1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * (C) 2001 Clemson University and The University of Chicago 4 * Copyright 2018 Omnibond Systems, L.L.C. 5 * 6 * See COPYING in top-level directory. 7 */ 8 9 /* 10 * Linux VFS inode operations. 11 */ 12 13 #include <linux/blkdev.h> 14 #include <linux/fileattr.h> 15 #include "protocol.h" 16 #include "orangefs-kernel.h" 17 #include "orangefs-bufmap.h" 18 19 static int orangefs_writepage_locked(struct folio *folio, 20 struct writeback_control *wbc) 21 { 22 struct inode *inode = folio->mapping->host; 23 struct orangefs_write_range *wr = NULL; 24 struct iov_iter iter; 25 struct bio_vec bv; 26 size_t wlen; 27 ssize_t ret; 28 loff_t len, off; 29 30 folio_start_writeback(folio); 31 32 len = i_size_read(inode); 33 if (folio->private) { 34 wr = folio->private; 35 WARN_ON(wr->pos >= len); 36 off = wr->pos; 37 if (off + wr->len > len) 38 wlen = len - off; 39 else 40 wlen = wr->len; 41 } else { 42 WARN_ON(1); 43 off = folio_pos(folio); 44 wlen = folio_size(folio); 45 46 if (wlen > len - off) 47 wlen = len - off; 48 } 49 /* Should've been handled in orangefs_invalidate_folio. */ 50 WARN_ON(off == len || off + wlen > len); 51 52 WARN_ON(wlen == 0); 53 bvec_set_folio(&bv, folio, wlen, offset_in_folio(folio, off)); 54 iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen); 55 56 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen, 57 len, wr, NULL, NULL); 58 if (ret < 0) { 59 mapping_set_error(folio->mapping, ret); 60 } else { 61 ret = 0; 62 } 63 kfree(folio_detach_private(folio)); 64 return ret; 65 } 66 67 struct orangefs_writepages { 68 loff_t off; 69 size_t len; 70 kuid_t uid; 71 kgid_t gid; 72 int maxpages; 73 int nfolios; 74 struct address_space *mapping; 75 struct folio **folios; 76 struct bio_vec *bv; 77 }; 78 79 static int orangefs_writepages_work(struct orangefs_writepages *ow, 80 struct writeback_control *wbc) 81 { 82 struct inode *inode = ow->mapping->host; 83 struct orangefs_write_range *wrp, wr; 84 struct iov_iter iter; 85 ssize_t ret; 86 size_t start; 87 loff_t len, off; 88 int i; 89 90 len = i_size_read(inode); 91 92 start = offset_in_folio(ow->folios[0], ow->off); 93 for (i = 0; i < ow->nfolios; i++) { 94 folio_start_writeback(ow->folios[i]); 95 bvec_set_folio(&ow->bv[i], ow->folios[i], 96 folio_size(ow->folios[i]) - start, start); 97 start = 0; 98 } 99 iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->nfolios, ow->len); 100 101 WARN_ON(ow->off >= len); 102 if (ow->off + ow->len > len) 103 ow->len = len - ow->off; 104 105 off = ow->off; 106 wr.uid = ow->uid; 107 wr.gid = ow->gid; 108 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len, 109 0, &wr, NULL, NULL); 110 if (ret < 0) 111 mapping_set_error(ow->mapping, ret); 112 else 113 ret = 0; 114 115 for (i = 0; i < ow->nfolios; i++) { 116 wrp = folio_detach_private(ow->folios[i]); 117 kfree(wrp); 118 folio_end_writeback(ow->folios[i]); 119 folio_unlock(ow->folios[i]); 120 } 121 122 return ret; 123 } 124 125 static int orangefs_writepages_callback(struct folio *folio, 126 struct writeback_control *wbc, struct orangefs_writepages *ow) 127 { 128 struct orangefs_write_range *wr = folio->private; 129 int ret; 130 131 if (!wr) { 132 folio_unlock(folio); 133 /* It's not private so there's nothing to write, right? */ 134 printk("writepages_callback not private!\n"); 135 BUG(); 136 return 0; 137 } 138 139 ret = -1; 140 if (ow->nfolios == 0) { 141 ow->off = wr->pos; 142 ow->len = wr->len; 143 ow->uid = wr->uid; 144 ow->gid = wr->gid; 145 ow->folios[ow->nfolios++] = folio; 146 ret = 0; 147 goto done; 148 } 149 if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) { 150 orangefs_writepages_work(ow, wbc); 151 ow->nfolios = 0; 152 ret = -1; 153 goto done; 154 } 155 if (ow->off + ow->len == wr->pos) { 156 ow->len += wr->len; 157 ow->folios[ow->nfolios++] = folio; 158 ret = 0; 159 goto done; 160 } 161 done: 162 if (ret == -1) { 163 if (ow->nfolios) { 164 orangefs_writepages_work(ow, wbc); 165 ow->nfolios = 0; 166 } 167 ret = orangefs_writepage_locked(folio, wbc); 168 mapping_set_error(folio->mapping, ret); 169 folio_unlock(folio); 170 folio_end_writeback(folio); 171 } else { 172 if (ow->nfolios == ow->maxpages) { 173 orangefs_writepages_work(ow, wbc); 174 ow->nfolios = 0; 175 } 176 } 177 return ret; 178 } 179 180 static int orangefs_writepages(struct address_space *mapping, 181 struct writeback_control *wbc) 182 { 183 struct orangefs_writepages *ow; 184 struct blk_plug plug; 185 int error; 186 struct folio *folio = NULL; 187 188 ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL); 189 if (!ow) 190 return -ENOMEM; 191 ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE; 192 ow->folios = kcalloc(ow->maxpages, sizeof(struct folio *), GFP_KERNEL); 193 if (!ow->folios) { 194 kfree(ow); 195 return -ENOMEM; 196 } 197 ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL); 198 if (!ow->bv) { 199 kfree(ow->folios); 200 kfree(ow); 201 return -ENOMEM; 202 } 203 ow->mapping = mapping; 204 blk_start_plug(&plug); 205 while ((folio = writeback_iter(mapping, wbc, folio, &error))) 206 error = orangefs_writepages_callback(folio, wbc, ow); 207 if (ow->nfolios) 208 error = orangefs_writepages_work(ow, wbc); 209 blk_finish_plug(&plug); 210 kfree(ow->folios); 211 kfree(ow->bv); 212 kfree(ow); 213 return error; 214 } 215 216 static int orangefs_launder_folio(struct folio *); 217 218 static void orangefs_readahead(struct readahead_control *rac) 219 { 220 loff_t offset; 221 struct iov_iter iter; 222 struct inode *inode = rac->mapping->host; 223 struct xarray *i_pages; 224 struct folio *folio; 225 loff_t new_start = readahead_pos(rac); 226 int ret; 227 size_t new_len = 0; 228 229 loff_t bytes_remaining = inode->i_size - readahead_pos(rac); 230 loff_t pages_remaining = bytes_remaining / PAGE_SIZE; 231 232 if (pages_remaining >= 1024) 233 new_len = 4194304; 234 else if (pages_remaining > readahead_count(rac)) 235 new_len = bytes_remaining; 236 237 if (new_len) 238 readahead_expand(rac, new_start, new_len); 239 240 offset = readahead_pos(rac); 241 i_pages = &rac->mapping->i_pages; 242 243 iov_iter_xarray(&iter, ITER_DEST, i_pages, offset, readahead_length(rac)); 244 245 /* read in the pages. */ 246 if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, 247 &offset, &iter, readahead_length(rac), 248 inode->i_size, NULL, NULL, rac->file)) < 0) 249 gossip_debug(GOSSIP_FILE_DEBUG, 250 "%s: wait_for_direct_io failed. \n", __func__); 251 else 252 ret = 0; 253 254 /* clean up. */ 255 while ((folio = readahead_folio(rac))) { 256 if (!ret) 257 folio_mark_uptodate(folio); 258 folio_unlock(folio); 259 } 260 } 261 262 static int orangefs_read_folio(struct file *file, struct folio *folio) 263 { 264 struct inode *inode = folio->mapping->host; 265 struct iov_iter iter; 266 struct bio_vec bv; 267 ssize_t ret; 268 loff_t off; /* offset of this folio in the file */ 269 270 if (folio_test_dirty(folio)) 271 orangefs_launder_folio(folio); 272 273 off = folio_pos(folio); 274 bvec_set_folio(&bv, folio, folio_size(folio), 0); 275 iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio)); 276 277 ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter, 278 folio_size(folio), inode->i_size, NULL, NULL, file); 279 /* this will only zero remaining unread portions of the folio data */ 280 iov_iter_zero(~0U, &iter); 281 /* takes care of potential aliasing */ 282 flush_dcache_folio(folio); 283 if (ret > 0) 284 ret = 0; 285 folio_end_read(folio, ret == 0); 286 return ret; 287 } 288 289 static int orangefs_write_begin(struct file *file, 290 struct address_space *mapping, loff_t pos, unsigned len, 291 struct folio **foliop, void **fsdata) 292 { 293 struct orangefs_write_range *wr; 294 struct folio *folio; 295 int ret; 296 297 folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN, 298 mapping_gfp_mask(mapping)); 299 if (IS_ERR(folio)) 300 return PTR_ERR(folio); 301 302 *foliop = folio; 303 304 if (folio_test_dirty(folio) && !folio_test_private(folio)) { 305 /* 306 * Should be impossible. If it happens, launder the page 307 * since we don't know what's dirty. This will WARN in 308 * orangefs_writepage_locked. 309 */ 310 ret = orangefs_launder_folio(folio); 311 if (ret) 312 return ret; 313 } 314 if (folio_test_private(folio)) { 315 struct orangefs_write_range *wr; 316 wr = folio_get_private(folio); 317 if (wr->pos + wr->len == pos && 318 uid_eq(wr->uid, current_fsuid()) && 319 gid_eq(wr->gid, current_fsgid())) { 320 wr->len += len; 321 goto okay; 322 } else { 323 ret = orangefs_launder_folio(folio); 324 if (ret) 325 return ret; 326 } 327 } 328 329 wr = kmalloc(sizeof *wr, GFP_KERNEL); 330 if (!wr) 331 return -ENOMEM; 332 333 wr->pos = pos; 334 wr->len = len; 335 wr->uid = current_fsuid(); 336 wr->gid = current_fsgid(); 337 folio_attach_private(folio, wr); 338 okay: 339 return 0; 340 } 341 342 static int orangefs_write_end(struct file *file, struct address_space *mapping, 343 loff_t pos, unsigned len, unsigned copied, struct folio *folio, 344 void *fsdata) 345 { 346 struct inode *inode = folio->mapping->host; 347 loff_t last_pos = pos + copied; 348 349 /* 350 * No need to use i_size_read() here, the i_size 351 * cannot change under us because we hold the i_mutex. 352 */ 353 if (last_pos > inode->i_size) 354 i_size_write(inode, last_pos); 355 356 /* zero the stale part of the folio if we did a short copy */ 357 if (!folio_test_uptodate(folio)) { 358 unsigned from = pos & (PAGE_SIZE - 1); 359 if (copied < len) { 360 folio_zero_range(folio, from + copied, len - copied); 361 } 362 /* Set fully written pages uptodate. */ 363 if (pos == folio_pos(folio) && 364 (len == PAGE_SIZE || pos + len == inode->i_size)) { 365 folio_zero_segment(folio, from + copied, PAGE_SIZE); 366 folio_mark_uptodate(folio); 367 } 368 } 369 370 folio_mark_dirty(folio); 371 folio_unlock(folio); 372 folio_put(folio); 373 374 mark_inode_dirty_sync(file_inode(file)); 375 return copied; 376 } 377 378 static void orangefs_invalidate_folio(struct folio *folio, 379 size_t offset, size_t length) 380 { 381 struct orangefs_write_range *wr = folio_get_private(folio); 382 383 if (offset == 0 && length == PAGE_SIZE) { 384 kfree(folio_detach_private(folio)); 385 return; 386 /* write range entirely within invalidate range (or equal) */ 387 } else if (folio_pos(folio) + offset <= wr->pos && 388 wr->pos + wr->len <= folio_pos(folio) + offset + length) { 389 kfree(folio_detach_private(folio)); 390 /* XXX is this right? only caller in fs */ 391 folio_cancel_dirty(folio); 392 return; 393 /* invalidate range chops off end of write range */ 394 } else if (wr->pos < folio_pos(folio) + offset && 395 wr->pos + wr->len <= folio_pos(folio) + offset + length && 396 folio_pos(folio) + offset < wr->pos + wr->len) { 397 size_t x; 398 x = wr->pos + wr->len - (folio_pos(folio) + offset); 399 WARN_ON(x > wr->len); 400 wr->len -= x; 401 wr->uid = current_fsuid(); 402 wr->gid = current_fsgid(); 403 /* invalidate range chops off beginning of write range */ 404 } else if (folio_pos(folio) + offset <= wr->pos && 405 folio_pos(folio) + offset + length < wr->pos + wr->len && 406 wr->pos < folio_pos(folio) + offset + length) { 407 size_t x; 408 x = folio_pos(folio) + offset + length - wr->pos; 409 WARN_ON(x > wr->len); 410 wr->pos += x; 411 wr->len -= x; 412 wr->uid = current_fsuid(); 413 wr->gid = current_fsgid(); 414 /* invalidate range entirely within write range (punch hole) */ 415 } else if (wr->pos < folio_pos(folio) + offset && 416 folio_pos(folio) + offset + length < wr->pos + wr->len) { 417 /* XXX what do we do here... should not WARN_ON */ 418 WARN_ON(1); 419 /* punch hole */ 420 /* 421 * should we just ignore this and write it out anyway? 422 * it hardly makes sense 423 */ 424 return; 425 /* non-overlapping ranges */ 426 } else { 427 /* WARN if they do overlap */ 428 if (!((folio_pos(folio) + offset + length <= wr->pos) ^ 429 (wr->pos + wr->len <= folio_pos(folio) + offset))) { 430 WARN_ON(1); 431 printk("invalidate range offset %llu length %zu\n", 432 folio_pos(folio) + offset, length); 433 printk("write range offset %llu length %zu\n", 434 wr->pos, wr->len); 435 } 436 return; 437 } 438 439 /* 440 * Above there are returns where wr is freed or where we WARN. 441 * Thus the following runs if wr was modified above. 442 */ 443 444 orangefs_launder_folio(folio); 445 } 446 447 static bool orangefs_release_folio(struct folio *folio, gfp_t foo) 448 { 449 return !folio_test_private(folio); 450 } 451 452 static void orangefs_free_folio(struct folio *folio) 453 { 454 kfree(folio_detach_private(folio)); 455 } 456 457 static int orangefs_launder_folio(struct folio *folio) 458 { 459 int r = 0; 460 struct writeback_control wbc = { 461 .sync_mode = WB_SYNC_ALL, 462 .nr_to_write = 0, 463 }; 464 folio_wait_writeback(folio); 465 if (folio_clear_dirty_for_io(folio)) { 466 r = orangefs_writepage_locked(folio, &wbc); 467 folio_end_writeback(folio); 468 } 469 return r; 470 } 471 472 static ssize_t orangefs_direct_IO(struct kiocb *iocb, 473 struct iov_iter *iter) 474 { 475 /* 476 * Comment from original do_readv_writev: 477 * Common entry point for read/write/readv/writev 478 * This function will dispatch it to either the direct I/O 479 * or buffered I/O path depending on the mount options and/or 480 * augmented/extended metadata attached to the file. 481 * Note: File extended attributes override any mount options. 482 */ 483 struct file *file = iocb->ki_filp; 484 loff_t pos = iocb->ki_pos; 485 enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ? 486 ORANGEFS_IO_WRITE : ORANGEFS_IO_READ; 487 loff_t *offset = &pos; 488 struct inode *inode = file->f_mapping->host; 489 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 490 struct orangefs_khandle *handle = &orangefs_inode->refn.khandle; 491 size_t count = iov_iter_count(iter); 492 ssize_t total_count = 0; 493 ssize_t ret = -EINVAL; 494 495 gossip_debug(GOSSIP_FILE_DEBUG, 496 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n", 497 __func__, 498 handle, 499 (int)count); 500 501 if (type == ORANGEFS_IO_WRITE) { 502 gossip_debug(GOSSIP_FILE_DEBUG, 503 "%s(%pU): proceeding with offset : %llu, " 504 "size %d\n", 505 __func__, 506 handle, 507 llu(*offset), 508 (int)count); 509 } 510 511 if (count == 0) { 512 ret = 0; 513 goto out; 514 } 515 516 while (iov_iter_count(iter)) { 517 size_t each_count = iov_iter_count(iter); 518 size_t amt_complete; 519 520 /* how much to transfer in this loop iteration */ 521 if (each_count > orangefs_bufmap_size_query()) 522 each_count = orangefs_bufmap_size_query(); 523 524 gossip_debug(GOSSIP_FILE_DEBUG, 525 "%s(%pU): size of each_count(%d)\n", 526 __func__, 527 handle, 528 (int)each_count); 529 gossip_debug(GOSSIP_FILE_DEBUG, 530 "%s(%pU): BEFORE wait_for_io: offset is %d\n", 531 __func__, 532 handle, 533 (int)*offset); 534 535 ret = wait_for_direct_io(type, inode, offset, iter, 536 each_count, 0, NULL, NULL, file); 537 gossip_debug(GOSSIP_FILE_DEBUG, 538 "%s(%pU): return from wait_for_io:%d\n", 539 __func__, 540 handle, 541 (int)ret); 542 543 if (ret < 0) 544 goto out; 545 546 *offset += ret; 547 total_count += ret; 548 amt_complete = ret; 549 550 gossip_debug(GOSSIP_FILE_DEBUG, 551 "%s(%pU): AFTER wait_for_io: offset is %d\n", 552 __func__, 553 handle, 554 (int)*offset); 555 556 /* 557 * if we got a short I/O operations, 558 * fall out and return what we got so far 559 */ 560 if (amt_complete < each_count) 561 break; 562 } /*end while */ 563 564 out: 565 if (total_count > 0) 566 ret = total_count; 567 if (ret > 0) { 568 if (type == ORANGEFS_IO_READ) { 569 file_accessed(file); 570 } else { 571 file_update_time(file); 572 if (*offset > i_size_read(inode)) 573 i_size_write(inode, *offset); 574 } 575 } 576 577 gossip_debug(GOSSIP_FILE_DEBUG, 578 "%s(%pU): Value(%d) returned.\n", 579 __func__, 580 handle, 581 (int)ret); 582 583 return ret; 584 } 585 586 /** ORANGEFS2 implementation of address space operations */ 587 static const struct address_space_operations orangefs_address_operations = { 588 .readahead = orangefs_readahead, 589 .read_folio = orangefs_read_folio, 590 .writepages = orangefs_writepages, 591 .dirty_folio = filemap_dirty_folio, 592 .write_begin = orangefs_write_begin, 593 .write_end = orangefs_write_end, 594 .invalidate_folio = orangefs_invalidate_folio, 595 .release_folio = orangefs_release_folio, 596 .free_folio = orangefs_free_folio, 597 .migrate_folio = filemap_migrate_folio, 598 .launder_folio = orangefs_launder_folio, 599 .direct_IO = orangefs_direct_IO, 600 }; 601 602 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf) 603 { 604 struct folio *folio = page_folio(vmf->page); 605 struct inode *inode = file_inode(vmf->vma->vm_file); 606 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 607 unsigned long *bitlock = &orangefs_inode->bitlock; 608 vm_fault_t ret; 609 struct orangefs_write_range *wr; 610 611 sb_start_pagefault(inode->i_sb); 612 613 if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) { 614 ret = VM_FAULT_RETRY; 615 goto out; 616 } 617 618 folio_lock(folio); 619 if (folio_test_dirty(folio) && !folio_test_private(folio)) { 620 /* 621 * Should be impossible. If it happens, launder the folio 622 * since we don't know what's dirty. This will WARN in 623 * orangefs_writepage_locked. 624 */ 625 if (orangefs_launder_folio(folio)) { 626 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; 627 goto out; 628 } 629 } 630 if (folio_test_private(folio)) { 631 wr = folio_get_private(folio); 632 if (uid_eq(wr->uid, current_fsuid()) && 633 gid_eq(wr->gid, current_fsgid())) { 634 wr->pos = page_offset(vmf->page); 635 wr->len = PAGE_SIZE; 636 goto okay; 637 } else { 638 if (orangefs_launder_folio(folio)) { 639 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; 640 goto out; 641 } 642 } 643 } 644 wr = kmalloc(sizeof *wr, GFP_KERNEL); 645 if (!wr) { 646 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; 647 goto out; 648 } 649 wr->pos = page_offset(vmf->page); 650 wr->len = PAGE_SIZE; 651 wr->uid = current_fsuid(); 652 wr->gid = current_fsgid(); 653 folio_attach_private(folio, wr); 654 okay: 655 656 file_update_time(vmf->vma->vm_file); 657 if (folio->mapping != inode->i_mapping) { 658 folio_unlock(folio); 659 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE; 660 goto out; 661 } 662 663 /* 664 * We mark the folio dirty already here so that when freeze is in 665 * progress, we are guaranteed that writeback during freezing will 666 * see the dirty folio and writeprotect it again. 667 */ 668 folio_mark_dirty(folio); 669 folio_wait_stable(folio); 670 ret = VM_FAULT_LOCKED; 671 out: 672 sb_end_pagefault(inode->i_sb); 673 return ret; 674 } 675 676 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr) 677 { 678 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 679 struct orangefs_kernel_op_s *new_op; 680 loff_t orig_size; 681 int ret = -EINVAL; 682 683 gossip_debug(GOSSIP_INODE_DEBUG, 684 "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", 685 __func__, 686 get_khandle_from_ino(inode), 687 &orangefs_inode->refn.khandle, 688 orangefs_inode->refn.fs_id, 689 iattr->ia_size); 690 691 /* Ensure that we have a up to date size, so we know if it changed. */ 692 ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE); 693 if (ret == -ESTALE) 694 ret = -EIO; 695 if (ret) { 696 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n", 697 __func__, ret); 698 return ret; 699 } 700 orig_size = i_size_read(inode); 701 702 /* This is truncate_setsize in a different order. */ 703 truncate_pagecache(inode, iattr->ia_size); 704 i_size_write(inode, iattr->ia_size); 705 if (iattr->ia_size > orig_size) 706 pagecache_isize_extended(inode, orig_size, iattr->ia_size); 707 708 new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE); 709 if (!new_op) 710 return -ENOMEM; 711 712 new_op->upcall.req.truncate.refn = orangefs_inode->refn; 713 new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; 714 715 ret = service_operation(new_op, 716 __func__, 717 get_interruptible_flag(inode)); 718 719 /* 720 * the truncate has no downcall members to retrieve, but 721 * the status value tells us if it went through ok or not 722 */ 723 gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret); 724 725 op_release(new_op); 726 727 if (ret != 0) 728 return ret; 729 730 if (orig_size != i_size_read(inode)) 731 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 732 733 return ret; 734 } 735 736 int __orangefs_setattr(struct inode *inode, struct iattr *iattr) 737 { 738 int ret; 739 740 if (iattr->ia_valid & ATTR_MODE) { 741 if (iattr->ia_mode & (S_ISVTX)) { 742 if (is_root_handle(inode)) { 743 /* 744 * allow sticky bit to be set on root (since 745 * it shows up that way by default anyhow), 746 * but don't show it to the server 747 */ 748 iattr->ia_mode -= S_ISVTX; 749 } else { 750 gossip_debug(GOSSIP_UTILS_DEBUG, 751 "User attempted to set sticky bit on non-root directory; returning EINVAL.\n"); 752 ret = -EINVAL; 753 goto out; 754 } 755 } 756 if (iattr->ia_mode & (S_ISUID)) { 757 gossip_debug(GOSSIP_UTILS_DEBUG, 758 "Attempting to set setuid bit (not supported); returning EINVAL.\n"); 759 ret = -EINVAL; 760 goto out; 761 } 762 } 763 764 if (iattr->ia_valid & ATTR_SIZE) { 765 ret = orangefs_setattr_size(inode, iattr); 766 if (ret) 767 goto out; 768 } 769 770 again: 771 spin_lock(&inode->i_lock); 772 if (ORANGEFS_I(inode)->attr_valid) { 773 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) && 774 gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) { 775 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid; 776 } else { 777 spin_unlock(&inode->i_lock); 778 write_inode_now(inode, 1); 779 goto again; 780 } 781 } else { 782 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid; 783 ORANGEFS_I(inode)->attr_uid = current_fsuid(); 784 ORANGEFS_I(inode)->attr_gid = current_fsgid(); 785 } 786 setattr_copy(&nop_mnt_idmap, inode, iattr); 787 spin_unlock(&inode->i_lock); 788 mark_inode_dirty(inode); 789 790 ret = 0; 791 out: 792 return ret; 793 } 794 795 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr) 796 { 797 int ret; 798 struct inode *inode = d_inode(dentry); 799 800 ret = __orangefs_setattr(inode, iattr); 801 /* change mode on a file that has ACLs */ 802 if (!ret && (iattr->ia_valid & ATTR_MODE)) 803 ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode); 804 return ret; 805 } 806 807 /* 808 * Change attributes of an object referenced by dentry. 809 */ 810 int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 811 struct iattr *iattr) 812 { 813 int ret; 814 gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n", 815 dentry); 816 ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr); 817 if (ret) 818 goto out; 819 ret = __orangefs_setattr_mode(dentry, iattr); 820 sync_inode_metadata(d_inode(dentry), 1); 821 out: 822 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", 823 ret); 824 return ret; 825 } 826 827 /* 828 * Obtain attributes of an object given a dentry 829 */ 830 int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path, 831 struct kstat *stat, u32 request_mask, unsigned int flags) 832 { 833 int ret; 834 struct inode *inode = path->dentry->d_inode; 835 836 gossip_debug(GOSSIP_INODE_DEBUG, 837 "orangefs_getattr: called on %pd mask %u\n", 838 path->dentry, request_mask); 839 840 ret = orangefs_inode_getattr(inode, 841 request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0); 842 if (ret == 0) { 843 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 844 845 /* override block size reported to stat */ 846 if (!(request_mask & STATX_SIZE)) 847 stat->result_mask &= ~STATX_SIZE; 848 849 generic_fill_statx_attr(inode, stat); 850 } 851 return ret; 852 } 853 854 int orangefs_permission(struct mnt_idmap *idmap, 855 struct inode *inode, int mask) 856 { 857 int ret; 858 859 if (mask & MAY_NOT_BLOCK) 860 return -ECHILD; 861 862 gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__); 863 864 /* Make sure the permission (and other common attrs) are up to date. */ 865 ret = orangefs_inode_getattr(inode, 0); 866 if (ret < 0) 867 return ret; 868 869 return generic_permission(&nop_mnt_idmap, inode, mask); 870 } 871 872 int orangefs_update_time(struct inode *inode, int flags) 873 { 874 struct iattr iattr; 875 876 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n", 877 get_khandle_from_ino(inode)); 878 flags = generic_update_time(inode, flags); 879 memset(&iattr, 0, sizeof iattr); 880 if (flags & S_ATIME) 881 iattr.ia_valid |= ATTR_ATIME; 882 if (flags & S_CTIME) 883 iattr.ia_valid |= ATTR_CTIME; 884 if (flags & S_MTIME) 885 iattr.ia_valid |= ATTR_MTIME; 886 return __orangefs_setattr(inode, &iattr); 887 } 888 889 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa) 890 { 891 u64 val = 0; 892 int ret; 893 894 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__, 895 dentry); 896 897 ret = orangefs_inode_getxattr(d_inode(dentry), 898 "user.pvfs2.meta_hint", 899 &val, sizeof(val)); 900 if (ret < 0 && ret != -ENODATA) 901 return ret; 902 903 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val); 904 905 fileattr_fill_flags(fa, val); 906 return 0; 907 } 908 909 static int orangefs_fileattr_set(struct mnt_idmap *idmap, 910 struct dentry *dentry, struct fileattr *fa) 911 { 912 u64 val = 0; 913 914 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__, 915 dentry); 916 /* 917 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is 918 * turned on for a file. The user is not allowed to turn on this bit, 919 * but the bit is present if the user first gets the flags and then 920 * updates the flags with some new settings. So, we ignore it in the 921 * following edit. bligon. 922 */ 923 if (fileattr_has_fsx(fa) || 924 (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) { 925 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n", 926 __func__); 927 return -EOPNOTSUPP; 928 } 929 val = fa->flags; 930 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val); 931 return orangefs_inode_setxattr(d_inode(dentry), 932 "user.pvfs2.meta_hint", 933 &val, sizeof(val), 0); 934 } 935 936 /* ORANGEFS2 implementation of VFS inode operations for files */ 937 static const struct inode_operations orangefs_file_inode_operations = { 938 .get_inode_acl = orangefs_get_acl, 939 .set_acl = orangefs_set_acl, 940 .setattr = orangefs_setattr, 941 .getattr = orangefs_getattr, 942 .listxattr = orangefs_listxattr, 943 .permission = orangefs_permission, 944 .update_time = orangefs_update_time, 945 .fileattr_get = orangefs_fileattr_get, 946 .fileattr_set = orangefs_fileattr_set, 947 }; 948 949 static int orangefs_init_iops(struct inode *inode) 950 { 951 inode->i_mapping->a_ops = &orangefs_address_operations; 952 953 switch (inode->i_mode & S_IFMT) { 954 case S_IFREG: 955 inode->i_op = &orangefs_file_inode_operations; 956 inode->i_fop = &orangefs_file_operations; 957 break; 958 case S_IFLNK: 959 inode->i_op = &orangefs_symlink_inode_operations; 960 break; 961 case S_IFDIR: 962 inode->i_op = &orangefs_dir_inode_operations; 963 inode->i_fop = &orangefs_dir_operations; 964 break; 965 default: 966 gossip_debug(GOSSIP_INODE_DEBUG, 967 "%s: unsupported mode\n", 968 __func__); 969 return -EINVAL; 970 } 971 972 return 0; 973 } 974 975 /* 976 * Given an ORANGEFS object identifier (fsid, handle), convert it into 977 * a ino_t type that will be used as a hash-index from where the handle will 978 * be searched for in the VFS hash table of inodes. 979 */ 980 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref) 981 { 982 if (!ref) 983 return 0; 984 return orangefs_khandle_to_ino(&(ref->khandle)); 985 } 986 987 /* 988 * Called to set up an inode from iget5_locked. 989 */ 990 static int orangefs_set_inode(struct inode *inode, void *data) 991 { 992 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; 993 ORANGEFS_I(inode)->refn.fs_id = ref->fs_id; 994 ORANGEFS_I(inode)->refn.khandle = ref->khandle; 995 ORANGEFS_I(inode)->attr_valid = 0; 996 hash_init(ORANGEFS_I(inode)->xattr_cache); 997 ORANGEFS_I(inode)->mapping_time = jiffies - 1; 998 ORANGEFS_I(inode)->bitlock = 0; 999 return 0; 1000 } 1001 1002 /* 1003 * Called to determine if handles match. 1004 */ 1005 static int orangefs_test_inode(struct inode *inode, void *data) 1006 { 1007 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; 1008 struct orangefs_inode_s *orangefs_inode = NULL; 1009 1010 orangefs_inode = ORANGEFS_I(inode); 1011 /* test handles and fs_ids... */ 1012 return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), 1013 &(ref->khandle)) && 1014 orangefs_inode->refn.fs_id == ref->fs_id); 1015 } 1016 1017 /* 1018 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS 1019 * file handle. 1020 * 1021 * @sb: the file system super block instance. 1022 * @ref: The ORANGEFS object for which we are trying to locate an inode. 1023 */ 1024 struct inode *orangefs_iget(struct super_block *sb, 1025 struct orangefs_object_kref *ref) 1026 { 1027 struct inode *inode = NULL; 1028 unsigned long hash; 1029 int error; 1030 1031 hash = orangefs_handle_hash(ref); 1032 inode = iget5_locked(sb, 1033 hash, 1034 orangefs_test_inode, 1035 orangefs_set_inode, 1036 ref); 1037 1038 if (!inode) 1039 return ERR_PTR(-ENOMEM); 1040 1041 if (!(inode->i_state & I_NEW)) 1042 return inode; 1043 1044 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW); 1045 if (error) { 1046 iget_failed(inode); 1047 return ERR_PTR(error); 1048 } 1049 1050 inode->i_ino = hash; /* needed for stat etc */ 1051 orangefs_init_iops(inode); 1052 unlock_new_inode(inode); 1053 1054 gossip_debug(GOSSIP_INODE_DEBUG, 1055 "iget handle %pU, fsid %d hash %ld i_ino %lu\n", 1056 &ref->khandle, 1057 ref->fs_id, 1058 hash, 1059 inode->i_ino); 1060 1061 return inode; 1062 } 1063 1064 /* 1065 * Allocate an inode for a newly created file and insert it into the inode hash. 1066 */ 1067 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir, 1068 umode_t mode, dev_t dev, struct orangefs_object_kref *ref) 1069 { 1070 struct posix_acl *acl = NULL, *default_acl = NULL; 1071 unsigned long hash = orangefs_handle_hash(ref); 1072 struct inode *inode; 1073 int error; 1074 1075 gossip_debug(GOSSIP_INODE_DEBUG, 1076 "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", 1077 __func__, 1078 sb, 1079 MAJOR(dev), 1080 MINOR(dev), 1081 mode); 1082 1083 inode = new_inode(sb); 1084 if (!inode) 1085 return ERR_PTR(-ENOMEM); 1086 1087 error = posix_acl_create(dir, &mode, &default_acl, &acl); 1088 if (error) 1089 goto out_iput; 1090 1091 orangefs_set_inode(inode, ref); 1092 inode->i_ino = hash; /* needed for stat etc */ 1093 1094 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW); 1095 if (error) 1096 goto out_iput; 1097 1098 orangefs_init_iops(inode); 1099 inode->i_rdev = dev; 1100 1101 if (default_acl) { 1102 error = __orangefs_set_acl(inode, default_acl, 1103 ACL_TYPE_DEFAULT); 1104 if (error) 1105 goto out_iput; 1106 } 1107 1108 if (acl) { 1109 error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS); 1110 if (error) 1111 goto out_iput; 1112 } 1113 1114 error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref); 1115 if (error < 0) 1116 goto out_iput; 1117 1118 gossip_debug(GOSSIP_INODE_DEBUG, 1119 "Initializing ACL's for inode %pU\n", 1120 get_khandle_from_ino(inode)); 1121 if (mode != inode->i_mode) { 1122 struct iattr iattr = { 1123 .ia_mode = mode, 1124 .ia_valid = ATTR_MODE, 1125 }; 1126 inode->i_mode = mode; 1127 __orangefs_setattr(inode, &iattr); 1128 __posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode); 1129 } 1130 posix_acl_release(acl); 1131 posix_acl_release(default_acl); 1132 return inode; 1133 1134 out_iput: 1135 iput(inode); 1136 posix_acl_release(acl); 1137 posix_acl_release(default_acl); 1138 return ERR_PTR(error); 1139 } 1140