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_readpage(struct file *file, struct page *page)
 292 {
 293         struct folio *folio = page_folio(page);
 294         struct inode *inode = page->mapping->host;
 295         struct iov_iter iter;
 296         struct bio_vec bv;
 297         ssize_t ret;
 298         loff_t off; /* offset into this page */
 299
 300         if (folio_test_dirty(folio))
 301                 orangefs_launder_folio(folio);
 302
 303         off = page_offset(page);
 304         bv.bv_page = page;
 305         bv.bv_len = PAGE_SIZE;
 306         bv.bv_offset = 0;
 307         iov_iter_bvec(&iter, READ, &bv, 1, PAGE_SIZE);
 308
 309         ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
 310             PAGE_SIZE, inode->i_size, NULL, NULL, file);
 311         /* this will only zero remaining unread portions of the page data */
 312         iov_iter_zero(~0U, &iter);
 313         /* takes care of potential aliasing */
 314         flush_dcache_page(page);
 315         if (ret < 0) {
 316                 SetPageError(page);
 317         } else {
 318                 SetPageUptodate(page);
 319                 if (PageError(page))
 320                         ClearPageError(page);
 321                 ret = 0;
 322         }
 323         /* unlock the page after the ->readpage() routine completes */
 324         unlock_page(page);
 325         return ret;
 326 }
 327
 328 static int orangefs_write_begin(struct file *file,
 329     struct address_space *mapping,
 330     loff_t pos, unsigned len, unsigned flags, struct page **pagep,
 331     void **fsdata)
 332 {
 333         struct orangefs_write_range *wr;
 334         struct folio *folio;
 335         struct page *page;
 336         pgoff_t index;
 337         int ret;
 338
 339         index = pos >> PAGE_SHIFT;
 340
 341         page = grab_cache_page_write_begin(mapping, index, flags);
 342         if (!page)
 343                 return -ENOMEM;
 344
 345         *pagep = page;
 346         folio = page_folio(page);
 347
 348         if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 349                 /*
 350                  * Should be impossible.  If it happens, launder the page
 351                  * since we don't know what's dirty.  This will WARN in
 352                  * orangefs_writepage_locked.
 353                  */
 354                 ret = orangefs_launder_folio(folio);
 355                 if (ret)
 356                         return ret;
 357         }
 358         if (folio_test_private(folio)) {
 359                 struct orangefs_write_range *wr;
 360                 wr = folio_get_private(folio);
 361                 if (wr->pos + wr->len == pos &&
 362                     uid_eq(wr->uid, current_fsuid()) &&
 363                     gid_eq(wr->gid, current_fsgid())) {
 364                         wr->len += len;
 365                         goto okay;
 366                 } else {
 367                         ret = orangefs_launder_folio(folio);
 368                         if (ret)
 369                                 return ret;
 370                 }
 371         }
 372
 373         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 374         if (!wr)
 375                 return -ENOMEM;
 376
 377         wr->pos = pos;
 378         wr->len = len;
 379         wr->uid = current_fsuid();
 380         wr->gid = current_fsgid();
 381         folio_attach_private(folio, wr);
 382 okay:
 383         return 0;
 384 }
 385
 386 static int orangefs_write_end(struct file *file, struct address_space *mapping,
 387     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
 388 {
 389         struct inode *inode = page->mapping->host;
 390         loff_t last_pos = pos + copied;
 391
 392         /*
 393          * No need to use i_size_read() here, the i_size
 394          * cannot change under us because we hold the i_mutex.
 395          */
 396         if (last_pos > inode->i_size)
 397                 i_size_write(inode, last_pos);
 398
 399         /* zero the stale part of the page if we did a short copy */
 400         if (!PageUptodate(page)) {
 401                 unsigned from = pos & (PAGE_SIZE - 1);
 402                 if (copied < len) {
 403                         zero_user(page, from + copied, len - copied);
 404                 }
 405                 /* Set fully written pages uptodate. */
 406                 if (pos == page_offset(page) &&
 407                     (len == PAGE_SIZE || pos + len == inode->i_size)) {
 408                         zero_user_segment(page, from + copied, PAGE_SIZE);
 409                         SetPageUptodate(page);
 410                 }
 411         }
 412
 413         set_page_dirty(page);
 414         unlock_page(page);
 415         put_page(page);
 416
 417         mark_inode_dirty_sync(file_inode(file));
 418         return copied;
 419 }
 420
 421 static void orangefs_invalidate_folio(struct folio *folio,
 422                                  size_t offset, size_t length)
 423 {
 424         struct orangefs_write_range *wr = folio_get_private(folio);
 425
 426         if (offset == 0 && length == PAGE_SIZE) {
 427                 kfree(folio_detach_private(folio));
 428                 return;
 429         /* write range entirely within invalidate range (or equal) */
 430         } else if (folio_pos(folio) + offset <= wr->pos &&
 431             wr->pos + wr->len <= folio_pos(folio) + offset + length) {
 432                 kfree(folio_detach_private(folio));
 433                 /* XXX is this right? only caller in fs */
 434                 folio_cancel_dirty(folio);
 435                 return;
 436         /* invalidate range chops off end of write range */
 437         } else if (wr->pos < folio_pos(folio) + offset &&
 438             wr->pos + wr->len <= folio_pos(folio) + offset + length &&
 439              folio_pos(folio) + offset < wr->pos + wr->len) {
 440                 size_t x;
 441                 x = wr->pos + wr->len - (folio_pos(folio) + offset);
 442                 WARN_ON(x > wr->len);
 443                 wr->len -= x;
 444                 wr->uid = current_fsuid();
 445                 wr->gid = current_fsgid();
 446         /* invalidate range chops off beginning of write range */
 447         } else if (folio_pos(folio) + offset <= wr->pos &&
 448             folio_pos(folio) + offset + length < wr->pos + wr->len &&
 449             wr->pos < folio_pos(folio) + offset + length) {
 450                 size_t x;
 451                 x = folio_pos(folio) + offset + length - wr->pos;
 452                 WARN_ON(x > wr->len);
 453                 wr->pos += x;
 454                 wr->len -= x;
 455                 wr->uid = current_fsuid();
 456                 wr->gid = current_fsgid();
 457         /* invalidate range entirely within write range (punch hole) */
 458         } else if (wr->pos < folio_pos(folio) + offset &&
 459             folio_pos(folio) + offset + length < wr->pos + wr->len) {
 460                 /* XXX what do we do here... should not WARN_ON */
 461                 WARN_ON(1);
 462                 /* punch hole */
 463                 /*
 464                  * should we just ignore this and write it out anyway?
 465                  * it hardly makes sense
 466                  */
 467                 return;
 468         /* non-overlapping ranges */
 469         } else {
 470                 /* WARN if they do overlap */
 471                 if (!((folio_pos(folio) + offset + length <= wr->pos) ^
 472                     (wr->pos + wr->len <= folio_pos(folio) + offset))) {
 473                         WARN_ON(1);
 474                         printk("invalidate range offset %llu length %zu\n",
 475                             folio_pos(folio) + offset, length);
 476                         printk("write range offset %llu length %zu\n",
 477                             wr->pos, wr->len);
 478                 }
 479                 return;
 480         }
 481
 482         /*
 483          * Above there are returns where wr is freed or where we WARN.
 484          * Thus the following runs if wr was modified above.
 485          */
 486
 487         orangefs_launder_folio(folio);
 488 }
 489
 490 static int orangefs_releasepage(struct page *page, gfp_t foo)
 491 {
 492         return !PagePrivate(page);
 493 }
 494
 495 static void orangefs_freepage(struct page *page)
 496 {
 497         kfree(detach_page_private(page));
 498 }
 499
 500 static int orangefs_launder_folio(struct folio *folio)
 501 {
 502         int r = 0;
 503         struct writeback_control wbc = {
 504                 .sync_mode = WB_SYNC_ALL,
 505                 .nr_to_write = 0,
 506         };
 507         folio_wait_writeback(folio);
 508         if (folio_clear_dirty_for_io(folio)) {
 509                 r = orangefs_writepage_locked(&folio->page, &wbc);
 510                 folio_end_writeback(folio);
 511         }
 512         return r;
 513 }
 514
 515 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
 516                                   struct iov_iter *iter)
 517 {
 518         /*
 519          * Comment from original do_readv_writev:
 520          * Common entry point for read/write/readv/writev
 521          * This function will dispatch it to either the direct I/O
 522          * or buffered I/O path depending on the mount options and/or
 523          * augmented/extended metadata attached to the file.
 524          * Note: File extended attributes override any mount options.
 525          */
 526         struct file *file = iocb->ki_filp;
 527         loff_t pos = iocb->ki_pos;
 528         enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
 529             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
 530         loff_t *offset = &pos;
 531         struct inode *inode = file->f_mapping->host;
 532         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 533         struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
 534         size_t count = iov_iter_count(iter);
 535         ssize_t total_count = 0;
 536         ssize_t ret = -EINVAL;
 537         int i = 0;
 538
 539         gossip_debug(GOSSIP_FILE_DEBUG,
 540                 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
 541                 __func__,
 542                 handle,
 543                 (int)count);
 544
 545         if (type == ORANGEFS_IO_WRITE) {
 546                 gossip_debug(GOSSIP_FILE_DEBUG,
 547                              "%s(%pU): proceeding with offset : %llu, "
 548                              "size %d\n",
 549                              __func__,
 550                              handle,
 551                              llu(*offset),
 552                              (int)count);
 553         }
 554
 555         if (count == 0) {
 556                 ret = 0;
 557                 goto out;
 558         }
 559
 560         while (iov_iter_count(iter)) {
 561                 size_t each_count = iov_iter_count(iter);
 562                 size_t amt_complete;
 563                 i++;
 564
 565                 /* how much to transfer in this loop iteration */
 566                 if (each_count > orangefs_bufmap_size_query())
 567                         each_count = orangefs_bufmap_size_query();
 568
 569                 gossip_debug(GOSSIP_FILE_DEBUG,
 570                              "%s(%pU): size of each_count(%d)\n",
 571                              __func__,
 572                              handle,
 573                              (int)each_count);
 574                 gossip_debug(GOSSIP_FILE_DEBUG,
 575                              "%s(%pU): BEFORE wait_for_io: offset is %d\n",
 576                              __func__,
 577                              handle,
 578                              (int)*offset);
 579
 580                 ret = wait_for_direct_io(type, inode, offset, iter,
 581                                 each_count, 0, NULL, NULL, file);
 582                 gossip_debug(GOSSIP_FILE_DEBUG,
 583                              "%s(%pU): return from wait_for_io:%d\n",
 584                              __func__,
 585                              handle,
 586                              (int)ret);
 587
 588                 if (ret < 0)
 589                         goto out;
 590
 591                 *offset += ret;
 592                 total_count += ret;
 593                 amt_complete = ret;
 594
 595                 gossip_debug(GOSSIP_FILE_DEBUG,
 596                              "%s(%pU): AFTER wait_for_io: offset is %d\n",
 597                              __func__,
 598                              handle,
 599                              (int)*offset);
 600
 601                 /*
 602                  * if we got a short I/O operations,
 603                  * fall out and return what we got so far
 604                  */
 605                 if (amt_complete < each_count)
 606                         break;
 607         } /*end while */
 608
 609 out:
 610         if (total_count > 0)
 611                 ret = total_count;
 612         if (ret > 0) {
 613                 if (type == ORANGEFS_IO_READ) {
 614                         file_accessed(file);
 615                 } else {
 616                         file_update_time(file);
 617                         if (*offset > i_size_read(inode))
 618                                 i_size_write(inode, *offset);
 619                 }
 620         }
 621
 622         gossip_debug(GOSSIP_FILE_DEBUG,
 623                      "%s(%pU): Value(%d) returned.\n",
 624                      __func__,
 625                      handle,
 626                      (int)ret);
 627
 628         return ret;
 629 }
 630
 631 /** ORANGEFS2 implementation of address space operations */
 632 static const struct address_space_operations orangefs_address_operations = {
 633         .writepage = orangefs_writepage,
 634         .readahead = orangefs_readahead,
 635         .readpage = orangefs_readpage,
 636         .writepages = orangefs_writepages,
 637         .dirty_folio = filemap_dirty_folio,
 638         .write_begin = orangefs_write_begin,
 639         .write_end = orangefs_write_end,
 640         .invalidate_folio = orangefs_invalidate_folio,
 641         .releasepage = orangefs_releasepage,
 642         .freepage = orangefs_freepage,
 643         .launder_folio = orangefs_launder_folio,
 644         .direct_IO = orangefs_direct_IO,
 645 };
 646
 647 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
 648 {
 649         struct folio *folio = page_folio(vmf->page);
 650         struct inode *inode = file_inode(vmf->vma->vm_file);
 651         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 652         unsigned long *bitlock = &orangefs_inode->bitlock;
 653         vm_fault_t ret;
 654         struct orangefs_write_range *wr;
 655
 656         sb_start_pagefault(inode->i_sb);
 657
 658         if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
 659                 ret = VM_FAULT_RETRY;
 660                 goto out;
 661         }
 662
 663         folio_lock(folio);
 664         if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 665                 /*
 666                  * Should be impossible.  If it happens, launder the folio
 667                  * since we don't know what's dirty.  This will WARN in
 668                  * orangefs_writepage_locked.
 669                  */
 670                 if (orangefs_launder_folio(folio)) {
 671                         ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 672                         goto out;
 673                 }
 674         }
 675         if (folio_test_private(folio)) {
 676                 wr = folio_get_private(folio);
 677                 if (uid_eq(wr->uid, current_fsuid()) &&
 678                     gid_eq(wr->gid, current_fsgid())) {
 679                         wr->pos = page_offset(vmf->page);
 680                         wr->len = PAGE_SIZE;
 681                         goto okay;
 682                 } else {
 683                         if (orangefs_launder_folio(folio)) {
 684                                 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 685                                 goto out;
 686                         }
 687                 }
 688         }
 689         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 690         if (!wr) {
 691                 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 692                 goto out;
 693         }
 694         wr->pos = page_offset(vmf->page);
 695         wr->len = PAGE_SIZE;
 696         wr->uid = current_fsuid();
 697         wr->gid = current_fsgid();
 698         folio_attach_private(folio, wr);
 699 okay:
 700
 701         file_update_time(vmf->vma->vm_file);
 702         if (folio->mapping != inode->i_mapping) {
 703                 folio_unlock(folio);
 704                 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
 705                 goto out;
 706         }
 707
 708         /*
 709          * We mark the folio dirty already here so that when freeze is in
 710          * progress, we are guaranteed that writeback during freezing will
 711          * see the dirty folio and writeprotect it again.
 712          */
 713         folio_mark_dirty(folio);
 714         folio_wait_stable(folio);
 715         ret = VM_FAULT_LOCKED;
 716 out:
 717         sb_end_pagefault(inode->i_sb);
 718         return ret;
 719 }
 720
 721 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 722 {
 723         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 724         struct orangefs_kernel_op_s *new_op;
 725         loff_t orig_size;
 726         int ret = -EINVAL;
 727
 728         gossip_debug(GOSSIP_INODE_DEBUG,
 729                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 730                      __func__,
 731                      get_khandle_from_ino(inode),
 732                      &orangefs_inode->refn.khandle,
 733                      orangefs_inode->refn.fs_id,
 734                      iattr->ia_size);
 735
 736         /* Ensure that we have a up to date size, so we know if it changed. */
 737         ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
 738         if (ret == -ESTALE)
 739                 ret = -EIO;
 740         if (ret) {
 741                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 742                     __func__, ret);
 743                 return ret;
 744         }
 745         orig_size = i_size_read(inode);
 746
 747         /* This is truncate_setsize in a different order. */
 748         truncate_pagecache(inode, iattr->ia_size);
 749         i_size_write(inode, iattr->ia_size);
 750         if (iattr->ia_size > orig_size)
 751                 pagecache_isize_extended(inode, orig_size, iattr->ia_size);
 752
 753         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 754         if (!new_op)
 755                 return -ENOMEM;
 756
 757         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 758         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 759
 760         ret = service_operation(new_op,
 761                 __func__,
 762                 get_interruptible_flag(inode));
 763
 764         /*
 765          * the truncate has no downcall members to retrieve, but
 766          * the status value tells us if it went through ok or not
 767          */
 768         gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
 769
 770         op_release(new_op);
 771
 772         if (ret != 0)
 773                 return ret;
 774
 775         if (orig_size != i_size_read(inode))
 776                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 777
 778         return ret;
 779 }
 780
 781 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
 782 {
 783         int ret;
 784
 785         if (iattr->ia_valid & ATTR_MODE) {
 786                 if (iattr->ia_mode & (S_ISVTX)) {
 787                         if (is_root_handle(inode)) {
 788                                 /*
 789                                  * allow sticky bit to be set on root (since
 790                                  * it shows up that way by default anyhow),
 791                                  * but don't show it to the server
 792                                  */
 793                                 iattr->ia_mode -= S_ISVTX;
 794                         } else {
 795                                 gossip_debug(GOSSIP_UTILS_DEBUG,
 796                                              "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
 797                                 ret = -EINVAL;
 798                                 goto out;
 799                         }
 800                 }
 801                 if (iattr->ia_mode & (S_ISUID)) {
 802                         gossip_debug(GOSSIP_UTILS_DEBUG,
 803                                      "Attempting to set setuid bit (not supported); returning EINVAL.\n");
 804                         ret = -EINVAL;
 805                         goto out;
 806                 }
 807         }
 808
 809         if (iattr->ia_valid & ATTR_SIZE) {
 810                 ret = orangefs_setattr_size(inode, iattr);
 811                 if (ret)
 812                         goto out;
 813         }
 814
 815 again:
 816         spin_lock(&inode->i_lock);
 817         if (ORANGEFS_I(inode)->attr_valid) {
 818                 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
 819                     gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
 820                         ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 821                 } else {
 822                         spin_unlock(&inode->i_lock);
 823                         write_inode_now(inode, 1);
 824                         goto again;
 825                 }
 826         } else {
 827                 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 828                 ORANGEFS_I(inode)->attr_uid = current_fsuid();
 829                 ORANGEFS_I(inode)->attr_gid = current_fsgid();
 830         }
 831         setattr_copy(&init_user_ns, inode, iattr);
 832         spin_unlock(&inode->i_lock);
 833         mark_inode_dirty(inode);
 834
 835         if (iattr->ia_valid & ATTR_MODE)
 836                 /* change mod on a file that has ACLs */
 837                 ret = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
 838
 839         ret = 0;
 840 out:
 841         return ret;
 842 }
 843
 844 /*
 845  * Change attributes of an object referenced by dentry.
 846  */
 847 int orangefs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 848                      struct iattr *iattr)
 849 {
 850         int ret;
 851         gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
 852             dentry);
 853         ret = setattr_prepare(&init_user_ns, dentry, iattr);
 854         if (ret)
 855                 goto out;
 856         ret = __orangefs_setattr(d_inode(dentry), iattr);
 857         sync_inode_metadata(d_inode(dentry), 1);
 858 out:
 859         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
 860             ret);
 861         return ret;
 862 }
 863
 864 /*
 865  * Obtain attributes of an object given a dentry
 866  */
 867 int orangefs_getattr(struct user_namespace *mnt_userns, const struct path *path,
 868                      struct kstat *stat, u32 request_mask, unsigned int flags)
 869 {
 870         int ret;
 871         struct inode *inode = path->dentry->d_inode;
 872
 873         gossip_debug(GOSSIP_INODE_DEBUG,
 874                      "orangefs_getattr: called on %pd mask %u\n",
 875                      path->dentry, request_mask);
 876
 877         ret = orangefs_inode_getattr(inode,
 878             request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
 879         if (ret == 0) {
 880                 generic_fillattr(&init_user_ns, inode, stat);
 881
 882                 /* override block size reported to stat */
 883                 if (!(request_mask & STATX_SIZE))
 884                         stat->result_mask &= ~STATX_SIZE;
 885
 886                 generic_fill_statx_attr(inode, stat);
 887         }
 888         return ret;
 889 }
 890
 891 int orangefs_permission(struct user_namespace *mnt_userns,
 892                         struct inode *inode, int mask)
 893 {
 894         int ret;
 895
 896         if (mask & MAY_NOT_BLOCK)
 897                 return -ECHILD;
 898
 899         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 900
 901         /* Make sure the permission (and other common attrs) are up to date. */
 902         ret = orangefs_inode_getattr(inode, 0);
 903         if (ret < 0)
 904                 return ret;
 905
 906         return generic_permission(&init_user_ns, inode, mask);
 907 }
 908
 909 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
 910 {
 911         struct iattr iattr;
 912         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
 913             get_khandle_from_ino(inode));
 914         generic_update_time(inode, time, flags);
 915         memset(&iattr, 0, sizeof iattr);
 916         if (flags & S_ATIME)
 917                 iattr.ia_valid |= ATTR_ATIME;
 918         if (flags & S_CTIME)
 919                 iattr.ia_valid |= ATTR_CTIME;
 920         if (flags & S_MTIME)
 921                 iattr.ia_valid |= ATTR_MTIME;
 922         return __orangefs_setattr(inode, &iattr);
 923 }
 924
 925 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 926 {
 927         u64 val = 0;
 928         int ret;
 929
 930         gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 931                      dentry);
 932
 933         ret = orangefs_inode_getxattr(d_inode(dentry),
 934                                       "user.pvfs2.meta_hint",
 935                                       &val, sizeof(val));
 936         if (ret < 0 && ret != -ENODATA)
 937                 return ret;
 938
 939         gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 940
 941         fileattr_fill_flags(fa, val);
 942         return 0;
 943 }
 944
 945 static int orangefs_fileattr_set(struct user_namespace *mnt_userns,
 946                                  struct dentry *dentry, struct fileattr *fa)
 947 {
 948         u64 val = 0;
 949
 950         gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 951                      dentry);
 952         /*
 953          * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
 954          * turned on for a file. The user is not allowed to turn on this bit,
 955          * but the bit is present if the user first gets the flags and then
 956          * updates the flags with some new settings. So, we ignore it in the
 957          * following edit. bligon.
 958          */
 959         if (fileattr_has_fsx(fa) ||
 960             (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
 961                 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
 962                            __func__);
 963                 return -EOPNOTSUPP;
 964         }
 965         val = fa->flags;
 966         gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 967         return orangefs_inode_setxattr(d_inode(dentry),
 968                                        "user.pvfs2.meta_hint",
 969                                        &val, sizeof(val), 0);
 970 }
 971
 972 /* ORANGEFS2 implementation of VFS inode operations for files */
 973 static const struct inode_operations orangefs_file_inode_operations = {
 974         .get_acl = orangefs_get_acl,
 975         .set_acl = orangefs_set_acl,
 976         .setattr = orangefs_setattr,
 977         .getattr = orangefs_getattr,
 978         .listxattr = orangefs_listxattr,
 979         .permission = orangefs_permission,
 980         .update_time = orangefs_update_time,
 981         .fileattr_get = orangefs_fileattr_get,
 982         .fileattr_set = orangefs_fileattr_set,
 983 };
 984
 985 static int orangefs_init_iops(struct inode *inode)
 986 {
 987         inode->i_mapping->a_ops = &orangefs_address_operations;
 988
 989         switch (inode->i_mode & S_IFMT) {
 990         case S_IFREG:
 991                 inode->i_op = &orangefs_file_inode_operations;
 992                 inode->i_fop = &orangefs_file_operations;
 993                 break;
 994         case S_IFLNK:
 995                 inode->i_op = &orangefs_symlink_inode_operations;
 996                 break;
 997         case S_IFDIR:
 998                 inode->i_op = &orangefs_dir_inode_operations;
 999                 inode->i_fop = &orangefs_dir_operations;
1000                 break;
1001         default:
1002                 gossip_debug(GOSSIP_INODE_DEBUG,
1003                              "%s: unsupported mode\n",
1004                              __func__);
1005                 return -EINVAL;
1006         }
1007
1008         return 0;
1009 }
1010
1011 /*
1012  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1013  * a ino_t type that will be used as a hash-index from where the handle will
1014  * be searched for in the VFS hash table of inodes.
1015  */
1016 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1017 {
1018         if (!ref)
1019                 return 0;
1020         return orangefs_khandle_to_ino(&(ref->khandle));
1021 }
1022
1023 /*
1024  * Called to set up an inode from iget5_locked.
1025  */
1026 static int orangefs_set_inode(struct inode *inode, void *data)
1027 {
1028         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1029         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1030         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1031         ORANGEFS_I(inode)->attr_valid = 0;
1032         hash_init(ORANGEFS_I(inode)->xattr_cache);
1033         ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1034         ORANGEFS_I(inode)->bitlock = 0;
1035         return 0;
1036 }
1037
1038 /*
1039  * Called to determine if handles match.
1040  */
1041 static int orangefs_test_inode(struct inode *inode, void *data)
1042 {
1043         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1044         struct orangefs_inode_s *orangefs_inode = NULL;
1045
1046         orangefs_inode = ORANGEFS_I(inode);
1047         /* test handles and fs_ids... */
1048         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1049                                 &(ref->khandle)) &&
1050                         orangefs_inode->refn.fs_id == ref->fs_id);
1051 }
1052
1053 /*
1054  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1055  * file handle.
1056  *
1057  * @sb: the file system super block instance.
1058  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1059  */
1060 struct inode *orangefs_iget(struct super_block *sb,
1061                 struct orangefs_object_kref *ref)
1062 {
1063         struct inode *inode = NULL;
1064         unsigned long hash;
1065         int error;
1066
1067         hash = orangefs_handle_hash(ref);
1068         inode = iget5_locked(sb,
1069                         hash,
1070                         orangefs_test_inode,
1071                         orangefs_set_inode,
1072                         ref);
1073
1074         if (!inode)
1075                 return ERR_PTR(-ENOMEM);
1076
1077         if (!(inode->i_state & I_NEW))
1078                 return inode;
1079
1080         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1081         if (error) {
1082                 iget_failed(inode);
1083                 return ERR_PTR(error);
1084         }
1085
1086         inode->i_ino = hash;    /* needed for stat etc */
1087         orangefs_init_iops(inode);
1088         unlock_new_inode(inode);
1089
1090         gossip_debug(GOSSIP_INODE_DEBUG,
1091                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1092                      &ref->khandle,
1093                      ref->fs_id,
1094                      hash,
1095                      inode->i_ino);
1096
1097         return inode;
1098 }
1099
1100 /*
1101  * Allocate an inode for a newly created file and insert it into the inode hash.
1102  */
1103 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1104                 int mode, dev_t dev, struct orangefs_object_kref *ref)
1105 {
1106         unsigned long hash = orangefs_handle_hash(ref);
1107         struct inode *inode;
1108         int error;
1109
1110         gossip_debug(GOSSIP_INODE_DEBUG,
1111                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1112                      __func__,
1113                      sb,
1114                      MAJOR(dev),
1115                      MINOR(dev),
1116                      mode);
1117
1118         inode = new_inode(sb);
1119         if (!inode)
1120                 return ERR_PTR(-ENOMEM);
1121
1122         orangefs_set_inode(inode, ref);
1123         inode->i_ino = hash;    /* needed for stat etc */
1124
1125         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1126         if (error)
1127                 goto out_iput;
1128
1129         orangefs_init_iops(inode);
1130         inode->i_rdev = dev;
1131
1132         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1133         if (error < 0)
1134                 goto out_iput;
1135
1136         gossip_debug(GOSSIP_INODE_DEBUG,
1137                      "Initializing ACL's for inode %pU\n",
1138                      get_khandle_from_ino(inode));
1139         orangefs_init_acl(inode, dir);
1140         return inode;
1141
1142 out_iput:
1143         iput(inode);
1144         return ERR_PTR(error);
1145 }