xref: /openbmc/linux/fs/orangefs/inode.c (revision 2cf1c348)
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_invalidatepage. */
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_page(struct page *);
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_readpage(struct file *file, struct page *page)
292 {
293 	struct inode *inode = page->mapping->host;
294 	struct iov_iter iter;
295 	struct bio_vec bv;
296 	ssize_t ret;
297 	loff_t off; /* offset into this page */
298 
299 	if (PageDirty(page))
300 		orangefs_launder_page(page);
301 
302 	off = page_offset(page);
303 	bv.bv_page = page;
304 	bv.bv_len = PAGE_SIZE;
305 	bv.bv_offset = 0;
306 	iov_iter_bvec(&iter, READ, &bv, 1, PAGE_SIZE);
307 
308 	ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
309 	    PAGE_SIZE, inode->i_size, NULL, NULL, file);
310 	/* this will only zero remaining unread portions of the page data */
311 	iov_iter_zero(~0U, &iter);
312 	/* takes care of potential aliasing */
313 	flush_dcache_page(page);
314 	if (ret < 0) {
315 		SetPageError(page);
316 	} else {
317 		SetPageUptodate(page);
318 		if (PageError(page))
319 			ClearPageError(page);
320 		ret = 0;
321 	}
322 	/* unlock the page after the ->readpage() routine completes */
323 	unlock_page(page);
324         return ret;
325 }
326 
327 static int orangefs_write_begin(struct file *file,
328     struct address_space *mapping,
329     loff_t pos, unsigned len, unsigned flags, struct page **pagep,
330     void **fsdata)
331 {
332 	struct orangefs_write_range *wr;
333 	struct page *page;
334 	pgoff_t index;
335 	int ret;
336 
337 	index = pos >> PAGE_SHIFT;
338 
339 	page = grab_cache_page_write_begin(mapping, index, flags);
340 	if (!page)
341 		return -ENOMEM;
342 
343 	*pagep = page;
344 
345 	if (PageDirty(page) && !PagePrivate(page)) {
346 		/*
347 		 * Should be impossible.  If it happens, launder the page
348 		 * since we don't know what's dirty.  This will WARN in
349 		 * orangefs_writepage_locked.
350 		 */
351 		ret = orangefs_launder_page(page);
352 		if (ret)
353 			return ret;
354 	}
355 	if (PagePrivate(page)) {
356 		struct orangefs_write_range *wr;
357 		wr = (struct orangefs_write_range *)page_private(page);
358 		if (wr->pos + wr->len == pos &&
359 		    uid_eq(wr->uid, current_fsuid()) &&
360 		    gid_eq(wr->gid, current_fsgid())) {
361 			wr->len += len;
362 			goto okay;
363 		} else {
364 			ret = orangefs_launder_page(page);
365 			if (ret)
366 				return ret;
367 		}
368 	}
369 
370 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
371 	if (!wr)
372 		return -ENOMEM;
373 
374 	wr->pos = pos;
375 	wr->len = len;
376 	wr->uid = current_fsuid();
377 	wr->gid = current_fsgid();
378 	attach_page_private(page, wr);
379 okay:
380 	return 0;
381 }
382 
383 static int orangefs_write_end(struct file *file, struct address_space *mapping,
384     loff_t pos, unsigned len, unsigned copied, struct page *page, void *fsdata)
385 {
386 	struct inode *inode = page->mapping->host;
387 	loff_t last_pos = pos + copied;
388 
389 	/*
390 	 * No need to use i_size_read() here, the i_size
391 	 * cannot change under us because we hold the i_mutex.
392 	 */
393 	if (last_pos > inode->i_size)
394 		i_size_write(inode, last_pos);
395 
396 	/* zero the stale part of the page if we did a short copy */
397 	if (!PageUptodate(page)) {
398 		unsigned from = pos & (PAGE_SIZE - 1);
399 		if (copied < len) {
400 			zero_user(page, from + copied, len - copied);
401 		}
402 		/* Set fully written pages uptodate. */
403 		if (pos == page_offset(page) &&
404 		    (len == PAGE_SIZE || pos + len == inode->i_size)) {
405 			zero_user_segment(page, from + copied, PAGE_SIZE);
406 			SetPageUptodate(page);
407 		}
408 	}
409 
410 	set_page_dirty(page);
411 	unlock_page(page);
412 	put_page(page);
413 
414 	mark_inode_dirty_sync(file_inode(file));
415 	return copied;
416 }
417 
418 static void orangefs_invalidatepage(struct page *page,
419 				 unsigned int offset,
420 				 unsigned int length)
421 {
422 	struct orangefs_write_range *wr;
423 	wr = (struct orangefs_write_range *)page_private(page);
424 
425 	if (offset == 0 && length == PAGE_SIZE) {
426 		kfree(detach_page_private(page));
427 		return;
428 	/* write range entirely within invalidate range (or equal) */
429 	} else if (page_offset(page) + offset <= wr->pos &&
430 	    wr->pos + wr->len <= page_offset(page) + offset + length) {
431 		kfree(detach_page_private(page));
432 		/* XXX is this right? only caller in fs */
433 		cancel_dirty_page(page);
434 		return;
435 	/* invalidate range chops off end of write range */
436 	} else if (wr->pos < page_offset(page) + offset &&
437 	    wr->pos + wr->len <= page_offset(page) + offset + length &&
438 	     page_offset(page) + offset < wr->pos + wr->len) {
439 		size_t x;
440 		x = wr->pos + wr->len - (page_offset(page) + offset);
441 		WARN_ON(x > wr->len);
442 		wr->len -= x;
443 		wr->uid = current_fsuid();
444 		wr->gid = current_fsgid();
445 	/* invalidate range chops off beginning of write range */
446 	} else if (page_offset(page) + offset <= wr->pos &&
447 	    page_offset(page) + offset + length < wr->pos + wr->len &&
448 	    wr->pos < page_offset(page) + offset + length) {
449 		size_t x;
450 		x = page_offset(page) + offset + length - wr->pos;
451 		WARN_ON(x > wr->len);
452 		wr->pos += x;
453 		wr->len -= x;
454 		wr->uid = current_fsuid();
455 		wr->gid = current_fsgid();
456 	/* invalidate range entirely within write range (punch hole) */
457 	} else if (wr->pos < page_offset(page) + offset &&
458 	    page_offset(page) + offset + length < wr->pos + wr->len) {
459 		/* XXX what do we do here... should not WARN_ON */
460 		WARN_ON(1);
461 		/* punch hole */
462 		/*
463 		 * should we just ignore this and write it out anyway?
464 		 * it hardly makes sense
465 		 */
466 		return;
467 	/* non-overlapping ranges */
468 	} else {
469 		/* WARN if they do overlap */
470 		if (!((page_offset(page) + offset + length <= wr->pos) ^
471 		    (wr->pos + wr->len <= page_offset(page) + offset))) {
472 			WARN_ON(1);
473 			printk("invalidate range offset %llu length %u\n",
474 			    page_offset(page) + offset, length);
475 			printk("write range offset %llu length %zu\n",
476 			    wr->pos, wr->len);
477 		}
478 		return;
479 	}
480 
481 	/*
482 	 * Above there are returns where wr is freed or where we WARN.
483 	 * Thus the following runs if wr was modified above.
484 	 */
485 
486 	orangefs_launder_page(page);
487 }
488 
489 static int orangefs_releasepage(struct page *page, gfp_t foo)
490 {
491 	return !PagePrivate(page);
492 }
493 
494 static void orangefs_freepage(struct page *page)
495 {
496 	kfree(detach_page_private(page));
497 }
498 
499 static int orangefs_launder_page(struct page *page)
500 {
501 	int r = 0;
502 	struct writeback_control wbc = {
503 		.sync_mode = WB_SYNC_ALL,
504 		.nr_to_write = 0,
505 	};
506 	wait_on_page_writeback(page);
507 	if (clear_page_dirty_for_io(page)) {
508 		r = orangefs_writepage_locked(page, &wbc);
509 		end_page_writeback(page);
510 	}
511 	return r;
512 }
513 
514 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
515 				  struct iov_iter *iter)
516 {
517 	/*
518 	 * Comment from original do_readv_writev:
519 	 * Common entry point for read/write/readv/writev
520 	 * This function will dispatch it to either the direct I/O
521 	 * or buffered I/O path depending on the mount options and/or
522 	 * augmented/extended metadata attached to the file.
523 	 * Note: File extended attributes override any mount options.
524 	 */
525 	struct file *file = iocb->ki_filp;
526 	loff_t pos = iocb->ki_pos;
527 	enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
528             ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
529 	loff_t *offset = &pos;
530 	struct inode *inode = file->f_mapping->host;
531 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
532 	struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
533 	size_t count = iov_iter_count(iter);
534 	ssize_t total_count = 0;
535 	ssize_t ret = -EINVAL;
536 	int i = 0;
537 
538 	gossip_debug(GOSSIP_FILE_DEBUG,
539 		"%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
540 		__func__,
541 		handle,
542 		(int)count);
543 
544 	if (type == ORANGEFS_IO_WRITE) {
545 		gossip_debug(GOSSIP_FILE_DEBUG,
546 			     "%s(%pU): proceeding with offset : %llu, "
547 			     "size %d\n",
548 			     __func__,
549 			     handle,
550 			     llu(*offset),
551 			     (int)count);
552 	}
553 
554 	if (count == 0) {
555 		ret = 0;
556 		goto out;
557 	}
558 
559 	while (iov_iter_count(iter)) {
560 		size_t each_count = iov_iter_count(iter);
561 		size_t amt_complete;
562 		i++;
563 
564 		/* how much to transfer in this loop iteration */
565 		if (each_count > orangefs_bufmap_size_query())
566 			each_count = orangefs_bufmap_size_query();
567 
568 		gossip_debug(GOSSIP_FILE_DEBUG,
569 			     "%s(%pU): size of each_count(%d)\n",
570 			     __func__,
571 			     handle,
572 			     (int)each_count);
573 		gossip_debug(GOSSIP_FILE_DEBUG,
574 			     "%s(%pU): BEFORE wait_for_io: offset is %d\n",
575 			     __func__,
576 			     handle,
577 			     (int)*offset);
578 
579 		ret = wait_for_direct_io(type, inode, offset, iter,
580 				each_count, 0, NULL, NULL, file);
581 		gossip_debug(GOSSIP_FILE_DEBUG,
582 			     "%s(%pU): return from wait_for_io:%d\n",
583 			     __func__,
584 			     handle,
585 			     (int)ret);
586 
587 		if (ret < 0)
588 			goto out;
589 
590 		*offset += ret;
591 		total_count += ret;
592 		amt_complete = ret;
593 
594 		gossip_debug(GOSSIP_FILE_DEBUG,
595 			     "%s(%pU): AFTER wait_for_io: offset is %d\n",
596 			     __func__,
597 			     handle,
598 			     (int)*offset);
599 
600 		/*
601 		 * if we got a short I/O operations,
602 		 * fall out and return what we got so far
603 		 */
604 		if (amt_complete < each_count)
605 			break;
606 	} /*end while */
607 
608 out:
609 	if (total_count > 0)
610 		ret = total_count;
611 	if (ret > 0) {
612 		if (type == ORANGEFS_IO_READ) {
613 			file_accessed(file);
614 		} else {
615 			file_update_time(file);
616 			if (*offset > i_size_read(inode))
617 				i_size_write(inode, *offset);
618 		}
619 	}
620 
621 	gossip_debug(GOSSIP_FILE_DEBUG,
622 		     "%s(%pU): Value(%d) returned.\n",
623 		     __func__,
624 		     handle,
625 		     (int)ret);
626 
627 	return ret;
628 }
629 
630 /** ORANGEFS2 implementation of address space operations */
631 static const struct address_space_operations orangefs_address_operations = {
632 	.writepage = orangefs_writepage,
633 	.readahead = orangefs_readahead,
634 	.readpage = orangefs_readpage,
635 	.writepages = orangefs_writepages,
636 	.set_page_dirty = __set_page_dirty_nobuffers,
637 	.write_begin = orangefs_write_begin,
638 	.write_end = orangefs_write_end,
639 	.invalidatepage = orangefs_invalidatepage,
640 	.releasepage = orangefs_releasepage,
641 	.freepage = orangefs_freepage,
642 	.launder_page = orangefs_launder_page,
643 	.direct_IO = orangefs_direct_IO,
644 };
645 
646 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
647 {
648 	struct page *page = vmf->page;
649 	struct inode *inode = file_inode(vmf->vma->vm_file);
650 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
651 	unsigned long *bitlock = &orangefs_inode->bitlock;
652 	vm_fault_t ret;
653 	struct orangefs_write_range *wr;
654 
655 	sb_start_pagefault(inode->i_sb);
656 
657 	if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
658 		ret = VM_FAULT_RETRY;
659 		goto out;
660 	}
661 
662 	lock_page(page);
663 	if (PageDirty(page) && !PagePrivate(page)) {
664 		/*
665 		 * Should be impossible.  If it happens, launder the page
666 		 * since we don't know what's dirty.  This will WARN in
667 		 * orangefs_writepage_locked.
668 		 */
669 		if (orangefs_launder_page(page)) {
670 			ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
671 			goto out;
672 		}
673 	}
674 	if (PagePrivate(page)) {
675 		wr = (struct orangefs_write_range *)page_private(page);
676 		if (uid_eq(wr->uid, current_fsuid()) &&
677 		    gid_eq(wr->gid, current_fsgid())) {
678 			wr->pos = page_offset(page);
679 			wr->len = PAGE_SIZE;
680 			goto okay;
681 		} else {
682 			if (orangefs_launder_page(page)) {
683 				ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
684 				goto out;
685 			}
686 		}
687 	}
688 	wr = kmalloc(sizeof *wr, GFP_KERNEL);
689 	if (!wr) {
690 		ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
691 		goto out;
692 	}
693 	wr->pos = page_offset(page);
694 	wr->len = PAGE_SIZE;
695 	wr->uid = current_fsuid();
696 	wr->gid = current_fsgid();
697 	attach_page_private(page, wr);
698 okay:
699 
700 	file_update_time(vmf->vma->vm_file);
701 	if (page->mapping != inode->i_mapping) {
702 		unlock_page(page);
703 		ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
704 		goto out;
705 	}
706 
707 	/*
708 	 * We mark the page dirty already here so that when freeze is in
709 	 * progress, we are guaranteed that writeback during freezing will
710 	 * see the dirty page and writeprotect it again.
711 	 */
712 	set_page_dirty(page);
713 	wait_for_stable_page(page);
714 	ret = VM_FAULT_LOCKED;
715 out:
716 	sb_end_pagefault(inode->i_sb);
717 	return ret;
718 }
719 
720 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
721 {
722 	struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
723 	struct orangefs_kernel_op_s *new_op;
724 	loff_t orig_size;
725 	int ret = -EINVAL;
726 
727 	gossip_debug(GOSSIP_INODE_DEBUG,
728 		     "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
729 		     __func__,
730 		     get_khandle_from_ino(inode),
731 		     &orangefs_inode->refn.khandle,
732 		     orangefs_inode->refn.fs_id,
733 		     iattr->ia_size);
734 
735 	/* Ensure that we have a up to date size, so we know if it changed. */
736 	ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
737 	if (ret == -ESTALE)
738 		ret = -EIO;
739 	if (ret) {
740 		gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
741 		    __func__, ret);
742 		return ret;
743 	}
744 	orig_size = i_size_read(inode);
745 
746 	/* This is truncate_setsize in a different order. */
747 	truncate_pagecache(inode, iattr->ia_size);
748 	i_size_write(inode, iattr->ia_size);
749 	if (iattr->ia_size > orig_size)
750 		pagecache_isize_extended(inode, orig_size, iattr->ia_size);
751 
752 	new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
753 	if (!new_op)
754 		return -ENOMEM;
755 
756 	new_op->upcall.req.truncate.refn = orangefs_inode->refn;
757 	new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
758 
759 	ret = service_operation(new_op,
760 		__func__,
761 		get_interruptible_flag(inode));
762 
763 	/*
764 	 * the truncate has no downcall members to retrieve, but
765 	 * the status value tells us if it went through ok or not
766 	 */
767 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
768 
769 	op_release(new_op);
770 
771 	if (ret != 0)
772 		return ret;
773 
774 	if (orig_size != i_size_read(inode))
775 		iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
776 
777 	return ret;
778 }
779 
780 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
781 {
782 	int ret;
783 
784 	if (iattr->ia_valid & ATTR_MODE) {
785 		if (iattr->ia_mode & (S_ISVTX)) {
786 			if (is_root_handle(inode)) {
787 				/*
788 				 * allow sticky bit to be set on root (since
789 				 * it shows up that way by default anyhow),
790 				 * but don't show it to the server
791 				 */
792 				iattr->ia_mode -= S_ISVTX;
793 			} else {
794 				gossip_debug(GOSSIP_UTILS_DEBUG,
795 					     "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
796 				ret = -EINVAL;
797 				goto out;
798 			}
799 		}
800 		if (iattr->ia_mode & (S_ISUID)) {
801 			gossip_debug(GOSSIP_UTILS_DEBUG,
802 				     "Attempting to set setuid bit (not supported); returning EINVAL.\n");
803 			ret = -EINVAL;
804 			goto out;
805 		}
806 	}
807 
808 	if (iattr->ia_valid & ATTR_SIZE) {
809 		ret = orangefs_setattr_size(inode, iattr);
810 		if (ret)
811 			goto out;
812 	}
813 
814 again:
815 	spin_lock(&inode->i_lock);
816 	if (ORANGEFS_I(inode)->attr_valid) {
817 		if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
818 		    gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
819 			ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
820 		} else {
821 			spin_unlock(&inode->i_lock);
822 			write_inode_now(inode, 1);
823 			goto again;
824 		}
825 	} else {
826 		ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
827 		ORANGEFS_I(inode)->attr_uid = current_fsuid();
828 		ORANGEFS_I(inode)->attr_gid = current_fsgid();
829 	}
830 	setattr_copy(&init_user_ns, inode, iattr);
831 	spin_unlock(&inode->i_lock);
832 	mark_inode_dirty(inode);
833 
834 	if (iattr->ia_valid & ATTR_MODE)
835 		/* change mod on a file that has ACLs */
836 		ret = posix_acl_chmod(&init_user_ns, inode, inode->i_mode);
837 
838 	ret = 0;
839 out:
840 	return ret;
841 }
842 
843 /*
844  * Change attributes of an object referenced by dentry.
845  */
846 int orangefs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
847 		     struct iattr *iattr)
848 {
849 	int ret;
850 	gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
851 	    dentry);
852 	ret = setattr_prepare(&init_user_ns, dentry, iattr);
853 	if (ret)
854 	        goto out;
855 	ret = __orangefs_setattr(d_inode(dentry), iattr);
856 	sync_inode_metadata(d_inode(dentry), 1);
857 out:
858 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
859 	    ret);
860 	return ret;
861 }
862 
863 /*
864  * Obtain attributes of an object given a dentry
865  */
866 int orangefs_getattr(struct user_namespace *mnt_userns, const struct path *path,
867 		     struct kstat *stat, u32 request_mask, unsigned int flags)
868 {
869 	int ret;
870 	struct inode *inode = path->dentry->d_inode;
871 
872 	gossip_debug(GOSSIP_INODE_DEBUG,
873 		     "orangefs_getattr: called on %pd mask %u\n",
874 		     path->dentry, request_mask);
875 
876 	ret = orangefs_inode_getattr(inode,
877 	    request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
878 	if (ret == 0) {
879 		generic_fillattr(&init_user_ns, inode, stat);
880 
881 		/* override block size reported to stat */
882 		if (!(request_mask & STATX_SIZE))
883 			stat->result_mask &= ~STATX_SIZE;
884 
885 		generic_fill_statx_attr(inode, stat);
886 	}
887 	return ret;
888 }
889 
890 int orangefs_permission(struct user_namespace *mnt_userns,
891 			struct inode *inode, int mask)
892 {
893 	int ret;
894 
895 	if (mask & MAY_NOT_BLOCK)
896 		return -ECHILD;
897 
898 	gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
899 
900 	/* Make sure the permission (and other common attrs) are up to date. */
901 	ret = orangefs_inode_getattr(inode, 0);
902 	if (ret < 0)
903 		return ret;
904 
905 	return generic_permission(&init_user_ns, inode, mask);
906 }
907 
908 int orangefs_update_time(struct inode *inode, struct timespec64 *time, int flags)
909 {
910 	struct iattr iattr;
911 	gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
912 	    get_khandle_from_ino(inode));
913 	generic_update_time(inode, time, flags);
914 	memset(&iattr, 0, sizeof iattr);
915         if (flags & S_ATIME)
916 		iattr.ia_valid |= ATTR_ATIME;
917 	if (flags & S_CTIME)
918 		iattr.ia_valid |= ATTR_CTIME;
919 	if (flags & S_MTIME)
920 		iattr.ia_valid |= ATTR_MTIME;
921 	return __orangefs_setattr(inode, &iattr);
922 }
923 
924 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
925 {
926 	u64 val = 0;
927 	int ret;
928 
929 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
930 		     dentry);
931 
932 	ret = orangefs_inode_getxattr(d_inode(dentry),
933 				      "user.pvfs2.meta_hint",
934 				      &val, sizeof(val));
935 	if (ret < 0 && ret != -ENODATA)
936 		return ret;
937 
938 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
939 
940 	fileattr_fill_flags(fa, val);
941 	return 0;
942 }
943 
944 static int orangefs_fileattr_set(struct user_namespace *mnt_userns,
945 				 struct dentry *dentry, struct fileattr *fa)
946 {
947 	u64 val = 0;
948 
949 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
950 		     dentry);
951 	/*
952 	 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
953 	 * turned on for a file. The user is not allowed to turn on this bit,
954 	 * but the bit is present if the user first gets the flags and then
955 	 * updates the flags with some new settings. So, we ignore it in the
956 	 * following edit. bligon.
957 	 */
958 	if (fileattr_has_fsx(fa) ||
959 	    (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
960 		gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
961 			   __func__);
962 		return -EOPNOTSUPP;
963 	}
964 	val = fa->flags;
965 	gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
966 	return orangefs_inode_setxattr(d_inode(dentry),
967 				       "user.pvfs2.meta_hint",
968 				       &val, sizeof(val), 0);
969 }
970 
971 /* ORANGEFS2 implementation of VFS inode operations for files */
972 static const struct inode_operations orangefs_file_inode_operations = {
973 	.get_acl = orangefs_get_acl,
974 	.set_acl = orangefs_set_acl,
975 	.setattr = orangefs_setattr,
976 	.getattr = orangefs_getattr,
977 	.listxattr = orangefs_listxattr,
978 	.permission = orangefs_permission,
979 	.update_time = orangefs_update_time,
980 	.fileattr_get = orangefs_fileattr_get,
981 	.fileattr_set = orangefs_fileattr_set,
982 };
983 
984 static int orangefs_init_iops(struct inode *inode)
985 {
986 	inode->i_mapping->a_ops = &orangefs_address_operations;
987 
988 	switch (inode->i_mode & S_IFMT) {
989 	case S_IFREG:
990 		inode->i_op = &orangefs_file_inode_operations;
991 		inode->i_fop = &orangefs_file_operations;
992 		break;
993 	case S_IFLNK:
994 		inode->i_op = &orangefs_symlink_inode_operations;
995 		break;
996 	case S_IFDIR:
997 		inode->i_op = &orangefs_dir_inode_operations;
998 		inode->i_fop = &orangefs_dir_operations;
999 		break;
1000 	default:
1001 		gossip_debug(GOSSIP_INODE_DEBUG,
1002 			     "%s: unsupported mode\n",
1003 			     __func__);
1004 		return -EINVAL;
1005 	}
1006 
1007 	return 0;
1008 }
1009 
1010 /*
1011  * Given an ORANGEFS object identifier (fsid, handle), convert it into
1012  * a ino_t type that will be used as a hash-index from where the handle will
1013  * be searched for in the VFS hash table of inodes.
1014  */
1015 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1016 {
1017 	if (!ref)
1018 		return 0;
1019 	return orangefs_khandle_to_ino(&(ref->khandle));
1020 }
1021 
1022 /*
1023  * Called to set up an inode from iget5_locked.
1024  */
1025 static int orangefs_set_inode(struct inode *inode, void *data)
1026 {
1027 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1028 	ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1029 	ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1030 	ORANGEFS_I(inode)->attr_valid = 0;
1031 	hash_init(ORANGEFS_I(inode)->xattr_cache);
1032 	ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1033 	ORANGEFS_I(inode)->bitlock = 0;
1034 	return 0;
1035 }
1036 
1037 /*
1038  * Called to determine if handles match.
1039  */
1040 static int orangefs_test_inode(struct inode *inode, void *data)
1041 {
1042 	struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1043 	struct orangefs_inode_s *orangefs_inode = NULL;
1044 
1045 	orangefs_inode = ORANGEFS_I(inode);
1046 	/* test handles and fs_ids... */
1047 	return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1048 				&(ref->khandle)) &&
1049 			orangefs_inode->refn.fs_id == ref->fs_id);
1050 }
1051 
1052 /*
1053  * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1054  * file handle.
1055  *
1056  * @sb: the file system super block instance.
1057  * @ref: The ORANGEFS object for which we are trying to locate an inode.
1058  */
1059 struct inode *orangefs_iget(struct super_block *sb,
1060 		struct orangefs_object_kref *ref)
1061 {
1062 	struct inode *inode = NULL;
1063 	unsigned long hash;
1064 	int error;
1065 
1066 	hash = orangefs_handle_hash(ref);
1067 	inode = iget5_locked(sb,
1068 			hash,
1069 			orangefs_test_inode,
1070 			orangefs_set_inode,
1071 			ref);
1072 
1073 	if (!inode)
1074 		return ERR_PTR(-ENOMEM);
1075 
1076 	if (!(inode->i_state & I_NEW))
1077 		return inode;
1078 
1079 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1080 	if (error) {
1081 		iget_failed(inode);
1082 		return ERR_PTR(error);
1083 	}
1084 
1085 	inode->i_ino = hash;	/* needed for stat etc */
1086 	orangefs_init_iops(inode);
1087 	unlock_new_inode(inode);
1088 
1089 	gossip_debug(GOSSIP_INODE_DEBUG,
1090 		     "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1091 		     &ref->khandle,
1092 		     ref->fs_id,
1093 		     hash,
1094 		     inode->i_ino);
1095 
1096 	return inode;
1097 }
1098 
1099 /*
1100  * Allocate an inode for a newly created file and insert it into the inode hash.
1101  */
1102 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1103 		int mode, dev_t dev, struct orangefs_object_kref *ref)
1104 {
1105 	unsigned long hash = orangefs_handle_hash(ref);
1106 	struct inode *inode;
1107 	int error;
1108 
1109 	gossip_debug(GOSSIP_INODE_DEBUG,
1110 		     "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1111 		     __func__,
1112 		     sb,
1113 		     MAJOR(dev),
1114 		     MINOR(dev),
1115 		     mode);
1116 
1117 	inode = new_inode(sb);
1118 	if (!inode)
1119 		return ERR_PTR(-ENOMEM);
1120 
1121 	orangefs_set_inode(inode, ref);
1122 	inode->i_ino = hash;	/* needed for stat etc */
1123 
1124 	error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1125 	if (error)
1126 		goto out_iput;
1127 
1128 	orangefs_init_iops(inode);
1129 	inode->i_rdev = dev;
1130 
1131 	error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1132 	if (error < 0)
1133 		goto out_iput;
1134 
1135 	gossip_debug(GOSSIP_INODE_DEBUG,
1136 		     "Initializing ACL's for inode %pU\n",
1137 		     get_khandle_from_ino(inode));
1138 	orangefs_init_acl(inode, dir);
1139 	return inode;
1140 
1141 out_iput:
1142 	iput(inode);
1143 	return ERR_PTR(error);
1144 }
1145