xref: /openbmc/linux/fs/orangefs/inode.c (revision 236bdb88)
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 page *page;
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 ((page = readahead_page(rac))) {
279 		page_endio(page, false, ret);
280 		put_page(page);
281 	}
282 }
283 
284 static int orangefs_read_folio(struct file *file, struct folio *folio)
285 {
286 	struct inode *inode = folio->mapping->host;
287 	struct iov_iter iter;
288 	struct bio_vec bv;
289 	ssize_t ret;
290 	loff_t off; /* offset of this folio in the file */
291 
292 	if (folio_test_dirty(folio))
293 		orangefs_launder_folio(folio);
294 
295 	off = folio_pos(folio);
296 	bvec_set_folio(&bv, folio, folio_size(folio), 0);
297 	iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio));
298 
299 	ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
300 			folio_size(folio), inode->i_size, NULL, NULL, file);
301 	/* this will only zero remaining unread portions of the folio data */
302 	iov_iter_zero(~0U, &iter);
303 	/* takes care of potential aliasing */
304 	flush_dcache_folio(folio);
305 	if (ret < 0) {
306 		folio_set_error(folio);
307 	} else {
308 		folio_mark_uptodate(folio);
309 		ret = 0;
310 	}
311 	/* unlock the folio after the ->read_folio() routine completes */
312 	folio_unlock(folio);
313         return ret;
314 }
315 
316 static int orangefs_write_begin(struct file *file,
317 		struct address_space *mapping, loff_t pos, unsigned len,
318 		struct page **pagep, void **fsdata)
319 {
320 	struct orangefs_write_range *wr;
321 	struct folio *folio;
322 	struct page *page;
323 	pgoff_t index;
324 	int ret;
325 
326 	index = pos >> PAGE_SHIFT;
327 
328 	page = grab_cache_page_write_begin(mapping, index);
329 	if (!page)
330 		return -ENOMEM;
331 
332 	*pagep = page;
333 	folio = page_folio(page);
334 
335 	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
336 		/*
337 		 * Should be impossible.  If it happens, launder the page
338 		 * since we don't know what's dirty.  This will WARN in
339 		 * orangefs_writepage_locked.
340 		 */
341 		ret = orangefs_launder_folio(folio);
342 		if (ret)
343 			return ret;
344 	}
345 	if (folio_test_private(folio)) {
346 		struct orangefs_write_range *wr;
347 		wr = folio_get_private(folio);
348 		if (wr->pos + wr->len == pos &&
349 		    uid_eq(wr->uid, current_fsuid()) &&
350 		    gid_eq(wr->gid, current_fsgid())) {
351 			wr->len += len;
352 			goto okay;
353 		} else {
354 			ret = orangefs_launder_folio(folio);
355 			if (ret)
356 				return ret;
357 		}
358 	}
359 
360 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
361 	if (!wr)
362 		return -ENOMEM;
363 
364 	wr->pos = pos;
365 	wr->len = len;
366 	wr->uid = current_fsuid();
367 	wr->gid = current_fsgid();
368 	folio_attach_private(folio, wr);
369 okay:
370 	return 0;
371 }
372 
373 static int orangefs_write_end(struct file *file, struct address_space *mapping,
374     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
375 {
376 	struct inode *inode = page->mapping->host;
377 	loff_t last_pos = pos + copied;
378 
379 	/*
380 	 * No need to use i_size_read() here, the i_size
381 	 * cannot change under us because we hold the i_mutex.
382 	 */
383 	if (last_pos > inode->i_size)
384 		i_size_write(inode, last_pos);
385 
386 	/* zero the stale part of the page if we did a short copy */
387 	if (!PageUptodate(page)) {
388 		unsigned from = pos & (PAGE_SIZE - 1);
389 		if (copied < len) {
390 			zero_user(page, from + copied, len - copied);
391 		}
392 		/* Set fully written pages uptodate. */
393 		if (pos == page_offset(page) &&
394 		    (len == PAGE_SIZE || pos + len == inode->i_size)) {
395 			zero_user_segment(page, from + copied, PAGE_SIZE);
396 			SetPageUptodate(page);
397 		}
398 	}
399 
400 	set_page_dirty(page);
401 	unlock_page(page);
402 	put_page(page);
403 
404 	mark_inode_dirty_sync(file_inode(file));
405 	return copied;
406 }
407 
408 static void orangefs_invalidate_folio(struct folio *folio,
409 				 size_t offset, size_t length)
410 {
411 	struct orangefs_write_range *wr = folio_get_private(folio);
412 
413 	if (offset == 0 && length == PAGE_SIZE) {
414 		kfree(folio_detach_private(folio));
415 		return;
416 	/* write range entirely within invalidate range (or equal) */
417 	} else if (folio_pos(folio) + offset <= wr->pos &&
418 	    wr->pos + wr->len <= folio_pos(folio) + offset + length) {
419 		kfree(folio_detach_private(folio));
420 		/* XXX is this right? only caller in fs */
421 		folio_cancel_dirty(folio);
422 		return;
423 	/* invalidate range chops off end of write range */
424 	} else if (wr->pos < folio_pos(folio) + offset &&
425 	    wr->pos + wr->len <= folio_pos(folio) + offset + length &&
426 	     folio_pos(folio) + offset < wr->pos + wr->len) {
427 		size_t x;
428 		x = wr->pos + wr->len - (folio_pos(folio) + offset);
429 		WARN_ON(x > wr->len);
430 		wr->len -= x;
431 		wr->uid = current_fsuid();
432 		wr->gid = current_fsgid();
433 	/* invalidate range chops off beginning of write range */
434 	} else if (folio_pos(folio) + offset <= wr->pos &&
435 	    folio_pos(folio) + offset + length < wr->pos + wr->len &&
436 	    wr->pos < folio_pos(folio) + offset + length) {
437 		size_t x;
438 		x = folio_pos(folio) + offset + length - wr->pos;
439 		WARN_ON(x > wr->len);
440 		wr->pos += x;
441 		wr->len -= x;
442 		wr->uid = current_fsuid();
443 		wr->gid = current_fsgid();
444 	/* invalidate range entirely within write range (punch hole) */
445 	} else if (wr->pos < folio_pos(folio) + offset &&
446 	    folio_pos(folio) + offset + length < wr->pos + wr->len) {
447 		/* XXX what do we do here... should not WARN_ON */
448 		WARN_ON(1);
449 		/* punch hole */
450 		/*
451 		 * should we just ignore this and write it out anyway?
452 		 * it hardly makes sense
453 		 */
454 		return;
455 	/* non-overlapping ranges */
456 	} else {
457 		/* WARN if they do overlap */
458 		if (!((folio_pos(folio) + offset + length <= wr->pos) ^
459 		    (wr->pos + wr->len <= folio_pos(folio) + offset))) {
460 			WARN_ON(1);
461 			printk("invalidate range offset %llu length %zu\n",
462 			    folio_pos(folio) + offset, length);
463 			printk("write range offset %llu length %zu\n",
464 			    wr->pos, wr->len);
465 		}
466 		return;
467 	}
468 
469 	/*
470 	 * Above there are returns where wr is freed or where we WARN.
471 	 * Thus the following runs if wr was modified above.
472 	 */
473 
474 	orangefs_launder_folio(folio);
475 }
476 
477 static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
478 {
479 	return !folio_test_private(folio);
480 }
481 
482 static void orangefs_free_folio(struct folio *folio)
483 {
484 	kfree(folio_detach_private(folio));
485 }
486 
487 static int orangefs_launder_folio(struct folio *folio)
488 {
489 	int r = 0;
490 	struct writeback_control wbc = {
491 		.sync_mode = WB_SYNC_ALL,
492 		.nr_to_write = 0,
493 	};
494 	folio_wait_writeback(folio);
495 	if (folio_clear_dirty_for_io(folio)) {
496 		r = orangefs_writepage_locked(&folio->page, &wbc);
497 		folio_end_writeback(folio);
498 	}
499 	return r;
500 }
501 
502 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
503 				  struct iov_iter *iter)
504 {
505 	/*
506 	 * Comment from original do_readv_writev:
507 	 * Common entry point for read/write/readv/writev
508 	 * This function will dispatch it to either the direct I/O
509 	 * or buffered I/O path depending on the mount options and/or
510 	 * augmented/extended metadata attached to the file.
511 	 * Note: File extended attributes override any mount options.
512 	 */
513 	struct file *file = iocb->ki_filp;
514 	loff_t pos = iocb->ki_pos;
515 	enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
516             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
517 	loff_t *offset = &pos;
518 	struct inode *inode = file->f_mapping->host;
519 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
520 	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
521 	size_t count = iov_iter_count(iter);
522 	ssize_t total_count = 0;
523 	ssize_t ret = -EINVAL;
524 
525 	gossip_debug(GOSSIP_FILE_DEBUG,
526 		"%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
527 		__func__,
528 		handle,
529 		(int)count);
530 
531 	if (type == ORANGEFS_IO_WRITE) {
532 		gossip_debug(GOSSIP_FILE_DEBUG,
533 			     "%s(%pU): proceeding with offset : %llu, "
534 			     "size %d\n",
535 			     __func__,
536 			     handle,
537 			     llu(*offset),
538 			     (int)count);
539 	}
540 
541 	if (count == 0) {
542 		ret = 0;
543 		goto out;
544 	}
545 
546 	while (iov_iter_count(iter)) {
547 		size_t each_count = iov_iter_count(iter);
548 		size_t amt_complete;
549 
550 		/* how much to transfer in this loop iteration */
551 		if (each_count > orangefs_bufmap_size_query())
552 			each_count = orangefs_bufmap_size_query();
553 
554 		gossip_debug(GOSSIP_FILE_DEBUG,
555 			     "%s(%pU): size of each_count(%d)\n",
556 			     __func__,
557 			     handle,
558 			     (int)each_count);
559 		gossip_debug(GOSSIP_FILE_DEBUG,
560 			     "%s(%pU): BEFORE wait_for_io: offset is %d\n",
561 			     __func__,
562 			     handle,
563 			     (int)*offset);
564 
565 		ret = wait_for_direct_io(type, inode, offset, iter,
566 				each_count, 0, NULL, NULL, file);
567 		gossip_debug(GOSSIP_FILE_DEBUG,
568 			     "%s(%pU): return from wait_for_io:%d\n",
569 			     __func__,
570 			     handle,
571 			     (int)ret);
572 
573 		if (ret < 0)
574 			goto out;
575 
576 		*offset += ret;
577 		total_count += ret;
578 		amt_complete = ret;
579 
580 		gossip_debug(GOSSIP_FILE_DEBUG,
581 			     "%s(%pU): AFTER wait_for_io: offset is %d\n",
582 			     __func__,
583 			     handle,
584 			     (int)*offset);
585 
586 		/*
587 		 * if we got a short I/O operations,
588 		 * fall out and return what we got so far
589 		 */
590 		if (amt_complete < each_count)
591 			break;
592 	} /*end while */
593 
594 out:
595 	if (total_count > 0)
596 		ret = total_count;
597 	if (ret > 0) {
598 		if (type == ORANGEFS_IO_READ) {
599 			file_accessed(file);
600 		} else {
601 			file_update_time(file);
602 			if (*offset > i_size_read(inode))
603 				i_size_write(inode, *offset);
604 		}
605 	}
606 
607 	gossip_debug(GOSSIP_FILE_DEBUG,
608 		     "%s(%pU): Value(%d) returned.\n",
609 		     __func__,
610 		     handle,
611 		     (int)ret);
612 
613 	return ret;
614 }
615 
616 /** ORANGEFS2 implementation of address space operations */
617 static const struct address_space_operations orangefs_address_operations = {
618 	.writepage = orangefs_writepage,
619 	.readahead = orangefs_readahead,
620 	.read_folio = orangefs_read_folio,
621 	.writepages = orangefs_writepages,
622 	.dirty_folio = filemap_dirty_folio,
623 	.write_begin = orangefs_write_begin,
624 	.write_end = orangefs_write_end,
625 	.invalidate_folio = orangefs_invalidate_folio,
626 	.release_folio = orangefs_release_folio,
627 	.free_folio = orangefs_free_folio,
628 	.launder_folio = orangefs_launder_folio,
629 	.direct_IO = orangefs_direct_IO,
630 };
631 
632 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
633 {
634 	struct folio *folio = page_folio(vmf->page);
635 	struct inode *inode = file_inode(vmf->vma->vm_file);
636 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
637 	unsigned long *bitlock = &orangefs_inode->bitlock;
638 	vm_fault_t ret;
639 	struct orangefs_write_range *wr;
640 
641 	sb_start_pagefault(inode->i_sb);
642 
643 	if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
644 		ret = VM_FAULT_RETRY;
645 		goto out;
646 	}
647 
648 	folio_lock(folio);
649 	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
650 		/*
651 		 * Should be impossible.  If it happens, launder the folio
652 		 * since we don't know what's dirty.  This will WARN in
653 		 * orangefs_writepage_locked.
654 		 */
655 		if (orangefs_launder_folio(folio)) {
656 			ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
657 			goto out;
658 		}
659 	}
660 	if (folio_test_private(folio)) {
661 		wr = folio_get_private(folio);
662 		if (uid_eq(wr->uid, current_fsuid()) &&
663 		    gid_eq(wr->gid, current_fsgid())) {
664 			wr->pos = page_offset(vmf->page);
665 			wr->len = PAGE_SIZE;
666 			goto okay;
667 		} else {
668 			if (orangefs_launder_folio(folio)) {
669 				ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
670 				goto out;
671 			}
672 		}
673 	}
674 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
675 	if (!wr) {
676 		ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
677 		goto out;
678 	}
679 	wr->pos = page_offset(vmf->page);
680 	wr->len = PAGE_SIZE;
681 	wr->uid = current_fsuid();
682 	wr->gid = current_fsgid();
683 	folio_attach_private(folio, wr);
684 okay:
685 
686 	file_update_time(vmf->vma->vm_file);
687 	if (folio->mapping != inode->i_mapping) {
688 		folio_unlock(folio);
689 		ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
690 		goto out;
691 	}
692 
693 	/*
694 	 * We mark the folio dirty already here so that when freeze is in
695 	 * progress, we are guaranteed that writeback during freezing will
696 	 * see the dirty folio and writeprotect it again.
697 	 */
698 	folio_mark_dirty(folio);
699 	folio_wait_stable(folio);
700 	ret = VM_FAULT_LOCKED;
701 out:
702 	sb_end_pagefault(inode->i_sb);
703 	return ret;
704 }
705 
706 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
707 {
708 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
709 	struct orangefs_kernel_op_s *new_op;
710 	loff_t orig_size;
711 	int ret = -EINVAL;
712 
713 	gossip_debug(GOSSIP_INODE_DEBUG,
714 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
715 		     __func__,
716 		     get_khandle_from_ino(inode),
717 		     &orangefs_inode->refn.khandle,
718 		     orangefs_inode->refn.fs_id,
719 		     iattr->ia_size);
720 
721 	/* Ensure that we have a up to date size, so we know if it changed. */
722 	ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
723 	if (ret == -ESTALE)
724 		ret = -EIO;
725 	if (ret) {
726 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
727 		    __func__, ret);
728 		return ret;
729 	}
730 	orig_size = i_size_read(inode);
731 
732 	/* This is truncate_setsize in a different order. */
733 	truncate_pagecache(inode, iattr->ia_size);
734 	i_size_write(inode, iattr->ia_size);
735 	if (iattr->ia_size > orig_size)
736 		pagecache_isize_extended(inode, orig_size, iattr->ia_size);
737 
738 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
739 	if (!new_op)
740 		return -ENOMEM;
741 
742 	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
743 	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
744 
745 	ret = service_operation(new_op,
746 		__func__,
747 		get_interruptible_flag(inode));
748 
749 	/*
750 	 * the truncate has no downcall members to retrieve, but
751 	 * the status value tells us if it went through ok or not
752 	 */
753 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
754 
755 	op_release(new_op);
756 
757 	if (ret != 0)
758 		return ret;
759 
760 	if (orig_size != i_size_read(inode))
761 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
762 
763 	return ret;
764 }
765 
766 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
767 {
768 	int ret;
769 
770 	if (iattr->ia_valid & ATTR_MODE) {
771 		if (iattr->ia_mode & (S_ISVTX)) {
772 			if (is_root_handle(inode)) {
773 				/*
774 				 * allow sticky bit to be set on root (since
775 				 * it shows up that way by default anyhow),
776 				 * but don't show it to the server
777 				 */
778 				iattr->ia_mode -= S_ISVTX;
779 			} else {
780 				gossip_debug(GOSSIP_UTILS_DEBUG,
781 					     "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
782 				ret = -EINVAL;
783 				goto out;
784 			}
785 		}
786 		if (iattr->ia_mode & (S_ISUID)) {
787 			gossip_debug(GOSSIP_UTILS_DEBUG,
788 				     "Attempting to set setuid bit (not supported); returning EINVAL.\n");
789 			ret = -EINVAL;
790 			goto out;
791 		}
792 	}
793 
794 	if (iattr->ia_valid & ATTR_SIZE) {
795 		ret = orangefs_setattr_size(inode, iattr);
796 		if (ret)
797 			goto out;
798 	}
799 
800 again:
801 	spin_lock(&inode->i_lock);
802 	if (ORANGEFS_I(inode)->attr_valid) {
803 		if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
804 		    gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
805 			ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
806 		} else {
807 			spin_unlock(&inode->i_lock);
808 			write_inode_now(inode, 1);
809 			goto again;
810 		}
811 	} else {
812 		ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
813 		ORANGEFS_I(inode)->attr_uid = current_fsuid();
814 		ORANGEFS_I(inode)->attr_gid = current_fsgid();
815 	}
816 	setattr_copy(&nop_mnt_idmap, inode, iattr);
817 	spin_unlock(&inode->i_lock);
818 	mark_inode_dirty(inode);
819 
820 	ret = 0;
821 out:
822 	return ret;
823 }
824 
825 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr)
826 {
827 	int ret;
828 	struct inode *inode = d_inode(dentry);
829 
830 	ret = __orangefs_setattr(inode, iattr);
831 	/* change mode on a file that has ACLs */
832 	if (!ret && (iattr->ia_valid & ATTR_MODE))
833 		ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
834 	return ret;
835 }
836 
837 /*
838  * Change attributes of an object referenced by dentry.
839  */
840 int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
841 		     struct iattr *iattr)
842 {
843 	int ret;
844 	gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
845 	    dentry);
846 	ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
847 	if (ret)
848 	        goto out;
849 	ret = __orangefs_setattr_mode(dentry, iattr);
850 	sync_inode_metadata(d_inode(dentry), 1);
851 out:
852 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
853 	    ret);
854 	return ret;
855 }
856 
857 /*
858  * Obtain attributes of an object given a dentry
859  */
860 int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path,
861 		     struct kstat *stat, u32 request_mask, unsigned int flags)
862 {
863 	int ret;
864 	struct inode *inode = path->dentry->d_inode;
865 
866 	gossip_debug(GOSSIP_INODE_DEBUG,
867 		     "orangefs_getattr: called on %pd mask %u\n",
868 		     path->dentry, request_mask);
869 
870 	ret = orangefs_inode_getattr(inode,
871 	    request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
872 	if (ret == 0) {
873 		generic_fillattr(&nop_mnt_idmap, inode, stat);
874 
875 		/* override block size reported to stat */
876 		if (!(request_mask & STATX_SIZE))
877 			stat->result_mask &= ~STATX_SIZE;
878 
879 		generic_fill_statx_attr(inode, stat);
880 	}
881 	return ret;
882 }
883 
884 int orangefs_permission(struct mnt_idmap *idmap,
885 			struct inode *inode, int mask)
886 {
887 	int ret;
888 
889 	if (mask & MAY_NOT_BLOCK)
890 		return -ECHILD;
891 
892 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
893 
894 	/* Make sure the permission (and other common attrs) are up to date. */
895 	ret = orangefs_inode_getattr(inode, 0);
896 	if (ret < 0)
897 		return ret;
898 
899 	return generic_permission(&nop_mnt_idmap, inode, mask);
900 }
901 
902 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
903 {
904 	struct iattr iattr;
905 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
906 	    get_khandle_from_ino(inode));
907 	generic_update_time(inode, time, flags);
908 	memset(&iattr, 0, sizeof iattr);
909         if (flags & S_ATIME)
910 		iattr.ia_valid |= ATTR_ATIME;
911 	if (flags & S_CTIME)
912 		iattr.ia_valid |= ATTR_CTIME;
913 	if (flags & S_MTIME)
914 		iattr.ia_valid |= ATTR_MTIME;
915 	return __orangefs_setattr(inode, &iattr);
916 }
917 
918 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
919 {
920 	u64 val = 0;
921 	int ret;
922 
923 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
924 		     dentry);
925 
926 	ret = orangefs_inode_getxattr(d_inode(dentry),
927 				      "user.pvfs2.meta_hint",
928 				      &val, sizeof(val));
929 	if (ret < 0 && ret != -ENODATA)
930 		return ret;
931 
932 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
933 
934 	fileattr_fill_flags(fa, val);
935 	return 0;
936 }
937 
938 static int orangefs_fileattr_set(struct mnt_idmap *idmap,
939 				 struct dentry *dentry, struct fileattr *fa)
940 {
941 	u64 val = 0;
942 
943 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
944 		     dentry);
945 	/*
946 	 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
947 	 * turned on for a file. The user is not allowed to turn on this bit,
948 	 * but the bit is present if the user first gets the flags and then
949 	 * updates the flags with some new settings. So, we ignore it in the
950 	 * following edit. bligon.
951 	 */
952 	if (fileattr_has_fsx(fa) ||
953 	    (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
954 		gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
955 			   __func__);
956 		return -EOPNOTSUPP;
957 	}
958 	val = fa->flags;
959 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
960 	return orangefs_inode_setxattr(d_inode(dentry),
961 				       "user.pvfs2.meta_hint",
962 				       &val, sizeof(val), 0);
963 }
964 
965 /* ORANGEFS2 implementation of VFS inode operations for files */
966 static const struct inode_operations orangefs_file_inode_operations = {
967 	.get_inode_acl = orangefs_get_acl,
968 	.set_acl = orangefs_set_acl,
969 	.setattr = orangefs_setattr,
970 	.getattr = orangefs_getattr,
971 	.listxattr = orangefs_listxattr,
972 	.permission = orangefs_permission,
973 	.update_time = orangefs_update_time,
974 	.fileattr_get = orangefs_fileattr_get,
975 	.fileattr_set = orangefs_fileattr_set,
976 };
977 
978 static int orangefs_init_iops(struct inode *inode)
979 {
980 	inode->i_mapping->a_ops = &orangefs_address_operations;
981 
982 	switch (inode->i_mode & S_IFMT) {
983 	case S_IFREG:
984 		inode->i_op = &orangefs_file_inode_operations;
985 		inode->i_fop = &orangefs_file_operations;
986 		break;
987 	case S_IFLNK:
988 		inode->i_op = &orangefs_symlink_inode_operations;
989 		break;
990 	case S_IFDIR:
991 		inode->i_op = &orangefs_dir_inode_operations;
992 		inode->i_fop = &orangefs_dir_operations;
993 		break;
994 	default:
995 		gossip_debug(GOSSIP_INODE_DEBUG,
996 			     "%s: unsupported mode\n",
997 			     __func__);
998 		return -EINVAL;
999 	}
1000 
1001 	return 0;
1002 }
1003 
1004 /*
1005  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1006  * a ino_t type that will be used as a hash-index from where the handle will
1007  * be searched for in the VFS hash table of inodes.
1008  */
1009 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1010 {
1011 	if (!ref)
1012 		return 0;
1013 	return orangefs_khandle_to_ino(&(ref->khandle));
1014 }
1015 
1016 /*
1017  * Called to set up an inode from iget5_locked.
1018  */
1019 static int orangefs_set_inode(struct inode *inode, void *data)
1020 {
1021 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1022 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1023 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1024 	ORANGEFS_I(inode)->attr_valid = 0;
1025 	hash_init(ORANGEFS_I(inode)->xattr_cache);
1026 	ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1027 	ORANGEFS_I(inode)->bitlock = 0;
1028 	return 0;
1029 }
1030 
1031 /*
1032  * Called to determine if handles match.
1033  */
1034 static int orangefs_test_inode(struct inode *inode, void *data)
1035 {
1036 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1037 	struct orangefs_inode_s *orangefs_inode = NULL;
1038 
1039 	orangefs_inode = ORANGEFS_I(inode);
1040 	/* test handles and fs_ids... */
1041 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1042 				&(ref->khandle)) &&
1043 			orangefs_inode->refn.fs_id == ref->fs_id);
1044 }
1045 
1046 /*
1047  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1048  * file handle.
1049  *
1050  * @sb: the file system super block instance.
1051  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1052  */
1053 struct inode *orangefs_iget(struct super_block *sb,
1054 		struct orangefs_object_kref *ref)
1055 {
1056 	struct inode *inode = NULL;
1057 	unsigned long hash;
1058 	int error;
1059 
1060 	hash = orangefs_handle_hash(ref);
1061 	inode = iget5_locked(sb,
1062 			hash,
1063 			orangefs_test_inode,
1064 			orangefs_set_inode,
1065 			ref);
1066 
1067 	if (!inode)
1068 		return ERR_PTR(-ENOMEM);
1069 
1070 	if (!(inode->i_state & I_NEW))
1071 		return inode;
1072 
1073 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1074 	if (error) {
1075 		iget_failed(inode);
1076 		return ERR_PTR(error);
1077 	}
1078 
1079 	inode->i_ino = hash;	/* needed for stat etc */
1080 	orangefs_init_iops(inode);
1081 	unlock_new_inode(inode);
1082 
1083 	gossip_debug(GOSSIP_INODE_DEBUG,
1084 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1085 		     &ref->khandle,
1086 		     ref->fs_id,
1087 		     hash,
1088 		     inode->i_ino);
1089 
1090 	return inode;
1091 }
1092 
1093 /*
1094  * Allocate an inode for a newly created file and insert it into the inode hash.
1095  */
1096 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1097 		umode_t mode, dev_t dev, struct orangefs_object_kref *ref)
1098 {
1099 	struct posix_acl *acl = NULL, *default_acl = NULL;
1100 	unsigned long hash = orangefs_handle_hash(ref);
1101 	struct inode *inode;
1102 	int error;
1103 
1104 	gossip_debug(GOSSIP_INODE_DEBUG,
1105 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1106 		     __func__,
1107 		     sb,
1108 		     MAJOR(dev),
1109 		     MINOR(dev),
1110 		     mode);
1111 
1112 	inode = new_inode(sb);
1113 	if (!inode)
1114 		return ERR_PTR(-ENOMEM);
1115 
1116 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
1117 	if (error)
1118 		goto out_iput;
1119 
1120 	orangefs_set_inode(inode, ref);
1121 	inode->i_ino = hash;	/* needed for stat etc */
1122 
1123 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1124 	if (error)
1125 		goto out_iput;
1126 
1127 	orangefs_init_iops(inode);
1128 	inode->i_rdev = dev;
1129 
1130 	if (default_acl) {
1131 		error = __orangefs_set_acl(inode, default_acl,
1132 					   ACL_TYPE_DEFAULT);
1133 		if (error)
1134 			goto out_iput;
1135 	}
1136 
1137 	if (acl) {
1138 		error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS);
1139 		if (error)
1140 			goto out_iput;
1141 	}
1142 
1143 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1144 	if (error < 0)
1145 		goto out_iput;
1146 
1147 	gossip_debug(GOSSIP_INODE_DEBUG,
1148 		     "Initializing ACL's for inode %pU\n",
1149 		     get_khandle_from_ino(inode));
1150 	if (mode != inode->i_mode) {
1151 		struct iattr iattr = {
1152 			.ia_mode = mode,
1153 			.ia_valid = ATTR_MODE,
1154 		};
1155 		inode->i_mode = mode;
1156 		__orangefs_setattr(inode, &iattr);
1157 		__posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode);
1158 	}
1159 	posix_acl_release(acl);
1160 	posix_acl_release(default_acl);
1161 	return inode;
1162 
1163 out_iput:
1164 	iput(inode);
1165 	posix_acl_release(acl);
1166 	posix_acl_release(default_acl);
1167 	return ERR_PTR(error);
1168 }
1169