xref: /linux/fs/orangefs/inode.c (revision 818448e9cf92e5c6b3c10320372eefcbe4174e4f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * (C) 2001 Clemson University and The University of Chicago
4  * Copyright 2018 Omnibond Systems, L.L.C.
5  *
6  * See COPYING in top-level directory.
7  */
8 
9 /*
10  *  Linux VFS inode operations.
11  */
12 
13 #include <linux/blkdev.h>
14 #include <linux/fileattr.h>
15 #include "protocol.h"
16 #include "orangefs-kernel.h"
17 #include "orangefs-bufmap.h"
18 
19 static int orangefs_writepage_locked(struct page *page,
20     struct writeback_control *wbc)
21 {
22 	struct inode *inode = page->mapping->host;
23 	struct orangefs_write_range *wr = NULL;
24 	struct iov_iter iter;
25 	struct bio_vec bv;
26 	size_t len, wlen;
27 	ssize_t ret;
28 	loff_t off;
29 
30 	set_page_writeback(page);
31 
32 	len = i_size_read(inode);
33 	if (PagePrivate(page)) {
34 		wr = (struct orangefs_write_range *)page_private(page);
35 		WARN_ON(wr->pos >= len);
36 		off = wr->pos;
37 		if (off + wr->len > len)
38 			wlen = len - off;
39 		else
40 			wlen = wr->len;
41 	} else {
42 		WARN_ON(1);
43 		off = page_offset(page);
44 		if (off + PAGE_SIZE > len)
45 			wlen = len - off;
46 		else
47 			wlen = PAGE_SIZE;
48 	}
49 	/* Should've been handled in orangefs_invalidate_folio. */
50 	WARN_ON(off == len || off + wlen > len);
51 
52 	bv.bv_page = page;
53 	bv.bv_len = wlen;
54 	bv.bv_offset = off % PAGE_SIZE;
55 	WARN_ON(wlen == 0);
56 	iov_iter_bvec(&iter, WRITE, &bv, 1, wlen);
57 
58 	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
59 	    len, wr, NULL, NULL);
60 	if (ret < 0) {
61 		SetPageError(page);
62 		mapping_set_error(page->mapping, ret);
63 	} else {
64 		ret = 0;
65 	}
66 	kfree(detach_page_private(page));
67 	return ret;
68 }
69 
70 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
71 {
72 	int ret;
73 	ret = orangefs_writepage_locked(page, wbc);
74 	unlock_page(page);
75 	end_page_writeback(page);
76 	return ret;
77 }
78 
79 struct orangefs_writepages {
80 	loff_t off;
81 	size_t len;
82 	kuid_t uid;
83 	kgid_t gid;
84 	int maxpages;
85 	int npages;
86 	struct page **pages;
87 	struct bio_vec *bv;
88 };
89 
90 static int orangefs_writepages_work(struct orangefs_writepages *ow,
91     struct writeback_control *wbc)
92 {
93 	struct inode *inode = ow->pages[0]->mapping->host;
94 	struct orangefs_write_range *wrp, wr;
95 	struct iov_iter iter;
96 	ssize_t ret;
97 	size_t len;
98 	loff_t off;
99 	int i;
100 
101 	len = i_size_read(inode);
102 
103 	for (i = 0; i < ow->npages; i++) {
104 		set_page_writeback(ow->pages[i]);
105 		ow->bv[i].bv_page = ow->pages[i];
106 		ow->bv[i].bv_len = min(page_offset(ow->pages[i]) + PAGE_SIZE,
107 		    ow->off + ow->len) -
108 		    max(ow->off, page_offset(ow->pages[i]));
109 		if (i == 0)
110 			ow->bv[i].bv_offset = ow->off -
111 			    page_offset(ow->pages[i]);
112 		else
113 			ow->bv[i].bv_offset = 0;
114 	}
115 	iov_iter_bvec(&iter, WRITE, ow->bv, ow->npages, ow->len);
116 
117 	WARN_ON(ow->off >= len);
118 	if (ow->off + ow->len > len)
119 		ow->len = len - ow->off;
120 
121 	off = ow->off;
122 	wr.uid = ow->uid;
123 	wr.gid = ow->gid;
124 	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
125 	    0, &wr, NULL, NULL);
126 	if (ret < 0) {
127 		for (i = 0; i < ow->npages; i++) {
128 			SetPageError(ow->pages[i]);
129 			mapping_set_error(ow->pages[i]->mapping, ret);
130 			if (PagePrivate(ow->pages[i])) {
131 				wrp = (struct orangefs_write_range *)
132 				    page_private(ow->pages[i]);
133 				ClearPagePrivate(ow->pages[i]);
134 				put_page(ow->pages[i]);
135 				kfree(wrp);
136 			}
137 			end_page_writeback(ow->pages[i]);
138 			unlock_page(ow->pages[i]);
139 		}
140 	} else {
141 		ret = 0;
142 		for (i = 0; i < ow->npages; i++) {
143 			if (PagePrivate(ow->pages[i])) {
144 				wrp = (struct orangefs_write_range *)
145 				    page_private(ow->pages[i]);
146 				ClearPagePrivate(ow->pages[i]);
147 				put_page(ow->pages[i]);
148 				kfree(wrp);
149 			}
150 			end_page_writeback(ow->pages[i]);
151 			unlock_page(ow->pages[i]);
152 		}
153 	}
154 	return ret;
155 }
156 
157 static int orangefs_writepages_callback(struct page *page,
158     struct writeback_control *wbc, void *data)
159 {
160 	struct orangefs_writepages *ow = data;
161 	struct orangefs_write_range *wr;
162 	int ret;
163 
164 	if (!PagePrivate(page)) {
165 		unlock_page(page);
166 		/* It's not private so there's nothing to write, right? */
167 		printk("writepages_callback not private!\n");
168 		BUG();
169 		return 0;
170 	}
171 	wr = (struct orangefs_write_range *)page_private(page);
172 
173 	ret = -1;
174 	if (ow->npages == 0) {
175 		ow->off = wr->pos;
176 		ow->len = wr->len;
177 		ow->uid = wr->uid;
178 		ow->gid = wr->gid;
179 		ow->pages[ow->npages++] = page;
180 		ret = 0;
181 		goto done;
182 	}
183 	if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
184 		orangefs_writepages_work(ow, wbc);
185 		ow->npages = 0;
186 		ret = -1;
187 		goto done;
188 	}
189 	if (ow->off + ow->len == wr->pos) {
190 		ow->len += wr->len;
191 		ow->pages[ow->npages++] = page;
192 		ret = 0;
193 		goto done;
194 	}
195 done:
196 	if (ret == -1) {
197 		if (ow->npages) {
198 			orangefs_writepages_work(ow, wbc);
199 			ow->npages = 0;
200 		}
201 		ret = orangefs_writepage_locked(page, wbc);
202 		mapping_set_error(page->mapping, ret);
203 		unlock_page(page);
204 		end_page_writeback(page);
205 	} else {
206 		if (ow->npages == ow->maxpages) {
207 			orangefs_writepages_work(ow, wbc);
208 			ow->npages = 0;
209 		}
210 	}
211 	return ret;
212 }
213 
214 static int orangefs_writepages(struct address_space *mapping,
215     struct writeback_control *wbc)
216 {
217 	struct orangefs_writepages *ow;
218 	struct blk_plug plug;
219 	int ret;
220 	ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
221 	if (!ow)
222 		return -ENOMEM;
223 	ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
224 	ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
225 	if (!ow->pages) {
226 		kfree(ow);
227 		return -ENOMEM;
228 	}
229 	ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
230 	if (!ow->bv) {
231 		kfree(ow->pages);
232 		kfree(ow);
233 		return -ENOMEM;
234 	}
235 	blk_start_plug(&plug);
236 	ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
237 	if (ow->npages)
238 		ret = orangefs_writepages_work(ow, wbc);
239 	blk_finish_plug(&plug);
240 	kfree(ow->pages);
241 	kfree(ow->bv);
242 	kfree(ow);
243 	return ret;
244 }
245 
246 static int orangefs_launder_folio(struct folio *);
247 
248 static void orangefs_readahead(struct readahead_control *rac)
249 {
250 	loff_t offset;
251 	struct iov_iter iter;
252 	struct inode *inode = rac->mapping->host;
253 	struct xarray *i_pages;
254 	struct page *page;
255 	loff_t new_start = readahead_pos(rac);
256 	int ret;
257 	size_t new_len = 0;
258 
259 	loff_t bytes_remaining = inode->i_size - readahead_pos(rac);
260 	loff_t pages_remaining = bytes_remaining / PAGE_SIZE;
261 
262 	if (pages_remaining >= 1024)
263 		new_len = 4194304;
264 	else if (pages_remaining > readahead_count(rac))
265 		new_len = bytes_remaining;
266 
267 	if (new_len)
268 		readahead_expand(rac, new_start, new_len);
269 
270 	offset = readahead_pos(rac);
271 	i_pages = &rac->mapping->i_pages;
272 
273 	iov_iter_xarray(&iter, READ, i_pages, offset, readahead_length(rac));
274 
275 	/* read in the pages. */
276 	if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode,
277 			&offset, &iter, readahead_length(rac),
278 			inode->i_size, NULL, NULL, rac->file)) < 0)
279 		gossip_debug(GOSSIP_FILE_DEBUG,
280 			"%s: wait_for_direct_io failed. \n", __func__);
281 	else
282 		ret = 0;
283 
284 	/* clean up. */
285 	while ((page = readahead_page(rac))) {
286 		page_endio(page, false, ret);
287 		put_page(page);
288 	}
289 }
290 
291 static int orangefs_read_folio(struct file *file, struct folio *folio)
292 {
293 	struct inode *inode = folio->mapping->host;
294 	struct iov_iter iter;
295 	struct bio_vec bv;
296 	ssize_t ret;
297 	loff_t off; /* offset of this folio in the file */
298 
299 	if (folio_test_dirty(folio))
300 		orangefs_launder_folio(folio);
301 
302 	off = folio_pos(folio);
303 	bv.bv_page = &folio->page;
304 	bv.bv_len = folio_size(folio);
305 	bv.bv_offset = 0;
306 	iov_iter_bvec(&iter, READ, &bv, 1, folio_size(folio));
307 
308 	ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
309 			folio_size(folio), inode->i_size, NULL, NULL, file);
310 	/* this will only zero remaining unread portions of the folio data */
311 	iov_iter_zero(~0U, &iter);
312 	/* takes care of potential aliasing */
313 	flush_dcache_folio(folio);
314 	if (ret < 0) {
315 		folio_set_error(folio);
316 	} else {
317 		folio_mark_uptodate(folio);
318 		ret = 0;
319 	}
320 	/* unlock the folio after the ->read_folio() routine completes */
321 	folio_unlock(folio);
322         return ret;
323 }
324 
325 static int orangefs_write_begin(struct file *file,
326 		struct address_space *mapping, loff_t pos, unsigned len,
327 		struct page **pagep, void **fsdata)
328 {
329 	struct orangefs_write_range *wr;
330 	struct folio *folio;
331 	struct page *page;
332 	pgoff_t index;
333 	int ret;
334 
335 	index = pos >> PAGE_SHIFT;
336 
337 	page = grab_cache_page_write_begin(mapping, index);
338 	if (!page)
339 		return -ENOMEM;
340 
341 	*pagep = page;
342 	folio = page_folio(page);
343 
344 	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
345 		/*
346 		 * Should be impossible.  If it happens, launder the page
347 		 * since we don't know what's dirty.  This will WARN in
348 		 * orangefs_writepage_locked.
349 		 */
350 		ret = orangefs_launder_folio(folio);
351 		if (ret)
352 			return ret;
353 	}
354 	if (folio_test_private(folio)) {
355 		struct orangefs_write_range *wr;
356 		wr = folio_get_private(folio);
357 		if (wr->pos + wr->len == pos &&
358 		    uid_eq(wr->uid, current_fsuid()) &&
359 		    gid_eq(wr->gid, current_fsgid())) {
360 			wr->len += len;
361 			goto okay;
362 		} else {
363 			ret = orangefs_launder_folio(folio);
364 			if (ret)
365 				return ret;
366 		}
367 	}
368 
369 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
370 	if (!wr)
371 		return -ENOMEM;
372 
373 	wr->pos = pos;
374 	wr->len = len;
375 	wr->uid = current_fsuid();
376 	wr->gid = current_fsgid();
377 	folio_attach_private(folio, wr);
378 okay:
379 	return 0;
380 }
381 
382 static int orangefs_write_end(struct file *file, struct address_space *mapping,
383     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
384 {
385 	struct inode *inode = page->mapping->host;
386 	loff_t last_pos = pos + copied;
387 
388 	/*
389 	 * No need to use i_size_read() here, the i_size
390 	 * cannot change under us because we hold the i_mutex.
391 	 */
392 	if (last_pos > inode->i_size)
393 		i_size_write(inode, last_pos);
394 
395 	/* zero the stale part of the page if we did a short copy */
396 	if (!PageUptodate(page)) {
397 		unsigned from = pos & (PAGE_SIZE - 1);
398 		if (copied < len) {
399 			zero_user(page, from + copied, len - copied);
400 		}
401 		/* Set fully written pages uptodate. */
402 		if (pos == page_offset(page) &&
403 		    (len == PAGE_SIZE || pos + len == inode->i_size)) {
404 			zero_user_segment(page, from + copied, PAGE_SIZE);
405 			SetPageUptodate(page);
406 		}
407 	}
408 
409 	set_page_dirty(page);
410 	unlock_page(page);
411 	put_page(page);
412 
413 	mark_inode_dirty_sync(file_inode(file));
414 	return copied;
415 }
416 
417 static void orangefs_invalidate_folio(struct folio *folio,
418 				 size_t offset, size_t length)
419 {
420 	struct orangefs_write_range *wr = folio_get_private(folio);
421 
422 	if (offset == 0 && length == PAGE_SIZE) {
423 		kfree(folio_detach_private(folio));
424 		return;
425 	/* write range entirely within invalidate range (or equal) */
426 	} else if (folio_pos(folio) + offset <= wr->pos &&
427 	    wr->pos + wr->len <= folio_pos(folio) + offset + length) {
428 		kfree(folio_detach_private(folio));
429 		/* XXX is this right? only caller in fs */
430 		folio_cancel_dirty(folio);
431 		return;
432 	/* invalidate range chops off end of write range */
433 	} else if (wr->pos < folio_pos(folio) + offset &&
434 	    wr->pos + wr->len <= folio_pos(folio) + offset + length &&
435 	     folio_pos(folio) + offset < wr->pos + wr->len) {
436 		size_t x;
437 		x = wr->pos + wr->len - (folio_pos(folio) + offset);
438 		WARN_ON(x > wr->len);
439 		wr->len -= x;
440 		wr->uid = current_fsuid();
441 		wr->gid = current_fsgid();
442 	/* invalidate range chops off beginning of write range */
443 	} else if (folio_pos(folio) + offset <= wr->pos &&
444 	    folio_pos(folio) + offset + length < wr->pos + wr->len &&
445 	    wr->pos < folio_pos(folio) + offset + length) {
446 		size_t x;
447 		x = folio_pos(folio) + offset + length - wr->pos;
448 		WARN_ON(x > wr->len);
449 		wr->pos += x;
450 		wr->len -= x;
451 		wr->uid = current_fsuid();
452 		wr->gid = current_fsgid();
453 	/* invalidate range entirely within write range (punch hole) */
454 	} else if (wr->pos < folio_pos(folio) + offset &&
455 	    folio_pos(folio) + offset + length < wr->pos + wr->len) {
456 		/* XXX what do we do here... should not WARN_ON */
457 		WARN_ON(1);
458 		/* punch hole */
459 		/*
460 		 * should we just ignore this and write it out anyway?
461 		 * it hardly makes sense
462 		 */
463 		return;
464 	/* non-overlapping ranges */
465 	} else {
466 		/* WARN if they do overlap */
467 		if (!((folio_pos(folio) + offset + length <= wr->pos) ^
468 		    (wr->pos + wr->len <= folio_pos(folio) + offset))) {
469 			WARN_ON(1);
470 			printk("invalidate range offset %llu length %zu\n",
471 			    folio_pos(folio) + offset, length);
472 			printk("write range offset %llu length %zu\n",
473 			    wr->pos, wr->len);
474 		}
475 		return;
476 	}
477 
478 	/*
479 	 * Above there are returns where wr is freed or where we WARN.
480 	 * Thus the following runs if wr was modified above.
481 	 */
482 
483 	orangefs_launder_folio(folio);
484 }
485 
486 static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
487 {
488 	return !folio_test_private(folio);
489 }
490 
491 static void orangefs_free_folio(struct folio *folio)
492 {
493 	kfree(folio_detach_private(folio));
494 }
495 
496 static int orangefs_launder_folio(struct folio *folio)
497 {
498 	int r = 0;
499 	struct writeback_control wbc = {
500 		.sync_mode = WB_SYNC_ALL,
501 		.nr_to_write = 0,
502 	};
503 	folio_wait_writeback(folio);
504 	if (folio_clear_dirty_for_io(folio)) {
505 		r = orangefs_writepage_locked(&folio->page, &wbc);
506 		folio_end_writeback(folio);
507 	}
508 	return r;
509 }
510 
511 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
512 				  struct iov_iter *iter)
513 {
514 	/*
515 	 * Comment from original do_readv_writev:
516 	 * Common entry point for read/write/readv/writev
517 	 * This function will dispatch it to either the direct I/O
518 	 * or buffered I/O path depending on the mount options and/or
519 	 * augmented/extended metadata attached to the file.
520 	 * Note: File extended attributes override any mount options.
521 	 */
522 	struct file *file = iocb->ki_filp;
523 	loff_t pos = iocb->ki_pos;
524 	enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
525             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
526 	loff_t *offset = &pos;
527 	struct inode *inode = file->f_mapping->host;
528 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
529 	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
530 	size_t count = iov_iter_count(iter);
531 	ssize_t total_count = 0;
532 	ssize_t ret = -EINVAL;
533 	int i = 0;
534 
535 	gossip_debug(GOSSIP_FILE_DEBUG,
536 		"%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
537 		__func__,
538 		handle,
539 		(int)count);
540 
541 	if (type == ORANGEFS_IO_WRITE) {
542 		gossip_debug(GOSSIP_FILE_DEBUG,
543 			     "%s(%pU): proceeding with offset : %llu, "
544 			     "size %d\n",
545 			     __func__,
546 			     handle,
547 			     llu(*offset),
548 			     (int)count);
549 	}
550 
551 	if (count == 0) {
552 		ret = 0;
553 		goto out;
554 	}
555 
556 	while (iov_iter_count(iter)) {
557 		size_t each_count = iov_iter_count(iter);
558 		size_t amt_complete;
559 		i++;
560 
561 		/* how much to transfer in this loop iteration */
562 		if (each_count > orangefs_bufmap_size_query())
563 			each_count = orangefs_bufmap_size_query();
564 
565 		gossip_debug(GOSSIP_FILE_DEBUG,
566 			     "%s(%pU): size of each_count(%d)\n",
567 			     __func__,
568 			     handle,
569 			     (int)each_count);
570 		gossip_debug(GOSSIP_FILE_DEBUG,
571 			     "%s(%pU): BEFORE wait_for_io: offset is %d\n",
572 			     __func__,
573 			     handle,
574 			     (int)*offset);
575 
576 		ret = wait_for_direct_io(type, inode, offset, iter,
577 				each_count, 0, NULL, NULL, file);
578 		gossip_debug(GOSSIP_FILE_DEBUG,
579 			     "%s(%pU): return from wait_for_io:%d\n",
580 			     __func__,
581 			     handle,
582 			     (int)ret);
583 
584 		if (ret < 0)
585 			goto out;
586 
587 		*offset += ret;
588 		total_count += ret;
589 		amt_complete = ret;
590 
591 		gossip_debug(GOSSIP_FILE_DEBUG,
592 			     "%s(%pU): AFTER wait_for_io: offset is %d\n",
593 			     __func__,
594 			     handle,
595 			     (int)*offset);
596 
597 		/*
598 		 * if we got a short I/O operations,
599 		 * fall out and return what we got so far
600 		 */
601 		if (amt_complete < each_count)
602 			break;
603 	} /*end while */
604 
605 out:
606 	if (total_count > 0)
607 		ret = total_count;
608 	if (ret > 0) {
609 		if (type == ORANGEFS_IO_READ) {
610 			file_accessed(file);
611 		} else {
612 			file_update_time(file);
613 			if (*offset > i_size_read(inode))
614 				i_size_write(inode, *offset);
615 		}
616 	}
617 
618 	gossip_debug(GOSSIP_FILE_DEBUG,
619 		     "%s(%pU): Value(%d) returned.\n",
620 		     __func__,
621 		     handle,
622 		     (int)ret);
623 
624 	return ret;
625 }
626 
627 /** ORANGEFS2 implementation of address space operations */
628 static const struct address_space_operations orangefs_address_operations = {
629 	.writepage = orangefs_writepage,
630 	.readahead = orangefs_readahead,
631 	.read_folio = orangefs_read_folio,
632 	.writepages = orangefs_writepages,
633 	.dirty_folio = filemap_dirty_folio,
634 	.write_begin = orangefs_write_begin,
635 	.write_end = orangefs_write_end,
636 	.invalidate_folio = orangefs_invalidate_folio,
637 	.release_folio = orangefs_release_folio,
638 	.free_folio = orangefs_free_folio,
639 	.launder_folio = orangefs_launder_folio,
640 	.direct_IO = orangefs_direct_IO,
641 };
642 
643 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
644 {
645 	struct folio *folio = page_folio(vmf->page);
646 	struct inode *inode = file_inode(vmf->vma->vm_file);
647 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
648 	unsigned long *bitlock = &orangefs_inode->bitlock;
649 	vm_fault_t ret;
650 	struct orangefs_write_range *wr;
651 
652 	sb_start_pagefault(inode->i_sb);
653 
654 	if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
655 		ret = VM_FAULT_RETRY;
656 		goto out;
657 	}
658 
659 	folio_lock(folio);
660 	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
661 		/*
662 		 * Should be impossible.  If it happens, launder the folio
663 		 * since we don't know what's dirty.  This will WARN in
664 		 * orangefs_writepage_locked.
665 		 */
666 		if (orangefs_launder_folio(folio)) {
667 			ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
668 			goto out;
669 		}
670 	}
671 	if (folio_test_private(folio)) {
672 		wr = folio_get_private(folio);
673 		if (uid_eq(wr->uid, current_fsuid()) &&
674 		    gid_eq(wr->gid, current_fsgid())) {
675 			wr->pos = page_offset(vmf->page);
676 			wr->len = PAGE_SIZE;
677 			goto okay;
678 		} else {
679 			if (orangefs_launder_folio(folio)) {
680 				ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
681 				goto out;
682 			}
683 		}
684 	}
685 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
686 	if (!wr) {
687 		ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
688 		goto out;
689 	}
690 	wr->pos = page_offset(vmf->page);
691 	wr->len = PAGE_SIZE;
692 	wr->uid = current_fsuid();
693 	wr->gid = current_fsgid();
694 	folio_attach_private(folio, wr);
695 okay:
696 
697 	file_update_time(vmf->vma->vm_file);
698 	if (folio->mapping != inode->i_mapping) {
699 		folio_unlock(folio);
700 		ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
701 		goto out;
702 	}
703 
704 	/*
705 	 * We mark the folio dirty already here so that when freeze is in
706 	 * progress, we are guaranteed that writeback during freezing will
707 	 * see the dirty folio and writeprotect it again.
708 	 */
709 	folio_mark_dirty(folio);
710 	folio_wait_stable(folio);
711 	ret = VM_FAULT_LOCKED;
712 out:
713 	sb_end_pagefault(inode->i_sb);
714 	return ret;
715 }
716 
717 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
718 {
719 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
720 	struct orangefs_kernel_op_s *new_op;
721 	loff_t orig_size;
722 	int ret = -EINVAL;
723 
724 	gossip_debug(GOSSIP_INODE_DEBUG,
725 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
726 		     __func__,
727 		     get_khandle_from_ino(inode),
728 		     &orangefs_inode->refn.khandle,
729 		     orangefs_inode->refn.fs_id,
730 		     iattr->ia_size);
731 
732 	/* Ensure that we have a up to date size, so we know if it changed. */
733 	ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
734 	if (ret == -ESTALE)
735 		ret = -EIO;
736 	if (ret) {
737 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
738 		    __func__, ret);
739 		return ret;
740 	}
741 	orig_size = i_size_read(inode);
742 
743 	/* This is truncate_setsize in a different order. */
744 	truncate_pagecache(inode, iattr->ia_size);
745 	i_size_write(inode, iattr->ia_size);
746 	if (iattr->ia_size > orig_size)
747 		pagecache_isize_extended(inode, orig_size, iattr->ia_size);
748 
749 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
750 	if (!new_op)
751 		return -ENOMEM;
752 
753 	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
754 	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
755 
756 	ret = service_operation(new_op,
757 		__func__,
758 		get_interruptible_flag(inode));
759 
760 	/*
761 	 * the truncate has no downcall members to retrieve, but
762 	 * the status value tells us if it went through ok or not
763 	 */
764 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
765 
766 	op_release(new_op);
767 
768 	if (ret != 0)
769 		return ret;
770 
771 	if (orig_size != i_size_read(inode))
772 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
773 
774 	return ret;
775 }
776 
777 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
778 {
779 	int ret;
780 
781 	if (iattr->ia_valid & ATTR_MODE) {
782 		if (iattr->ia_mode & (S_ISVTX)) {
783 			if (is_root_handle(inode)) {
784 				/*
785 				 * allow sticky bit to be set on root (since
786 				 * it shows up that way by default anyhow),
787 				 * but don't show it to the server
788 				 */
789 				iattr->ia_mode -= S_ISVTX;
790 			} else {
791 				gossip_debug(GOSSIP_UTILS_DEBUG,
792 					     "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
793 				ret = -EINVAL;
794 				goto out;
795 			}
796 		}
797 		if (iattr->ia_mode & (S_ISUID)) {
798 			gossip_debug(GOSSIP_UTILS_DEBUG,
799 				     "Attempting to set setuid bit (not supported); returning EINVAL.\n");
800 			ret = -EINVAL;
801 			goto out;
802 		}
803 	}
804 
805 	if (iattr->ia_valid & ATTR_SIZE) {
806 		ret = orangefs_setattr_size(inode, iattr);
807 		if (ret)
808 			goto out;
809 	}
810 
811 again:
812 	spin_lock(&inode->i_lock);
813 	if (ORANGEFS_I(inode)->attr_valid) {
814 		if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
815 		    gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
816 			ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
817 		} else {
818 			spin_unlock(&inode->i_lock);
819 			write_inode_now(inode, 1);
820 			goto again;
821 		}
822 	} else {
823 		ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
824 		ORANGEFS_I(inode)->attr_uid = current_fsuid();
825 		ORANGEFS_I(inode)->attr_gid = current_fsgid();
826 	}
827 	setattr_copy(&init_user_ns, inode, iattr);
828 	spin_unlock(&inode->i_lock);
829 	mark_inode_dirty(inode);
830 
831 	if (iattr->ia_valid & ATTR_MODE)
832 		/* change mod on a file that has ACLs */
833 		ret = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
834 
835 	ret = 0;
836 out:
837 	return ret;
838 }
839 
840 /*
841  * Change attributes of an object referenced by dentry.
842  */
843 int orangefs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
844 		     struct iattr *iattr)
845 {
846 	int ret;
847 	gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
848 	    dentry);
849 	ret = setattr_prepare(&init_user_ns, dentry, iattr);
850 	if (ret)
851 	        goto out;
852 	ret = __orangefs_setattr(d_inode(dentry), iattr);
853 	sync_inode_metadata(d_inode(dentry), 1);
854 out:
855 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
856 	    ret);
857 	return ret;
858 }
859 
860 /*
861  * Obtain attributes of an object given a dentry
862  */
863 int orangefs_getattr(struct user_namespace *mnt_userns, const struct path *path,
864 		     struct kstat *stat, u32 request_mask, unsigned int flags)
865 {
866 	int ret;
867 	struct inode *inode = path->dentry->d_inode;
868 
869 	gossip_debug(GOSSIP_INODE_DEBUG,
870 		     "orangefs_getattr: called on %pd mask %u\n",
871 		     path->dentry, request_mask);
872 
873 	ret = orangefs_inode_getattr(inode,
874 	    request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
875 	if (ret == 0) {
876 		generic_fillattr(&init_user_ns, inode, stat);
877 
878 		/* override block size reported to stat */
879 		if (!(request_mask & STATX_SIZE))
880 			stat->result_mask &= ~STATX_SIZE;
881 
882 		generic_fill_statx_attr(inode, stat);
883 	}
884 	return ret;
885 }
886 
887 int orangefs_permission(struct user_namespace *mnt_userns,
888 			struct inode *inode, int mask)
889 {
890 	int ret;
891 
892 	if (mask & MAY_NOT_BLOCK)
893 		return -ECHILD;
894 
895 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
896 
897 	/* Make sure the permission (and other common attrs) are up to date. */
898 	ret = orangefs_inode_getattr(inode, 0);
899 	if (ret < 0)
900 		return ret;
901 
902 	return generic_permission(&init_user_ns, inode, mask);
903 }
904 
905 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
906 {
907 	struct iattr iattr;
908 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
909 	    get_khandle_from_ino(inode));
910 	generic_update_time(inode, time, flags);
911 	memset(&iattr, 0, sizeof iattr);
912         if (flags & S_ATIME)
913 		iattr.ia_valid |= ATTR_ATIME;
914 	if (flags & S_CTIME)
915 		iattr.ia_valid |= ATTR_CTIME;
916 	if (flags & S_MTIME)
917 		iattr.ia_valid |= ATTR_MTIME;
918 	return __orangefs_setattr(inode, &iattr);
919 }
920 
921 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
922 {
923 	u64 val = 0;
924 	int ret;
925 
926 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
927 		     dentry);
928 
929 	ret = orangefs_inode_getxattr(d_inode(dentry),
930 				      "user.pvfs2.meta_hint",
931 				      &val, sizeof(val));
932 	if (ret < 0 && ret != -ENODATA)
933 		return ret;
934 
935 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
936 
937 	fileattr_fill_flags(fa, val);
938 	return 0;
939 }
940 
941 static int orangefs_fileattr_set(struct user_namespace *mnt_userns,
942 				 struct dentry *dentry, struct fileattr *fa)
943 {
944 	u64 val = 0;
945 
946 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
947 		     dentry);
948 	/*
949 	 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
950 	 * turned on for a file. The user is not allowed to turn on this bit,
951 	 * but the bit is present if the user first gets the flags and then
952 	 * updates the flags with some new settings. So, we ignore it in the
953 	 * following edit. bligon.
954 	 */
955 	if (fileattr_has_fsx(fa) ||
956 	    (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
957 		gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
958 			   __func__);
959 		return -EOPNOTSUPP;
960 	}
961 	val = fa->flags;
962 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
963 	return orangefs_inode_setxattr(d_inode(dentry),
964 				       "user.pvfs2.meta_hint",
965 				       &val, sizeof(val), 0);
966 }
967 
968 /* ORANGEFS2 implementation of VFS inode operations for files */
969 static const struct inode_operations orangefs_file_inode_operations = {
970 	.get_acl = orangefs_get_acl,
971 	.set_acl = orangefs_set_acl,
972 	.setattr = orangefs_setattr,
973 	.getattr = orangefs_getattr,
974 	.listxattr = orangefs_listxattr,
975 	.permission = orangefs_permission,
976 	.update_time = orangefs_update_time,
977 	.fileattr_get = orangefs_fileattr_get,
978 	.fileattr_set = orangefs_fileattr_set,
979 };
980 
981 static int orangefs_init_iops(struct inode *inode)
982 {
983 	inode->i_mapping->a_ops = &orangefs_address_operations;
984 
985 	switch (inode->i_mode & S_IFMT) {
986 	case S_IFREG:
987 		inode->i_op = &orangefs_file_inode_operations;
988 		inode->i_fop = &orangefs_file_operations;
989 		break;
990 	case S_IFLNK:
991 		inode->i_op = &orangefs_symlink_inode_operations;
992 		break;
993 	case S_IFDIR:
994 		inode->i_op = &orangefs_dir_inode_operations;
995 		inode->i_fop = &orangefs_dir_operations;
996 		break;
997 	default:
998 		gossip_debug(GOSSIP_INODE_DEBUG,
999 			     "%s: unsupported mode\n",
1000 			     __func__);
1001 		return -EINVAL;
1002 	}
1003 
1004 	return 0;
1005 }
1006 
1007 /*
1008  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1009  * a ino_t type that will be used as a hash-index from where the handle will
1010  * be searched for in the VFS hash table of inodes.
1011  */
1012 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1013 {
1014 	if (!ref)
1015 		return 0;
1016 	return orangefs_khandle_to_ino(&(ref->khandle));
1017 }
1018 
1019 /*
1020  * Called to set up an inode from iget5_locked.
1021  */
1022 static int orangefs_set_inode(struct inode *inode, void *data)
1023 {
1024 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1025 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1026 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1027 	ORANGEFS_I(inode)->attr_valid = 0;
1028 	hash_init(ORANGEFS_I(inode)->xattr_cache);
1029 	ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1030 	ORANGEFS_I(inode)->bitlock = 0;
1031 	return 0;
1032 }
1033 
1034 /*
1035  * Called to determine if handles match.
1036  */
1037 static int orangefs_test_inode(struct inode *inode, void *data)
1038 {
1039 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1040 	struct orangefs_inode_s *orangefs_inode = NULL;
1041 
1042 	orangefs_inode = ORANGEFS_I(inode);
1043 	/* test handles and fs_ids... */
1044 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1045 				&(ref->khandle)) &&
1046 			orangefs_inode->refn.fs_id == ref->fs_id);
1047 }
1048 
1049 /*
1050  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1051  * file handle.
1052  *
1053  * @sb: the file system super block instance.
1054  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1055  */
1056 struct inode *orangefs_iget(struct super_block *sb,
1057 		struct orangefs_object_kref *ref)
1058 {
1059 	struct inode *inode = NULL;
1060 	unsigned long hash;
1061 	int error;
1062 
1063 	hash = orangefs_handle_hash(ref);
1064 	inode = iget5_locked(sb,
1065 			hash,
1066 			orangefs_test_inode,
1067 			orangefs_set_inode,
1068 			ref);
1069 
1070 	if (!inode)
1071 		return ERR_PTR(-ENOMEM);
1072 
1073 	if (!(inode->i_state & I_NEW))
1074 		return inode;
1075 
1076 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1077 	if (error) {
1078 		iget_failed(inode);
1079 		return ERR_PTR(error);
1080 	}
1081 
1082 	inode->i_ino = hash;	/* needed for stat etc */
1083 	orangefs_init_iops(inode);
1084 	unlock_new_inode(inode);
1085 
1086 	gossip_debug(GOSSIP_INODE_DEBUG,
1087 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1088 		     &ref->khandle,
1089 		     ref->fs_id,
1090 		     hash,
1091 		     inode->i_ino);
1092 
1093 	return inode;
1094 }
1095 
1096 /*
1097  * Allocate an inode for a newly created file and insert it into the inode hash.
1098  */
1099 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1100 		int mode, dev_t dev, struct orangefs_object_kref *ref)
1101 {
1102 	unsigned long hash = orangefs_handle_hash(ref);
1103 	struct inode *inode;
1104 	int error;
1105 
1106 	gossip_debug(GOSSIP_INODE_DEBUG,
1107 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1108 		     __func__,
1109 		     sb,
1110 		     MAJOR(dev),
1111 		     MINOR(dev),
1112 		     mode);
1113 
1114 	inode = new_inode(sb);
1115 	if (!inode)
1116 		return ERR_PTR(-ENOMEM);
1117 
1118 	orangefs_set_inode(inode, ref);
1119 	inode->i_ino = hash;	/* needed for stat etc */
1120 
1121 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1122 	if (error)
1123 		goto out_iput;
1124 
1125 	orangefs_init_iops(inode);
1126 	inode->i_rdev = dev;
1127 
1128 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1129 	if (error < 0)
1130 		goto out_iput;
1131 
1132 	gossip_debug(GOSSIP_INODE_DEBUG,
1133 		     "Initializing ACL's for inode %pU\n",
1134 		     get_khandle_from_ino(inode));
1135 	orangefs_init_acl(inode, dir);
1136 	return inode;
1137 
1138 out_iput:
1139 	iput(inode);
1140 	return ERR_PTR(error);
1141 }
1142