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 off = wr->pos; 36 if ((off + wr->len > len) && (off <= len)) 37 wlen = len - off; 38 else 39 wlen = wr->len; 40 if (wlen == 0) 41 wlen = wr->len; 42 } else { 43 WARN_ON(1); 44 off = folio_pos(folio); 45 wlen = folio_size(folio); 46 47 if (wlen > len - off) 48 wlen = len - off; 49 } 50 51 WARN_ON(wlen == 0); 52 bvec_set_folio(&bv, folio, wlen, offset_in_folio(folio, off)); 53 iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen); 54 55 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen, 56 len, wr, NULL, NULL); 57 if (ret < 0) { 58 mapping_set_error(folio->mapping, ret); 59 } else { 60 ret = 0; 61 } 62 kfree(folio_detach_private(folio)); 63 return ret; 64 } 65 66 struct orangefs_writepages { 67 loff_t off; 68 size_t len; 69 kuid_t uid; 70 kgid_t gid; 71 int maxpages; 72 int nfolios; 73 struct address_space *mapping; 74 struct folio **folios; 75 struct bio_vec *bv; 76 }; 77 78 static int orangefs_writepages_work(struct orangefs_writepages *ow, 79 struct writeback_control *wbc) 80 { 81 struct inode *inode = ow->mapping->host; 82 struct orangefs_write_range *wrp, wr; 83 struct iov_iter iter; 84 ssize_t ret; 85 size_t start; 86 loff_t len, off; 87 int i; 88 89 len = i_size_read(inode); 90 91 start = offset_in_folio(ow->folios[0], ow->off); 92 for (i = 0; i < ow->nfolios; i++) { 93 folio_start_writeback(ow->folios[i]); 94 bvec_set_folio(&ow->bv[i], ow->folios[i], 95 folio_size(ow->folios[i]) - start, start); 96 start = 0; 97 } 98 iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->nfolios, ow->len); 99 100 WARN_ON(ow->off >= len); 101 if (ow->off + ow->len > len) 102 ow->len = len - ow->off; 103 104 off = ow->off; 105 wr.uid = ow->uid; 106 wr.gid = ow->gid; 107 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len, 108 0, &wr, NULL, NULL); 109 if (ret < 0) 110 mapping_set_error(ow->mapping, ret); 111 else 112 ret = 0; 113 114 for (i = 0; i < ow->nfolios; i++) { 115 wrp = folio_detach_private(ow->folios[i]); 116 kfree(wrp); 117 folio_end_writeback(ow->folios[i]); 118 folio_unlock(ow->folios[i]); 119 } 120 121 return ret; 122 } 123 124 static int orangefs_writepages_callback(struct folio *folio, 125 struct writeback_control *wbc, struct orangefs_writepages *ow) 126 { 127 struct orangefs_write_range *wr = folio->private; 128 int ret; 129 130 if (!wr) { 131 folio_unlock(folio); 132 /* It's not private so there's nothing to write, right? */ 133 printk("writepages_callback not private!\n"); 134 BUG(); 135 return 0; 136 } 137 138 ret = -1; 139 if (ow->nfolios == 0) { 140 ow->off = wr->pos; 141 ow->len = wr->len; 142 ow->uid = wr->uid; 143 ow->gid = wr->gid; 144 ow->folios[ow->nfolios++] = folio; 145 ret = 0; 146 goto done; 147 } 148 if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) { 149 orangefs_writepages_work(ow, wbc); 150 ow->nfolios = 0; 151 ret = -1; 152 goto done; 153 } 154 if (ow->off + ow->len == wr->pos) { 155 ow->len += wr->len; 156 ow->folios[ow->nfolios++] = folio; 157 ret = 0; 158 goto done; 159 } 160 done: 161 if (ret == -1) { 162 if (ow->nfolios) { 163 orangefs_writepages_work(ow, wbc); 164 ow->nfolios = 0; 165 } 166 ret = orangefs_writepage_locked(folio, wbc); 167 mapping_set_error(folio->mapping, ret); 168 folio_unlock(folio); 169 folio_end_writeback(folio); 170 } else { 171 if (ow->nfolios == ow->maxpages) { 172 orangefs_writepages_work(ow, wbc); 173 ow->nfolios = 0; 174 } 175 } 176 return ret; 177 } 178 179 static int orangefs_writepages(struct address_space *mapping, 180 struct writeback_control *wbc) 181 { 182 struct orangefs_writepages *ow; 183 struct blk_plug plug; 184 int error; 185 struct folio *folio = NULL; 186 187 ow = kzalloc_obj(struct orangefs_writepages); 188 if (!ow) 189 return -ENOMEM; 190 ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE; 191 ow->folios = kzalloc_objs(struct folio *, ow->maxpages); 192 if (!ow->folios) { 193 kfree(ow); 194 return -ENOMEM; 195 } 196 ow->bv = kzalloc_objs(struct bio_vec, ow->maxpages); 197 if (!ow->bv) { 198 kfree(ow->folios); 199 kfree(ow); 200 return -ENOMEM; 201 } 202 ow->mapping = mapping; 203 blk_start_plug(&plug); 204 while ((folio = writeback_iter(mapping, wbc, folio, &error))) 205 error = orangefs_writepages_callback(folio, wbc, ow); 206 if (ow->nfolios) 207 error = orangefs_writepages_work(ow, wbc); 208 blk_finish_plug(&plug); 209 kfree(ow->folios); 210 kfree(ow->bv); 211 kfree(ow); 212 return error; 213 } 214 215 static int orangefs_launder_folio(struct folio *); 216 217 static void orangefs_readahead(struct readahead_control *rac) 218 { 219 loff_t offset; 220 struct iov_iter iter; 221 struct inode *inode = rac->mapping->host; 222 struct xarray *i_pages; 223 struct folio *folio; 224 loff_t new_start = readahead_pos(rac); 225 int ret; 226 size_t new_len = 0; 227 size_t this_size; 228 size_t remaining; 229 230 loff_t bytes_remaining = inode->i_size - readahead_pos(rac); 231 loff_t pages_remaining = bytes_remaining / PAGE_SIZE; 232 233 if (pages_remaining >= 1024) 234 new_len = 4194304; 235 else if (pages_remaining > readahead_count(rac)) 236 new_len = bytes_remaining; 237 238 if (new_len) 239 readahead_expand(rac, new_start, new_len); 240 241 offset = readahead_pos(rac); 242 i_pages = &rac->mapping->i_pages; 243 244 iov_iter_xarray(&iter, ITER_DEST, i_pages, 245 offset, readahead_length(rac)); 246 247 remaining = readahead_length(rac); 248 while (remaining) { 249 if (remaining > 4194304) 250 this_size = 4194304; 251 else 252 this_size = remaining; 253 254 /* read in the pages. */ 255 if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, 256 &offset, &iter, this_size, 257 inode->i_size, NULL, NULL, rac->file)) < 0) { 258 gossip_debug(GOSSIP_FILE_DEBUG, 259 "%s: wait_for_direct_io failed. :%d: \n", 260 __func__, ret); 261 goto cleanup; 262 } else { 263 ret = 0; 264 } 265 266 remaining -= this_size; 267 offset += this_size; 268 } 269 270 cleanup: 271 /* clean up. */ 272 while ((folio = readahead_folio(rac))) { 273 if (!ret) 274 folio_mark_uptodate(folio); 275 folio_unlock(folio); 276 } 277 } 278 279 static int orangefs_read_folio(struct file *file, struct folio *folio) 280 { 281 struct inode *inode = folio->mapping->host; 282 struct iov_iter iter; 283 struct bio_vec bv; 284 ssize_t ret; 285 loff_t off; /* offset of this folio in the file */ 286 287 if (folio_test_dirty(folio)) 288 orangefs_launder_folio(folio); 289 290 off = folio_pos(folio); 291 bvec_set_folio(&bv, folio, folio_size(folio), 0); 292 iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio)); 293 294 ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter, 295 folio_size(folio), inode->i_size, NULL, NULL, file); 296 /* this will only zero remaining unread portions of the folio data */ 297 iov_iter_zero(~0U, &iter); 298 /* takes care of potential aliasing */ 299 flush_dcache_folio(folio); 300 if (ret > 0) 301 ret = 0; 302 folio_end_read(folio, ret == 0); 303 return ret; 304 } 305 306 static int orangefs_write_begin(const struct kiocb *iocb, 307 struct address_space *mapping, loff_t pos, 308 unsigned len, struct folio **foliop, 309 void **fsdata) 310 { 311 struct orangefs_write_range *wr; 312 struct folio *folio; 313 int ret; 314 315 folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN, 316 mapping_gfp_mask(mapping)); 317 if (IS_ERR(folio)) 318 return PTR_ERR(folio); 319 320 *foliop = folio; 321 322 if (folio_test_dirty(folio) && !folio_test_private(folio)) { 323 /* 324 * Should be impossible. If it happens, launder the page 325 * since we don't know what's dirty. This will WARN in 326 * orangefs_writepage_locked. 327 */ 328 ret = orangefs_launder_folio(folio); 329 if (ret) 330 return ret; 331 } 332 if (folio_test_private(folio)) { 333 struct orangefs_write_range *wr; 334 wr = folio_get_private(folio); 335 if (wr->pos + wr->len == pos && 336 uid_eq(wr->uid, current_fsuid()) && 337 gid_eq(wr->gid, current_fsgid())) { 338 wr->len += len; 339 goto okay; 340 } else { 341 wr->pos = pos; 342 wr->len = len; 343 ret = orangefs_launder_folio(folio); 344 if (ret) 345 return ret; 346 } 347 } 348 349 wr = kmalloc_obj(*wr); 350 if (!wr) 351 return -ENOMEM; 352 353 wr->pos = pos; 354 wr->len = len; 355 wr->uid = current_fsuid(); 356 wr->gid = current_fsgid(); 357 folio_attach_private(folio, wr); 358 okay: 359 return 0; 360 } 361 362 static int orangefs_write_end(const struct kiocb *iocb, 363 struct address_space *mapping, 364 loff_t pos, unsigned len, unsigned copied, 365 struct folio *folio, void *fsdata) 366 { 367 struct inode *inode = folio->mapping->host; 368 loff_t last_pos = pos + copied; 369 370 /* 371 * No need to use i_size_read() here, the i_size 372 * cannot change under us because we hold the i_mutex. 373 */ 374 if (last_pos > inode->i_size) 375 i_size_write(inode, last_pos); 376 377 /* zero the stale part of the folio if we did a short copy */ 378 if (!folio_test_uptodate(folio)) { 379 unsigned from = pos & (PAGE_SIZE - 1); 380 if (copied < len) { 381 folio_zero_range(folio, from + copied, len - copied); 382 } 383 /* Set fully written pages uptodate. */ 384 if (pos == folio_pos(folio) && 385 (len == PAGE_SIZE || pos + len == inode->i_size)) { 386 folio_zero_segment(folio, from + copied, PAGE_SIZE); 387 folio_mark_uptodate(folio); 388 } 389 } 390 391 folio_mark_dirty(folio); 392 folio_unlock(folio); 393 folio_put(folio); 394 395 mark_inode_dirty_sync(file_inode(iocb->ki_filp)); 396 return copied; 397 } 398 399 static void orangefs_invalidate_folio(struct folio *folio, 400 size_t offset, size_t length) 401 { 402 struct orangefs_write_range *wr = folio_get_private(folio); 403 404 if (offset == 0 && length == PAGE_SIZE) { 405 kfree(folio_detach_private(folio)); 406 return; 407 /* write range entirely within invalidate range (or equal) */ 408 } else if (folio_pos(folio) + offset <= wr->pos && 409 wr->pos + wr->len <= folio_pos(folio) + offset + length) { 410 kfree(folio_detach_private(folio)); 411 /* XXX is this right? only caller in fs */ 412 folio_cancel_dirty(folio); 413 return; 414 /* invalidate range chops off end of write range */ 415 } else if (wr->pos < folio_pos(folio) + offset && 416 wr->pos + wr->len <= folio_pos(folio) + offset + length && 417 folio_pos(folio) + offset < wr->pos + wr->len) { 418 size_t x; 419 x = wr->pos + wr->len - (folio_pos(folio) + offset); 420 WARN_ON(x > wr->len); 421 wr->len -= x; 422 wr->uid = current_fsuid(); 423 wr->gid = current_fsgid(); 424 /* invalidate range chops off beginning of write range */ 425 } else if (folio_pos(folio) + offset <= wr->pos && 426 folio_pos(folio) + offset + length < wr->pos + wr->len && 427 wr->pos < folio_pos(folio) + offset + length) { 428 size_t x; 429 x = folio_pos(folio) + offset + length - wr->pos; 430 WARN_ON(x > wr->len); 431 wr->pos += x; 432 wr->len -= x; 433 wr->uid = current_fsuid(); 434 wr->gid = current_fsgid(); 435 /* invalidate range entirely within write range (punch hole) */ 436 } else if (wr->pos < folio_pos(folio) + offset && 437 folio_pos(folio) + offset + length < wr->pos + wr->len) { 438 /* XXX what do we do here... should not WARN_ON */ 439 WARN_ON(1); 440 /* punch hole */ 441 /* 442 * should we just ignore this and write it out anyway? 443 * it hardly makes sense 444 */ 445 return; 446 /* non-overlapping ranges */ 447 } else { 448 /* WARN if they do overlap */ 449 if (!((folio_pos(folio) + offset + length <= wr->pos) ^ 450 (wr->pos + wr->len <= folio_pos(folio) + offset))) { 451 WARN_ON(1); 452 printk("invalidate range offset %llu length %zu\n", 453 folio_pos(folio) + offset, length); 454 printk("write range offset %llu length %zu\n", 455 wr->pos, wr->len); 456 } 457 return; 458 } 459 460 /* 461 * Above there are returns where wr is freed or where we WARN. 462 * Thus the following runs if wr was modified above. 463 */ 464 465 orangefs_launder_folio(folio); 466 } 467 468 static bool orangefs_release_folio(struct folio *folio, gfp_t foo) 469 { 470 return !folio_test_private(folio); 471 } 472 473 static void orangefs_free_folio(struct folio *folio) 474 { 475 kfree(folio_detach_private(folio)); 476 } 477 478 static int orangefs_launder_folio(struct folio *folio) 479 { 480 int r = 0; 481 struct writeback_control wbc = { 482 .sync_mode = WB_SYNC_ALL, 483 .nr_to_write = 0, 484 }; 485 folio_wait_writeback(folio); 486 if (folio_clear_dirty_for_io(folio)) { 487 r = orangefs_writepage_locked(folio, &wbc); 488 folio_end_writeback(folio); 489 } 490 return r; 491 } 492 493 static ssize_t orangefs_direct_IO(struct kiocb *iocb, 494 struct iov_iter *iter) 495 { 496 /* 497 * Comment from original do_readv_writev: 498 * Common entry point for read/write/readv/writev 499 * This function will dispatch it to either the direct I/O 500 * or buffered I/O path depending on the mount options and/or 501 * augmented/extended metadata attached to the file. 502 * Note: File extended attributes override any mount options. 503 */ 504 struct file *file = iocb->ki_filp; 505 loff_t pos = iocb->ki_pos; 506 enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ? 507 ORANGEFS_IO_WRITE : ORANGEFS_IO_READ; 508 loff_t *offset = &pos; 509 struct inode *inode = file->f_mapping->host; 510 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 511 struct orangefs_khandle *handle = &orangefs_inode->refn.khandle; 512 size_t count = iov_iter_count(iter); 513 ssize_t total_count = 0; 514 ssize_t ret = -EINVAL; 515 516 gossip_debug(GOSSIP_FILE_DEBUG, 517 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n", 518 __func__, 519 handle, 520 (int)count); 521 522 if (type == ORANGEFS_IO_WRITE) { 523 gossip_debug(GOSSIP_FILE_DEBUG, 524 "%s(%pU): proceeding with offset : %llu, " 525 "size %d\n", 526 __func__, 527 handle, 528 llu(*offset), 529 (int)count); 530 } 531 532 if (count == 0) { 533 ret = 0; 534 goto out; 535 } 536 537 while (iov_iter_count(iter)) { 538 size_t each_count = iov_iter_count(iter); 539 size_t amt_complete; 540 541 /* how much to transfer in this loop iteration */ 542 if (each_count > orangefs_bufmap_size_query()) 543 each_count = orangefs_bufmap_size_query(); 544 545 gossip_debug(GOSSIP_FILE_DEBUG, 546 "%s(%pU): size of each_count(%d)\n", 547 __func__, 548 handle, 549 (int)each_count); 550 gossip_debug(GOSSIP_FILE_DEBUG, 551 "%s(%pU): BEFORE wait_for_io: offset is %d\n", 552 __func__, 553 handle, 554 (int)*offset); 555 556 ret = wait_for_direct_io(type, inode, offset, iter, 557 each_count, 0, NULL, NULL, file); 558 gossip_debug(GOSSIP_FILE_DEBUG, 559 "%s(%pU): return from wait_for_io:%d\n", 560 __func__, 561 handle, 562 (int)ret); 563 564 if (ret < 0) 565 goto out; 566 567 *offset += ret; 568 total_count += ret; 569 amt_complete = ret; 570 571 gossip_debug(GOSSIP_FILE_DEBUG, 572 "%s(%pU): AFTER wait_for_io: offset is %d\n", 573 __func__, 574 handle, 575 (int)*offset); 576 577 /* 578 * if we got a short I/O operations, 579 * fall out and return what we got so far 580 */ 581 if (amt_complete < each_count) 582 break; 583 } /*end while */ 584 585 out: 586 if (total_count > 0) 587 ret = total_count; 588 if (ret > 0) { 589 if (type == ORANGEFS_IO_READ) { 590 file_accessed(file); 591 } else { 592 file_update_time(file); 593 if (*offset > i_size_read(inode)) 594 i_size_write(inode, *offset); 595 } 596 } 597 598 gossip_debug(GOSSIP_FILE_DEBUG, 599 "%s(%pU): Value(%d) returned.\n", 600 __func__, 601 handle, 602 (int)ret); 603 604 return ret; 605 } 606 607 /** ORANGEFS2 implementation of address space operations */ 608 static const struct address_space_operations orangefs_address_operations = { 609 .readahead = orangefs_readahead, 610 .read_folio = orangefs_read_folio, 611 .writepages = orangefs_writepages, 612 .dirty_folio = filemap_dirty_folio, 613 .write_begin = orangefs_write_begin, 614 .write_end = orangefs_write_end, 615 .invalidate_folio = orangefs_invalidate_folio, 616 .release_folio = orangefs_release_folio, 617 .free_folio = orangefs_free_folio, 618 .migrate_folio = filemap_migrate_folio, 619 .launder_folio = orangefs_launder_folio, 620 .direct_IO = orangefs_direct_IO, 621 }; 622 623 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf) 624 { 625 struct folio *folio = page_folio(vmf->page); 626 struct inode *inode = file_inode(vmf->vma->vm_file); 627 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 628 unsigned long *bitlock = &orangefs_inode->bitlock; 629 vm_fault_t ret; 630 struct orangefs_write_range *wr; 631 632 sb_start_pagefault(inode->i_sb); 633 634 if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) { 635 ret = VM_FAULT_RETRY; 636 goto out; 637 } 638 639 folio_lock(folio); 640 if (folio_test_dirty(folio) && !folio_test_private(folio)) { 641 /* 642 * Should be impossible. If it happens, launder the folio 643 * since we don't know what's dirty. This will WARN in 644 * orangefs_writepage_locked. 645 */ 646 if (orangefs_launder_folio(folio)) { 647 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; 648 goto out; 649 } 650 } 651 if (folio_test_private(folio)) { 652 wr = folio_get_private(folio); 653 if (uid_eq(wr->uid, current_fsuid()) && 654 gid_eq(wr->gid, current_fsgid())) { 655 wr->pos = page_offset(vmf->page); 656 wr->len = PAGE_SIZE; 657 goto okay; 658 } else { 659 if (orangefs_launder_folio(folio)) { 660 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; 661 goto out; 662 } 663 } 664 } 665 wr = kmalloc_obj(*wr); 666 if (!wr) { 667 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY; 668 goto out; 669 } 670 wr->pos = page_offset(vmf->page); 671 wr->len = PAGE_SIZE; 672 wr->uid = current_fsuid(); 673 wr->gid = current_fsgid(); 674 folio_attach_private(folio, wr); 675 okay: 676 677 file_update_time(vmf->vma->vm_file); 678 if (folio->mapping != inode->i_mapping) { 679 folio_unlock(folio); 680 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE; 681 goto out; 682 } 683 684 /* 685 * We mark the folio dirty already here so that when freeze is in 686 * progress, we are guaranteed that writeback during freezing will 687 * see the dirty folio and writeprotect it again. 688 */ 689 folio_mark_dirty(folio); 690 folio_wait_stable(folio); 691 ret = VM_FAULT_LOCKED; 692 out: 693 sb_end_pagefault(inode->i_sb); 694 return ret; 695 } 696 697 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr) 698 { 699 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode); 700 struct orangefs_kernel_op_s *new_op; 701 loff_t orig_size; 702 int ret = -EINVAL; 703 704 gossip_debug(GOSSIP_INODE_DEBUG, 705 "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n", 706 __func__, 707 get_khandle_from_ino(inode), 708 &orangefs_inode->refn.khandle, 709 orangefs_inode->refn.fs_id, 710 iattr->ia_size); 711 712 /* Ensure that we have a up to date size, so we know if it changed. */ 713 ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE); 714 if (ret == -ESTALE) 715 ret = -EIO; 716 if (ret) { 717 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n", 718 __func__, ret); 719 return ret; 720 } 721 orig_size = i_size_read(inode); 722 723 /* This is truncate_setsize in a different order. */ 724 truncate_pagecache(inode, iattr->ia_size); 725 i_size_write(inode, iattr->ia_size); 726 if (iattr->ia_size > orig_size) 727 pagecache_isize_extended(inode, orig_size, iattr->ia_size); 728 729 new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE); 730 if (!new_op) 731 return -ENOMEM; 732 733 new_op->upcall.req.truncate.refn = orangefs_inode->refn; 734 new_op->upcall.req.truncate.size = (__s64) iattr->ia_size; 735 736 ret = service_operation(new_op, 737 __func__, 738 get_interruptible_flag(inode)); 739 740 /* 741 * the truncate has no downcall members to retrieve, but 742 * the status value tells us if it went through ok or not 743 */ 744 gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret); 745 746 op_release(new_op); 747 748 if (ret != 0) 749 return ret; 750 751 if (orig_size != i_size_read(inode)) 752 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 753 754 return ret; 755 } 756 757 int __orangefs_setattr(struct inode *inode, struct iattr *iattr) 758 { 759 int ret; 760 761 if (iattr->ia_valid & ATTR_MODE) { 762 if (iattr->ia_mode & (S_ISVTX)) { 763 if (is_root_handle(inode)) { 764 /* 765 * allow sticky bit to be set on root (since 766 * it shows up that way by default anyhow), 767 * but don't show it to the server 768 */ 769 iattr->ia_mode -= S_ISVTX; 770 } else { 771 gossip_debug(GOSSIP_UTILS_DEBUG, 772 "User attempted to set sticky bit on non-root directory; returning EINVAL.\n"); 773 ret = -EINVAL; 774 goto out; 775 } 776 } 777 if (iattr->ia_mode & (S_ISUID)) { 778 gossip_debug(GOSSIP_UTILS_DEBUG, 779 "Attempting to set setuid bit (not supported); returning EINVAL.\n"); 780 ret = -EINVAL; 781 goto out; 782 } 783 } 784 785 if (iattr->ia_valid & ATTR_SIZE) { 786 ret = orangefs_setattr_size(inode, iattr); 787 if (ret) 788 goto out; 789 } 790 791 again: 792 spin_lock(&inode->i_lock); 793 if (ORANGEFS_I(inode)->attr_valid) { 794 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) && 795 gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) { 796 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid; 797 } else { 798 spin_unlock(&inode->i_lock); 799 write_inode_now(inode, 1); 800 goto again; 801 } 802 } else { 803 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid; 804 ORANGEFS_I(inode)->attr_uid = current_fsuid(); 805 ORANGEFS_I(inode)->attr_gid = current_fsgid(); 806 } 807 setattr_copy(&nop_mnt_idmap, inode, iattr); 808 spin_unlock(&inode->i_lock); 809 mark_inode_dirty(inode); 810 811 ret = 0; 812 out: 813 return ret; 814 } 815 816 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr) 817 { 818 int ret; 819 struct inode *inode = d_inode(dentry); 820 821 ret = __orangefs_setattr(inode, iattr); 822 /* change mode on a file that has ACLs */ 823 if (!ret && (iattr->ia_valid & ATTR_MODE)) 824 ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode); 825 return ret; 826 } 827 828 /* 829 * Change attributes of an object referenced by dentry. 830 */ 831 int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry, 832 struct iattr *iattr) 833 { 834 int ret; 835 gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n", 836 dentry); 837 ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr); 838 if (ret) 839 goto out; 840 ret = __orangefs_setattr_mode(dentry, iattr); 841 sync_inode_metadata(d_inode(dentry), 1); 842 out: 843 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n", 844 ret); 845 return ret; 846 } 847 848 /* 849 * Obtain attributes of an object given a dentry 850 */ 851 int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path, 852 struct kstat *stat, u32 request_mask, unsigned int flags) 853 { 854 int ret; 855 struct inode *inode = path->dentry->d_inode; 856 857 gossip_debug(GOSSIP_INODE_DEBUG, 858 "orangefs_getattr: called on %pd mask %u\n", 859 path->dentry, request_mask); 860 861 ret = orangefs_inode_getattr(inode, 862 request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0); 863 if (ret == 0) { 864 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat); 865 866 /* override block size reported to stat */ 867 if (!(request_mask & STATX_SIZE)) 868 stat->result_mask &= ~STATX_SIZE; 869 870 generic_fill_statx_attr(inode, stat); 871 } 872 return ret; 873 } 874 875 int orangefs_permission(struct mnt_idmap *idmap, 876 struct inode *inode, int mask) 877 { 878 int ret; 879 880 if (mask & MAY_NOT_BLOCK) 881 return -ECHILD; 882 883 gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__); 884 885 /* Make sure the permission (and other common attrs) are up to date. */ 886 ret = orangefs_inode_getattr(inode, 0); 887 if (ret < 0) 888 return ret; 889 890 return generic_permission(&nop_mnt_idmap, inode, mask); 891 } 892 893 int orangefs_update_time(struct inode *inode, enum fs_update_time type, 894 unsigned int flags) 895 { 896 struct iattr iattr = { }; 897 int dirty; 898 899 if (flags & IOCB_NOWAIT) 900 return -EAGAIN; 901 902 switch (type) { 903 case FS_UPD_ATIME: 904 iattr.ia_valid = ATTR_ATIME; 905 break; 906 case FS_UPD_CMTIME: 907 iattr.ia_valid = ATTR_CTIME | ATTR_MTIME; 908 break; 909 } 910 911 dirty = inode_update_time(inode, type, flags); 912 if (dirty <= 0) 913 return dirty; 914 return __orangefs_setattr(inode, &iattr); 915 } 916 917 static int orangefs_fileattr_get(struct dentry *dentry, struct file_kattr *fa) 918 { 919 u64 val = 0; 920 int ret; 921 922 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__, 923 dentry); 924 925 ret = orangefs_inode_getxattr(d_inode(dentry), 926 "user.pvfs2.meta_hint", 927 &val, sizeof(val)); 928 if (ret < 0 && ret != -ENODATA) 929 return ret; 930 931 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val); 932 933 fileattr_fill_flags(fa, val); 934 return 0; 935 } 936 937 static int orangefs_fileattr_set(struct mnt_idmap *idmap, 938 struct dentry *dentry, struct file_kattr *fa) 939 { 940 u64 val = 0; 941 942 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__, 943 dentry); 944 /* 945 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is 946 * turned on for a file. The user is not allowed to turn on this bit, 947 * but the bit is present if the user first gets the flags and then 948 * updates the flags with some new settings. So, we ignore it in the 949 * following edit. bligon. 950 */ 951 if (fileattr_has_fsx(fa) || 952 (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) { 953 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n", 954 __func__); 955 return -EOPNOTSUPP; 956 } 957 val = fa->flags; 958 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val); 959 return orangefs_inode_setxattr(d_inode(dentry), 960 "user.pvfs2.meta_hint", 961 &val, sizeof(val), 0); 962 } 963 964 /* ORANGEFS2 implementation of VFS inode operations for files */ 965 static const struct inode_operations orangefs_file_inode_operations = { 966 .get_inode_acl = orangefs_get_acl, 967 .set_acl = orangefs_set_acl, 968 .setattr = orangefs_setattr, 969 .getattr = orangefs_getattr, 970 .listxattr = orangefs_listxattr, 971 .permission = orangefs_permission, 972 .update_time = orangefs_update_time, 973 .fileattr_get = orangefs_fileattr_get, 974 .fileattr_set = orangefs_fileattr_set, 975 }; 976 977 static int orangefs_init_iops(struct inode *inode) 978 { 979 inode->i_mapping->a_ops = &orangefs_address_operations; 980 981 switch (inode->i_mode & S_IFMT) { 982 case S_IFREG: 983 inode->i_op = &orangefs_file_inode_operations; 984 inode->i_fop = &orangefs_file_operations; 985 break; 986 case S_IFLNK: 987 inode->i_op = &orangefs_symlink_inode_operations; 988 break; 989 case S_IFDIR: 990 inode->i_op = &orangefs_dir_inode_operations; 991 inode->i_fop = &orangefs_dir_operations; 992 break; 993 default: 994 gossip_debug(GOSSIP_INODE_DEBUG, 995 "%s: unsupported mode\n", 996 __func__); 997 return -EINVAL; 998 } 999 1000 return 0; 1001 } 1002 1003 /* 1004 * Given an ORANGEFS object identifier (fsid, handle), convert it into 1005 * a ino_t type that will be used as a hash-index from where the handle will 1006 * be searched for in the VFS hash table of inodes. 1007 */ 1008 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref) 1009 { 1010 if (!ref) 1011 return 0; 1012 return orangefs_khandle_to_ino(&(ref->khandle)); 1013 } 1014 1015 /* 1016 * Called to set up an inode from iget5_locked. 1017 */ 1018 static int orangefs_set_inode(struct inode *inode, void *data) 1019 { 1020 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; 1021 ORANGEFS_I(inode)->refn.fs_id = ref->fs_id; 1022 ORANGEFS_I(inode)->refn.khandle = ref->khandle; 1023 ORANGEFS_I(inode)->attr_valid = 0; 1024 hash_init(ORANGEFS_I(inode)->xattr_cache); 1025 ORANGEFS_I(inode)->mapping_time = jiffies - 1; 1026 ORANGEFS_I(inode)->bitlock = 0; 1027 return 0; 1028 } 1029 1030 /* 1031 * Called to determine if handles match. 1032 */ 1033 static int orangefs_test_inode(struct inode *inode, void *data) 1034 { 1035 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data; 1036 struct orangefs_inode_s *orangefs_inode = NULL; 1037 1038 orangefs_inode = ORANGEFS_I(inode); 1039 /* test handles and fs_ids... */ 1040 return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle), 1041 &(ref->khandle)) && 1042 orangefs_inode->refn.fs_id == ref->fs_id); 1043 } 1044 1045 /* 1046 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS 1047 * file handle. 1048 * 1049 * @sb: the file system super block instance. 1050 * @ref: The ORANGEFS object for which we are trying to locate an inode. 1051 */ 1052 struct inode *orangefs_iget(struct super_block *sb, 1053 struct orangefs_object_kref *ref) 1054 { 1055 struct inode *inode = NULL; 1056 unsigned long hash; 1057 int error; 1058 1059 hash = orangefs_handle_hash(ref); 1060 inode = iget5_locked(sb, 1061 hash, 1062 orangefs_test_inode, 1063 orangefs_set_inode, 1064 ref); 1065 1066 if (!inode) 1067 return ERR_PTR(-ENOMEM); 1068 1069 if (!(inode_state_read_once(inode) & I_NEW)) 1070 return inode; 1071 1072 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW); 1073 if (error) { 1074 iget_failed(inode); 1075 return ERR_PTR(error); 1076 } 1077 1078 inode->i_ino = hash; /* needed for stat etc */ 1079 orangefs_init_iops(inode); 1080 unlock_new_inode(inode); 1081 1082 gossip_debug(GOSSIP_INODE_DEBUG, 1083 "iget handle %pU, fsid %d hash %ld i_ino %llu\n", 1084 &ref->khandle, 1085 ref->fs_id, 1086 hash, 1087 inode->i_ino); 1088 1089 return inode; 1090 } 1091 1092 /* 1093 * Allocate an inode for a newly created file and insert it into the inode hash. 1094 */ 1095 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir, 1096 umode_t mode, dev_t dev, struct orangefs_object_kref *ref) 1097 { 1098 struct posix_acl *acl = NULL, *default_acl = NULL; 1099 unsigned long hash = orangefs_handle_hash(ref); 1100 struct inode *inode; 1101 int error; 1102 1103 gossip_debug(GOSSIP_INODE_DEBUG, 1104 "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n", 1105 __func__, 1106 sb, 1107 MAJOR(dev), 1108 MINOR(dev), 1109 mode); 1110 1111 inode = new_inode(sb); 1112 if (!inode) 1113 return ERR_PTR(-ENOMEM); 1114 1115 error = posix_acl_create(dir, &mode, &default_acl, &acl); 1116 if (error) 1117 goto out_iput; 1118 1119 orangefs_set_inode(inode, ref); 1120 inode->i_ino = hash; /* needed for stat etc */ 1121 1122 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW); 1123 if (error) 1124 goto out_iput; 1125 1126 orangefs_init_iops(inode); 1127 inode->i_rdev = dev; 1128 1129 if (default_acl) { 1130 error = __orangefs_set_acl(inode, default_acl, 1131 ACL_TYPE_DEFAULT); 1132 if (error) 1133 goto out_iput; 1134 } 1135 1136 if (acl) { 1137 error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS); 1138 if (error) 1139 goto out_iput; 1140 } 1141 1142 error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref); 1143 if (error < 0) 1144 goto out_iput; 1145 1146 gossip_debug(GOSSIP_INODE_DEBUG, 1147 "Initializing ACL's for inode %pU\n", 1148 get_khandle_from_ino(inode)); 1149 if (mode != inode->i_mode) { 1150 struct iattr iattr = { 1151 .ia_mode = mode, 1152 .ia_valid = ATTR_MODE, 1153 }; 1154 inode->i_mode = mode; 1155 __orangefs_setattr(inode, &iattr); 1156 __posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode); 1157 } 1158 posix_acl_release(acl); 1159 posix_acl_release(default_acl); 1160 return inode; 1161 1162 out_iput: 1163 iput(inode); 1164 posix_acl_release(acl); 1165 posix_acl_release(default_acl); 1166 return ERR_PTR(error); 1167 } 1168