = gfs2_dir_del(dip, name); 941 if (error) 942 return error; 943 944 error = gfs2_change_nlink(dip, -1); 945 if (error) 946 return error; 947 948 gfs2_str2qstr(&dotname, "."); 949 error = gfs2_dir_del(ip, &dotname); 950 if (error) 951 return error; 952 953 gfs2_str2qstr(&dotname, ".."); 954 error = gfs2_dir_del(ip, &dotname); 955 if (error) 956 return error; 957 958 error = gfs2_change_nlink(ip, -2); 959 if (error) 960 return error; 961 962 return error; 963 } 964 965 /* 966 * gfs2_unlink_ok - check to see that a inode is still in a directory 967 * @dip: the directory 968 * @name: the name of the file 969 * @ip: the inode 970 * 971 * Assumes that the lock on (at least) @dip is held. 972 * 973 * Returns: 0 if the parent/child relationship is correct, errno if it isn't 974 */ 975 976 int gfs2_unlink_ok(struct gfs2_inode *dip, const struct qstr *name, 977 struct gfs2_inode *ip) 978 { 979 struct gfs2_inum inum; 980 unsigned int type; 981 int error; 982 983 if (IS_IMMUTABLE(&ip->i_inode) || IS_APPEND(&ip->i_inode)) 984 return -EPERM; 985 986 if ((dip->i_di.di_mode & S_ISVTX) && 987 dip->i_di.di_uid != current->fsuid && 988 ip->i_di.di_uid != current->fsuid && !capable(CAP_FOWNER)) 989 return -EPERM; 990 991 if (IS_APPEND(&dip->i_inode)) 992 return -EPERM; 993 994 error = permission(&dip->i_inode, MAY_WRITE | MAY_EXEC, NULL); 995 if (error) 996 return error; 997 998 error = gfs2_dir_search(&dip->i_inode, name, &inum, &type); 999 if (error) 1000 return error; 1001 1002 if (!gfs2_inum_equal(&inum, &ip->i_num)) 1003 return -ENOENT; 1004 1005 if (IF2DT(ip->i_di.di_mode) != type) { 1006 gfs2_consist_inode(dip); 1007 return -EIO; 1008 } 1009 1010 return 0; 1011 } 1012 1013 /* 1014 * gfs2_ok_to_move - check if it's ok to move a directory to another directory 1015 * @this: move this 1016 * @to: to here 1017 * 1018 * Follow @to back to the root and make sure we don't encounter @this 1019 * Assumes we already hold the rename lock. 1020 * 1021 * Returns: errno 1022 */ 1023 1024 int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to) 1025 { 1026 struct inode *dir = &to->i_inode; 1027 struct super_block *sb = dir->i_sb; 1028 struct inode *tmp; 1029 struct qstr dotdot; 1030 int error = 0; 1031 1032 gfs2_str2qstr(&dotdot, ".."); 1033 1034 igrab(dir); 1035 1036 for (;;) { 1037 if (dir == &this->i_inode) { 1038 error = -EINVAL; 1039 break; 1040 } 1041 if (dir == sb->s_root->d_inode) { 1042 error = 0; 1043 break; 1044 } 1045 1046 tmp = gfs2_lookupi(dir, &dotdot, 1, NULL); 1047 if (IS_ERR(tmp)) { 1048 error = PTR_ERR(tmp); 1049 break; 1050 } 1051 1052 iput(dir); 1053 dir = tmp; 1054 } 1055 1056 iput(dir); 1057 1058 return error; 1059 } 1060 1061 /** 1062 * gfs2_readlinki - return the contents of a symlink 1063 * @ip: the symlink's inode 1064 * @buf: a pointer to the buffer to be filled 1065 * @len: a pointer to the length of @buf 1066 * 1067 * If @buf is too small, a piece of memory is kmalloc()ed and needs 1068 * to be freed by the caller. 1069 * 1070 * Returns: errno 1071 */ 1072 1073 int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len) 1074 { 1075 struct gfs2_holder i_gh; 1076 struct buffer_head *dibh; 1077 unsigned int x; 1078 int error; 1079 1080 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); 1081 error = gfs2_glock_nq_atime(&i_gh); 1082 if (error) { 1083 gfs2_holder_uninit(&i_gh); 1084 return error; 1085 } 1086 1087 if (!ip->i_di.di_size) { 1088 gfs2_consist_inode(ip); 1089 error = -EIO; 1090 goto out; 1091 } 1092 1093 error = gfs2_meta_inode_buffer(ip, &dibh); 1094 if (error) 1095 goto out; 1096 1097 x = ip->i_di.di_size + 1; 1098 if (x > *len) { 1099 *buf = kmalloc(x, GFP_KERNEL); 1100 if (!*buf) { 1101 error = -ENOMEM; 1102 goto out_brelse; 1103 } 1104 } 1105 1106 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x); 1107 *len = x; 1108 1109 out_brelse: 1110 brelse(dibh); 1111 out: 1112 gfs2_glock_dq_uninit(&i_gh); 1113 return error; 1114 } 1115 1116 /** 1117 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and 1118 * conditionally update the inode's atime 1119 * @gh: the holder to acquire 1120 * 1121 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap 1122 * Update if the difference between the current time and the inode's current 1123 * atime is greater than an interval specified at mount. 1124 * 1125 * Returns: errno 1126 */ 1127 1128 int gfs2_glock_nq_atime(struct gfs2_holder *gh) 1129 { 1130 struct gfs2_glock *gl = gh->gh_gl; 1131 struct gfs2_sbd *sdp = gl->gl_sbd; 1132 struct gfs2_inode *ip = gl->gl_object; 1133 s64 curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum); 1134 unsigned int state; 1135 int flags; 1136 int error; 1137 1138 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) || 1139 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) || 1140 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops)) 1141 return -EINVAL; 1142 1143 state = gh->gh_state; 1144 flags = gh->gh_flags; 1145 1146 error = gfs2_glock_nq(gh); 1147 if (error) 1148 return error; 1149 1150 if (test_bit(SDF_NOATIME, &sdp->sd_flags) || 1151 (sdp->sd_vfs->s_flags & MS_RDONLY)) 1152 return 0; 1153 1154 curtime = get_seconds(); 1155 if (curtime - ip->i_di.di_atime >= quantum) { 1156 gfs2_glock_dq(gh); 1157 gfs2_holder_reinit(LM_ST_EXCLUSIVE, gh->gh_flags & ~LM_FLAG_ANY, 1158 gh); 1159 error = gfs2_glock_nq(gh); 1160 if (error) 1161 return error; 1162 1163 /* Verify that atime hasn't been updated while we were 1164 trying to get exclusive lock. */ 1165 1166 curtime = get_seconds(); 1167 if (curtime - ip->i_di.di_atime >= quantum) { 1168 struct buffer_head *dibh; 1169 1170 error = gfs2_trans_begin(sdp, RES_DINODE, 0); 1171 if (error == -EROFS) 1172 return 0; 1173 if (error) 1174 goto fail; 1175 1176 error = gfs2_meta_inode_buffer(ip, &dibh); 1177 if (error) 1178 goto fail_end_trans; 1179 1180 ip->i_di.di_atime = curtime; 1181 1182 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1183 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1184 brelse(dibh); 1185 1186 gfs2_trans_end(sdp); 1187 } 1188 1189 /* If someone else has asked for the glock, 1190 unlock and let them have it. Then reacquire 1191 in the original state. */ 1192 if (gfs2_glock_is_blocking(gl)) { 1193 gfs2_glock_dq(gh); 1194 gfs2_holder_reinit(state, flags, gh); 1195 return gfs2_glock_nq(gh); 1196 } 1197 } 1198 1199 return 0; 1200 1201 fail_end_trans: 1202 gfs2_trans_end(sdp); 1203 fail: 1204 gfs2_glock_dq(gh); 1205 return error; 1206 } 1207 1208 /** 1209 * glock_compare_atime - Compare two struct gfs2_glock structures for sort 1210 * @arg_a: the first structure 1211 * @arg_b: the second structure 1212 * 1213 * Returns: 1 if A > B 1214 * -1 if A < B 1215 * 0 if A == B 1216 */ 1217 1218 static int glock_compare_atime(const void *arg_a, const void *arg_b) 1219 { 1220 const struct gfs2_holder *gh_a = *(const struct gfs2_holder **)arg_a; 1221 const struct gfs2_holder *gh_b = *(const struct gfs2_holder **)arg_b; 1222 const struct lm_lockname *a = &gh_a->gh_gl->gl_name; 1223 const struct lm_lockname *b = &gh_b->gh_gl->gl_name; 1224 1225 if (a->ln_number > b->ln_number) 1226 return 1; 1227 if (a->ln_number < b->ln_number) 1228 return -1; 1229 if (gh_a->gh_state == LM_ST_SHARED && gh_b->gh_state == LM_ST_EXCLUSIVE) 1230 return 1; 1231 if (gh_a->gh_state == LM_ST_SHARED && (gh_b->gh_flags & GL_ATIME)) 1232 return 1; 1233 1234 return 0; 1235 } 1236 1237 /** 1238 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an 1239 * atime update 1240 * @num_gh: the number of structures 1241 * @ghs: an array of struct gfs2_holder structures 1242 * 1243 * Returns: 0 on success (all glocks acquired), 1244 * errno on failure (no glocks acquired) 1245 */ 1246 1247 int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs) 1248 { 1249 struct gfs2_holder **p; 1250 unsigned int x; 1251 int error = 0; 1252 1253 if (!num_gh) 1254 return 0; 1255 1256 if (num_gh == 1) { 1257 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1258 if (ghs->gh_flags & GL_ATIME) 1259 error = gfs2_glock_nq_atime(ghs); 1260 else 1261 error = gfs2_glock_nq(ghs); 1262 return error; 1263 } 1264 1265 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL); 1266 if (!p) 1267 return -ENOMEM; 1268 1269 for (x = 0; x < num_gh; x++) 1270 p[x] = &ghs[x]; 1271 1272 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL); 1273 1274 for (x = 0; x < num_gh; x++) { 1275 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC); 1276 1277 if (p[x]->gh_flags & GL_ATIME) 1278 error = gfs2_glock_nq_atime(p[x]); 1279 else 1280 error = gfs2_glock_nq(p[x]); 1281 1282 if (error) { 1283 while (x--) 1284 gfs2_glock_dq(p[x]); 1285 break; 1286 } 1287 } 1288 1289 kfree(p); 1290 return error; 1291 } 1292 1293 1294 static int 1295 __gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1296 { 1297 struct buffer_head *dibh; 1298 int error; 1299 1300 error = gfs2_meta_inode_buffer(ip, &dibh); 1301 if (!error) { 1302 error = inode_setattr(&ip->i_inode, attr); 1303 gfs2_assert_warn(GFS2_SB(&ip->i_inode), !error); 1304 gfs2_inode_attr_out(ip); 1305 1306 gfs2_trans_add_bh(ip->i_gl, dibh, 1); 1307 gfs2_dinode_out(&ip->i_di, dibh->b_data); 1308 brelse(dibh); 1309 } 1310 return error; 1311 } 1312 1313 /** 1314 * gfs2_setattr_simple - 1315 * @ip: 1316 * @attr: 1317 * 1318 * Called with a reference on the vnode. 1319 * 1320 * Returns: errno 1321 */ 1322 1323 int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr) 1324 { 1325 int error; 1326 1327 if (current->journal_info) 1328 return __gfs2_setattr_simple(ip, attr); 1329 1330 error = gfs2_trans_begin(GFS2_SB(&ip->i_inode), RES_DINODE, 0); 1331 if (error) 1332 return error; 1333 1334 error = __gfs2_setattr_simple(ip, attr); 1335 gfs2_trans_end(GFS2_SB(&ip->i_inode)); 1336 return error; 1337 } 1338 1339