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         WARN_ON(wlen == 0);
  53         bvec_set_page(&bv, page, wlen, off % PAGE_SIZE);
  54         iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen);
  55
  56         ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
  57             len, wr, NULL, NULL);
  58         if (ret < 0) {
  59                 SetPageError(page);
  60                 mapping_set_error(page->mapping, ret);
  61         } else {
  62                 ret = 0;
  63         }
  64         kfree(detach_page_private(page));
  65         return ret;
  66 }
  67
  68 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
  69 {
  70         int ret;
  71         ret = orangefs_writepage_locked(page, wbc);
  72         unlock_page(page);
  73         end_page_writeback(page);
  74         return ret;
  75 }
  76
  77 struct orangefs_writepages {
  78         loff_t off;
  79         size_t len;
  80         kuid_t uid;
  81         kgid_t gid;
  82         int maxpages;
  83         int npages;
  84         struct page **pages;
  85         struct bio_vec *bv;
  86 };
  87
  88 static int orangefs_writepages_work(struct orangefs_writepages *ow,
  89     struct writeback_control *wbc)
  90 {
  91         struct inode *inode = ow->pages[0]->mapping->host;
  92         struct orangefs_write_range *wrp, wr;
  93         struct iov_iter iter;
  94         ssize_t ret;
  95         size_t len;
  96         loff_t off;
  97         int i;
  98
  99         len = i_size_read(inode);
 100
 101         for (i = 0; i < ow->npages; i++) {
 102                 set_page_writeback(ow->pages[i]);
 103                 bvec_set_page(&ow->bv[i], ow->pages[i],
 104                               min(page_offset(ow->pages[i]) + PAGE_SIZE,
 105                                   ow->off + ow->len) -
 106                               max(ow->off, page_offset(ow->pages[i])),
 107                               i == 0 ? ow->off - page_offset(ow->pages[i]) : 0);
 108         }
 109         iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->npages, ow->len);
 110
 111         WARN_ON(ow->off >= len);
 112         if (ow->off + ow->len > len)
 113                 ow->len = len - ow->off;
 114
 115         off = ow->off;
 116         wr.uid = ow->uid;
 117         wr.gid = ow->gid;
 118         ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
 119             0, &wr, NULL, NULL);
 120         if (ret < 0) {
 121                 for (i = 0; i < ow->npages; i++) {
 122                         SetPageError(ow->pages[i]);
 123                         mapping_set_error(ow->pages[i]->mapping, ret);
 124                         if (PagePrivate(ow->pages[i])) {
 125                                 wrp = (struct orangefs_write_range *)
 126                                     page_private(ow->pages[i]);
 127                                 ClearPagePrivate(ow->pages[i]);
 128                                 put_page(ow->pages[i]);
 129                                 kfree(wrp);
 130                         }
 131                         end_page_writeback(ow->pages[i]);
 132                         unlock_page(ow->pages[i]);
 133                 }
 134         } else {
 135                 ret = 0;
 136                 for (i = 0; i < ow->npages; i++) {
 137                         if (PagePrivate(ow->pages[i])) {
 138                                 wrp = (struct orangefs_write_range *)
 139                                     page_private(ow->pages[i]);
 140                                 ClearPagePrivate(ow->pages[i]);
 141                                 put_page(ow->pages[i]);
 142                                 kfree(wrp);
 143                         }
 144                         end_page_writeback(ow->pages[i]);
 145                         unlock_page(ow->pages[i]);
 146                 }
 147         }
 148         return ret;
 149 }
 150
 151 static int orangefs_writepages_callback(struct folio *folio,
 152                 struct writeback_control *wbc, void *data)
 153 {
 154         struct orangefs_writepages *ow = data;
 155         struct orangefs_write_range *wr = folio->private;
 156         int ret;
 157
 158         if (!wr) {
 159                 folio_unlock(folio);
 160                 /* It's not private so there's nothing to write, right? */
 161                 printk("writepages_callback not private!\n");
 162                 BUG();
 163                 return 0;
 164         }
 165
 166         ret = -1;
 167         if (ow->npages == 0) {
 168                 ow->off = wr->pos;
 169                 ow->len = wr->len;
 170                 ow->uid = wr->uid;
 171                 ow->gid = wr->gid;
 172                 ow->pages[ow->npages++] = &folio->page;
 173                 ret = 0;
 174                 goto done;
 175         }
 176         if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
 177                 orangefs_writepages_work(ow, wbc);
 178                 ow->npages = 0;
 179                 ret = -1;
 180                 goto done;
 181         }
 182         if (ow->off + ow->len == wr->pos) {
 183                 ow->len += wr->len;
 184                 ow->pages[ow->npages++] = &folio->page;
 185                 ret = 0;
 186                 goto done;
 187         }
 188 done:
 189         if (ret == -1) {
 190                 if (ow->npages) {
 191                         orangefs_writepages_work(ow, wbc);
 192                         ow->npages = 0;
 193                 }
 194                 ret = orangefs_writepage_locked(&folio->page, wbc);
 195                 mapping_set_error(folio->mapping, ret);
 196                 folio_unlock(folio);
 197                 folio_end_writeback(folio);
 198         } else {
 199                 if (ow->npages == ow->maxpages) {
 200                         orangefs_writepages_work(ow, wbc);
 201                         ow->npages = 0;
 202                 }
 203         }
 204         return ret;
 205 }
 206
 207 static int orangefs_writepages(struct address_space *mapping,
 208     struct writeback_control *wbc)
 209 {
 210         struct orangefs_writepages *ow;
 211         struct blk_plug plug;
 212         int ret;
 213         ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
 214         if (!ow)
 215                 return -ENOMEM;
 216         ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
 217         ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
 218         if (!ow->pages) {
 219                 kfree(ow);
 220                 return -ENOMEM;
 221         }
 222         ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
 223         if (!ow->bv) {
 224                 kfree(ow->pages);
 225                 kfree(ow);
 226                 return -ENOMEM;
 227         }
 228         blk_start_plug(&plug);
 229         ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
 230         if (ow->npages)
 231                 ret = orangefs_writepages_work(ow, wbc);
 232         blk_finish_plug(&plug);
 233         kfree(ow->pages);
 234         kfree(ow->bv);
 235         kfree(ow);
 236         return ret;
 237 }
 238
 239 static int orangefs_launder_folio(struct folio *);
 240
 241 static void orangefs_readahead(struct readahead_control *rac)
 242 {
 243         loff_t offset;
 244         struct iov_iter iter;
 245         struct inode *inode = rac->mapping->host;
 246         struct xarray *i_pages;
 247         struct folio *folio;
 248         loff_t new_start = readahead_pos(rac);
 249         int ret;
 250         size_t new_len = 0;
 251
 252         loff_t bytes_remaining = inode->i_size - readahead_pos(rac);
 253         loff_t pages_remaining = bytes_remaining / PAGE_SIZE;
 254
 255         if (pages_remaining >= 1024)
 256                 new_len = 4194304;
 257         else if (pages_remaining > readahead_count(rac))
 258                 new_len = bytes_remaining;
 259
 260         if (new_len)
 261                 readahead_expand(rac, new_start, new_len);
 262
 263         offset = readahead_pos(rac);
 264         i_pages = &rac->mapping->i_pages;
 265
 266         iov_iter_xarray(&iter, ITER_DEST, i_pages, offset, readahead_length(rac));
 267
 268         /* read in the pages. */
 269         if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode,
 270                         &offset, &iter, readahead_length(rac),
 271                         inode->i_size, NULL, NULL, rac->file)) < 0)
 272                 gossip_debug(GOSSIP_FILE_DEBUG,
 273                         "%s: wait_for_direct_io failed. \n", __func__);
 274         else
 275                 ret = 0;
 276
 277         /* clean up. */
 278         while ((folio = readahead_folio(rac))) {
 279                 if (!ret)
 280                         folio_mark_uptodate(folio);
 281                 folio_unlock(folio);
 282         }
 283 }
 284
 285 static int orangefs_read_folio(struct file *file, struct folio *folio)
 286 {
 287         struct inode *inode = folio->mapping->host;
 288         struct iov_iter iter;
 289         struct bio_vec bv;
 290         ssize_t ret;
 291         loff_t off; /* offset of this folio in the file */
 292
 293         if (folio_test_dirty(folio))
 294                 orangefs_launder_folio(folio);
 295
 296         off = folio_pos(folio);
 297         bvec_set_folio(&bv, folio, folio_size(folio), 0);
 298         iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio));
 299
 300         ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
 301                         folio_size(folio), inode->i_size, NULL, NULL, file);
 302         /* this will only zero remaining unread portions of the folio data */
 303         iov_iter_zero(~0U, &iter);
 304         /* takes care of potential aliasing */
 305         flush_dcache_folio(folio);
 306         if (ret < 0) {
 307                 folio_set_error(folio);
 308         } else {
 309                 folio_mark_uptodate(folio);
 310                 ret = 0;
 311         }
 312         /* unlock the folio after the ->read_folio() routine completes */
 313         folio_unlock(folio);
 314         return ret;
 315 }
 316
 317 static int orangefs_write_begin(struct file *file,
 318                 struct address_space *mapping, loff_t pos, unsigned len,
 319                 struct page **pagep, void **fsdata)
 320 {
 321         struct orangefs_write_range *wr;
 322         struct folio *folio;
 323         struct page *page;
 324         pgoff_t index;
 325         int ret;
 326
 327         index = pos >> PAGE_SHIFT;
 328
 329         page = grab_cache_page_write_begin(mapping, index);
 330         if (!page)
 331                 return -ENOMEM;
 332
 333         *pagep = page;
 334         folio = page_folio(page);
 335
 336         if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 337                 /*
 338                  * Should be impossible.  If it happens, launder the page
 339                  * since we don't know what's dirty.  This will WARN in
 340                  * orangefs_writepage_locked.
 341                  */
 342                 ret = orangefs_launder_folio(folio);
 343                 if (ret)
 344                         return ret;
 345         }
 346         if (folio_test_private(folio)) {
 347                 struct orangefs_write_range *wr;
 348                 wr = folio_get_private(folio);
 349                 if (wr->pos + wr->len == pos &&
 350                     uid_eq(wr->uid, current_fsuid()) &&
 351                     gid_eq(wr->gid, current_fsgid())) {
 352                         wr->len += len;
 353                         goto okay;
 354                 } else {
 355                         ret = orangefs_launder_folio(folio);
 356                         if (ret)
 357                                 return ret;
 358                 }
 359         }
 360
 361         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 362         if (!wr)
 363                 return -ENOMEM;
 364
 365         wr->pos = pos;
 366         wr->len = len;
 367         wr->uid = current_fsuid();
 368         wr->gid = current_fsgid();
 369         folio_attach_private(folio, wr);
 370 okay:
 371         return 0;
 372 }
 373
 374 static int orangefs_write_end(struct file *file, struct address_space *mapping,
 375     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
 376 {
 377         struct inode *inode = page->mapping->host;
 378         loff_t last_pos = pos + copied;
 379
 380         /*
 381          * No need to use i_size_read() here, the i_size
 382          * cannot change under us because we hold the i_mutex.
 383          */
 384         if (last_pos > inode->i_size)
 385                 i_size_write(inode, last_pos);
 386
 387         /* zero the stale part of the page if we did a short copy */
 388         if (!PageUptodate(page)) {
 389                 unsigned from = pos & (PAGE_SIZE - 1);
 390                 if (copied < len) {
 391                         zero_user(page, from + copied, len - copied);
 392                 }
 393                 /* Set fully written pages uptodate. */
 394                 if (pos == page_offset(page) &&
 395                     (len == PAGE_SIZE || pos + len == inode->i_size)) {
 396                         zero_user_segment(page, from + copied, PAGE_SIZE);
 397                         SetPageUptodate(page);
 398                 }
 399         }
 400
 401         set_page_dirty(page);
 402         unlock_page(page);
 403         put_page(page);
 404
 405         mark_inode_dirty_sync(file_inode(file));
 406         return copied;
 407 }
 408
 409 static void orangefs_invalidate_folio(struct folio *folio,
 410                                  size_t offset, size_t length)
 411 {
 412         struct orangefs_write_range *wr = folio_get_private(folio);
 413
 414         if (offset == 0 && length == PAGE_SIZE) {
 415                 kfree(folio_detach_private(folio));
 416                 return;
 417         /* write range entirely within invalidate range (or equal) */
 418         } else if (folio_pos(folio) + offset <= wr->pos &&
 419             wr->pos + wr->len <= folio_pos(folio) + offset + length) {
 420                 kfree(folio_detach_private(folio));
 421                 /* XXX is this right? only caller in fs */
 422                 folio_cancel_dirty(folio);
 423                 return;
 424         /* invalidate range chops off end of write range */
 425         } else if (wr->pos < folio_pos(folio) + offset &&
 426             wr->pos + wr->len <= folio_pos(folio) + offset + length &&
 427              folio_pos(folio) + offset < wr->pos + wr->len) {
 428                 size_t x;
 429                 x = wr->pos + wr->len - (folio_pos(folio) + offset);
 430                 WARN_ON(x > wr->len);
 431                 wr->len -= x;
 432                 wr->uid = current_fsuid();
 433                 wr->gid = current_fsgid();
 434         /* invalidate range chops off beginning of write range */
 435         } else if (folio_pos(folio) + offset <= wr->pos &&
 436             folio_pos(folio) + offset + length < wr->pos + wr->len &&
 437             wr->pos < folio_pos(folio) + offset + length) {
 438                 size_t x;
 439                 x = folio_pos(folio) + offset + length - wr->pos;
 440                 WARN_ON(x > wr->len);
 441                 wr->pos += x;
 442                 wr->len -= x;
 443                 wr->uid = current_fsuid();
 444                 wr->gid = current_fsgid();
 445         /* invalidate range entirely within write range (punch hole) */
 446         } else if (wr->pos < folio_pos(folio) + offset &&
 447             folio_pos(folio) + offset + length < wr->pos + wr->len) {
 448                 /* XXX what do we do here... should not WARN_ON */
 449                 WARN_ON(1);
 450                 /* punch hole */
 451                 /*
 452                  * should we just ignore this and write it out anyway?
 453                  * it hardly makes sense
 454                  */
 455                 return;
 456         /* non-overlapping ranges */
 457         } else {
 458                 /* WARN if they do overlap */
 459                 if (!((folio_pos(folio) + offset + length <= wr->pos) ^
 460                     (wr->pos + wr->len <= folio_pos(folio) + offset))) {
 461                         WARN_ON(1);
 462                         printk("invalidate range offset %llu length %zu\n",
 463                             folio_pos(folio) + offset, length);
 464                         printk("write range offset %llu length %zu\n",
 465                             wr->pos, wr->len);
 466                 }
 467                 return;
 468         }
 469
 470         /*
 471          * Above there are returns where wr is freed or where we WARN.
 472          * Thus the following runs if wr was modified above.
 473          */
 474
 475         orangefs_launder_folio(folio);
 476 }
 477
 478 static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
 479 {
 480         return !folio_test_private(folio);
 481 }
 482
 483 static void orangefs_free_folio(struct folio *folio)
 484 {
 485         kfree(folio_detach_private(folio));
 486 }
 487
 488 static int orangefs_launder_folio(struct folio *folio)
 489 {
 490         int r = 0;
 491         struct writeback_control wbc = {
 492                 .sync_mode = WB_SYNC_ALL,
 493                 .nr_to_write = 0,
 494         };
 495         folio_wait_writeback(folio);
 496         if (folio_clear_dirty_for_io(folio)) {
 497                 r = orangefs_writepage_locked(&folio->page, &wbc);
 498                 folio_end_writeback(folio);
 499         }
 500         return r;
 501 }
 502
 503 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
 504                                   struct iov_iter *iter)
 505 {
 506         /*
 507          * Comment from original do_readv_writev:
 508          * Common entry point for read/write/readv/writev
 509          * This function will dispatch it to either the direct I/O
 510          * or buffered I/O path depending on the mount options and/or
 511          * augmented/extended metadata attached to the file.
 512          * Note: File extended attributes override any mount options.
 513          */
 514         struct file *file = iocb->ki_filp;
 515         loff_t pos = iocb->ki_pos;
 516         enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
 517             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
 518         loff_t *offset = &pos;
 519         struct inode *inode = file->f_mapping->host;
 520         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 521         struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
 522         size_t count = iov_iter_count(iter);
 523         ssize_t total_count = 0;
 524         ssize_t ret = -EINVAL;
 525
 526         gossip_debug(GOSSIP_FILE_DEBUG,
 527                 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
 528                 __func__,
 529                 handle,
 530                 (int)count);
 531
 532         if (type == ORANGEFS_IO_WRITE) {
 533                 gossip_debug(GOSSIP_FILE_DEBUG,
 534                              "%s(%pU): proceeding with offset : %llu, "
 535                              "size %d\n",
 536                              __func__,
 537                              handle,
 538                              llu(*offset),
 539                              (int)count);
 540         }
 541
 542         if (count == 0) {
 543                 ret = 0;
 544                 goto out;
 545         }
 546
 547         while (iov_iter_count(iter)) {
 548                 size_t each_count = iov_iter_count(iter);
 549                 size_t amt_complete;
 550
 551                 /* how much to transfer in this loop iteration */
 552                 if (each_count > orangefs_bufmap_size_query())
 553                         each_count = orangefs_bufmap_size_query();
 554
 555                 gossip_debug(GOSSIP_FILE_DEBUG,
 556                              "%s(%pU): size of each_count(%d)\n",
 557                              __func__,
 558                              handle,
 559                              (int)each_count);
 560                 gossip_debug(GOSSIP_FILE_DEBUG,
 561                              "%s(%pU): BEFORE wait_for_io: offset is %d\n",
 562                              __func__,
 563                              handle,
 564                              (int)*offset);
 565
 566                 ret = wait_for_direct_io(type, inode, offset, iter,
 567                                 each_count, 0, NULL, NULL, file);
 568                 gossip_debug(GOSSIP_FILE_DEBUG,
 569                              "%s(%pU): return from wait_for_io:%d\n",
 570                              __func__,
 571                              handle,
 572                              (int)ret);
 573
 574                 if (ret < 0)
 575                         goto out;
 576
 577                 *offset += ret;
 578                 total_count += ret;
 579                 amt_complete = ret;
 580
 581                 gossip_debug(GOSSIP_FILE_DEBUG,
 582                              "%s(%pU): AFTER wait_for_io: offset is %d\n",
 583                              __func__,
 584                              handle,
 585                              (int)*offset);
 586
 587                 /*
 588                  * if we got a short I/O operations,
 589                  * fall out and return what we got so far
 590                  */
 591                 if (amt_complete < each_count)
 592                         break;
 593         } /*end while */
 594
 595 out:
 596         if (total_count > 0)
 597                 ret = total_count;
 598         if (ret > 0) {
 599                 if (type == ORANGEFS_IO_READ) {
 600                         file_accessed(file);
 601                 } else {
 602                         file_update_time(file);
 603                         if (*offset > i_size_read(inode))
 604                                 i_size_write(inode, *offset);
 605                 }
 606         }
 607
 608         gossip_debug(GOSSIP_FILE_DEBUG,
 609                      "%s(%pU): Value(%d) returned.\n",
 610                      __func__,
 611                      handle,
 612                      (int)ret);
 613
 614         return ret;
 615 }
 616
 617 /** ORANGEFS2 implementation of address space operations */
 618 static const struct address_space_operations orangefs_address_operations = {
 619         .writepage = orangefs_writepage,
 620         .readahead = orangefs_readahead,
 621         .read_folio = orangefs_read_folio,
 622         .writepages = orangefs_writepages,
 623         .dirty_folio = filemap_dirty_folio,
 624         .write_begin = orangefs_write_begin,
 625         .write_end = orangefs_write_end,
 626         .invalidate_folio = orangefs_invalidate_folio,
 627         .release_folio = orangefs_release_folio,
 628         .free_folio = orangefs_free_folio,
 629         .launder_folio = orangefs_launder_folio,
 630         .direct_IO = orangefs_direct_IO,
 631 };
 632
 633 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
 634 {
 635         struct folio *folio = page_folio(vmf->page);
 636         struct inode *inode = file_inode(vmf->vma->vm_file);
 637         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 638         unsigned long *bitlock = &orangefs_inode->bitlock;
 639         vm_fault_t ret;
 640         struct orangefs_write_range *wr;
 641
 642         sb_start_pagefault(inode->i_sb);
 643
 644         if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
 645                 ret = VM_FAULT_RETRY;
 646                 goto out;
 647         }
 648
 649         folio_lock(folio);
 650         if (folio_test_dirty(folio) && !folio_test_private(folio)) {
 651                 /*
 652                  * Should be impossible.  If it happens, launder the folio
 653                  * since we don't know what's dirty.  This will WARN in
 654                  * orangefs_writepage_locked.
 655                  */
 656                 if (orangefs_launder_folio(folio)) {
 657                         ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 658                         goto out;
 659                 }
 660         }
 661         if (folio_test_private(folio)) {
 662                 wr = folio_get_private(folio);
 663                 if (uid_eq(wr->uid, current_fsuid()) &&
 664                     gid_eq(wr->gid, current_fsgid())) {
 665                         wr->pos = page_offset(vmf->page);
 666                         wr->len = PAGE_SIZE;
 667                         goto okay;
 668                 } else {
 669                         if (orangefs_launder_folio(folio)) {
 670                                 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 671                                 goto out;
 672                         }
 673                 }
 674         }
 675         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 676         if (!wr) {
 677                 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
 678                 goto out;
 679         }
 680         wr->pos = page_offset(vmf->page);
 681         wr->len = PAGE_SIZE;
 682         wr->uid = current_fsuid();
 683         wr->gid = current_fsgid();
 684         folio_attach_private(folio, wr);
 685 okay:
 686
 687         file_update_time(vmf->vma->vm_file);
 688         if (folio->mapping != inode->i_mapping) {
 689                 folio_unlock(folio);
 690                 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
 691                 goto out;
 692         }
 693
 694         /*
 695          * We mark the folio dirty already here so that when freeze is in
 696          * progress, we are guaranteed that writeback during freezing will
 697          * see the dirty folio and writeprotect it again.
 698          */
 699         folio_mark_dirty(folio);
 700         folio_wait_stable(folio);
 701         ret = VM_FAULT_LOCKED;
 702 out:
 703         sb_end_pagefault(inode->i_sb);
 704         return ret;
 705 }
 706
 707 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 708 {
 709         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 710         struct orangefs_kernel_op_s *new_op;
 711         loff_t orig_size;
 712         int ret = -EINVAL;
 713
 714         gossip_debug(GOSSIP_INODE_DEBUG,
 715                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 716                      __func__,
 717                      get_khandle_from_ino(inode),
 718                      &orangefs_inode->refn.khandle,
 719                      orangefs_inode->refn.fs_id,
 720                      iattr->ia_size);
 721
 722         /* Ensure that we have a up to date size, so we know if it changed. */
 723         ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
 724         if (ret == -ESTALE)
 725                 ret = -EIO;
 726         if (ret) {
 727                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 728                     __func__, ret);
 729                 return ret;
 730         }
 731         orig_size = i_size_read(inode);
 732
 733         /* This is truncate_setsize in a different order. */
 734         truncate_pagecache(inode, iattr->ia_size);
 735         i_size_write(inode, iattr->ia_size);
 736         if (iattr->ia_size > orig_size)
 737                 pagecache_isize_extended(inode, orig_size, iattr->ia_size);
 738
 739         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 740         if (!new_op)
 741                 return -ENOMEM;
 742
 743         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 744         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 745
 746         ret = service_operation(new_op,
 747                 __func__,
 748                 get_interruptible_flag(inode));
 749
 750         /*
 751          * the truncate has no downcall members to retrieve, but
 752          * the status value tells us if it went through ok or not
 753          */
 754         gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
 755
 756         op_release(new_op);
 757
 758         if (ret != 0)
 759                 return ret;
 760
 761         if (orig_size != i_size_read(inode))
 762                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 763
 764         return ret;
 765 }
 766
 767 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
 768 {
 769         int ret;
 770
 771         if (iattr->ia_valid & ATTR_MODE) {
 772                 if (iattr->ia_mode & (S_ISVTX)) {
 773                         if (is_root_handle(inode)) {
 774                                 /*
 775                                  * allow sticky bit to be set on root (since
 776                                  * it shows up that way by default anyhow),
 777                                  * but don't show it to the server
 778                                  */
 779                                 iattr->ia_mode -= S_ISVTX;
 780                         } else {
 781                                 gossip_debug(GOSSIP_UTILS_DEBUG,
 782                                              "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
 783                                 ret = -EINVAL;
 784                                 goto out;
 785                         }
 786                 }
 787                 if (iattr->ia_mode & (S_ISUID)) {
 788                         gossip_debug(GOSSIP_UTILS_DEBUG,
 789                                      "Attempting to set setuid bit (not supported); returning EINVAL.\n");
 790                         ret = -EINVAL;
 791                         goto out;
 792                 }
 793         }
 794
 795         if (iattr->ia_valid & ATTR_SIZE) {
 796                 ret = orangefs_setattr_size(inode, iattr);
 797                 if (ret)
 798                         goto out;
 799         }
 800
 801 again:
 802         spin_lock(&inode->i_lock);
 803         if (ORANGEFS_I(inode)->attr_valid) {
 804                 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
 805                     gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
 806                         ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 807                 } else {
 808                         spin_unlock(&inode->i_lock);
 809                         write_inode_now(inode, 1);
 810                         goto again;
 811                 }
 812         } else {
 813                 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 814                 ORANGEFS_I(inode)->attr_uid = current_fsuid();
 815                 ORANGEFS_I(inode)->attr_gid = current_fsgid();
 816         }
 817         setattr_copy(&nop_mnt_idmap, inode, iattr);
 818         spin_unlock(&inode->i_lock);
 819         mark_inode_dirty(inode);
 820
 821         ret = 0;
 822 out:
 823         return ret;
 824 }
 825
 826 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr)
 827 {
 828         int ret;
 829         struct inode *inode = d_inode(dentry);
 830
 831         ret = __orangefs_setattr(inode, iattr);
 832         /* change mode on a file that has ACLs */
 833         if (!ret && (iattr->ia_valid & ATTR_MODE))
 834                 ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
 835         return ret;
 836 }
 837
 838 /*
 839  * Change attributes of an object referenced by dentry.
 840  */
 841 int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 842                      struct iattr *iattr)
 843 {
 844         int ret;
 845         gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
 846             dentry);
 847         ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
 848         if (ret)
 849                 goto out;
 850         ret = __orangefs_setattr_mode(dentry, iattr);
 851         sync_inode_metadata(d_inode(dentry), 1);
 852 out:
 853         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
 854             ret);
 855         return ret;
 856 }
 857
 858 /*
 859  * Obtain attributes of an object given a dentry
 860  */
 861 int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path,
 862                      struct kstat *stat, u32 request_mask, unsigned int flags)
 863 {
 864         int ret;
 865         struct inode *inode = path->dentry->d_inode;
 866
 867         gossip_debug(GOSSIP_INODE_DEBUG,
 868                      "orangefs_getattr: called on %pd mask %u\n",
 869                      path->dentry, request_mask);
 870
 871         ret = orangefs_inode_getattr(inode,
 872             request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
 873         if (ret == 0) {
 874                 generic_fillattr(&nop_mnt_idmap, inode, stat);
 875
 876                 /* override block size reported to stat */
 877                 if (!(request_mask & STATX_SIZE))
 878                         stat->result_mask &= ~STATX_SIZE;
 879
 880                 generic_fill_statx_attr(inode, stat);
 881         }
 882         return ret;
 883 }
 884
 885 int orangefs_permission(struct mnt_idmap *idmap,
 886                         struct inode *inode, int mask)
 887 {
 888         int ret;
 889
 890         if (mask & MAY_NOT_BLOCK)
 891                 return -ECHILD;
 892
 893         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 894
 895         /* Make sure the permission (and other common attrs) are up to date. */
 896         ret = orangefs_inode_getattr(inode, 0);
 897         if (ret < 0)
 898                 return ret;
 899
 900         return generic_permission(&nop_mnt_idmap, inode, mask);
 901 }
 902
 903 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
 904 {
 905         struct iattr iattr;
 906         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
 907             get_khandle_from_ino(inode));
 908         generic_update_time(inode, time, flags);
 909         memset(&iattr, 0, sizeof iattr);
 910         if (flags & S_ATIME)
 911                 iattr.ia_valid |= ATTR_ATIME;
 912         if (flags & S_CTIME)
 913                 iattr.ia_valid |= ATTR_CTIME;
 914         if (flags & S_MTIME)
 915                 iattr.ia_valid |= ATTR_MTIME;
 916         return __orangefs_setattr(inode, &iattr);
 917 }
 918
 919 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
 920 {
 921         u64 val = 0;
 922         int ret;
 923
 924         gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 925                      dentry);
 926
 927         ret = orangefs_inode_getxattr(d_inode(dentry),
 928                                       "user.pvfs2.meta_hint",
 929                                       &val, sizeof(val));
 930         if (ret < 0 && ret != -ENODATA)
 931                 return ret;
 932
 933         gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 934
 935         fileattr_fill_flags(fa, val);
 936         return 0;
 937 }
 938
 939 static int orangefs_fileattr_set(struct mnt_idmap *idmap,
 940                                  struct dentry *dentry, struct fileattr *fa)
 941 {
 942         u64 val = 0;
 943
 944         gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
 945                      dentry);
 946         /*
 947          * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
 948          * turned on for a file. The user is not allowed to turn on this bit,
 949          * but the bit is present if the user first gets the flags and then
 950          * updates the flags with some new settings. So, we ignore it in the
 951          * following edit. bligon.
 952          */
 953         if (fileattr_has_fsx(fa) ||
 954             (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
 955                 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
 956                            __func__);
 957                 return -EOPNOTSUPP;
 958         }
 959         val = fa->flags;
 960         gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
 961         return orangefs_inode_setxattr(d_inode(dentry),
 962                                        "user.pvfs2.meta_hint",
 963                                        &val, sizeof(val), 0);
 964 }
 965
 966 /* ORANGEFS2 implementation of VFS inode operations for files */
 967 static const struct inode_operations orangefs_file_inode_operations = {
 968         .get_inode_acl = orangefs_get_acl,
 969         .set_acl = orangefs_set_acl,
 970         .setattr = orangefs_setattr,
 971         .getattr = orangefs_getattr,
 972         .listxattr = orangefs_listxattr,
 973         .permission = orangefs_permission,
 974         .update_time = orangefs_update_time,
 975         .fileattr_get = orangefs_fileattr_get,
 976         .fileattr_set = orangefs_fileattr_set,
 977 };
 978
 979 static int orangefs_init_iops(struct inode *inode)
 980 {
 981         inode->i_mapping->a_ops = &orangefs_address_operations;
 982
 983         switch (inode->i_mode & S_IFMT) {
 984         case S_IFREG:
 985                 inode->i_op = &orangefs_file_inode_operations;
 986                 inode->i_fop = &orangefs_file_operations;
 987                 break;
 988         case S_IFLNK:
 989                 inode->i_op = &orangefs_symlink_inode_operations;
 990                 break;
 991         case S_IFDIR:
 992                 inode->i_op = &orangefs_dir_inode_operations;
 993                 inode->i_fop = &orangefs_dir_operations;
 994                 break;
 995         default:
 996                 gossip_debug(GOSSIP_INODE_DEBUG,
 997                              "%s: unsupported mode\n",
 998                              __func__);
 999                 return -EINVAL;
1000         }
1001
1002         return 0;
1003 }
1004
1005 /*
1006  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1007  * a ino_t type that will be used as a hash-index from where the handle will
1008  * be searched for in the VFS hash table of inodes.
1009  */
1010 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1011 {
1012         if (!ref)
1013                 return 0;
1014         return orangefs_khandle_to_ino(&(ref->khandle));
1015 }
1016
1017 /*
1018  * Called to set up an inode from iget5_locked.
1019  */
1020 static int orangefs_set_inode(struct inode *inode, void *data)
1021 {
1022         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1023         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1024         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1025         ORANGEFS_I(inode)->attr_valid = 0;
1026         hash_init(ORANGEFS_I(inode)->xattr_cache);
1027         ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1028         ORANGEFS_I(inode)->bitlock = 0;
1029         return 0;
1030 }
1031
1032 /*
1033  * Called to determine if handles match.
1034  */
1035 static int orangefs_test_inode(struct inode *inode, void *data)
1036 {
1037         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1038         struct orangefs_inode_s *orangefs_inode = NULL;
1039
1040         orangefs_inode = ORANGEFS_I(inode);
1041         /* test handles and fs_ids... */
1042         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1043                                 &(ref->khandle)) &&
1044                         orangefs_inode->refn.fs_id == ref->fs_id);
1045 }
1046
1047 /*
1048  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1049  * file handle.
1050  *
1051  * @sb: the file system super block instance.
1052  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1053  */
1054 struct inode *orangefs_iget(struct super_block *sb,
1055                 struct orangefs_object_kref *ref)
1056 {
1057         struct inode *inode = NULL;
1058         unsigned long hash;
1059         int error;
1060
1061         hash = orangefs_handle_hash(ref);
1062         inode = iget5_locked(sb,
1063                         hash,
1064                         orangefs_test_inode,
1065                         orangefs_set_inode,
1066                         ref);
1067
1068         if (!inode)
1069                 return ERR_PTR(-ENOMEM);
1070
1071         if (!(inode->i_state & I_NEW))
1072                 return inode;
1073
1074         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1075         if (error) {
1076                 iget_failed(inode);
1077                 return ERR_PTR(error);
1078         }
1079
1080         inode->i_ino = hash;    /* needed for stat etc */
1081         orangefs_init_iops(inode);
1082         unlock_new_inode(inode);
1083
1084         gossip_debug(GOSSIP_INODE_DEBUG,
1085                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1086                      &ref->khandle,
1087                      ref->fs_id,
1088                      hash,
1089                      inode->i_ino);
1090
1091         return inode;
1092 }
1093
1094 /*
1095  * Allocate an inode for a newly created file and insert it into the inode hash.
1096  */
1097 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1098                 umode_t mode, dev_t dev, struct orangefs_object_kref *ref)
1099 {
1100         struct posix_acl *acl = NULL, *default_acl = NULL;
1101         unsigned long hash = orangefs_handle_hash(ref);
1102         struct inode *inode;
1103         int error;
1104
1105         gossip_debug(GOSSIP_INODE_DEBUG,
1106                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1107                      __func__,
1108                      sb,
1109                      MAJOR(dev),
1110                      MINOR(dev),
1111                      mode);
1112
1113         inode = new_inode(sb);
1114         if (!inode)
1115                 return ERR_PTR(-ENOMEM);
1116
1117         error = posix_acl_create(dir, &mode, &default_acl, &acl);
1118         if (error)
1119                 goto out_iput;
1120
1121         orangefs_set_inode(inode, ref);
1122         inode->i_ino = hash;    /* needed for stat etc */
1123
1124         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1125         if (error)
1126                 goto out_iput;
1127
1128         orangefs_init_iops(inode);
1129         inode->i_rdev = dev;
1130
1131         if (default_acl) {
1132                 error = __orangefs_set_acl(inode, default_acl,
1133                                            ACL_TYPE_DEFAULT);
1134                 if (error)
1135                         goto out_iput;
1136         }
1137
1138         if (acl) {
1139                 error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS);
1140                 if (error)
1141                         goto out_iput;
1142         }
1143
1144         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1145         if (error < 0)
1146                 goto out_iput;
1147
1148         gossip_debug(GOSSIP_INODE_DEBUG,
1149                      "Initializing ACL's for inode %pU\n",
1150                      get_khandle_from_ino(inode));
1151         if (mode != inode->i_mode) {
1152                 struct iattr iattr = {
1153                         .ia_mode = mode,
1154                         .ia_valid = ATTR_MODE,
1155                 };
1156                 inode->i_mode = mode;
1157                 __orangefs_setattr(inode, &iattr);
1158                 __posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode);
1159         }
1160         posix_acl_release(acl);
1161         posix_acl_release(default_acl);
1162         return inode;
1163
1164 out_iput:
1165         iput(inode);
1166         posix_acl_release(acl);
1167         posix_acl_release(default_acl);
1168         return ERR_PTR(error);
1169 }