fs/orangefs/inode.c

   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 page *page,
  20     struct writeback_control *wbc)
  21 {
  22         struct inode *inode = page->mapping->host;
  23         struct orangefs_write_range *wr = NULL;
  24         struct iov_iter iter;
  25         struct bio_vec bv;
  26         size_t len, wlen;
  27         ssize_t ret;
  28         loff_t off;
  29
  30         set_page_writeback(page);
  31
  32         len = i_size_read(inode);
  33         if (PagePrivate(page)) {
  34                 wr = (struct orangefs_write_range *)page_private(page);
  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 = page_offset(page);
  44                 if (off + PAGE_SIZE > len)
  45                         wlen = len - off;
  46                 else
  47                         wlen = PAGE_SIZE;
  48         }
  49         /* Should've been handled in orangefs_invalidate_folio. */
  50         WARN_ON(off == len || off + wlen > len);
  51
  52         bv.bv_page = page;
  53         bv.bv_len = wlen;
  54         bv.bv_offset = off % PAGE_SIZE;
  55         WARN_ON(wlen == 0);
  56         iov_iter_bvec(&iter, WRITE, &bv, 1, wlen);
  57
  58         ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
  59             len, wr, NULL, NULL);
  60         if (ret < 0) {
  61                 SetPageError(page);
  62                 mapping_set_error(page->mapping, ret);
  63         } else {
  64                 ret = 0;
  65         }
  66         kfree(detach_page_private(page));
  67         return ret;
  68 }
  69
  70 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
  71 {
  72         int ret;
  73         ret = orangefs_writepage_locked(page, wbc);
  74         unlock_page(page);
  75         end_page_writeback(page);
  76         return ret;
  77 }
  78
  79 struct orangefs_writepages {
  80         loff_t off;
  81         size_t len;
  82         kuid_t uid;
  83         kgid_t gid;
  84         int maxpages;
  85         int npages;
  86         struct page **pages;
  87         struct bio_vec *bv;
  88 };
  89
  90 static int orangefs_writepages_work(struct orangefs_writepages *ow,
  91     struct writeback_control *wbc)
  92 {
  93         struct inode *inode = ow->pages[0]->mapping->host;
  94         struct orangefs_write_range *wrp, wr;
  95         struct iov_iter iter;
  96         ssize_t ret;
  97         size_t len;
  98         loff_t off;
  99         int i;
 100
 101         len = i_size_read(inode);
 102
 103         for (i = 0; i < ow->npages; i++) {
 104                 set_page_writeback(ow->pages[i]);
 105                 ow->bv[i].bv_page = ow->pages[i];
 106                 ow->bv[i].bv_len = min(page_offset(ow->pages[i]) + PAGE_SIZE,
 107                     ow->off + ow->len) -
 108                     max(ow->off, page_offset(ow->pages[i]));
 109                 if (i == 0)
 110                         ow->bv[i].bv_offset = ow->off -
 111                             page_offset(ow->pages[i]);
 112                 else
 113                         ow->bv[i].bv_offset = 0;
 114         }
 115         iov_iter_bvec(&iter, WRITE, ow->bv, ow->npages, ow->len);
 116
 117         WARN_ON(ow->off >= len);
 118         if (ow->off + ow->len > len)
 119                 ow->len = len - ow->off;
 120
 121         off = ow->off;
 122         wr.uid = ow->uid;
 123         wr.gid = ow->gid;
 124         ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
 125             0, &wr, NULL, NULL);
 126         if (ret < 0) {
 127                 for (i = 0; i < ow->npages; i++) {
 128                         SetPageError(ow->pages[i]);
 129                         mapping_set_error(ow->pages[i]->mapping, ret);
 130                         if (PagePrivate(ow->pages[i])) {
 131                                 wrp = (struct orangefs_write_range *)
 132                                     page_private(ow->pages[i]);
 133                                 ClearPagePrivate(ow->pages[i]);
 134                                 put_page(ow->pages[i]);
 135                                 kfree(wrp);
 136                         }
 137                         end_page_writeback(ow->pages[i]);
 138                         unlock_page(ow->pages[i]);
 139                 }
 140         } else {
 141                 ret = 0;
 142                 for (i = 0; i < ow->npages; i++) {
 143                         if (PagePrivate(ow->pages[i])) {
 144                                 wrp = (struct orangefs_write_range *)
 145                                     page_private(ow->pages[i]);
 146                                 ClearPagePrivate(ow->pages[i]);
 147                                 put_page(ow->pages[i]);
 148                                 kfree(wrp);
 149                         }
 150                         end_page_writeback(ow->pages[i]);
 151                         unlock_page(ow->pages[i]);
 152                 }
 153         }
 154         return ret;
 155 }
 156
 157 static int orangefs_writepages_callback(struct page *page,
 158     struct writeback_control *wbc, void *data)
 159 {
 160         struct orangefs_writepages *ow = data;
 161         struct orangefs_write_range *wr;
 162         int ret;
 163
 164         if (!PagePrivate(page)) {
 165                 unlock_page(page);
 166                 /* It's not private so there's nothing to write, right? */
 167                 printk("writepages_callback not private!\n");
 168                 BUG();
 169                 return 0;
 170         }
 171         wr = (struct orangefs_write_range *)page_private(page);
 172
 173         ret = -1;
 174         if (ow->npages == 0) {
 175                 ow->off = wr->pos;
 176                 ow->len = wr->len;
 177                 ow->uid = wr->uid;
 178                 ow->gid = wr->gid;
 179                 ow->pages[ow->npages++] = page;
 180                 ret = 0;
 181                 goto done;
 182         }
 183         if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
 184                 orangefs_writepages_work(ow, wbc);
 185                 ow->npages = 0;
 186                 ret = -1;
 187                 goto done;
 188         }
 189         if (ow->off + ow->len == wr->pos) {
 190                 ow->len += wr->len;
 191                 ow->pages[ow->npages++] = page;
 192                 ret = 0;
 193                 goto done;
 194         }
 195 done:
 196         if (ret == -1) {
 197                 if (ow->npages) {
 198                         orangefs_writepages_work(ow, wbc);
 199                         ow->npages = 0;
 200                 }
 201                 ret = orangefs_writepage_locked(page, wbc);
 202                 mapping_set_error(page->mapping, ret);
 203                 unlock_page(page);
 204                 end_page_writeback(page);
 205         } else {
 206                 if (ow->npages == ow->maxpages) {
 207                         orangefs_writepages_work(ow, wbc);
 208                         ow->npages = 0;
 209                 }
 210         }
 211         return ret;
 212 }
 213
 214 static int orangefs_writepages(struct address_space *mapping,
 215     struct writeback_control *wbc)
 216 {
 217         struct orangefs_writepages *ow;
 218         struct blk_plug plug;
 219         int ret;
 220         ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
 221         if (!ow)
 222                 return -ENOMEM;
 223         ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
 224         ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
 225         if (!ow->pages) {
 226                 kfree(ow);
 227                 return -ENOMEM;
 228         }
 229         ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
 230         if (!ow->bv) {
 231                 kfree(ow->pages);
 232                 kfree(ow);
 233                 return -ENOMEM;
 234         }
 235         blk_start_plug(&plug);
 236         ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
 237         if (ow->npages)
 238                 ret = orangefs_writepages_work(ow, wbc);
 239         blk_finish_plug(&plug);
 240         kfree(ow->pages);
 241         kfree(ow->bv);
 242         kfree(ow);
 243         return ret;
 244 }
 245
 246 static int orangefs_launder_folio(struct folio *);
 247
 248 static void orangefs_readahead(struct readahead_control *rac)
 249 {
 250         loff_t offset;
 251         struct iov_iter iter;
 252         struct inode *inode = rac->mapping->host;
 253         struct xarray *i_pages;
 254         struct page *page;
 255         loff_t new_start = readahead_pos(rac);
 256         int ret;
 257         size_t new_len = 0;
 258
 259         loff_t bytes_remaining = inode->i_size - readahead_pos(rac);
 260         loff_t pages_remaining = bytes_remaining / PAGE_SIZE;
 261
 262         if (pages_remaining >= 1024)
 263                 new_len = 4194304;
 264         else if (pages_remaining > readahead_count(rac))
 265                 new_len = bytes_remaining;
 266
 267         if (new_len)
 268                 readahead_expand(rac, new_start, new_len);
 269
 270         offset = readahead_pos(rac);
 271         i_pages = &rac->mapping->i_pages;
 272
 273         iov_iter_xarray(&iter, READ, i_pages, offset, readahead_length(rac));
 274
 275         /* read in the pages. */
 276         if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode,
 277                         &offset, &iter, readahead_length(rac),
 278                         inode->i_size, NULL, NULL, rac->file)) < 0)
 279                 gossip_debug(GOSSIP_FILE_DEBUG,
 280                         "%s: wait_for_direct_io failed. \n", __func__);
 281         else
 282                 ret = 0;
 283
 284         /* clean up. */
 285         while ((page = readahead_page(rac))) {
 286                 page_endio(page, false, ret);
 287                 put_page(page);
 288         }
 289 }
 290
 291 static int orangefs_read_folio(struct file *file, struct folio *folio)
 292 {
 293         struct inode *inode = folio->mapping->host;
 294         struct iov_iter iter;
 295         struct bio_vec bv;
 296         ssize_t ret;
 297         loff_t off; /* offset of this folio in the file */
 298
 299         if (folio_test_dirty(folio))
 300                 orangefs_launder_folio(folio);
 301
 302         off = folio_pos(folio);
 303         bv.bv_page = &folio->page;
 304         bv.bv_len = folio_size(folio);
 305         bv.bv_offset = 0;
 306         iov_iter_bvec(&iter, READ, &bv, 1, folio_size(folio));
 307
 308         ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
 309                         folio_size(folio), inode->i_size, NULL, NULL, file);
 310         /* this will only zero remaining unread portions of the page data */
 311         iov_iter_zero(~0U, &iter);
 312         /* takes care of potential aliasing */
 313         flush_dcache_folio(folio);
 314         if (ret < 0) {
 315                 folio_set_error(folio);
 316         } else {
 317                 folio_mark_uptodate(folio);
 318                 if (folio_test_error(folio))
 319                         folio_clear_error(folio);
 320                 ret = 0;
 321         }
 322         /* unlock the folio after the ->read_folio() routine completes */
 323         folio_unlock(folio);
 324         return ret;
 325 }
 326
 327 static int orangefs_write_begin(struct file *file,
 328                 struct address_space *mapping, loff_t pos, unsigned len,
 329                 struct page **pagep, void **fsdata)
 330 {
 331         struct orangefs_write_range *wr;
 332         struct folio *folio;
 333         struct page *page;
 334         pgoff_t index;
 335         int ret;
 336
 337         index = pos >> PAGE_SHIFT;
 338
 339         page = grab_cache_page_write_begin(mapping, index);
 340         if (!page)
 341                 return -ENOMEM;
 342
 343         *pagep = page;
 344         folio = page_folio(page);
 345
 346         if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 347                 /*
 348                  * Should be impossible.  If it happens, launder the page
 349                  * since we don't know what's dirty.  This will WARN in
 350                  * orangefs_writepage_locked.
 351                  */
 352                 ret = orangefs_launder_folio(folio);
 353                 if (ret)
 354                         return ret;
 355         }
 356         if (folio_test_private(folio)) {
 357                 struct orangefs_write_range *wr;
 358                 wr = folio_get_private(folio);
 359                 if (wr->pos + wr->len == pos &&
 360                     uid_eq(wr->uid, current_fsuid()) &&
 361                     gid_eq(wr->gid, current_fsgid())) {
 362                         wr->len += len;
 363                         goto okay;
 364                 } else {
 365                         ret = orangefs_launder_folio(folio);
 366                         if (ret)
 367                                 return ret;
 368                 }
 369         }
 370
 371         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 372         if (!wr)
 373                 return -ENOMEM;
 374
 375         wr->pos = pos;
 376         wr->len = len;
 377         wr->uid = current_fsuid();
 378         wr->gid = current_fsgid();
 379         folio_attach_private(folio, wr);
 380 okay:
 381         return 0;
 382 }
 383
 384 static int orangefs_write_end(struct file *file, struct address_space *mapping,
 385     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
 386 {
 387         struct inode *inode = page->mapping->host;
 388         loff_t last_pos = pos + copied;
 389
 390         /*
 391          * No need to use i_size_read() here, the i_size
 392          * cannot change under us because we hold the i_mutex.
 393          */
 394         if (last_pos > inode->i_size)
 395                 i_size_write(inode, last_pos);
 396
 397         /* zero the stale part of the page if we did a short copy */
 398         if (!PageUptodate(page)) {
 399                 unsigned from = pos & (PAGE_SIZE - 1);
 400                 if (copied < len) {
 401                         zero_user(page, from + copied, len - copied);
 402                 }
 403                 /* Set fully written pages uptodate. */
 404                 if (pos == page_offset(page) &&
 405                     (len == PAGE_SIZE || pos + len == inode->i_size)) {
 406                         zero_user_segment(page, from + copied, PAGE_SIZE);
 407                         SetPageUptodate(page);
 408                 }
 409         }
 410
 411         set_page_dirty(page);
 412         unlock_page(page);
 413         put_page(page);
 414
 415         mark_inode_dirty_sync(file_inode(file));
 416         return copied;
 417 }
 418
 419 static void orangefs_invalidate_folio(struct folio *folio,
 420                                  size_t offset, size_t length)
 421 {
 422         struct orangefs_write_range *wr = folio_get_private(folio);
 423
 424         if (offset == 0 && length == PAGE_SIZE) {
 425                 kfree(folio_detach_private(folio));
 426                 return;
 427         /* write range entirely within invalidate range (or equal) */
 428         } else if (folio_pos(folio) + offset <= wr->pos &&
 429             wr->pos + wr->len <= folio_pos(folio) + offset + length) {
 430                 kfree(folio_detach_private(folio));
 431                 /* XXX is this right? only caller in fs */
 432                 folio_cancel_dirty(folio);
 433                 return;
 434         /* invalidate range chops off end of write range */
 435         } else if (wr->pos < folio_pos(folio) + offset &&
 436             wr->pos + wr->len <= folio_pos(folio) + offset + length &&
 437              folio_pos(folio) + offset < wr->pos + wr->len) {
 438                 size_t x;
 439                 x = wr->pos + wr->len - (folio_pos(folio) + offset);
 440                 WARN_ON(x > wr->len);
 441                 wr->len -= x;
 442                 wr->uid = current_fsuid();
 443                 wr->gid = current_fsgid();
 444         /* invalidate range chops off beginning of write range */
 445         } else if (folio_pos(folio) + offset <= wr->pos &&
 446             folio_pos(folio) + offset + length < wr->pos + wr->len &&
 447             wr->pos < folio_pos(folio) + offset + length) {
 448                 size_t x;
 449                 x = folio_pos(folio) + offset + length - wr->pos;
 450                 WARN_ON(x > wr->len);
 451                 wr->pos += x;
 452                 wr->len -= x;
 453                 wr->uid = current_fsuid();
 454                 wr->gid = current_fsgid();
 455         /* invalidate range entirely within write range (punch hole) */
 456         } else if (wr->pos < folio_pos(folio) + offset &&
 457             folio_pos(folio) + offset + length < wr->pos + wr->len) {
 458                 /* XXX what do we do here... should not WARN_ON */
 459                 WARN_ON(1);
 460                 /* punch hole */
 461                 /*
 462                  * should we just ignore this and write it out anyway?
 463                  * it hardly makes sense
 464                  */
 465                 return;
 466         /* non-overlapping ranges */
 467         } else {
 468                 /* WARN if they do overlap */
 469                 if (!((folio_pos(folio) + offset + length <= wr->pos) ^
 470                     (wr->pos + wr->len <= folio_pos(folio) + offset))) {
 471                         WARN_ON(1);
 472                         printk("invalidate range offset %llu length %zu\n",
 473                             folio_pos(folio) + offset, length);
 474                         printk("write range offset %llu length %zu\n",
 475                             wr->pos, wr->len);
 476                 }
 477                 return;
 478         }
 479
 480         /*
 481          * Above there are returns where wr is freed or where we WARN.
 482          * Thus the following runs if wr was modified above.
 483          */
 484
 485         orangefs_launder_folio(folio);
 486 }
 487
 488 static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
 489 {
 490         return !folio_test_private(folio);
 491 }
 492
 493 static void orangefs_free_folio(struct folio *folio)
 494 {
 495         kfree(folio_detach_private(folio));
 496 }
 497
 498 static int orangefs_launder_folio(struct folio *folio)
 499 {
 500         int r = 0;
 501         struct writeback_control wbc = {
 502                 .sync_mode = WB_SYNC_ALL,
 503                 .nr_to_write = 0,
 504         };
 505         folio_wait_writeback(folio);
 506         if (folio_clear_dirty_for_io(folio)) {
 507                 r = orangefs_writepage_locked(&folio->page, &wbc);
 508                 folio_end_writeback(folio);
 509         }
 510         return r;
 511 }
 512
 513 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
 514                                   struct iov_iter *iter)
 515 {
 516         /*
 517          * Comment from original do_readv_writev:
 518          * Common entry point for read/write/readv/writev
 519          * This function will dispatch it to either the direct I/O
 520          * or buffered I/O path depending on the mount options and/or
 521          * augmented/extended metadata attached to the file.
 522          * Note: File extended attributes override any mount options.
 523          */
 524         struct file *file = iocb->ki_filp;
 525         loff_t pos = iocb->ki_pos;
 526         enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
 527             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
 528         loff_t *offset = &pos;
 529         struct inode *inode = file->f_mapping->host;
 530         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 531         struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
 532         size_t count = iov_iter_count(iter);
 533         ssize_t total_count = 0;
 534         ssize_t ret = -EINVAL;
 535         int i = 0;
 536
 537         gossip_debug(GOSSIP_FILE_DEBUG,
 538                 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
 539                 __func__,
 540                 handle,
 541                 (int)count);
 542
 543         if (type == ORANGEFS_IO_WRITE) {
 544                 gossip_debug(GOSSIP_FILE_DEBUG,
 545                              "%s(%pU): proceeding with offset : %llu, "
 546                              "size %d\n",
 547                              __func__,
 548                              handle,
 549                              llu(*offset),
 550                              (int)count);
 551         }
 552
 553         if (count == 0) {
 554                 ret = 0;
 555                 goto out;
 556         }
 557
 558         while (iov_iter_count(iter)) {
 559                 size_t each_count = iov_iter_count(iter);
 560                 size_t amt_complete;
 561                 i++;
 562
 563                 /* how much to transfer in this loop iteration */
 564                 if (each_count > orangefs_bufmap_size_query())
 565                         each_count = orangefs_bufmap_size_query();
 566
 567                 gossip_debug(GOSSIP_FILE_DEBUG,
 568                              "%s(%pU): size of each_count(%d)\n",
 569                              __func__,
 570                              handle,
 571                              (int)each_count);
 572                 gossip_debug(GOSSIP_FILE_DEBUG,
 573                              "%s(%pU): BEFORE wait_for_io: offset is %d\n",
 574                              __func__,
 575                              handle,
 576                              (int)*offset);
 577
 578                 ret = wait_for_direct_io(type, inode, offset, iter,
 579                                 each_count, 0, NULL, NULL, file);
 580                 gossip_debug(GOSSIP_FILE_DEBUG,
 581                              "%s(%pU): return from wait_for_io:%d\n",
 582                              __func__,
 583                              handle,
 584                              (int)ret);
 585
 586                 if (ret < 0)
 587                         goto out;
 588
 589                 *offset += ret;
 590                 total_count += ret;
 591                 amt_complete = ret;
 592
 593                 gossip_debug(GOSSIP_FILE_DEBUG,
 594                              "%s(%pU): AFTER wait_for_io: offset is %d\n",
 595                              __func__,
 596                              handle,
 597                              (int)*offset);
 598
 599                 /*
 600                  * if we got a short I/O operations,
 601                  * fall out and return what we got so far
 602                  */
 603                 if (amt_complete < each_count)
 604                         break;
 605         } /*end while */
 606
 607 out:
 608         if (total_count > 0)
 609                 ret = total_count;
 610         if (ret > 0) {
 611                 if (type == ORANGEFS_IO_READ) {
 612                         file_accessed(file);
 613                 } else {
 614                         file_update_time(file);
 615                         if (*offset > i_size_read(inode))
 616                                 i_size_write(inode, *offset);
 617                 }
 618         }
 619
 620         gossip_debug(GOSSIP_FILE_DEBUG,
 621                      "%s(%pU): Value(%d) returned.\n",
 622                      __func__,
 623                      handle,
 624                      (int)ret);
 625
 626         return ret;
 627 }
 628
 629 /** ORANGEFS2 implementation of address space operations */
 630 static const struct address_space_operations orangefs_address_operations = {
 631         .writepage = orangefs_writepage,
 632         .readahead = orangefs_readahead,
 633         .read_folio = orangefs_read_folio,
 634         .writepages = orangefs_writepages,
 635         .dirty_folio = filemap_dirty_folio,
 636         .write_begin = orangefs_write_begin,
 637         .write_end = orangefs_write_end,
 638         .invalidate_folio = orangefs_invalidate_folio,
 639         .release_folio = orangefs_release_folio,
 640         .free_folio = orangefs_free_folio,
 641         .launder_folio = orangefs_launder_folio,
 642         .direct_IO = orangefs_direct_IO,
 643 };
 644
 645 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
 646 {
 647         struct folio *folio = page_folio(vmf->page);
 648         struct inode *inode = file_inode(vmf->vma->vm_file);
 649         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 650         unsigned long *bitlock = &orangefs_inode->bitlock;
 651         vm_fault_t ret;
 652         struct orangefs_write_range *wr;
 653
 654         sb_start_pagefault(inode->i_sb);
 655
 656         if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
 657                 ret = VM_FAULT_RETRY;
 658                 goto out;
 659         }
 660
 661         folio_lock(folio);
 662         if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 663                 /*
 664                  * Should be impossible.  If it happens, launder the folio
 665                  * since we don't know what's dirty.  This will WARN in
 666                  * orangefs_writepage_locked.
 667                  */
 668                 if (orangefs_launder_folio(folio)) {
 669                         ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 670                         goto out;
 671                 }
 672         }
 673         if (folio_test_private(folio)) {
 674                 wr = folio_get_private(folio);
 675                 if (uid_eq(wr->uid, current_fsuid()) &&
 676                     gid_eq(wr->gid, current_fsgid())) {
 677                         wr->pos = page_offset(vmf->page);
 678                         wr->len = PAGE_SIZE;
 679                         goto okay;
 680                 } else {
 681                         if (orangefs_launder_folio(folio)) {
 682                                 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 683                                 goto out;
 684                         }
 685                 }
 686         }
 687         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 688         if (!wr) {
 689                 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 690                 goto out;
 691         }
 692         wr->pos = page_offset(vmf->page);
 693         wr->len = PAGE_SIZE;
 694         wr->uid = current_fsuid();
 695         wr->gid = current_fsgid();
 696         folio_attach_private(folio, wr);
 697 okay:
 698
 699         file_update_time(vmf->vma->vm_file);
 700         if (folio->mapping != inode->i_mapping) {
 701                 folio_unlock(folio);
 702                 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
 703                 goto out;
 704         }
 705
 706         /*
 707          * We mark the folio dirty already here so that when freeze is in
 708          * progress, we are guaranteed that writeback during freezing will
 709          * see the dirty folio and writeprotect it again.
 710          */
 711         folio_mark_dirty(folio);
 712         folio_wait_stable(folio);
 713         ret = VM_FAULT_LOCKED;
 714 out:
 715         sb_end_pagefault(inode->i_sb);
 716         return ret;
 717 }
 718
 719 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 720 {
 721         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 722         struct orangefs_kernel_op_s *new_op;
 723         loff_t orig_size;
 724         int ret = -EINVAL;
 725
 726         gossip_debug(GOSSIP_INODE_DEBUG,
 727                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 728                      __func__,
 729                      get_khandle_from_ino(inode),
 730                      &orangefs_inode->refn.khandle,
 731                      orangefs_inode->refn.fs_id,
 732                      iattr->ia_size);
 733
 734         /* Ensure that we have a up to date size, so we know if it changed. */
 735         ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
 736         if (ret == -ESTALE)
 737                 ret = -EIO;
 738         if (ret) {
 739                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 740                     __func__, ret);
 741                 return ret;
 742         }
 743         orig_size = i_size_read(inode);
 744
 745         /* This is truncate_setsize in a different order. */
 746         truncate_pagecache(inode, iattr->ia_size);
 747         i_size_write(inode, iattr->ia_size);
 748         if (iattr->ia_size > orig_size)
 749                 pagecache_isize_extended(inode, orig_size, iattr->ia_size);
 750
 751         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 752         if (!new_op)
 753                 return -ENOMEM;
 754
 755         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 756         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 757
 758         ret = service_operation(new_op,
 759                 __func__,
 760                 get_interruptible_flag(inode));
 761
 762         /*
 763          * the truncate has no downcall members to retrieve, but
 764          * the status value tells us if it went through ok or not
 765          */
 766         gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
 767
 768         op_release(new_op);
 769
 770         if (ret != 0)
 771                 return ret;
 772
 773         if (orig_size != i_size_read(inode))
 774                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 775
 776         return ret;
 777 }
 778
 779 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
 780 {
 781         int ret;
 782
 783         if (iattr->ia_valid & ATTR_MODE) {
 784                 if (iattr->ia_mode & (S_ISVTX)) {
 785                         if (is_root_handle(inode)) {
 786                                 /*
 787                                  * allow sticky bit to be set on root (since
 788                                  * it shows up that way by default anyhow),
 789                                  * but don't show it to the server
 790                                  */
 791                                 iattr->ia_mode -= S_ISVTX;
 792                         } else {
 793                                 gossip_debug(GOSSIP_UTILS_DEBUG,
 794                                              "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
 795                                 ret = -EINVAL;
 796                                 goto out;
 797                         }
 798                 }
 799                 if (iattr->ia_mode & (S_ISUID)) {
 800                         gossip_debug(GOSSIP_UTILS_DEBUG,
 801                                      "Attempting to set setuid bit (not supported); returning EINVAL.\n");
 802                         ret = -EINVAL;
 803                         goto out;
 804                 }
 805         }
 806
 807         if (iattr->ia_valid & ATTR_SIZE) {
 808                 ret = orangefs_setattr_size(inode, iattr);
 809                 if (ret)
 810                         goto out;
 811         }
 812
 813 again:
 814         spin_lock(&inode->i_lock);
 815         if (ORANGEFS_I(inode)->attr_valid) {
 816                 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
 817                     gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
 818                         ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 819                 } else {
 820                         spin_unlock(&inode->i_lock);
 821                         write_inode_now(inode, 1);
 822                         goto again;
 823                 }
 824         } else {
 825                 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 826                 ORANGEFS_I(inode)->attr_uid = current_fsuid();
 827                 ORANGEFS_I(inode)->attr_gid = current_fsgid();
 828         }
 829         setattr_copy(&init_user_ns, inode, iattr);
 830         spin_unlock(&inode->i_lock);
 831         mark_inode_dirty(inode);
 832
 833         if (iattr->ia_valid & ATTR_MODE)
 834                 /* change mod on a file that has ACLs */
 835                 ret = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
 836
 837         ret = 0;
 838 out:
 839         return ret;
 840 }
 841
 842 /*
 843  * Change attributes of an object referenced by dentry.
 844  */
 845 int orangefs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 846                      struct iattr *iattr)
 847 {
 848         int ret;
 849         gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
 850             dentry);
 851         ret = setattr_prepare(&init_user_ns, dentry, iattr);
 852         if (ret)
 853                 goto out;
 854         ret = __orangefs_setattr(d_inode(dentry), iattr);
 855         sync_inode_metadata(d_inode(dentry), 1);
 856 out:
 857         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
 858             ret);
 859         return ret;
 860 }
 861
 862 /*
 863  * Obtain attributes of an object given a dentry
 864  */
 865 int orangefs_getattr(struct user_namespace *mnt_userns, const struct path *path,
 866                      struct kstat *stat, u32 request_mask, unsigned int flags)
 867 {
 868         int ret;
 869         struct inode *inode = path->dentry->d_inode;
 870
 871         gossip_debug(GOSSIP_INODE_DEBUG,
 872                      "orangefs_getattr: called on %pd mask %u\n",
 873                      path->dentry, request_mask);
 874
 875         ret = orangefs_inode_getattr(inode,
 876             request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
 877         if (ret == 0) {
 878                 generic_fillattr(&init_user_ns, inode, stat);
 879
 880                 /* override block size reported to stat */
 881                 if (!(request_mask & STATX_SIZE))
 882                         stat->result_mask &= ~STATX_SIZE;
 883
 884                 generic_fill_statx_attr(inode, stat);
 885         }
 886         return ret;
 887 }
 888
 889 int orangefs_permission(struct user_namespace *mnt_userns,
 890                         struct inode *inode, int mask)
 891 {
 892         int ret;
 893
 894         if (mask & MAY_NOT_BLOCK)
 895                 return -ECHILD;
 896
 897         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 898
 899         /* Make sure the permission (and other common attrs) are up to date. */
 900         ret = orangefs_inode_getattr(inode, 0);
 901         if (ret < 0)
 902                 return ret;
 903
 904         return generic_permission(&init_user_ns, inode, mask);
 905 }
 906
 907 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
 908 {
 909         struct iattr iattr;
 910         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
 911             get_khandle_from_ino(inode));
 912         generic_update_time(inode, time, flags);
 913         memset(&iattr, 0, sizeof iattr);
 914         if (flags & S_ATIME)
 915                 iattr.ia_valid |= ATTR_ATIME;
 916         if (flags & S_CTIME)
 917                 iattr.ia_valid |= ATTR_CTIME;
 918         if (flags & S_MTIME)
 919                 iattr.ia_valid |= ATTR_MTIME;
 920         return __orangefs_setattr(inode, &iattr);
 921 }
 922
 923 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 924 {
 925         u64 val = 0;
 926         int ret;
 927
 928         gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 929                      dentry);
 930
 931         ret = orangefs_inode_getxattr(d_inode(dentry),
 932                                       "user.pvfs2.meta_hint",
 933                                       &val, sizeof(val));
 934         if (ret < 0 && ret != -ENODATA)
 935                 return ret;
 936
 937         gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 938
 939         fileattr_fill_flags(fa, val);
 940         return 0;
 941 }
 942
 943 static int orangefs_fileattr_set(struct user_namespace *mnt_userns,
 944                                  struct dentry *dentry, struct fileattr *fa)
 945 {
 946         u64 val = 0;
 947
 948         gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 949                      dentry);
 950         /*
 951          * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
 952          * turned on for a file. The user is not allowed to turn on this bit,
 953          * but the bit is present if the user first gets the flags and then
 954          * updates the flags with some new settings. So, we ignore it in the
 955          * following edit. bligon.
 956          */
 957         if (fileattr_has_fsx(fa) ||
 958             (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
 959                 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
 960                            __func__);
 961                 return -EOPNOTSUPP;
 962         }
 963         val = fa->flags;
 964         gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 965         return orangefs_inode_setxattr(d_inode(dentry),
 966                                        "user.pvfs2.meta_hint",
 967                                        &val, sizeof(val), 0);
 968 }
 969
 970 /* ORANGEFS2 implementation of VFS inode operations for files */
 971 static const struct inode_operations orangefs_file_inode_operations = {
 972         .get_acl = orangefs_get_acl,
 973         .set_acl = orangefs_set_acl,
 974         .setattr = orangefs_setattr,
 975         .getattr = orangefs_getattr,
 976         .listxattr = orangefs_listxattr,
 977         .permission = orangefs_permission,
 978         .update_time = orangefs_update_time,
 979         .fileattr_get = orangefs_fileattr_get,
 980         .fileattr_set = orangefs_fileattr_set,
 981 };
 982
 983 static int orangefs_init_iops(struct inode *inode)
 984 {
 985         inode->i_mapping->a_ops = &orangefs_address_operations;
 986
 987         switch (inode->i_mode & S_IFMT) {
 988         case S_IFREG:
 989                 inode->i_op = &orangefs_file_inode_operations;
 990                 inode->i_fop = &orangefs_file_operations;
 991                 break;
 992         case S_IFLNK:
 993                 inode->i_op = &orangefs_symlink_inode_operations;
 994                 break;
 995         case S_IFDIR:
 996                 inode->i_op = &orangefs_dir_inode_operations;
 997                 inode->i_fop = &orangefs_dir_operations;
 998                 break;
 999         default:
1000                 gossip_debug(GOSSIP_INODE_DEBUG,
1001                              "%s: unsupported mode\n",
1002                              __func__);
1003                 return -EINVAL;
1004         }
1005
1006         return 0;
1007 }
1008
1009 /*
1010  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1011  * a ino_t type that will be used as a hash-index from where the handle will
1012  * be searched for in the VFS hash table of inodes.
1013  */
1014 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1015 {
1016         if (!ref)
1017                 return 0;
1018         return orangefs_khandle_to_ino(&(ref->khandle));
1019 }
1020
1021 /*
1022  * Called to set up an inode from iget5_locked.
1023  */
1024 static int orangefs_set_inode(struct inode *inode, void *data)
1025 {
1026         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1027         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1028         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1029         ORANGEFS_I(inode)->attr_valid = 0;
1030         hash_init(ORANGEFS_I(inode)->xattr_cache);
1031         ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1032         ORANGEFS_I(inode)->bitlock = 0;
1033         return 0;
1034 }
1035
1036 /*
1037  * Called to determine if handles match.
1038  */
1039 static int orangefs_test_inode(struct inode *inode, void *data)
1040 {
1041         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1042         struct orangefs_inode_s *orangefs_inode = NULL;
1043
1044         orangefs_inode = ORANGEFS_I(inode);
1045         /* test handles and fs_ids... */
1046         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1047                                 &(ref->khandle)) &&
1048                         orangefs_inode->refn.fs_id == ref->fs_id);
1049 }
1050
1051 /*
1052  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1053  * file handle.
1054  *
1055  * @sb: the file system super block instance.
1056  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1057  */
1058 struct inode *orangefs_iget(struct super_block *sb,
1059                 struct orangefs_object_kref *ref)
1060 {
1061         struct inode *inode = NULL;
1062         unsigned long hash;
1063         int error;
1064
1065         hash = orangefs_handle_hash(ref);
1066         inode = iget5_locked(sb,
1067                         hash,
1068                         orangefs_test_inode,
1069                         orangefs_set_inode,
1070                         ref);
1071
1072         if (!inode)
1073                 return ERR_PTR(-ENOMEM);
1074
1075         if (!(inode->i_state & I_NEW))
1076                 return inode;
1077
1078         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1079         if (error) {
1080                 iget_failed(inode);
1081                 return ERR_PTR(error);
1082         }
1083
1084         inode->i_ino = hash;    /* needed for stat etc */
1085         orangefs_init_iops(inode);
1086         unlock_new_inode(inode);
1087
1088         gossip_debug(GOSSIP_INODE_DEBUG,
1089                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1090                      &ref->khandle,
1091                      ref->fs_id,
1092                      hash,
1093                      inode->i_ino);
1094
1095         return inode;
1096 }
1097
1098 /*
1099  * Allocate an inode for a newly created file and insert it into the inode hash.
1100  */
1101 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1102                 int mode, dev_t dev, struct orangefs_object_kref *ref)
1103 {
1104         unsigned long hash = orangefs_handle_hash(ref);
1105         struct inode *inode;
1106         int error;
1107
1108         gossip_debug(GOSSIP_INODE_DEBUG,
1109                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1110                      __func__,
1111                      sb,
1112                      MAJOR(dev),
1113                      MINOR(dev),
1114                      mode);
1115
1116         inode = new_inode(sb);
1117         if (!inode)
1118                 return ERR_PTR(-ENOMEM);
1119
1120         orangefs_set_inode(inode, ref);
1121         inode->i_ino = hash;    /* needed for stat etc */
1122
1123         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1124         if (error)
1125                 goto out_iput;
1126
1127         orangefs_init_iops(inode);
1128         inode->i_rdev = dev;
1129
1130         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1131         if (error < 0)
1132                 goto out_iput;
1133
1134         gossip_debug(GOSSIP_INODE_DEBUG,
1135                      "Initializing ACL's for inode %pU\n",
1136                      get_khandle_from_ino(inode));
1137         orangefs_init_acl(inode, dir);
1138         return inode;
1139
1140 out_iput:
1141         iput(inode);
1142         return ERR_PTR(error);
1143 }