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/bvec.h>
  14 #include "protocol.h"
  15 #include "orangefs-kernel.h"
  16 #include "orangefs-bufmap.h"
  17
  18 static int orangefs_writepage_locked(struct page *page,
  19     struct writeback_control *wbc)
  20 {
  21         struct inode *inode = page->mapping->host;
  22         struct orangefs_write_range *wr = NULL;
  23         struct iov_iter iter;
  24         struct bio_vec bv;
  25         size_t len, wlen;
  26         ssize_t ret;
  27         loff_t off;
  28
  29         set_page_writeback(page);
  30
  31         len = i_size_read(inode);
  32         if (PagePrivate(page)) {
  33                 wr = (struct orangefs_write_range *)page_private(page);
  34                 off = wr->pos;
  35                 if (off + wr->len > len)
  36                         wlen = len - off;
  37                 else
  38                         wlen = wr->len;
  39         } else {
  40                 WARN_ON(1);
  41                 off = page_offset(page);
  42                 if (off + PAGE_SIZE > len)
  43                         wlen = len - off;
  44                 else
  45                         wlen = PAGE_SIZE;
  46         }
  47         /* Should've been handled in orangefs_invalidatepage. */
  48         WARN_ON(off == len || off + wlen > len);
  49
  50         bv.bv_page = page;
  51         bv.bv_len = wlen;
  52         bv.bv_offset = off % PAGE_SIZE;
  53         WARN_ON(wlen == 0);
  54         iov_iter_bvec(&iter, WRITE, &bv, 1, wlen);
  55
  56         ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
  57             len, wr);
  58         if (ret < 0) {
  59                 SetPageError(page);
  60                 mapping_set_error(page->mapping, ret);
  61         } else {
  62                 ret = 0;
  63         }
  64         if (wr) {
  65                 kfree(wr);
  66                 set_page_private(page, 0);
  67                 ClearPagePrivate(page);
  68                 put_page(page);
  69         }
  70         return ret;
  71 }
  72
  73 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
  74 {
  75         int ret;
  76         ret = orangefs_writepage_locked(page, wbc);
  77         unlock_page(page);
  78         end_page_writeback(page);
  79         return ret;
  80 }
  81
  82 static int orangefs_readpage(struct file *file, struct page *page)
  83 {
  84         struct inode *inode = page->mapping->host;
  85         struct iov_iter iter;
  86         struct bio_vec bv;
  87         ssize_t ret;
  88         loff_t off;
  89
  90         off = page_offset(page);
  91         bv.bv_page = page;
  92         bv.bv_len = PAGE_SIZE;
  93         bv.bv_offset = 0;
  94         iov_iter_bvec(&iter, READ, &bv, 1, PAGE_SIZE);
  95
  96         ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
  97             PAGE_SIZE, inode->i_size, NULL);
  98         /* this will only zero remaining unread portions of the page data */
  99         iov_iter_zero(~0U, &iter);
 100         /* takes care of potential aliasing */
 101         flush_dcache_page(page);
 102         if (ret < 0) {
 103                 SetPageError(page);
 104         } else {
 105                 SetPageUptodate(page);
 106                 if (PageError(page))
 107                         ClearPageError(page);
 108                 ret = 0;
 109         }
 110         /* unlock the page after the ->readpage() routine completes */
 111         unlock_page(page);
 112         return ret;
 113 }
 114
 115 static int orangefs_launder_page(struct page *);
 116
 117 static int orangefs_write_begin(struct file *file,
 118     struct address_space *mapping,
 119     loff_t pos, unsigned len, unsigned flags, struct page **pagep,
 120     void **fsdata)
 121 {
 122         struct orangefs_write_range *wr;
 123         struct page *page;
 124         pgoff_t index;
 125         int ret;
 126
 127         index = pos >> PAGE_SHIFT;
 128
 129         page = grab_cache_page_write_begin(mapping, index, flags);
 130         if (!page)
 131                 return -ENOMEM;
 132
 133         *pagep = page;
 134
 135         if (PageDirty(page) && !PagePrivate(page)) {
 136                 /*
 137                  * Should be impossible.  If it happens, launder the page
 138                  * since we don't know what's dirty.  This will WARN in
 139                  * orangefs_writepage_locked.
 140                  */
 141                 ret = orangefs_launder_page(page);
 142                 if (ret)
 143                         return ret;
 144         }
 145         if (PagePrivate(page)) {
 146                 struct orangefs_write_range *wr;
 147                 wr = (struct orangefs_write_range *)page_private(page);
 148                 if (wr->pos + wr->len == pos &&
 149                     uid_eq(wr->uid, current_fsuid()) &&
 150                     gid_eq(wr->gid, current_fsgid())) {
 151                         wr->len += len;
 152                         goto okay;
 153                 } else {
 154                         ret = orangefs_launder_page(page);
 155                         if (ret)
 156                                 return ret;
 157                 }
 158
 159         }
 160
 161         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 162         if (!wr)
 163                 return -ENOMEM;
 164
 165         wr->pos = pos;
 166         wr->len = len;
 167         wr->uid = current_fsuid();
 168         wr->gid = current_fsgid();
 169         SetPagePrivate(page);
 170         set_page_private(page, (unsigned long)wr);
 171         get_page(page);
 172 okay:
 173
 174         if (!PageUptodate(page) && (len != PAGE_SIZE)) {
 175                 unsigned from = pos & (PAGE_SIZE - 1);
 176
 177                 zero_user_segments(page, 0, from, from + len, PAGE_SIZE);
 178         }
 179         return 0;
 180 }
 181
 182 static int orangefs_write_end(struct file *file, struct address_space *mapping,
 183     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
 184 {
 185         int r;
 186         r = simple_write_end(file, mapping, pos, len, copied, page, fsdata);
 187         mark_inode_dirty_sync(file_inode(file));
 188         return r;
 189 }
 190
 191 static void orangefs_invalidatepage(struct page *page,
 192                                  unsigned int offset,
 193                                  unsigned int length)
 194 {
 195         struct orangefs_write_range *wr;
 196         wr = (struct orangefs_write_range *)page_private(page);
 197
 198         if (offset == 0 && length == PAGE_SIZE) {
 199                 kfree((struct orangefs_write_range *)page_private(page));
 200                 set_page_private(page, 0);
 201                 ClearPagePrivate(page);
 202                 put_page(page);
 203         /* write range entirely within invalidate range (or equal) */
 204         } else if (page_offset(page) + offset <= wr->pos &&
 205             wr->pos + wr->len <= page_offset(page) + offset + length) {
 206                 kfree((struct orangefs_write_range *)page_private(page));
 207                 set_page_private(page, 0);
 208                 ClearPagePrivate(page);
 209                 put_page(page);
 210                 /* XXX is this right? only caller in fs */
 211                 cancel_dirty_page(page);
 212         /* invalidate range chops off end of write range */
 213         } else if (wr->pos < page_offset(page) + offset &&
 214             wr->pos + wr->len <= page_offset(page) + offset + length &&
 215              page_offset(page) + offset < wr->pos + wr->len) {
 216                 size_t x;
 217                 x = wr->pos + wr->len - (page_offset(page) + offset);
 218                 WARN_ON(x > wr->len);
 219                 wr->len -= x;
 220                 wr->uid = current_fsuid();
 221                 wr->gid = current_fsgid();
 222         /* invalidate range chops off beginning of write range */
 223         } else if (page_offset(page) + offset <= wr->pos &&
 224             page_offset(page) + offset + length < wr->pos + wr->len &&
 225             wr->pos < page_offset(page) + offset + length) {
 226                 size_t x;
 227                 x = page_offset(page) + offset + length - wr->pos;
 228                 WARN_ON(x > wr->len);
 229                 wr->pos += x;
 230                 wr->len -= x;
 231                 wr->uid = current_fsuid();
 232                 wr->gid = current_fsgid();
 233         /* invalidate range entirely within write range (punch hole) */
 234         } else if (wr->pos < page_offset(page) + offset &&
 235             page_offset(page) + offset + length < wr->pos + wr->len) {
 236                 /* XXX what do we do here... should not WARN_ON */
 237                 WARN_ON(1);
 238                 /* punch hole */
 239                 /*
 240                  * should we just ignore this and write it out anyway?
 241                  * it hardly makes sense
 242                  */
 243         /* non-overlapping ranges */
 244         } else {
 245                 /* WARN if they do overlap */
 246                 if (!((page_offset(page) + offset + length <= wr->pos) ^
 247                     (wr->pos + wr->len <= page_offset(page) + offset))) {
 248                         WARN_ON(1);
 249                         printk("invalidate range offset %llu length %u\n",
 250                             page_offset(page) + offset, length);
 251                         printk("write range offset %llu length %zu\n",
 252                             wr->pos, wr->len);
 253                 }
 254         }
 255 }
 256
 257 static int orangefs_releasepage(struct page *page, gfp_t foo)
 258 {
 259         return !PagePrivate(page);
 260 }
 261
 262 static void orangefs_freepage(struct page *page)
 263 {
 264         if (PagePrivate(page)) {
 265                 kfree((struct orangefs_write_range *)page_private(page));
 266                 set_page_private(page, 0);
 267                 ClearPagePrivate(page);
 268                 put_page(page);
 269         }
 270 }
 271
 272 static int orangefs_launder_page(struct page *page)
 273 {
 274         int r = 0;
 275         struct writeback_control wbc = {
 276                 .sync_mode = WB_SYNC_ALL,
 277                 .nr_to_write = 0,
 278         };
 279         wait_on_page_writeback(page);
 280         if (clear_page_dirty_for_io(page)) {
 281                 r = orangefs_writepage_locked(page, &wbc);
 282                 end_page_writeback(page);
 283         }
 284         return r;
 285 }
 286
 287 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
 288                                   struct iov_iter *iter)
 289 {
 290         /*
 291          * Comment from original do_readv_writev:
 292          * Common entry point for read/write/readv/writev
 293          * This function will dispatch it to either the direct I/O
 294          * or buffered I/O path depending on the mount options and/or
 295          * augmented/extended metadata attached to the file.
 296          * Note: File extended attributes override any mount options.
 297          */
 298         struct file *file = iocb->ki_filp;
 299         loff_t pos = iocb->ki_pos;
 300         enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
 301             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
 302         loff_t *offset = &pos;
 303         struct inode *inode = file->f_mapping->host;
 304         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 305         struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
 306         size_t count = iov_iter_count(iter);
 307         ssize_t total_count = 0;
 308         ssize_t ret = -EINVAL;
 309         int i = 0;
 310
 311         gossip_debug(GOSSIP_FILE_DEBUG,
 312                 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
 313                 __func__,
 314                 handle,
 315                 (int)count);
 316
 317         if (type == ORANGEFS_IO_WRITE) {
 318                 gossip_debug(GOSSIP_FILE_DEBUG,
 319                              "%s(%pU): proceeding with offset : %llu, "
 320                              "size %d\n",
 321                              __func__,
 322                              handle,
 323                              llu(*offset),
 324                              (int)count);
 325         }
 326
 327         if (count == 0) {
 328                 ret = 0;
 329                 goto out;
 330         }
 331
 332         while (iov_iter_count(iter)) {
 333                 size_t each_count = iov_iter_count(iter);
 334                 size_t amt_complete;
 335                 i++;
 336
 337                 /* how much to transfer in this loop iteration */
 338                 if (each_count > orangefs_bufmap_size_query())
 339                         each_count = orangefs_bufmap_size_query();
 340
 341                 gossip_debug(GOSSIP_FILE_DEBUG,
 342                              "%s(%pU): size of each_count(%d)\n",
 343                              __func__,
 344                              handle,
 345                              (int)each_count);
 346                 gossip_debug(GOSSIP_FILE_DEBUG,
 347                              "%s(%pU): BEFORE wait_for_io: offset is %d\n",
 348                              __func__,
 349                              handle,
 350                              (int)*offset);
 351
 352                 ret = wait_for_direct_io(type, inode, offset, iter,
 353                                 each_count, 0, NULL);
 354                 gossip_debug(GOSSIP_FILE_DEBUG,
 355                              "%s(%pU): return from wait_for_io:%d\n",
 356                              __func__,
 357                              handle,
 358                              (int)ret);
 359
 360                 if (ret < 0)
 361                         goto out;
 362
 363                 *offset += ret;
 364                 total_count += ret;
 365                 amt_complete = ret;
 366
 367                 gossip_debug(GOSSIP_FILE_DEBUG,
 368                              "%s(%pU): AFTER wait_for_io: offset is %d\n",
 369                              __func__,
 370                              handle,
 371                              (int)*offset);
 372
 373                 /*
 374                  * if we got a short I/O operations,
 375                  * fall out and return what we got so far
 376                  */
 377                 if (amt_complete < each_count)
 378                         break;
 379         } /*end while */
 380
 381 out:
 382         if (total_count > 0)
 383                 ret = total_count;
 384         if (ret > 0) {
 385                 if (type == ORANGEFS_IO_READ) {
 386                         file_accessed(file);
 387                 } else {
 388                         file_update_time(file);
 389                         if (*offset > i_size_read(inode))
 390                                 i_size_write(inode, *offset);
 391                 }
 392         }
 393
 394         gossip_debug(GOSSIP_FILE_DEBUG,
 395                      "%s(%pU): Value(%d) returned.\n",
 396                      __func__,
 397                      handle,
 398                      (int)ret);
 399
 400         return ret;
 401 }
 402
 403 /** ORANGEFS2 implementation of address space operations */
 404 static const struct address_space_operations orangefs_address_operations = {
 405         .writepage = orangefs_writepage,
 406         .readpage = orangefs_readpage,
 407         .set_page_dirty = __set_page_dirty_nobuffers,
 408         .write_begin = orangefs_write_begin,
 409         .write_end = orangefs_write_end,
 410         .invalidatepage = orangefs_invalidatepage,
 411         .releasepage = orangefs_releasepage,
 412         .freepage = orangefs_freepage,
 413         .launder_page = orangefs_launder_page,
 414         .direct_IO = orangefs_direct_IO,
 415 };
 416
 417 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
 418 {
 419         struct page *page = vmf->page;
 420         struct inode *inode = file_inode(vmf->vma->vm_file);
 421         vm_fault_t ret = VM_FAULT_LOCKED;
 422         struct orangefs_write_range *wr;
 423
 424         lock_page(page);
 425         if (PageDirty(page) && !PagePrivate(page)) {
 426                 /*
 427                  * Should be impossible.  If it happens, launder the page
 428                  * since we don't know what's dirty.  This will WARN in
 429                  * orangefs_writepage_locked.
 430                  */
 431                 if (orangefs_launder_page(page)) {
 432                         ret = VM_FAULT_RETRY;
 433                         goto out;
 434                 }
 435         }
 436         if (PagePrivate(page)) {
 437                 wr = (struct orangefs_write_range *)page_private(page);
 438                 if (uid_eq(wr->uid, current_fsuid()) &&
 439                     gid_eq(wr->gid, current_fsgid())) {
 440                         wr->pos = page_offset(page);
 441                         wr->len = PAGE_SIZE;
 442                         goto okay;
 443                 } else {
 444                         if (orangefs_launder_page(page)) {
 445                                 ret = VM_FAULT_RETRY;
 446                                 goto out;
 447                         }
 448                 }
 449         }
 450         wr = kmalloc(sizeof *wr, GFP_KERNEL);
 451         if (!wr) {
 452                 ret = VM_FAULT_RETRY;
 453                 goto out;
 454         }
 455         wr->pos = page_offset(page);
 456         wr->len = PAGE_SIZE;
 457         wr->uid = current_fsuid();
 458         wr->gid = current_fsgid();
 459         SetPagePrivate(page);
 460         set_page_private(page, (unsigned long)wr);
 461         get_page(page);
 462 okay:
 463
 464         sb_start_pagefault(inode->i_sb);
 465         file_update_time(vmf->vma->vm_file);
 466         if (page->mapping != inode->i_mapping) {
 467                 unlock_page(page);
 468                 ret = VM_FAULT_NOPAGE;
 469                 goto out;
 470         }
 471
 472         /*
 473          * We mark the page dirty already here so that when freeze is in
 474          * progress, we are guaranteed that writeback during freezing will
 475          * see the dirty page and writeprotect it again.
 476          */
 477         set_page_dirty(page);
 478         wait_for_stable_page(page);
 479 out:
 480         sb_end_pagefault(inode->i_sb);
 481         return ret;
 482 }
 483
 484 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
 485 {
 486         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
 487         struct orangefs_kernel_op_s *new_op;
 488         loff_t orig_size;
 489         int ret = -EINVAL;
 490
 491         gossip_debug(GOSSIP_INODE_DEBUG,
 492                      "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
 493                      __func__,
 494                      get_khandle_from_ino(inode),
 495                      &orangefs_inode->refn.khandle,
 496                      orangefs_inode->refn.fs_id,
 497                      iattr->ia_size);
 498
 499         /* Ensure that we have a up to date size, so we know if it changed. */
 500         ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
 501         if (ret == -ESTALE)
 502                 ret = -EIO;
 503         if (ret) {
 504                 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
 505                     __func__, ret);
 506                 return ret;
 507         }
 508         orig_size = i_size_read(inode);
 509
 510         truncate_setsize(inode, iattr->ia_size);
 511
 512         new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
 513         if (!new_op)
 514                 return -ENOMEM;
 515
 516         new_op->upcall.req.truncate.refn = orangefs_inode->refn;
 517         new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
 518
 519         ret = service_operation(new_op,
 520                 __func__,
 521                 get_interruptible_flag(inode));
 522
 523         /*
 524          * the truncate has no downcall members to retrieve, but
 525          * the status value tells us if it went through ok or not
 526          */
 527         gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
 528
 529         op_release(new_op);
 530
 531         if (ret != 0)
 532                 return ret;
 533
 534         if (orig_size != i_size_read(inode))
 535                 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
 536
 537         return ret;
 538 }
 539
 540 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
 541 {
 542         int ret;
 543
 544         if (iattr->ia_valid & ATTR_MODE) {
 545                 if (iattr->ia_mode & (S_ISVTX)) {
 546                         if (is_root_handle(inode)) {
 547                                 /*
 548                                  * allow sticky bit to be set on root (since
 549                                  * it shows up that way by default anyhow),
 550                                  * but don't show it to the server
 551                                  */
 552                                 iattr->ia_mode -= S_ISVTX;
 553                         } else {
 554                                 gossip_debug(GOSSIP_UTILS_DEBUG,
 555                                              "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
 556                                 return -EINVAL;
 557                         }
 558                 }
 559                 if (iattr->ia_mode & (S_ISUID)) {
 560                         gossip_debug(GOSSIP_UTILS_DEBUG,
 561                                      "Attempting to set setuid bit (not supported); returning EINVAL.\n");
 562                         return -EINVAL;
 563                 }
 564         }
 565
 566         if (iattr->ia_valid & ATTR_SIZE) {
 567                 ret = orangefs_setattr_size(inode, iattr);
 568                 if (ret)
 569                         goto out;
 570         }
 571
 572 again:
 573         spin_lock(&inode->i_lock);
 574         if (ORANGEFS_I(inode)->attr_valid) {
 575                 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
 576                     gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
 577                         ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 578                 } else {
 579                         spin_unlock(&inode->i_lock);
 580                         write_inode_now(inode, 1);
 581                         goto again;
 582                 }
 583         } else {
 584                 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
 585                 ORANGEFS_I(inode)->attr_uid = current_fsuid();
 586                 ORANGEFS_I(inode)->attr_gid = current_fsgid();
 587         }
 588         setattr_copy(inode, iattr);
 589         spin_unlock(&inode->i_lock);
 590         mark_inode_dirty(inode);
 591
 592         if (iattr->ia_valid & ATTR_MODE)
 593                 /* change mod on a file that has ACLs */
 594                 ret = posix_acl_chmod(inode, inode->i_mode);
 595
 596         ret = 0;
 597 out:
 598         return ret;
 599 }
 600
 601 /*
 602  * Change attributes of an object referenced by dentry.
 603  */
 604 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
 605 {
 606         int ret;
 607         gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
 608             dentry);
 609         ret = setattr_prepare(dentry, iattr);
 610         if (ret)
 611                 goto out;
 612         ret = __orangefs_setattr(d_inode(dentry), iattr);
 613         sync_inode_metadata(d_inode(dentry), 1);
 614 out:
 615         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
 616             ret);
 617         return ret;
 618 }
 619
 620 /*
 621  * Obtain attributes of an object given a dentry
 622  */
 623 int orangefs_getattr(const struct path *path, struct kstat *stat,
 624                      u32 request_mask, unsigned int flags)
 625 {
 626         int ret = -ENOENT;
 627         struct inode *inode = path->dentry->d_inode;
 628
 629         gossip_debug(GOSSIP_INODE_DEBUG,
 630                      "orangefs_getattr: called on %pd mask %u\n",
 631                      path->dentry, request_mask);
 632
 633         ret = orangefs_inode_getattr(inode,
 634             request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
 635         if (ret == 0) {
 636                 generic_fillattr(inode, stat);
 637
 638                 /* override block size reported to stat */
 639                 if (!(request_mask & STATX_SIZE))
 640                         stat->result_mask &= ~STATX_SIZE;
 641
 642                 stat->attributes_mask = STATX_ATTR_IMMUTABLE |
 643                     STATX_ATTR_APPEND;
 644                 if (inode->i_flags & S_IMMUTABLE)
 645                         stat->attributes |= STATX_ATTR_IMMUTABLE;
 646                 if (inode->i_flags & S_APPEND)
 647                         stat->attributes |= STATX_ATTR_APPEND;
 648         }
 649         return ret;
 650 }
 651
 652 int orangefs_permission(struct inode *inode, int mask)
 653 {
 654         int ret;
 655
 656         if (mask & MAY_NOT_BLOCK)
 657                 return -ECHILD;
 658
 659         gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
 660
 661         /* Make sure the permission (and other common attrs) are up to date. */
 662         ret = orangefs_inode_getattr(inode, 0);
 663         if (ret < 0)
 664                 return ret;
 665
 666         return generic_permission(inode, mask);
 667 }
 668
 669 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
 670 {
 671         struct iattr iattr;
 672         gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
 673             get_khandle_from_ino(inode));
 674         generic_update_time(inode, time, flags);
 675         memset(&iattr, 0, sizeof iattr);
 676         if (flags & S_ATIME)
 677                 iattr.ia_valid |= ATTR_ATIME;
 678         if (flags & S_CTIME)
 679                 iattr.ia_valid |= ATTR_CTIME;
 680         if (flags & S_MTIME)
 681                 iattr.ia_valid |= ATTR_MTIME;
 682         return __orangefs_setattr(inode, &iattr);
 683 }
 684
 685 /* ORANGEFS2 implementation of VFS inode operations for files */
 686 static const struct inode_operations orangefs_file_inode_operations = {
 687         .get_acl = orangefs_get_acl,
 688         .set_acl = orangefs_set_acl,
 689         .setattr = orangefs_setattr,
 690         .getattr = orangefs_getattr,
 691         .listxattr = orangefs_listxattr,
 692         .permission = orangefs_permission,
 693         .update_time = orangefs_update_time,
 694 };
 695
 696 static int orangefs_init_iops(struct inode *inode)
 697 {
 698         inode->i_mapping->a_ops = &orangefs_address_operations;
 699
 700         switch (inode->i_mode & S_IFMT) {
 701         case S_IFREG:
 702                 inode->i_op = &orangefs_file_inode_operations;
 703                 inode->i_fop = &orangefs_file_operations;
 704                 break;
 705         case S_IFLNK:
 706                 inode->i_op = &orangefs_symlink_inode_operations;
 707                 break;
 708         case S_IFDIR:
 709                 inode->i_op = &orangefs_dir_inode_operations;
 710                 inode->i_fop = &orangefs_dir_operations;
 711                 break;
 712         default:
 713                 gossip_debug(GOSSIP_INODE_DEBUG,
 714                              "%s: unsupported mode\n",
 715                              __func__);
 716                 return -EINVAL;
 717         }
 718
 719         return 0;
 720 }
 721
 722 /*
 723  * Given an ORANGEFS object identifier (fsid, handle), convert it into
 724  * a ino_t type that will be used as a hash-index from where the handle will
 725  * be searched for in the VFS hash table of inodes.
 726  */
 727 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
 728 {
 729         if (!ref)
 730                 return 0;
 731         return orangefs_khandle_to_ino(&(ref->khandle));
 732 }
 733
 734 /*
 735  * Called to set up an inode from iget5_locked.
 736  */
 737 static int orangefs_set_inode(struct inode *inode, void *data)
 738 {
 739         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 740         ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
 741         ORANGEFS_I(inode)->refn.khandle = ref->khandle;
 742         ORANGEFS_I(inode)->attr_valid = 0;
 743         hash_init(ORANGEFS_I(inode)->xattr_cache);
 744         return 0;
 745 }
 746
 747 /*
 748  * Called to determine if handles match.
 749  */
 750 static int orangefs_test_inode(struct inode *inode, void *data)
 751 {
 752         struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
 753         struct orangefs_inode_s *orangefs_inode = NULL;
 754
 755         orangefs_inode = ORANGEFS_I(inode);
 756         /* test handles and fs_ids... */
 757         return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
 758                                 &(ref->khandle)) &&
 759                         orangefs_inode->refn.fs_id == ref->fs_id);
 760 }
 761
 762 /*
 763  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
 764  * file handle.
 765  *
 766  * @sb: the file system super block instance.
 767  * @ref: The ORANGEFS object for which we are trying to locate an inode.
 768  */
 769 struct inode *orangefs_iget(struct super_block *sb,
 770                 struct orangefs_object_kref *ref)
 771 {
 772         struct inode *inode = NULL;
 773         unsigned long hash;
 774         int error;
 775
 776         hash = orangefs_handle_hash(ref);
 777         inode = iget5_locked(sb,
 778                         hash,
 779                         orangefs_test_inode,
 780                         orangefs_set_inode,
 781                         ref);
 782
 783         if (!inode)
 784                 return ERR_PTR(-ENOMEM);
 785
 786         if (!(inode->i_state & I_NEW))
 787                 return inode;
 788
 789         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
 790         if (error) {
 791                 iget_failed(inode);
 792                 return ERR_PTR(error);
 793         }
 794
 795         inode->i_ino = hash;    /* needed for stat etc */
 796         orangefs_init_iops(inode);
 797         unlock_new_inode(inode);
 798
 799         gossip_debug(GOSSIP_INODE_DEBUG,
 800                      "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
 801                      &ref->khandle,
 802                      ref->fs_id,
 803                      hash,
 804                      inode->i_ino);
 805
 806         return inode;
 807 }
 808
 809 /*
 810  * Allocate an inode for a newly created file and insert it into the inode hash.
 811  */
 812 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
 813                 int mode, dev_t dev, struct orangefs_object_kref *ref)
 814 {
 815         unsigned long hash = orangefs_handle_hash(ref);
 816         struct inode *inode;
 817         int error;
 818
 819         gossip_debug(GOSSIP_INODE_DEBUG,
 820                      "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
 821                      __func__,
 822                      sb,
 823                      MAJOR(dev),
 824                      MINOR(dev),
 825                      mode);
 826
 827         inode = new_inode(sb);
 828         if (!inode)
 829                 return ERR_PTR(-ENOMEM);
 830
 831         orangefs_set_inode(inode, ref);
 832         inode->i_ino = hash;    /* needed for stat etc */
 833
 834         error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
 835         if (error)
 836                 goto out_iput;
 837
 838         orangefs_init_iops(inode);
 839         inode->i_rdev = dev;
 840
 841         error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
 842         if (error < 0)
 843                 goto out_iput;
 844
 845         gossip_debug(GOSSIP_INODE_DEBUG,
 846                      "Initializing ACL's for inode %pU\n",
 847                      get_khandle_from_ino(inode));
 848         orangefs_init_acl(inode, dir);
 849         return inode;
 850
 851 out_iput:
 852         iput(inode);
 853         return ERR_PTR(error);
 854 }