xref: /openbmc/linux/fs/orangefs/inode.c (revision b830f94f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * (C) 2001 Clemson University and The University of Chicago
4  * Copyright 2018 Omnibond Systems, L.L.C.
5  *
6  * See COPYING in top-level directory.
7  */
8 
9 /*
10  *  Linux VFS inode operations.
11  */
12 
13 #include <linux/bvec.h>
14 #include "protocol.h"
15 #include "orangefs-kernel.h"
16 #include "orangefs-bufmap.h"
17 
18 static int orangefs_writepage_locked(struct page *page,
19     struct writeback_control *wbc)
20 {
21 	struct inode *inode = page->mapping->host;
22 	struct orangefs_write_range *wr = NULL;
23 	struct iov_iter iter;
24 	struct bio_vec bv;
25 	size_t len, wlen;
26 	ssize_t ret;
27 	loff_t off;
28 
29 	set_page_writeback(page);
30 
31 	len = i_size_read(inode);
32 	if (PagePrivate(page)) {
33 		wr = (struct orangefs_write_range *)page_private(page);
34 		WARN_ON(wr->pos >= len);
35 		off = wr->pos;
36 		if (off + wr->len > len)
37 			wlen = len - off;
38 		else
39 			wlen = wr->len;
40 	} else {
41 		WARN_ON(1);
42 		off = page_offset(page);
43 		if (off + PAGE_SIZE > len)
44 			wlen = len - off;
45 		else
46 			wlen = PAGE_SIZE;
47 	}
48 	/* Should've been handled in orangefs_invalidatepage. */
49 	WARN_ON(off == len || off + wlen > len);
50 
51 	bv.bv_page = page;
52 	bv.bv_len = wlen;
53 	bv.bv_offset = off % PAGE_SIZE;
54 	WARN_ON(wlen == 0);
55 	iov_iter_bvec(&iter, WRITE, &bv, 1, wlen);
56 
57 	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
58 	    len, wr, NULL);
59 	if (ret < 0) {
60 		SetPageError(page);
61 		mapping_set_error(page->mapping, ret);
62 	} else {
63 		ret = 0;
64 	}
65 	if (wr) {
66 		kfree(wr);
67 		set_page_private(page, 0);
68 		ClearPagePrivate(page);
69 		put_page(page);
70 	}
71 	return ret;
72 }
73 
74 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
75 {
76 	int ret;
77 	ret = orangefs_writepage_locked(page, wbc);
78 	unlock_page(page);
79 	end_page_writeback(page);
80 	return ret;
81 }
82 
83 struct orangefs_writepages {
84 	loff_t off;
85 	size_t len;
86 	kuid_t uid;
87 	kgid_t gid;
88 	int maxpages;
89 	int npages;
90 	struct page **pages;
91 	struct bio_vec *bv;
92 };
93 
94 static int orangefs_writepages_work(struct orangefs_writepages *ow,
95     struct writeback_control *wbc)
96 {
97 	struct inode *inode = ow->pages[0]->mapping->host;
98 	struct orangefs_write_range *wrp, wr;
99 	struct iov_iter iter;
100 	ssize_t ret;
101 	size_t len;
102 	loff_t off;
103 	int i;
104 
105 	len = i_size_read(inode);
106 
107 	for (i = 0; i < ow->npages; i++) {
108 		set_page_writeback(ow->pages[i]);
109 		ow->bv[i].bv_page = ow->pages[i];
110 		ow->bv[i].bv_len = min(page_offset(ow->pages[i]) + PAGE_SIZE,
111 		    ow->off + ow->len) -
112 		    max(ow->off, page_offset(ow->pages[i]));
113 		if (i == 0)
114 			ow->bv[i].bv_offset = ow->off -
115 			    page_offset(ow->pages[i]);
116 		else
117 			ow->bv[i].bv_offset = 0;
118 	}
119 	iov_iter_bvec(&iter, WRITE, ow->bv, ow->npages, ow->len);
120 
121 	WARN_ON(ow->off >= len);
122 	if (ow->off + ow->len > len)
123 		ow->len = len - ow->off;
124 
125 	off = ow->off;
126 	wr.uid = ow->uid;
127 	wr.gid = ow->gid;
128 	ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
129 	    0, &wr, NULL);
130 	if (ret < 0) {
131 		for (i = 0; i < ow->npages; i++) {
132 			SetPageError(ow->pages[i]);
133 			mapping_set_error(ow->pages[i]->mapping, ret);
134 			if (PagePrivate(ow->pages[i])) {
135 				wrp = (struct orangefs_write_range *)
136 				    page_private(ow->pages[i]);
137 				ClearPagePrivate(ow->pages[i]);
138 				put_page(ow->pages[i]);
139 				kfree(wrp);
140 			}
141 			end_page_writeback(ow->pages[i]);
142 			unlock_page(ow->pages[i]);
143 		}
144 	} else {
145 		ret = 0;
146 		for (i = 0; i < ow->npages; i++) {
147 			if (PagePrivate(ow->pages[i])) {
148 				wrp = (struct orangefs_write_range *)
149 				    page_private(ow->pages[i]);
150 				ClearPagePrivate(ow->pages[i]);
151 				put_page(ow->pages[i]);
152 				kfree(wrp);
153 			}
154 			end_page_writeback(ow->pages[i]);
155 			unlock_page(ow->pages[i]);
156 		}
157 	}
158 	return ret;
159 }
160 
161 static int orangefs_writepages_callback(struct page *page,
162     struct writeback_control *wbc, void *data)
163 {
164 	struct orangefs_writepages *ow = data;
165 	struct orangefs_write_range *wr;
166 	int ret;
167 
168 	if (!PagePrivate(page)) {
169 		unlock_page(page);
170 		/* It's not private so there's nothing to write, right? */
171 		printk("writepages_callback not private!\n");
172 		BUG();
173 		return 0;
174 	}
175 	wr = (struct orangefs_write_range *)page_private(page);
176 
177 	ret = -1;
178 	if (ow->npages == 0) {
179 		ow->off = wr->pos;
180 		ow->len = wr->len;
181 		ow->uid = wr->uid;
182 		ow->gid = wr->gid;
183 		ow->pages[ow->npages++] = page;
184 		ret = 0;
185 		goto done;
186 	}
187 	if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
188 		orangefs_writepages_work(ow, wbc);
189 		ow->npages = 0;
190 		ret = -1;
191 		goto done;
192 	}
193 	if (ow->off + ow->len == wr->pos) {
194 		ow->len += wr->len;
195 		ow->pages[ow->npages++] = page;
196 		ret = 0;
197 		goto done;
198 	}
199 done:
200 	if (ret == -1) {
201 		if (ow->npages) {
202 			orangefs_writepages_work(ow, wbc);
203 			ow->npages = 0;
204 		}
205 		ret = orangefs_writepage_locked(page, wbc);
206 		mapping_set_error(page->mapping, ret);
207 		unlock_page(page);
208 		end_page_writeback(page);
209 	} else {
210 		if (ow->npages == ow->maxpages) {
211 			orangefs_writepages_work(ow, wbc);
212 			ow->npages = 0;
213 		}
214 	}
215 	return ret;
216 }
217 
218 static int orangefs_writepages(struct address_space *mapping,
219     struct writeback_control *wbc)
220 {
221 	struct orangefs_writepages *ow;
222 	struct blk_plug plug;
223 	int ret;
224 	ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
225 	if (!ow)
226 		return -ENOMEM;
227 	ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
228 	ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
229 	if (!ow->pages) {
230 		kfree(ow);
231 		return -ENOMEM;
232 	}
233 	ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
234 	if (!ow->bv) {
235 		kfree(ow->pages);
236 		kfree(ow);
237 		return -ENOMEM;
238 	}
239 	blk_start_plug(&plug);
240 	ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
241 	if (ow->npages)
242 		ret = orangefs_writepages_work(ow, wbc);
243 	blk_finish_plug(&plug);
244 	kfree(ow->pages);
245 	kfree(ow->bv);
246 	kfree(ow);
247 	return ret;
248 }
249 
250 static int orangefs_launder_page(struct page *);
251 
252 static int orangefs_readpage(struct file *file, struct page *page)
253 {
254 	struct inode *inode = page->mapping->host;
255 	struct iov_iter iter;
256 	struct bio_vec bv;
257 	ssize_t ret;
258 	loff_t off; /* offset into this page */
259 	pgoff_t index; /* which page */
260 	struct page *next_page;
261 	char *kaddr;
262 	struct orangefs_read_options *ro = file->private_data;
263 	loff_t read_size;
264 	loff_t roundedup;
265 	int buffer_index = -1; /* orangefs shared memory slot */
266 	int slot_index;   /* index into slot */
267 	int remaining;
268 
269 	/*
270 	 * If they set some miniscule size for "count" in read(2)
271 	 * (for example) then let's try to read a page, or the whole file
272 	 * if it is smaller than a page. Once "count" goes over a page
273 	 * then lets round up to the highest page size multiple that is
274 	 * less than or equal to "count" and do that much orangefs IO and
275 	 * try to fill as many pages as we can from it.
276 	 *
277 	 * "count" should be represented in ro->blksiz.
278 	 *
279 	 * inode->i_size = file size.
280 	 */
281 	if (ro) {
282 		if (ro->blksiz < PAGE_SIZE) {
283 			if (inode->i_size < PAGE_SIZE)
284 				read_size = inode->i_size;
285 			else
286 				read_size = PAGE_SIZE;
287 		} else {
288 			roundedup = ((PAGE_SIZE - 1) & ro->blksiz) ?
289 				((ro->blksiz + PAGE_SIZE) & ~(PAGE_SIZE -1)) :
290 				ro->blksiz;
291 			if (roundedup > inode->i_size)
292 				read_size = inode->i_size;
293 			else
294 				read_size = roundedup;
295 
296 		}
297 	} else {
298 		read_size = PAGE_SIZE;
299 	}
300 	if (!read_size)
301 		read_size = PAGE_SIZE;
302 
303 	if (PageDirty(page))
304 		orangefs_launder_page(page);
305 
306 	off = page_offset(page);
307 	index = off >> PAGE_SHIFT;
308 	bv.bv_page = page;
309 	bv.bv_len = PAGE_SIZE;
310 	bv.bv_offset = 0;
311 	iov_iter_bvec(&iter, READ, &bv, 1, PAGE_SIZE);
312 
313 	ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
314 	    read_size, inode->i_size, NULL, &buffer_index);
315 	remaining = ret;
316 	/* this will only zero remaining unread portions of the page data */
317 	iov_iter_zero(~0U, &iter);
318 	/* takes care of potential aliasing */
319 	flush_dcache_page(page);
320 	if (ret < 0) {
321 		SetPageError(page);
322 		unlock_page(page);
323 		goto out;
324 	} else {
325 		SetPageUptodate(page);
326 		if (PageError(page))
327 			ClearPageError(page);
328 		ret = 0;
329 	}
330 	/* unlock the page after the ->readpage() routine completes */
331 	unlock_page(page);
332 
333 	if (remaining > PAGE_SIZE) {
334 		slot_index = 0;
335 		while ((remaining - PAGE_SIZE) >= PAGE_SIZE) {
336 			remaining -= PAGE_SIZE;
337 			/*
338 			 * It is an optimization to try and fill more than one
339 			 * page... by now we've already gotten the single
340 			 * page we were after, if stuff doesn't seem to
341 			 * be going our way at this point just return
342 			 * and hope for the best.
343 			 *
344 			 * If we look for pages and they're already there is
345 			 * one reason to give up, and if they're not there
346 			 * and we can't create them is another reason.
347 			 */
348 
349 			index++;
350 			slot_index++;
351 			next_page = find_get_page(inode->i_mapping, index);
352 			if (next_page) {
353 				gossip_debug(GOSSIP_FILE_DEBUG,
354 					"%s: found next page, quitting\n",
355 					__func__);
356 				put_page(next_page);
357 				goto out;
358 			}
359 			next_page = find_or_create_page(inode->i_mapping,
360 							index,
361 							GFP_KERNEL);
362 			/*
363 			 * I've never hit this, leave it as a printk for
364 			 * now so it will be obvious.
365 			 */
366 			if (!next_page) {
367 				printk("%s: can't create next page, quitting\n",
368 					__func__);
369 				goto out;
370 			}
371 			kaddr = kmap_atomic(next_page);
372 			orangefs_bufmap_page_fill(kaddr,
373 						buffer_index,
374 						slot_index);
375 			kunmap_atomic(kaddr);
376 			SetPageUptodate(next_page);
377 			unlock_page(next_page);
378 			put_page(next_page);
379 		}
380 	}
381 
382 out:
383 	if (buffer_index != -1)
384 		orangefs_bufmap_put(buffer_index);
385 	return ret;
386 }
387 
388 static int orangefs_write_begin(struct file *file,
389     struct address_space *mapping,
390     loff_t pos, unsigned len, unsigned flags, struct page **pagep,
391     void **fsdata)
392 {
393 	struct orangefs_write_range *wr;
394 	struct page *page;
395 	pgoff_t index;
396 	int ret;
397 
398 	index = pos >> PAGE_SHIFT;
399 
400 	page = grab_cache_page_write_begin(mapping, index, flags);
401 	if (!page)
402 		return -ENOMEM;
403 
404 	*pagep = page;
405 
406 	if (PageDirty(page) && !PagePrivate(page)) {
407 		/*
408 		 * Should be impossible.  If it happens, launder the page
409 		 * since we don't know what's dirty.  This will WARN in
410 		 * orangefs_writepage_locked.
411 		 */
412 		ret = orangefs_launder_page(page);
413 		if (ret)
414 			return ret;
415 	}
416 	if (PagePrivate(page)) {
417 		struct orangefs_write_range *wr;
418 		wr = (struct orangefs_write_range *)page_private(page);
419 		if (wr->pos + wr->len == pos &&
420 		    uid_eq(wr->uid, current_fsuid()) &&
421 		    gid_eq(wr->gid, current_fsgid())) {
422 			wr->len += len;
423 			goto okay;
424 		} else {
425 			ret = orangefs_launder_page(page);
426 			if (ret)
427 				return ret;
428 		}
429 	}
430 
431 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
432 	if (!wr)
433 		return -ENOMEM;
434 
435 	wr->pos = pos;
436 	wr->len = len;
437 	wr->uid = current_fsuid();
438 	wr->gid = current_fsgid();
439 	SetPagePrivate(page);
440 	set_page_private(page, (unsigned long)wr);
441 	get_page(page);
442 okay:
443 	return 0;
444 }
445 
446 static int orangefs_write_end(struct file *file, struct address_space *mapping,
447     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
448 {
449 	struct inode *inode = page->mapping->host;
450 	loff_t last_pos = pos + copied;
451 
452 	/*
453 	 * No need to use i_size_read() here, the i_size
454 	 * cannot change under us because we hold the i_mutex.
455 	 */
456 	if (last_pos > inode->i_size)
457 		i_size_write(inode, last_pos);
458 
459 	/* zero the stale part of the page if we did a short copy */
460 	if (!PageUptodate(page)) {
461 		unsigned from = pos & (PAGE_SIZE - 1);
462 		if (copied < len) {
463 			zero_user(page, from + copied, len - copied);
464 		}
465 		/* Set fully written pages uptodate. */
466 		if (pos == page_offset(page) &&
467 		    (len == PAGE_SIZE || pos + len == inode->i_size)) {
468 			zero_user_segment(page, from + copied, PAGE_SIZE);
469 			SetPageUptodate(page);
470 		}
471 	}
472 
473 	set_page_dirty(page);
474 	unlock_page(page);
475 	put_page(page);
476 
477 	mark_inode_dirty_sync(file_inode(file));
478 	return copied;
479 }
480 
481 static void orangefs_invalidatepage(struct page *page,
482 				 unsigned int offset,
483 				 unsigned int length)
484 {
485 	struct orangefs_write_range *wr;
486 	wr = (struct orangefs_write_range *)page_private(page);
487 
488 	if (offset == 0 && length == PAGE_SIZE) {
489 		kfree((struct orangefs_write_range *)page_private(page));
490 		set_page_private(page, 0);
491 		ClearPagePrivate(page);
492 		put_page(page);
493 		return;
494 	/* write range entirely within invalidate range (or equal) */
495 	} else if (page_offset(page) + offset <= wr->pos &&
496 	    wr->pos + wr->len <= page_offset(page) + offset + length) {
497 		kfree((struct orangefs_write_range *)page_private(page));
498 		set_page_private(page, 0);
499 		ClearPagePrivate(page);
500 		put_page(page);
501 		/* XXX is this right? only caller in fs */
502 		cancel_dirty_page(page);
503 		return;
504 	/* invalidate range chops off end of write range */
505 	} else if (wr->pos < page_offset(page) + offset &&
506 	    wr->pos + wr->len <= page_offset(page) + offset + length &&
507 	     page_offset(page) + offset < wr->pos + wr->len) {
508 		size_t x;
509 		x = wr->pos + wr->len - (page_offset(page) + offset);
510 		WARN_ON(x > wr->len);
511 		wr->len -= x;
512 		wr->uid = current_fsuid();
513 		wr->gid = current_fsgid();
514 	/* invalidate range chops off beginning of write range */
515 	} else if (page_offset(page) + offset <= wr->pos &&
516 	    page_offset(page) + offset + length < wr->pos + wr->len &&
517 	    wr->pos < page_offset(page) + offset + length) {
518 		size_t x;
519 		x = page_offset(page) + offset + length - wr->pos;
520 		WARN_ON(x > wr->len);
521 		wr->pos += x;
522 		wr->len -= x;
523 		wr->uid = current_fsuid();
524 		wr->gid = current_fsgid();
525 	/* invalidate range entirely within write range (punch hole) */
526 	} else if (wr->pos < page_offset(page) + offset &&
527 	    page_offset(page) + offset + length < wr->pos + wr->len) {
528 		/* XXX what do we do here... should not WARN_ON */
529 		WARN_ON(1);
530 		/* punch hole */
531 		/*
532 		 * should we just ignore this and write it out anyway?
533 		 * it hardly makes sense
534 		 */
535 		return;
536 	/* non-overlapping ranges */
537 	} else {
538 		/* WARN if they do overlap */
539 		if (!((page_offset(page) + offset + length <= wr->pos) ^
540 		    (wr->pos + wr->len <= page_offset(page) + offset))) {
541 			WARN_ON(1);
542 			printk("invalidate range offset %llu length %u\n",
543 			    page_offset(page) + offset, length);
544 			printk("write range offset %llu length %zu\n",
545 			    wr->pos, wr->len);
546 		}
547 		return;
548 	}
549 
550 	/*
551 	 * Above there are returns where wr is freed or where we WARN.
552 	 * Thus the following runs if wr was modified above.
553 	 */
554 
555 	orangefs_launder_page(page);
556 }
557 
558 static int orangefs_releasepage(struct page *page, gfp_t foo)
559 {
560 	return !PagePrivate(page);
561 }
562 
563 static void orangefs_freepage(struct page *page)
564 {
565 	if (PagePrivate(page)) {
566 		kfree((struct orangefs_write_range *)page_private(page));
567 		set_page_private(page, 0);
568 		ClearPagePrivate(page);
569 		put_page(page);
570 	}
571 }
572 
573 static int orangefs_launder_page(struct page *page)
574 {
575 	int r = 0;
576 	struct writeback_control wbc = {
577 		.sync_mode = WB_SYNC_ALL,
578 		.nr_to_write = 0,
579 	};
580 	wait_on_page_writeback(page);
581 	if (clear_page_dirty_for_io(page)) {
582 		r = orangefs_writepage_locked(page, &wbc);
583 		end_page_writeback(page);
584 	}
585 	return r;
586 }
587 
588 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
589 				  struct iov_iter *iter)
590 {
591 	/*
592 	 * Comment from original do_readv_writev:
593 	 * Common entry point for read/write/readv/writev
594 	 * This function will dispatch it to either the direct I/O
595 	 * or buffered I/O path depending on the mount options and/or
596 	 * augmented/extended metadata attached to the file.
597 	 * Note: File extended attributes override any mount options.
598 	 */
599 	struct file *file = iocb->ki_filp;
600 	loff_t pos = iocb->ki_pos;
601 	enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
602             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
603 	loff_t *offset = &pos;
604 	struct inode *inode = file->f_mapping->host;
605 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
606 	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
607 	size_t count = iov_iter_count(iter);
608 	ssize_t total_count = 0;
609 	ssize_t ret = -EINVAL;
610 	int i = 0;
611 
612 	gossip_debug(GOSSIP_FILE_DEBUG,
613 		"%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
614 		__func__,
615 		handle,
616 		(int)count);
617 
618 	if (type == ORANGEFS_IO_WRITE) {
619 		gossip_debug(GOSSIP_FILE_DEBUG,
620 			     "%s(%pU): proceeding with offset : %llu, "
621 			     "size %d\n",
622 			     __func__,
623 			     handle,
624 			     llu(*offset),
625 			     (int)count);
626 	}
627 
628 	if (count == 0) {
629 		ret = 0;
630 		goto out;
631 	}
632 
633 	while (iov_iter_count(iter)) {
634 		size_t each_count = iov_iter_count(iter);
635 		size_t amt_complete;
636 		i++;
637 
638 		/* how much to transfer in this loop iteration */
639 		if (each_count > orangefs_bufmap_size_query())
640 			each_count = orangefs_bufmap_size_query();
641 
642 		gossip_debug(GOSSIP_FILE_DEBUG,
643 			     "%s(%pU): size of each_count(%d)\n",
644 			     __func__,
645 			     handle,
646 			     (int)each_count);
647 		gossip_debug(GOSSIP_FILE_DEBUG,
648 			     "%s(%pU): BEFORE wait_for_io: offset is %d\n",
649 			     __func__,
650 			     handle,
651 			     (int)*offset);
652 
653 		ret = wait_for_direct_io(type, inode, offset, iter,
654 				each_count, 0, NULL, NULL);
655 		gossip_debug(GOSSIP_FILE_DEBUG,
656 			     "%s(%pU): return from wait_for_io:%d\n",
657 			     __func__,
658 			     handle,
659 			     (int)ret);
660 
661 		if (ret < 0)
662 			goto out;
663 
664 		*offset += ret;
665 		total_count += ret;
666 		amt_complete = ret;
667 
668 		gossip_debug(GOSSIP_FILE_DEBUG,
669 			     "%s(%pU): AFTER wait_for_io: offset is %d\n",
670 			     __func__,
671 			     handle,
672 			     (int)*offset);
673 
674 		/*
675 		 * if we got a short I/O operations,
676 		 * fall out and return what we got so far
677 		 */
678 		if (amt_complete < each_count)
679 			break;
680 	} /*end while */
681 
682 out:
683 	if (total_count > 0)
684 		ret = total_count;
685 	if (ret > 0) {
686 		if (type == ORANGEFS_IO_READ) {
687 			file_accessed(file);
688 		} else {
689 			file_update_time(file);
690 			if (*offset > i_size_read(inode))
691 				i_size_write(inode, *offset);
692 		}
693 	}
694 
695 	gossip_debug(GOSSIP_FILE_DEBUG,
696 		     "%s(%pU): Value(%d) returned.\n",
697 		     __func__,
698 		     handle,
699 		     (int)ret);
700 
701 	return ret;
702 }
703 
704 /** ORANGEFS2 implementation of address space operations */
705 static const struct address_space_operations orangefs_address_operations = {
706 	.writepage = orangefs_writepage,
707 	.readpage = orangefs_readpage,
708 	.writepages = orangefs_writepages,
709 	.set_page_dirty = __set_page_dirty_nobuffers,
710 	.write_begin = orangefs_write_begin,
711 	.write_end = orangefs_write_end,
712 	.invalidatepage = orangefs_invalidatepage,
713 	.releasepage = orangefs_releasepage,
714 	.freepage = orangefs_freepage,
715 	.launder_page = orangefs_launder_page,
716 	.direct_IO = orangefs_direct_IO,
717 };
718 
719 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
720 {
721 	struct page *page = vmf->page;
722 	struct inode *inode = file_inode(vmf->vma->vm_file);
723 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
724 	unsigned long *bitlock = &orangefs_inode->bitlock;
725 	vm_fault_t ret;
726 	struct orangefs_write_range *wr;
727 
728 	sb_start_pagefault(inode->i_sb);
729 
730 	if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
731 		ret = VM_FAULT_RETRY;
732 		goto out;
733 	}
734 
735 	lock_page(page);
736 	if (PageDirty(page) && !PagePrivate(page)) {
737 		/*
738 		 * Should be impossible.  If it happens, launder the page
739 		 * since we don't know what's dirty.  This will WARN in
740 		 * orangefs_writepage_locked.
741 		 */
742 		if (orangefs_launder_page(page)) {
743 			ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
744 			goto out;
745 		}
746 	}
747 	if (PagePrivate(page)) {
748 		wr = (struct orangefs_write_range *)page_private(page);
749 		if (uid_eq(wr->uid, current_fsuid()) &&
750 		    gid_eq(wr->gid, current_fsgid())) {
751 			wr->pos = page_offset(page);
752 			wr->len = PAGE_SIZE;
753 			goto okay;
754 		} else {
755 			if (orangefs_launder_page(page)) {
756 				ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
757 				goto out;
758 			}
759 		}
760 	}
761 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
762 	if (!wr) {
763 		ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
764 		goto out;
765 	}
766 	wr->pos = page_offset(page);
767 	wr->len = PAGE_SIZE;
768 	wr->uid = current_fsuid();
769 	wr->gid = current_fsgid();
770 	SetPagePrivate(page);
771 	set_page_private(page, (unsigned long)wr);
772 	get_page(page);
773 okay:
774 
775 	file_update_time(vmf->vma->vm_file);
776 	if (page->mapping != inode->i_mapping) {
777 		unlock_page(page);
778 		ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
779 		goto out;
780 	}
781 
782 	/*
783 	 * We mark the page dirty already here so that when freeze is in
784 	 * progress, we are guaranteed that writeback during freezing will
785 	 * see the dirty page and writeprotect it again.
786 	 */
787 	set_page_dirty(page);
788 	wait_for_stable_page(page);
789 	ret = VM_FAULT_LOCKED;
790 out:
791 	sb_end_pagefault(inode->i_sb);
792 	return ret;
793 }
794 
795 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
796 {
797 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
798 	struct orangefs_kernel_op_s *new_op;
799 	loff_t orig_size;
800 	int ret = -EINVAL;
801 
802 	gossip_debug(GOSSIP_INODE_DEBUG,
803 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
804 		     __func__,
805 		     get_khandle_from_ino(inode),
806 		     &orangefs_inode->refn.khandle,
807 		     orangefs_inode->refn.fs_id,
808 		     iattr->ia_size);
809 
810 	/* Ensure that we have a up to date size, so we know if it changed. */
811 	ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
812 	if (ret == -ESTALE)
813 		ret = -EIO;
814 	if (ret) {
815 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
816 		    __func__, ret);
817 		return ret;
818 	}
819 	orig_size = i_size_read(inode);
820 
821 	/* This is truncate_setsize in a different order. */
822 	truncate_pagecache(inode, iattr->ia_size);
823 	i_size_write(inode, iattr->ia_size);
824 	if (iattr->ia_size > orig_size)
825 		pagecache_isize_extended(inode, orig_size, iattr->ia_size);
826 
827 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
828 	if (!new_op)
829 		return -ENOMEM;
830 
831 	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
832 	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
833 
834 	ret = service_operation(new_op,
835 		__func__,
836 		get_interruptible_flag(inode));
837 
838 	/*
839 	 * the truncate has no downcall members to retrieve, but
840 	 * the status value tells us if it went through ok or not
841 	 */
842 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
843 
844 	op_release(new_op);
845 
846 	if (ret != 0)
847 		return ret;
848 
849 	if (orig_size != i_size_read(inode))
850 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
851 
852 	return ret;
853 }
854 
855 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
856 {
857 	int ret;
858 
859 	if (iattr->ia_valid & ATTR_MODE) {
860 		if (iattr->ia_mode & (S_ISVTX)) {
861 			if (is_root_handle(inode)) {
862 				/*
863 				 * allow sticky bit to be set on root (since
864 				 * it shows up that way by default anyhow),
865 				 * but don't show it to the server
866 				 */
867 				iattr->ia_mode -= S_ISVTX;
868 			} else {
869 				gossip_debug(GOSSIP_UTILS_DEBUG,
870 					     "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
871 				ret = -EINVAL;
872 				goto out;
873 			}
874 		}
875 		if (iattr->ia_mode & (S_ISUID)) {
876 			gossip_debug(GOSSIP_UTILS_DEBUG,
877 				     "Attempting to set setuid bit (not supported); returning EINVAL.\n");
878 			ret = -EINVAL;
879 			goto out;
880 		}
881 	}
882 
883 	if (iattr->ia_valid & ATTR_SIZE) {
884 		ret = orangefs_setattr_size(inode, iattr);
885 		if (ret)
886 			goto out;
887 	}
888 
889 again:
890 	spin_lock(&inode->i_lock);
891 	if (ORANGEFS_I(inode)->attr_valid) {
892 		if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
893 		    gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
894 			ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
895 		} else {
896 			spin_unlock(&inode->i_lock);
897 			write_inode_now(inode, 1);
898 			goto again;
899 		}
900 	} else {
901 		ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
902 		ORANGEFS_I(inode)->attr_uid = current_fsuid();
903 		ORANGEFS_I(inode)->attr_gid = current_fsgid();
904 	}
905 	setattr_copy(inode, iattr);
906 	spin_unlock(&inode->i_lock);
907 	mark_inode_dirty(inode);
908 
909 	if (iattr->ia_valid & ATTR_MODE)
910 		/* change mod on a file that has ACLs */
911 		ret = posix_acl_chmod(inode, inode->i_mode);
912 
913 	ret = 0;
914 out:
915 	return ret;
916 }
917 
918 /*
919  * Change attributes of an object referenced by dentry.
920  */
921 int orangefs_setattr(struct dentry *dentry, struct iattr *iattr)
922 {
923 	int ret;
924 	gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
925 	    dentry);
926 	ret = setattr_prepare(dentry, iattr);
927 	if (ret)
928 	        goto out;
929 	ret = __orangefs_setattr(d_inode(dentry), iattr);
930 	sync_inode_metadata(d_inode(dentry), 1);
931 out:
932 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
933 	    ret);
934 	return ret;
935 }
936 
937 /*
938  * Obtain attributes of an object given a dentry
939  */
940 int orangefs_getattr(const struct path *path, struct kstat *stat,
941 		     u32 request_mask, unsigned int flags)
942 {
943 	int ret = -ENOENT;
944 	struct inode *inode = path->dentry->d_inode;
945 
946 	gossip_debug(GOSSIP_INODE_DEBUG,
947 		     "orangefs_getattr: called on %pd mask %u\n",
948 		     path->dentry, request_mask);
949 
950 	ret = orangefs_inode_getattr(inode,
951 	    request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
952 	if (ret == 0) {
953 		generic_fillattr(inode, stat);
954 
955 		/* override block size reported to stat */
956 		if (!(request_mask & STATX_SIZE))
957 			stat->result_mask &= ~STATX_SIZE;
958 
959 		stat->attributes_mask = STATX_ATTR_IMMUTABLE |
960 		    STATX_ATTR_APPEND;
961 		if (inode->i_flags & S_IMMUTABLE)
962 			stat->attributes |= STATX_ATTR_IMMUTABLE;
963 		if (inode->i_flags & S_APPEND)
964 			stat->attributes |= STATX_ATTR_APPEND;
965 	}
966 	return ret;
967 }
968 
969 int orangefs_permission(struct inode *inode, int mask)
970 {
971 	int ret;
972 
973 	if (mask & MAY_NOT_BLOCK)
974 		return -ECHILD;
975 
976 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
977 
978 	/* Make sure the permission (and other common attrs) are up to date. */
979 	ret = orangefs_inode_getattr(inode, 0);
980 	if (ret < 0)
981 		return ret;
982 
983 	return generic_permission(inode, mask);
984 }
985 
986 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
987 {
988 	struct iattr iattr;
989 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
990 	    get_khandle_from_ino(inode));
991 	generic_update_time(inode, time, flags);
992 	memset(&iattr, 0, sizeof iattr);
993         if (flags & S_ATIME)
994 		iattr.ia_valid |= ATTR_ATIME;
995 	if (flags & S_CTIME)
996 		iattr.ia_valid |= ATTR_CTIME;
997 	if (flags & S_MTIME)
998 		iattr.ia_valid |= ATTR_MTIME;
999 	return __orangefs_setattr(inode, &iattr);
1000 }
1001 
1002 /* ORANGEFS2 implementation of VFS inode operations for files */
1003 static const struct inode_operations orangefs_file_inode_operations = {
1004 	.get_acl = orangefs_get_acl,
1005 	.set_acl = orangefs_set_acl,
1006 	.setattr = orangefs_setattr,
1007 	.getattr = orangefs_getattr,
1008 	.listxattr = orangefs_listxattr,
1009 	.permission = orangefs_permission,
1010 	.update_time = orangefs_update_time,
1011 };
1012 
1013 static int orangefs_init_iops(struct inode *inode)
1014 {
1015 	inode->i_mapping->a_ops = &orangefs_address_operations;
1016 
1017 	switch (inode->i_mode & S_IFMT) {
1018 	case S_IFREG:
1019 		inode->i_op = &orangefs_file_inode_operations;
1020 		inode->i_fop = &orangefs_file_operations;
1021 		break;
1022 	case S_IFLNK:
1023 		inode->i_op = &orangefs_symlink_inode_operations;
1024 		break;
1025 	case S_IFDIR:
1026 		inode->i_op = &orangefs_dir_inode_operations;
1027 		inode->i_fop = &orangefs_dir_operations;
1028 		break;
1029 	default:
1030 		gossip_debug(GOSSIP_INODE_DEBUG,
1031 			     "%s: unsupported mode\n",
1032 			     __func__);
1033 		return -EINVAL;
1034 	}
1035 
1036 	return 0;
1037 }
1038 
1039 /*
1040  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1041  * a ino_t type that will be used as a hash-index from where the handle will
1042  * be searched for in the VFS hash table of inodes.
1043  */
1044 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1045 {
1046 	if (!ref)
1047 		return 0;
1048 	return orangefs_khandle_to_ino(&(ref->khandle));
1049 }
1050 
1051 /*
1052  * Called to set up an inode from iget5_locked.
1053  */
1054 static int orangefs_set_inode(struct inode *inode, void *data)
1055 {
1056 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1057 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1058 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1059 	ORANGEFS_I(inode)->attr_valid = 0;
1060 	hash_init(ORANGEFS_I(inode)->xattr_cache);
1061 	ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1062 	ORANGEFS_I(inode)->bitlock = 0;
1063 	return 0;
1064 }
1065 
1066 /*
1067  * Called to determine if handles match.
1068  */
1069 static int orangefs_test_inode(struct inode *inode, void *data)
1070 {
1071 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1072 	struct orangefs_inode_s *orangefs_inode = NULL;
1073 
1074 	orangefs_inode = ORANGEFS_I(inode);
1075 	/* test handles and fs_ids... */
1076 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1077 				&(ref->khandle)) &&
1078 			orangefs_inode->refn.fs_id == ref->fs_id);
1079 }
1080 
1081 /*
1082  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1083  * file handle.
1084  *
1085  * @sb: the file system super block instance.
1086  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1087  */
1088 struct inode *orangefs_iget(struct super_block *sb,
1089 		struct orangefs_object_kref *ref)
1090 {
1091 	struct inode *inode = NULL;
1092 	unsigned long hash;
1093 	int error;
1094 
1095 	hash = orangefs_handle_hash(ref);
1096 	inode = iget5_locked(sb,
1097 			hash,
1098 			orangefs_test_inode,
1099 			orangefs_set_inode,
1100 			ref);
1101 
1102 	if (!inode)
1103 		return ERR_PTR(-ENOMEM);
1104 
1105 	if (!(inode->i_state & I_NEW))
1106 		return inode;
1107 
1108 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1109 	if (error) {
1110 		iget_failed(inode);
1111 		return ERR_PTR(error);
1112 	}
1113 
1114 	inode->i_ino = hash;	/* needed for stat etc */
1115 	orangefs_init_iops(inode);
1116 	unlock_new_inode(inode);
1117 
1118 	gossip_debug(GOSSIP_INODE_DEBUG,
1119 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1120 		     &ref->khandle,
1121 		     ref->fs_id,
1122 		     hash,
1123 		     inode->i_ino);
1124 
1125 	return inode;
1126 }
1127 
1128 /*
1129  * Allocate an inode for a newly created file and insert it into the inode hash.
1130  */
1131 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1132 		int mode, dev_t dev, struct orangefs_object_kref *ref)
1133 {
1134 	unsigned long hash = orangefs_handle_hash(ref);
1135 	struct inode *inode;
1136 	int error;
1137 
1138 	gossip_debug(GOSSIP_INODE_DEBUG,
1139 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1140 		     __func__,
1141 		     sb,
1142 		     MAJOR(dev),
1143 		     MINOR(dev),
1144 		     mode);
1145 
1146 	inode = new_inode(sb);
1147 	if (!inode)
1148 		return ERR_PTR(-ENOMEM);
1149 
1150 	orangefs_set_inode(inode, ref);
1151 	inode->i_ino = hash;	/* needed for stat etc */
1152 
1153 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1154 	if (error)
1155 		goto out_iput;
1156 
1157 	orangefs_init_iops(inode);
1158 	inode->i_rdev = dev;
1159 
1160 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1161 	if (error < 0)
1162 		goto out_iput;
1163 
1164 	gossip_debug(GOSSIP_INODE_DEBUG,
1165 		     "Initializing ACL's for inode %pU\n",
1166 		     get_khandle_from_ino(inode));
1167 	orangefs_init_acl(inode, dir);
1168 	return inode;
1169 
1170 out_iput:
1171 	iput(inode);
1172 	return ERR_PTR(error);
1173 }
1174