1 // SPDX-License-Identifier: GPL-2.0
3 * (C) 2001 Clemson University and The University of Chicago
4 * Copyright 2018 Omnibond Systems, L.L.C.
6 * See COPYING in top-level directory.
10 * Linux VFS inode operations.
13 #include <linux/blkdev.h>
14 #include <linux/fileattr.h>
16 #include "orangefs-kernel.h"
17 #include "orangefs-bufmap.h"
19 static int orangefs_writepage_locked(struct page *page,
20 struct writeback_control *wbc)
22 struct inode *inode = page->mapping->host;
23 struct orangefs_write_range *wr = NULL;
30 set_page_writeback(page);
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);
37 if (off + wr->len > len)
43 off = page_offset(page);
44 if (off + PAGE_SIZE > len)
49 /* Should've been handled in orangefs_invalidate_folio. */
50 WARN_ON(off == len || off + wlen > len);
53 bvec_set_page(&bv, page, wlen, off % PAGE_SIZE);
54 iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen);
56 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
60 mapping_set_error(page->mapping, ret);
64 kfree(detach_page_private(page));
68 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
71 ret = orangefs_writepage_locked(page, wbc);
73 end_page_writeback(page);
77 struct orangefs_writepages {
88 static int orangefs_writepages_work(struct orangefs_writepages *ow,
89 struct writeback_control *wbc)
91 struct inode *inode = ow->pages[0]->mapping->host;
92 struct orangefs_write_range *wrp, wr;
99 len = i_size_read(inode);
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,
106 max(ow->off, page_offset(ow->pages[i])),
107 i == 0 ? ow->off - page_offset(ow->pages[i]) : 0);
109 iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->npages, ow->len);
111 WARN_ON(ow->off >= len);
112 if (ow->off + ow->len > len)
113 ow->len = len - ow->off;
118 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
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]);
131 end_page_writeback(ow->pages[i]);
132 unlock_page(ow->pages[i]);
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]);
144 end_page_writeback(ow->pages[i]);
145 unlock_page(ow->pages[i]);
151 static int orangefs_writepages_callback(struct folio *folio,
152 struct writeback_control *wbc, void *data)
154 struct orangefs_writepages *ow = data;
155 struct orangefs_write_range *wr = folio->private;
160 /* It's not private so there's nothing to write, right? */
161 printk("writepages_callback not private!\n");
167 if (ow->npages == 0) {
172 ow->pages[ow->npages++] = &folio->page;
176 if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
177 orangefs_writepages_work(ow, wbc);
182 if (ow->off + ow->len == wr->pos) {
184 ow->pages[ow->npages++] = &folio->page;
191 orangefs_writepages_work(ow, wbc);
194 ret = orangefs_writepage_locked(&folio->page, wbc);
195 mapping_set_error(folio->mapping, ret);
197 folio_end_writeback(folio);
199 if (ow->npages == ow->maxpages) {
200 orangefs_writepages_work(ow, wbc);
207 static int orangefs_writepages(struct address_space *mapping,
208 struct writeback_control *wbc)
210 struct orangefs_writepages *ow;
211 struct blk_plug plug;
213 ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
216 ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
217 ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
222 ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
228 blk_start_plug(&plug);
229 ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
231 ret = orangefs_writepages_work(ow, wbc);
232 blk_finish_plug(&plug);
239 static int orangefs_launder_folio(struct folio *);
241 static void orangefs_readahead(struct readahead_control *rac)
244 struct iov_iter iter;
245 struct inode *inode = rac->mapping->host;
246 struct xarray *i_pages;
248 loff_t new_start = readahead_pos(rac);
252 loff_t bytes_remaining = inode->i_size - readahead_pos(rac);
253 loff_t pages_remaining = bytes_remaining / PAGE_SIZE;
255 if (pages_remaining >= 1024)
257 else if (pages_remaining > readahead_count(rac))
258 new_len = bytes_remaining;
261 readahead_expand(rac, new_start, new_len);
263 offset = readahead_pos(rac);
264 i_pages = &rac->mapping->i_pages;
266 iov_iter_xarray(&iter, ITER_DEST, i_pages, offset, readahead_length(rac));
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__);
278 while ((folio = readahead_folio(rac))) {
280 folio_mark_uptodate(folio);
285 static int orangefs_read_folio(struct file *file, struct folio *folio)
287 struct inode *inode = folio->mapping->host;
288 struct iov_iter iter;
291 loff_t off; /* offset of this folio in the file */
293 if (folio_test_dirty(folio))
294 orangefs_launder_folio(folio);
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));
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);
307 folio_set_error(folio);
309 folio_mark_uptodate(folio);
312 /* unlock the folio after the ->read_folio() routine completes */
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)
321 struct orangefs_write_range *wr;
327 index = pos >> PAGE_SHIFT;
329 page = grab_cache_page_write_begin(mapping, index);
334 folio = page_folio(page);
336 if (folio_test_dirty(folio) && !folio_test_private(folio)) {
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.
342 ret = orangefs_launder_folio(folio);
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())) {
355 ret = orangefs_launder_folio(folio);
361 wr = kmalloc(sizeof *wr, GFP_KERNEL);
367 wr->uid = current_fsuid();
368 wr->gid = current_fsgid();
369 folio_attach_private(folio, wr);
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)
377 struct inode *inode = page->mapping->host;
378 loff_t last_pos = pos + copied;
381 * No need to use i_size_read() here, the i_size
382 * cannot change under us because we hold the i_mutex.
384 if (last_pos > inode->i_size)
385 i_size_write(inode, last_pos);
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);
391 zero_user(page, from + copied, len - copied);
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);
401 set_page_dirty(page);
405 mark_inode_dirty_sync(file_inode(file));
409 static void orangefs_invalidate_folio(struct folio *folio,
410 size_t offset, size_t length)
412 struct orangefs_write_range *wr = folio_get_private(folio);
414 if (offset == 0 && length == PAGE_SIZE) {
415 kfree(folio_detach_private(folio));
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);
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) {
429 x = wr->pos + wr->len - (folio_pos(folio) + offset);
430 WARN_ON(x > wr->len);
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) {
439 x = folio_pos(folio) + offset + length - wr->pos;
440 WARN_ON(x > wr->len);
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 */
452 * should we just ignore this and write it out anyway?
453 * it hardly makes sense
456 /* non-overlapping ranges */
458 /* WARN if they do overlap */
459 if (!((folio_pos(folio) + offset + length <= wr->pos) ^
460 (wr->pos + wr->len <= folio_pos(folio) + offset))) {
462 printk("invalidate range offset %llu length %zu\n",
463 folio_pos(folio) + offset, length);
464 printk("write range offset %llu length %zu\n",
471 * Above there are returns where wr is freed or where we WARN.
472 * Thus the following runs if wr was modified above.
475 orangefs_launder_folio(folio);
478 static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
480 return !folio_test_private(folio);
483 static void orangefs_free_folio(struct folio *folio)
485 kfree(folio_detach_private(folio));
488 static int orangefs_launder_folio(struct folio *folio)
491 struct writeback_control wbc = {
492 .sync_mode = WB_SYNC_ALL,
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);
503 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
504 struct iov_iter *iter)
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.
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;
526 gossip_debug(GOSSIP_FILE_DEBUG,
527 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
532 if (type == ORANGEFS_IO_WRITE) {
533 gossip_debug(GOSSIP_FILE_DEBUG,
534 "%s(%pU): proceeding with offset : %llu, "
547 while (iov_iter_count(iter)) {
548 size_t each_count = iov_iter_count(iter);
551 /* how much to transfer in this loop iteration */
552 if (each_count > orangefs_bufmap_size_query())
553 each_count = orangefs_bufmap_size_query();
555 gossip_debug(GOSSIP_FILE_DEBUG,
556 "%s(%pU): size of each_count(%d)\n",
560 gossip_debug(GOSSIP_FILE_DEBUG,
561 "%s(%pU): BEFORE wait_for_io: offset is %d\n",
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",
581 gossip_debug(GOSSIP_FILE_DEBUG,
582 "%s(%pU): AFTER wait_for_io: offset is %d\n",
588 * if we got a short I/O operations,
589 * fall out and return what we got so far
591 if (amt_complete < each_count)
599 if (type == ORANGEFS_IO_READ) {
602 file_update_time(file);
603 if (*offset > i_size_read(inode))
604 i_size_write(inode, *offset);
608 gossip_debug(GOSSIP_FILE_DEBUG,
609 "%s(%pU): Value(%d) returned.\n",
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,
633 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
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;
640 struct orangefs_write_range *wr;
642 sb_start_pagefault(inode->i_sb);
644 if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
645 ret = VM_FAULT_RETRY;
650 if (folio_test_dirty(folio) && !folio_test_private(folio)) {
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.
656 if (orangefs_launder_folio(folio)) {
657 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
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);
669 if (orangefs_launder_folio(folio)) {
670 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
675 wr = kmalloc(sizeof *wr, GFP_KERNEL);
677 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
680 wr->pos = page_offset(vmf->page);
682 wr->uid = current_fsuid();
683 wr->gid = current_fsgid();
684 folio_attach_private(folio, wr);
687 file_update_time(vmf->vma->vm_file);
688 if (folio->mapping != inode->i_mapping) {
690 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
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.
699 folio_mark_dirty(folio);
700 folio_wait_stable(folio);
701 ret = VM_FAULT_LOCKED;
703 sb_end_pagefault(inode->i_sb);
707 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
709 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
710 struct orangefs_kernel_op_s *new_op;
714 gossip_debug(GOSSIP_INODE_DEBUG,
715 "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
717 get_khandle_from_ino(inode),
718 &orangefs_inode->refn.khandle,
719 orangefs_inode->refn.fs_id,
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);
727 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
731 orig_size = i_size_read(inode);
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);
739 new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
743 new_op->upcall.req.truncate.refn = orangefs_inode->refn;
744 new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
746 ret = service_operation(new_op,
748 get_interruptible_flag(inode));
751 * the truncate has no downcall members to retrieve, but
752 * the status value tells us if it went through ok or not
754 gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
761 if (orig_size != i_size_read(inode))
762 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
767 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
771 if (iattr->ia_valid & ATTR_MODE) {
772 if (iattr->ia_mode & (S_ISVTX)) {
773 if (is_root_handle(inode)) {
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
779 iattr->ia_mode -= S_ISVTX;
781 gossip_debug(GOSSIP_UTILS_DEBUG,
782 "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
787 if (iattr->ia_mode & (S_ISUID)) {
788 gossip_debug(GOSSIP_UTILS_DEBUG,
789 "Attempting to set setuid bit (not supported); returning EINVAL.\n");
795 if (iattr->ia_valid & ATTR_SIZE) {
796 ret = orangefs_setattr_size(inode, iattr);
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;
808 spin_unlock(&inode->i_lock);
809 write_inode_now(inode, 1);
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();
817 setattr_copy(&nop_mnt_idmap, inode, iattr);
818 spin_unlock(&inode->i_lock);
819 mark_inode_dirty(inode);
826 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr)
829 struct inode *inode = d_inode(dentry);
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);
839 * Change attributes of an object referenced by dentry.
841 int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
845 gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
847 ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
850 ret = __orangefs_setattr_mode(dentry, iattr);
851 sync_inode_metadata(d_inode(dentry), 1);
853 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
859 * Obtain attributes of an object given a dentry
861 int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path,
862 struct kstat *stat, u32 request_mask, unsigned int flags)
865 struct inode *inode = path->dentry->d_inode;
867 gossip_debug(GOSSIP_INODE_DEBUG,
868 "orangefs_getattr: called on %pd mask %u\n",
869 path->dentry, request_mask);
871 ret = orangefs_inode_getattr(inode,
872 request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
874 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
876 /* override block size reported to stat */
877 if (!(request_mask & STATX_SIZE))
878 stat->result_mask &= ~STATX_SIZE;
880 generic_fill_statx_attr(inode, stat);
885 int orangefs_permission(struct mnt_idmap *idmap,
886 struct inode *inode, int mask)
890 if (mask & MAY_NOT_BLOCK)
893 gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
895 /* Make sure the permission (and other common attrs) are up to date. */
896 ret = orangefs_inode_getattr(inode, 0);
900 return generic_permission(&nop_mnt_idmap, inode, mask);
903 int orangefs_update_time(struct inode *inode, int flags)
907 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
908 get_khandle_from_ino(inode));
909 flags = generic_update_time(inode, flags);
910 memset(&iattr, 0, sizeof iattr);
912 iattr.ia_valid |= ATTR_ATIME;
914 iattr.ia_valid |= ATTR_CTIME;
916 iattr.ia_valid |= ATTR_MTIME;
917 return __orangefs_setattr(inode, &iattr);
920 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
925 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
928 ret = orangefs_inode_getxattr(d_inode(dentry),
929 "user.pvfs2.meta_hint",
931 if (ret < 0 && ret != -ENODATA)
934 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
936 fileattr_fill_flags(fa, val);
940 static int orangefs_fileattr_set(struct mnt_idmap *idmap,
941 struct dentry *dentry, struct fileattr *fa)
945 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
948 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
949 * turned on for a file. The user is not allowed to turn on this bit,
950 * but the bit is present if the user first gets the flags and then
951 * updates the flags with some new settings. So, we ignore it in the
952 * following edit. bligon.
954 if (fileattr_has_fsx(fa) ||
955 (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
956 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
961 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
962 return orangefs_inode_setxattr(d_inode(dentry),
963 "user.pvfs2.meta_hint",
964 &val, sizeof(val), 0);
967 /* ORANGEFS2 implementation of VFS inode operations for files */
968 static const struct inode_operations orangefs_file_inode_operations = {
969 .get_inode_acl = orangefs_get_acl,
970 .set_acl = orangefs_set_acl,
971 .setattr = orangefs_setattr,
972 .getattr = orangefs_getattr,
973 .listxattr = orangefs_listxattr,
974 .permission = orangefs_permission,
975 .update_time = orangefs_update_time,
976 .fileattr_get = orangefs_fileattr_get,
977 .fileattr_set = orangefs_fileattr_set,
980 static int orangefs_init_iops(struct inode *inode)
982 inode->i_mapping->a_ops = &orangefs_address_operations;
984 switch (inode->i_mode & S_IFMT) {
986 inode->i_op = &orangefs_file_inode_operations;
987 inode->i_fop = &orangefs_file_operations;
990 inode->i_op = &orangefs_symlink_inode_operations;
993 inode->i_op = &orangefs_dir_inode_operations;
994 inode->i_fop = &orangefs_dir_operations;
997 gossip_debug(GOSSIP_INODE_DEBUG,
998 "%s: unsupported mode\n",
1007 * Given an ORANGEFS object identifier (fsid, handle), convert it into
1008 * a ino_t type that will be used as a hash-index from where the handle will
1009 * be searched for in the VFS hash table of inodes.
1011 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1015 return orangefs_khandle_to_ino(&(ref->khandle));
1019 * Called to set up an inode from iget5_locked.
1021 static int orangefs_set_inode(struct inode *inode, void *data)
1023 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1024 ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1025 ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1026 ORANGEFS_I(inode)->attr_valid = 0;
1027 hash_init(ORANGEFS_I(inode)->xattr_cache);
1028 ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1029 ORANGEFS_I(inode)->bitlock = 0;
1034 * Called to determine if handles match.
1036 static int orangefs_test_inode(struct inode *inode, void *data)
1038 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1039 struct orangefs_inode_s *orangefs_inode = NULL;
1041 orangefs_inode = ORANGEFS_I(inode);
1042 /* test handles and fs_ids... */
1043 return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1045 orangefs_inode->refn.fs_id == ref->fs_id);
1049 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1052 * @sb: the file system super block instance.
1053 * @ref: The ORANGEFS object for which we are trying to locate an inode.
1055 struct inode *orangefs_iget(struct super_block *sb,
1056 struct orangefs_object_kref *ref)
1058 struct inode *inode = NULL;
1062 hash = orangefs_handle_hash(ref);
1063 inode = iget5_locked(sb,
1065 orangefs_test_inode,
1070 return ERR_PTR(-ENOMEM);
1072 if (!(inode->i_state & I_NEW))
1075 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1078 return ERR_PTR(error);
1081 inode->i_ino = hash; /* needed for stat etc */
1082 orangefs_init_iops(inode);
1083 unlock_new_inode(inode);
1085 gossip_debug(GOSSIP_INODE_DEBUG,
1086 "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1096 * Allocate an inode for a newly created file and insert it into the inode hash.
1098 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1099 umode_t mode, dev_t dev, struct orangefs_object_kref *ref)
1101 struct posix_acl *acl = NULL, *default_acl = NULL;
1102 unsigned long hash = orangefs_handle_hash(ref);
1103 struct inode *inode;
1106 gossip_debug(GOSSIP_INODE_DEBUG,
1107 "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1114 inode = new_inode(sb);
1116 return ERR_PTR(-ENOMEM);
1118 error = posix_acl_create(dir, &mode, &default_acl, &acl);
1122 orangefs_set_inode(inode, ref);
1123 inode->i_ino = hash; /* needed for stat etc */
1125 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1129 orangefs_init_iops(inode);
1130 inode->i_rdev = dev;
1133 error = __orangefs_set_acl(inode, default_acl,
1140 error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS);
1145 error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1149 gossip_debug(GOSSIP_INODE_DEBUG,
1150 "Initializing ACL's for inode %pU\n",
1151 get_khandle_from_ino(inode));
1152 if (mode != inode->i_mode) {
1153 struct iattr iattr = {
1155 .ia_valid = ATTR_MODE,
1157 inode->i_mode = mode;
1158 __orangefs_setattr(inode, &iattr);
1159 __posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode);
1161 posix_acl_release(acl);
1162 posix_acl_release(default_acl);
1167 posix_acl_release(acl);
1168 posix_acl_release(default_acl);
1169 return ERR_PTR(error);