xref: /openbmc/linux/fs/orangefs/inode.c (revision 5abbeebd)
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, ITER_SOURCE, &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, ITER_SOURCE, 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, ITER_DEST, 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, ITER_DEST, &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 
534 	gossip_debug(GOSSIP_FILE_DEBUG,
535 		"%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
536 		__func__,
537 		handle,
538 		(int)count);
539 
540 	if (type == ORANGEFS_IO_WRITE) {
541 		gossip_debug(GOSSIP_FILE_DEBUG,
542 			     "%s(%pU): proceeding with offset : %llu, "
543 			     "size %d\n",
544 			     __func__,
545 			     handle,
546 			     llu(*offset),
547 			     (int)count);
548 	}
549 
550 	if (count == 0) {
551 		ret = 0;
552 		goto out;
553 	}
554 
555 	while (iov_iter_count(iter)) {
556 		size_t each_count = iov_iter_count(iter);
557 		size_t amt_complete;
558 
559 		/* how much to transfer in this loop iteration */
560 		if (each_count > orangefs_bufmap_size_query())
561 			each_count = orangefs_bufmap_size_query();
562 
563 		gossip_debug(GOSSIP_FILE_DEBUG,
564 			     "%s(%pU): size of each_count(%d)\n",
565 			     __func__,
566 			     handle,
567 			     (int)each_count);
568 		gossip_debug(GOSSIP_FILE_DEBUG,
569 			     "%s(%pU): BEFORE wait_for_io: offset is %d\n",
570 			     __func__,
571 			     handle,
572 			     (int)*offset);
573 
574 		ret = wait_for_direct_io(type, inode, offset, iter,
575 				each_count, 0, NULL, NULL, file);
576 		gossip_debug(GOSSIP_FILE_DEBUG,
577 			     "%s(%pU): return from wait_for_io:%d\n",
578 			     __func__,
579 			     handle,
580 			     (int)ret);
581 
582 		if (ret < 0)
583 			goto out;
584 
585 		*offset += ret;
586 		total_count += ret;
587 		amt_complete = ret;
588 
589 		gossip_debug(GOSSIP_FILE_DEBUG,
590 			     "%s(%pU): AFTER wait_for_io: offset is %d\n",
591 			     __func__,
592 			     handle,
593 			     (int)*offset);
594 
595 		/*
596 		 * if we got a short I/O operations,
597 		 * fall out and return what we got so far
598 		 */
599 		if (amt_complete < each_count)
600 			break;
601 	} /*end while */
602 
603 out:
604 	if (total_count > 0)
605 		ret = total_count;
606 	if (ret > 0) {
607 		if (type == ORANGEFS_IO_READ) {
608 			file_accessed(file);
609 		} else {
610 			file_update_time(file);
611 			if (*offset > i_size_read(inode))
612 				i_size_write(inode, *offset);
613 		}
614 	}
615 
616 	gossip_debug(GOSSIP_FILE_DEBUG,
617 		     "%s(%pU): Value(%d) returned.\n",
618 		     __func__,
619 		     handle,
620 		     (int)ret);
621 
622 	return ret;
623 }
624 
625 /** ORANGEFS2 implementation of address space operations */
626 static const struct address_space_operations orangefs_address_operations = {
627 	.writepage = orangefs_writepage,
628 	.readahead = orangefs_readahead,
629 	.read_folio = orangefs_read_folio,
630 	.writepages = orangefs_writepages,
631 	.dirty_folio = filemap_dirty_folio,
632 	.write_begin = orangefs_write_begin,
633 	.write_end = orangefs_write_end,
634 	.invalidate_folio = orangefs_invalidate_folio,
635 	.release_folio = orangefs_release_folio,
636 	.free_folio = orangefs_free_folio,
637 	.launder_folio = orangefs_launder_folio,
638 	.direct_IO = orangefs_direct_IO,
639 };
640 
641 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
642 {
643 	struct folio *folio = page_folio(vmf->page);
644 	struct inode *inode = file_inode(vmf->vma->vm_file);
645 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
646 	unsigned long *bitlock = &orangefs_inode->bitlock;
647 	vm_fault_t ret;
648 	struct orangefs_write_range *wr;
649 
650 	sb_start_pagefault(inode->i_sb);
651 
652 	if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
653 		ret = VM_FAULT_RETRY;
654 		goto out;
655 	}
656 
657 	folio_lock(folio);
658 	if (folio_test_dirty(folio) && !folio_test_private(folio)) {
659 		/*
660 		 * Should be impossible.  If it happens, launder the folio
661 		 * since we don't know what's dirty.  This will WARN in
662 		 * orangefs_writepage_locked.
663 		 */
664 		if (orangefs_launder_folio(folio)) {
665 			ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
666 			goto out;
667 		}
668 	}
669 	if (folio_test_private(folio)) {
670 		wr = folio_get_private(folio);
671 		if (uid_eq(wr->uid, current_fsuid()) &&
672 		    gid_eq(wr->gid, current_fsgid())) {
673 			wr->pos = page_offset(vmf->page);
674 			wr->len = PAGE_SIZE;
675 			goto okay;
676 		} else {
677 			if (orangefs_launder_folio(folio)) {
678 				ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
679 				goto out;
680 			}
681 		}
682 	}
683 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
684 	if (!wr) {
685 		ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
686 		goto out;
687 	}
688 	wr->pos = page_offset(vmf->page);
689 	wr->len = PAGE_SIZE;
690 	wr->uid = current_fsuid();
691 	wr->gid = current_fsgid();
692 	folio_attach_private(folio, wr);
693 okay:
694 
695 	file_update_time(vmf->vma->vm_file);
696 	if (folio->mapping != inode->i_mapping) {
697 		folio_unlock(folio);
698 		ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
699 		goto out;
700 	}
701 
702 	/*
703 	 * We mark the folio dirty already here so that when freeze is in
704 	 * progress, we are guaranteed that writeback during freezing will
705 	 * see the dirty folio and writeprotect it again.
706 	 */
707 	folio_mark_dirty(folio);
708 	folio_wait_stable(folio);
709 	ret = VM_FAULT_LOCKED;
710 out:
711 	sb_end_pagefault(inode->i_sb);
712 	return ret;
713 }
714 
715 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
716 {
717 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
718 	struct orangefs_kernel_op_s *new_op;
719 	loff_t orig_size;
720 	int ret = -EINVAL;
721 
722 	gossip_debug(GOSSIP_INODE_DEBUG,
723 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
724 		     __func__,
725 		     get_khandle_from_ino(inode),
726 		     &orangefs_inode->refn.khandle,
727 		     orangefs_inode->refn.fs_id,
728 		     iattr->ia_size);
729 
730 	/* Ensure that we have a up to date size, so we know if it changed. */
731 	ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
732 	if (ret == -ESTALE)
733 		ret = -EIO;
734 	if (ret) {
735 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
736 		    __func__, ret);
737 		return ret;
738 	}
739 	orig_size = i_size_read(inode);
740 
741 	/* This is truncate_setsize in a different order. */
742 	truncate_pagecache(inode, iattr->ia_size);
743 	i_size_write(inode, iattr->ia_size);
744 	if (iattr->ia_size > orig_size)
745 		pagecache_isize_extended(inode, orig_size, iattr->ia_size);
746 
747 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
748 	if (!new_op)
749 		return -ENOMEM;
750 
751 	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
752 	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
753 
754 	ret = service_operation(new_op,
755 		__func__,
756 		get_interruptible_flag(inode));
757 
758 	/*
759 	 * the truncate has no downcall members to retrieve, but
760 	 * the status value tells us if it went through ok or not
761 	 */
762 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
763 
764 	op_release(new_op);
765 
766 	if (ret != 0)
767 		return ret;
768 
769 	if (orig_size != i_size_read(inode))
770 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
771 
772 	return ret;
773 }
774 
775 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
776 {
777 	int ret;
778 
779 	if (iattr->ia_valid & ATTR_MODE) {
780 		if (iattr->ia_mode & (S_ISVTX)) {
781 			if (is_root_handle(inode)) {
782 				/*
783 				 * allow sticky bit to be set on root (since
784 				 * it shows up that way by default anyhow),
785 				 * but don't show it to the server
786 				 */
787 				iattr->ia_mode -= S_ISVTX;
788 			} else {
789 				gossip_debug(GOSSIP_UTILS_DEBUG,
790 					     "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
791 				ret = -EINVAL;
792 				goto out;
793 			}
794 		}
795 		if (iattr->ia_mode & (S_ISUID)) {
796 			gossip_debug(GOSSIP_UTILS_DEBUG,
797 				     "Attempting to set setuid bit (not supported); returning EINVAL.\n");
798 			ret = -EINVAL;
799 			goto out;
800 		}
801 	}
802 
803 	if (iattr->ia_valid & ATTR_SIZE) {
804 		ret = orangefs_setattr_size(inode, iattr);
805 		if (ret)
806 			goto out;
807 	}
808 
809 again:
810 	spin_lock(&inode->i_lock);
811 	if (ORANGEFS_I(inode)->attr_valid) {
812 		if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
813 		    gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
814 			ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
815 		} else {
816 			spin_unlock(&inode->i_lock);
817 			write_inode_now(inode, 1);
818 			goto again;
819 		}
820 	} else {
821 		ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
822 		ORANGEFS_I(inode)->attr_uid = current_fsuid();
823 		ORANGEFS_I(inode)->attr_gid = current_fsgid();
824 	}
825 	setattr_copy(&init_user_ns, inode, iattr);
826 	spin_unlock(&inode->i_lock);
827 	mark_inode_dirty(inode);
828 
829 	ret = 0;
830 out:
831 	return ret;
832 }
833 
834 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr)
835 {
836 	int ret;
837 	struct inode *inode = d_inode(dentry);
838 
839 	ret = __orangefs_setattr(inode, iattr);
840 	/* change mode on a file that has ACLs */
841 	if (!ret && (iattr->ia_valid & ATTR_MODE))
842 		ret = posix_acl_chmod(&init_user_ns, dentry, inode->i_mode);
843 	return ret;
844 }
845 
846 /*
847  * Change attributes of an object referenced by dentry.
848  */
849 int orangefs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
850 		     struct iattr *iattr)
851 {
852 	int ret;
853 	gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
854 	    dentry);
855 	ret = setattr_prepare(&init_user_ns, dentry, iattr);
856 	if (ret)
857 	        goto out;
858 	ret = __orangefs_setattr_mode(dentry, iattr);
859 	sync_inode_metadata(d_inode(dentry), 1);
860 out:
861 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
862 	    ret);
863 	return ret;
864 }
865 
866 /*
867  * Obtain attributes of an object given a dentry
868  */
869 int orangefs_getattr(struct user_namespace *mnt_userns, const struct path *path,
870 		     struct kstat *stat, u32 request_mask, unsigned int flags)
871 {
872 	int ret;
873 	struct inode *inode = path->dentry->d_inode;
874 
875 	gossip_debug(GOSSIP_INODE_DEBUG,
876 		     "orangefs_getattr: called on %pd mask %u\n",
877 		     path->dentry, request_mask);
878 
879 	ret = orangefs_inode_getattr(inode,
880 	    request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
881 	if (ret == 0) {
882 		generic_fillattr(&init_user_ns, inode, stat);
883 
884 		/* override block size reported to stat */
885 		if (!(request_mask & STATX_SIZE))
886 			stat->result_mask &= ~STATX_SIZE;
887 
888 		generic_fill_statx_attr(inode, stat);
889 	}
890 	return ret;
891 }
892 
893 int orangefs_permission(struct user_namespace *mnt_userns,
894 			struct inode *inode, int mask)
895 {
896 	int ret;
897 
898 	if (mask & MAY_NOT_BLOCK)
899 		return -ECHILD;
900 
901 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
902 
903 	/* Make sure the permission (and other common attrs) are up to date. */
904 	ret = orangefs_inode_getattr(inode, 0);
905 	if (ret < 0)
906 		return ret;
907 
908 	return generic_permission(&init_user_ns, inode, mask);
909 }
910 
911 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
912 {
913 	struct iattr iattr;
914 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
915 	    get_khandle_from_ino(inode));
916 	generic_update_time(inode, time, flags);
917 	memset(&iattr, 0, sizeof iattr);
918         if (flags & S_ATIME)
919 		iattr.ia_valid |= ATTR_ATIME;
920 	if (flags & S_CTIME)
921 		iattr.ia_valid |= ATTR_CTIME;
922 	if (flags & S_MTIME)
923 		iattr.ia_valid |= ATTR_MTIME;
924 	return __orangefs_setattr(inode, &iattr);
925 }
926 
927 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
928 {
929 	u64 val = 0;
930 	int ret;
931 
932 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
933 		     dentry);
934 
935 	ret = orangefs_inode_getxattr(d_inode(dentry),
936 				      "user.pvfs2.meta_hint",
937 				      &val, sizeof(val));
938 	if (ret < 0 && ret != -ENODATA)
939 		return ret;
940 
941 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
942 
943 	fileattr_fill_flags(fa, val);
944 	return 0;
945 }
946 
947 static int orangefs_fileattr_set(struct user_namespace *mnt_userns,
948 				 struct dentry *dentry, struct fileattr *fa)
949 {
950 	u64 val = 0;
951 
952 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
953 		     dentry);
954 	/*
955 	 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
956 	 * turned on for a file. The user is not allowed to turn on this bit,
957 	 * but the bit is present if the user first gets the flags and then
958 	 * updates the flags with some new settings. So, we ignore it in the
959 	 * following edit. bligon.
960 	 */
961 	if (fileattr_has_fsx(fa) ||
962 	    (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
963 		gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
964 			   __func__);
965 		return -EOPNOTSUPP;
966 	}
967 	val = fa->flags;
968 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
969 	return orangefs_inode_setxattr(d_inode(dentry),
970 				       "user.pvfs2.meta_hint",
971 				       &val, sizeof(val), 0);
972 }
973 
974 /* ORANGEFS2 implementation of VFS inode operations for files */
975 static const struct inode_operations orangefs_file_inode_operations = {
976 	.get_inode_acl = orangefs_get_acl,
977 	.set_acl = orangefs_set_acl,
978 	.setattr = orangefs_setattr,
979 	.getattr = orangefs_getattr,
980 	.listxattr = orangefs_listxattr,
981 	.permission = orangefs_permission,
982 	.update_time = orangefs_update_time,
983 	.fileattr_get = orangefs_fileattr_get,
984 	.fileattr_set = orangefs_fileattr_set,
985 };
986 
987 static int orangefs_init_iops(struct inode *inode)
988 {
989 	inode->i_mapping->a_ops = &orangefs_address_operations;
990 
991 	switch (inode->i_mode & S_IFMT) {
992 	case S_IFREG:
993 		inode->i_op = &orangefs_file_inode_operations;
994 		inode->i_fop = &orangefs_file_operations;
995 		break;
996 	case S_IFLNK:
997 		inode->i_op = &orangefs_symlink_inode_operations;
998 		break;
999 	case S_IFDIR:
1000 		inode->i_op = &orangefs_dir_inode_operations;
1001 		inode->i_fop = &orangefs_dir_operations;
1002 		break;
1003 	default:
1004 		gossip_debug(GOSSIP_INODE_DEBUG,
1005 			     "%s: unsupported mode\n",
1006 			     __func__);
1007 		return -EINVAL;
1008 	}
1009 
1010 	return 0;
1011 }
1012 
1013 /*
1014  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1015  * a ino_t type that will be used as a hash-index from where the handle will
1016  * be searched for in the VFS hash table of inodes.
1017  */
1018 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1019 {
1020 	if (!ref)
1021 		return 0;
1022 	return orangefs_khandle_to_ino(&(ref->khandle));
1023 }
1024 
1025 /*
1026  * Called to set up an inode from iget5_locked.
1027  */
1028 static int orangefs_set_inode(struct inode *inode, void *data)
1029 {
1030 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1031 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1032 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1033 	ORANGEFS_I(inode)->attr_valid = 0;
1034 	hash_init(ORANGEFS_I(inode)->xattr_cache);
1035 	ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1036 	ORANGEFS_I(inode)->bitlock = 0;
1037 	return 0;
1038 }
1039 
1040 /*
1041  * Called to determine if handles match.
1042  */
1043 static int orangefs_test_inode(struct inode *inode, void *data)
1044 {
1045 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1046 	struct orangefs_inode_s *orangefs_inode = NULL;
1047 
1048 	orangefs_inode = ORANGEFS_I(inode);
1049 	/* test handles and fs_ids... */
1050 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1051 				&(ref->khandle)) &&
1052 			orangefs_inode->refn.fs_id == ref->fs_id);
1053 }
1054 
1055 /*
1056  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1057  * file handle.
1058  *
1059  * @sb: the file system super block instance.
1060  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1061  */
1062 struct inode *orangefs_iget(struct super_block *sb,
1063 		struct orangefs_object_kref *ref)
1064 {
1065 	struct inode *inode = NULL;
1066 	unsigned long hash;
1067 	int error;
1068 
1069 	hash = orangefs_handle_hash(ref);
1070 	inode = iget5_locked(sb,
1071 			hash,
1072 			orangefs_test_inode,
1073 			orangefs_set_inode,
1074 			ref);
1075 
1076 	if (!inode)
1077 		return ERR_PTR(-ENOMEM);
1078 
1079 	if (!(inode->i_state & I_NEW))
1080 		return inode;
1081 
1082 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1083 	if (error) {
1084 		iget_failed(inode);
1085 		return ERR_PTR(error);
1086 	}
1087 
1088 	inode->i_ino = hash;	/* needed for stat etc */
1089 	orangefs_init_iops(inode);
1090 	unlock_new_inode(inode);
1091 
1092 	gossip_debug(GOSSIP_INODE_DEBUG,
1093 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1094 		     &ref->khandle,
1095 		     ref->fs_id,
1096 		     hash,
1097 		     inode->i_ino);
1098 
1099 	return inode;
1100 }
1101 
1102 /*
1103  * Allocate an inode for a newly created file and insert it into the inode hash.
1104  */
1105 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1106 		umode_t mode, dev_t dev, struct orangefs_object_kref *ref)
1107 {
1108 	struct posix_acl *acl = NULL, *default_acl = NULL;
1109 	unsigned long hash = orangefs_handle_hash(ref);
1110 	struct inode *inode;
1111 	int error;
1112 
1113 	gossip_debug(GOSSIP_INODE_DEBUG,
1114 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1115 		     __func__,
1116 		     sb,
1117 		     MAJOR(dev),
1118 		     MINOR(dev),
1119 		     mode);
1120 
1121 	inode = new_inode(sb);
1122 	if (!inode)
1123 		return ERR_PTR(-ENOMEM);
1124 
1125 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
1126 	if (error)
1127 		goto out_iput;
1128 
1129 	orangefs_set_inode(inode, ref);
1130 	inode->i_ino = hash;	/* needed for stat etc */
1131 
1132 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1133 	if (error)
1134 		goto out_iput;
1135 
1136 	orangefs_init_iops(inode);
1137 	inode->i_rdev = dev;
1138 
1139 	if (default_acl) {
1140 		error = __orangefs_set_acl(inode, default_acl,
1141 					   ACL_TYPE_DEFAULT);
1142 		if (error)
1143 			goto out_iput;
1144 	}
1145 
1146 	if (acl) {
1147 		error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS);
1148 		if (error)
1149 			goto out_iput;
1150 	}
1151 
1152 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1153 	if (error < 0)
1154 		goto out_iput;
1155 
1156 	gossip_debug(GOSSIP_INODE_DEBUG,
1157 		     "Initializing ACL's for inode %pU\n",
1158 		     get_khandle_from_ino(inode));
1159 	if (mode != inode->i_mode) {
1160 		struct iattr iattr = {
1161 			.ia_mode = mode,
1162 			.ia_valid = ATTR_MODE,
1163 		};
1164 		inode->i_mode = mode;
1165 		__orangefs_setattr(inode, &iattr);
1166 		__posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode);
1167 	}
1168 	posix_acl_release(acl);
1169 	posix_acl_release(default_acl);
1170 	return inode;
1171 
1172 out_iput:
1173 	iput(inode);
1174 	posix_acl_release(acl);
1175 	posix_acl_release(default_acl);
1176 	return ERR_PTR(error);
1177 }
1178