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
orangefs_writepage_locked(struct page * page,struct writeback_control * wbc)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 wlen;
27 ssize_t ret;
28 loff_t len, off;
29
30 set_page_writeback(page);
31
32 len = i_size_read(inode);
33 if (PagePrivate(page)) {
34 wr = (struct orangefs_write_range *)page_private(page);
35 WARN_ON(wr->pos >= len);
36 off = wr->pos;
37 if (off + wr->len > len)
38 wlen = len - off;
39 else
40 wlen = wr->len;
41 } else {
42 WARN_ON(1);
43 off = page_offset(page);
44 if (off + PAGE_SIZE > len)
45 wlen = len - off;
46 else
47 wlen = PAGE_SIZE;
48 }
49 /* Should've been handled in orangefs_invalidate_folio. */
50 WARN_ON(off == len || off + wlen > len);
51
52 WARN_ON(wlen == 0);
53 bvec_set_page(&bv, page, wlen, off % PAGE_SIZE);
54 iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, wlen);
55
56 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, wlen,
57 len, wr, NULL, NULL);
58 if (ret < 0) {
59 mapping_set_error(page->mapping, ret);
60 } else {
61 ret = 0;
62 }
63 kfree(detach_page_private(page));
64 return ret;
65 }
66
orangefs_writepage(struct page * page,struct writeback_control * wbc)67 static int orangefs_writepage(struct page *page, struct writeback_control *wbc)
68 {
69 int ret;
70 ret = orangefs_writepage_locked(page, wbc);
71 unlock_page(page);
72 end_page_writeback(page);
73 return ret;
74 }
75
76 struct orangefs_writepages {
77 loff_t off;
78 size_t len;
79 kuid_t uid;
80 kgid_t gid;
81 int maxpages;
82 int npages;
83 struct page **pages;
84 struct bio_vec *bv;
85 };
86
orangefs_writepages_work(struct orangefs_writepages * ow,struct writeback_control * wbc)87 static int orangefs_writepages_work(struct orangefs_writepages *ow,
88 struct writeback_control *wbc)
89 {
90 struct inode *inode = ow->pages[0]->mapping->host;
91 struct orangefs_write_range *wrp, wr;
92 struct iov_iter iter;
93 ssize_t ret;
94 loff_t len, off;
95 int i;
96
97 len = i_size_read(inode);
98
99 for (i = 0; i < ow->npages; i++) {
100 set_page_writeback(ow->pages[i]);
101 bvec_set_page(&ow->bv[i], ow->pages[i],
102 min(page_offset(ow->pages[i]) + PAGE_SIZE,
103 ow->off + ow->len) -
104 max(ow->off, page_offset(ow->pages[i])),
105 i == 0 ? ow->off - page_offset(ow->pages[i]) : 0);
106 }
107 iov_iter_bvec(&iter, ITER_SOURCE, ow->bv, ow->npages, ow->len);
108
109 WARN_ON(ow->off >= len);
110 if (ow->off + ow->len > len)
111 ow->len = len - ow->off;
112
113 off = ow->off;
114 wr.uid = ow->uid;
115 wr.gid = ow->gid;
116 ret = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter, ow->len,
117 0, &wr, NULL, NULL);
118 if (ret < 0) {
119 for (i = 0; i < ow->npages; i++) {
120 mapping_set_error(ow->pages[i]->mapping, ret);
121 if (PagePrivate(ow->pages[i])) {
122 wrp = (struct orangefs_write_range *)
123 page_private(ow->pages[i]);
124 ClearPagePrivate(ow->pages[i]);
125 put_page(ow->pages[i]);
126 kfree(wrp);
127 }
128 end_page_writeback(ow->pages[i]);
129 unlock_page(ow->pages[i]);
130 }
131 } else {
132 ret = 0;
133 for (i = 0; i < ow->npages; i++) {
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 }
145 return ret;
146 }
147
orangefs_writepages_callback(struct folio * folio,struct writeback_control * wbc,void * data)148 static int orangefs_writepages_callback(struct folio *folio,
149 struct writeback_control *wbc, void *data)
150 {
151 struct orangefs_writepages *ow = data;
152 struct orangefs_write_range *wr = folio->private;
153 int ret;
154
155 if (!wr) {
156 folio_unlock(folio);
157 /* It's not private so there's nothing to write, right? */
158 printk("writepages_callback not private!\n");
159 BUG();
160 return 0;
161 }
162
163 ret = -1;
164 if (ow->npages == 0) {
165 ow->off = wr->pos;
166 ow->len = wr->len;
167 ow->uid = wr->uid;
168 ow->gid = wr->gid;
169 ow->pages[ow->npages++] = &folio->page;
170 ret = 0;
171 goto done;
172 }
173 if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
174 orangefs_writepages_work(ow, wbc);
175 ow->npages = 0;
176 ret = -1;
177 goto done;
178 }
179 if (ow->off + ow->len == wr->pos) {
180 ow->len += wr->len;
181 ow->pages[ow->npages++] = &folio->page;
182 ret = 0;
183 goto done;
184 }
185 done:
186 if (ret == -1) {
187 if (ow->npages) {
188 orangefs_writepages_work(ow, wbc);
189 ow->npages = 0;
190 }
191 ret = orangefs_writepage_locked(&folio->page, wbc);
192 mapping_set_error(folio->mapping, ret);
193 folio_unlock(folio);
194 folio_end_writeback(folio);
195 } else {
196 if (ow->npages == ow->maxpages) {
197 orangefs_writepages_work(ow, wbc);
198 ow->npages = 0;
199 }
200 }
201 return ret;
202 }
203
orangefs_writepages(struct address_space * mapping,struct writeback_control * wbc)204 static int orangefs_writepages(struct address_space *mapping,
205 struct writeback_control *wbc)
206 {
207 struct orangefs_writepages *ow;
208 struct blk_plug plug;
209 int ret;
210 ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
211 if (!ow)
212 return -ENOMEM;
213 ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
214 ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
215 if (!ow->pages) {
216 kfree(ow);
217 return -ENOMEM;
218 }
219 ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
220 if (!ow->bv) {
221 kfree(ow->pages);
222 kfree(ow);
223 return -ENOMEM;
224 }
225 blk_start_plug(&plug);
226 ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
227 if (ow->npages)
228 ret = orangefs_writepages_work(ow, wbc);
229 blk_finish_plug(&plug);
230 kfree(ow->pages);
231 kfree(ow->bv);
232 kfree(ow);
233 return ret;
234 }
235
236 static int orangefs_launder_folio(struct folio *);
237
orangefs_readahead(struct readahead_control * rac)238 static void orangefs_readahead(struct readahead_control *rac)
239 {
240 loff_t offset;
241 struct iov_iter iter;
242 struct inode *inode = rac->mapping->host;
243 struct xarray *i_pages;
244 struct folio *folio;
245 loff_t new_start = readahead_pos(rac);
246 int ret;
247 size_t new_len = 0;
248
249 loff_t bytes_remaining = inode->i_size - readahead_pos(rac);
250 loff_t pages_remaining = bytes_remaining / PAGE_SIZE;
251
252 if (pages_remaining >= 1024)
253 new_len = 4194304;
254 else if (pages_remaining > readahead_count(rac))
255 new_len = bytes_remaining;
256
257 if (new_len)
258 readahead_expand(rac, new_start, new_len);
259
260 offset = readahead_pos(rac);
261 i_pages = &rac->mapping->i_pages;
262
263 iov_iter_xarray(&iter, ITER_DEST, i_pages, offset, readahead_length(rac));
264
265 /* read in the pages. */
266 if ((ret = wait_for_direct_io(ORANGEFS_IO_READ, inode,
267 &offset, &iter, readahead_length(rac),
268 inode->i_size, NULL, NULL, rac->file)) < 0)
269 gossip_debug(GOSSIP_FILE_DEBUG,
270 "%s: wait_for_direct_io failed. \n", __func__);
271 else
272 ret = 0;
273
274 /* clean up. */
275 while ((folio = readahead_folio(rac))) {
276 if (!ret)
277 folio_mark_uptodate(folio);
278 folio_unlock(folio);
279 }
280 }
281
orangefs_read_folio(struct file * file,struct folio * folio)282 static int orangefs_read_folio(struct file *file, struct folio *folio)
283 {
284 struct inode *inode = folio->mapping->host;
285 struct iov_iter iter;
286 struct bio_vec bv;
287 ssize_t ret;
288 loff_t off; /* offset of this folio in the file */
289
290 if (folio_test_dirty(folio))
291 orangefs_launder_folio(folio);
292
293 off = folio_pos(folio);
294 bvec_set_folio(&bv, folio, folio_size(folio), 0);
295 iov_iter_bvec(&iter, ITER_DEST, &bv, 1, folio_size(folio));
296
297 ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, &off, &iter,
298 folio_size(folio), inode->i_size, NULL, NULL, file);
299 /* this will only zero remaining unread portions of the folio data */
300 iov_iter_zero(~0U, &iter);
301 /* takes care of potential aliasing */
302 flush_dcache_folio(folio);
303 if (ret > 0)
304 ret = 0;
305 folio_end_read(folio, ret == 0);
306 return ret;
307 }
308
orangefs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop,void ** fsdata)309 static int orangefs_write_begin(struct file *file,
310 struct address_space *mapping, loff_t pos, unsigned len,
311 struct folio **foliop, void **fsdata)
312 {
313 struct orangefs_write_range *wr;
314 struct folio *folio;
315 int ret;
316
317 folio = __filemap_get_folio(mapping, pos / PAGE_SIZE, FGP_WRITEBEGIN,
318 mapping_gfp_mask(mapping));
319 if (IS_ERR(folio))
320 return PTR_ERR(folio);
321
322 *foliop = folio;
323
324 if (folio_test_dirty(folio) && !folio_test_private(folio)) {
325 /*
326 * Should be impossible. If it happens, launder the page
327 * since we don't know what's dirty. This will WARN in
328 * orangefs_writepage_locked.
329 */
330 ret = orangefs_launder_folio(folio);
331 if (ret)
332 return ret;
333 }
334 if (folio_test_private(folio)) {
335 struct orangefs_write_range *wr;
336 wr = folio_get_private(folio);
337 if (wr->pos + wr->len == pos &&
338 uid_eq(wr->uid, current_fsuid()) &&
339 gid_eq(wr->gid, current_fsgid())) {
340 wr->len += len;
341 goto okay;
342 } else {
343 ret = orangefs_launder_folio(folio);
344 if (ret)
345 return ret;
346 }
347 }
348
349 wr = kmalloc(sizeof *wr, GFP_KERNEL);
350 if (!wr)
351 return -ENOMEM;
352
353 wr->pos = pos;
354 wr->len = len;
355 wr->uid = current_fsuid();
356 wr->gid = current_fsgid();
357 folio_attach_private(folio, wr);
358 okay:
359 return 0;
360 }
361
orangefs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio,void * fsdata)362 static int orangefs_write_end(struct file *file, struct address_space *mapping,
363 loff_t pos, unsigned len, unsigned copied, struct folio *folio,
364 void *fsdata)
365 {
366 struct inode *inode = folio->mapping->host;
367 loff_t last_pos = pos + copied;
368
369 /*
370 * No need to use i_size_read() here, the i_size
371 * cannot change under us because we hold the i_mutex.
372 */
373 if (last_pos > inode->i_size)
374 i_size_write(inode, last_pos);
375
376 /* zero the stale part of the folio if we did a short copy */
377 if (!folio_test_uptodate(folio)) {
378 unsigned from = pos & (PAGE_SIZE - 1);
379 if (copied < len) {
380 folio_zero_range(folio, from + copied, len - copied);
381 }
382 /* Set fully written pages uptodate. */
383 if (pos == folio_pos(folio) &&
384 (len == PAGE_SIZE || pos + len == inode->i_size)) {
385 folio_zero_segment(folio, from + copied, PAGE_SIZE);
386 folio_mark_uptodate(folio);
387 }
388 }
389
390 folio_mark_dirty(folio);
391 folio_unlock(folio);
392 folio_put(folio);
393
394 mark_inode_dirty_sync(file_inode(file));
395 return copied;
396 }
397
orangefs_invalidate_folio(struct folio * folio,size_t offset,size_t length)398 static void orangefs_invalidate_folio(struct folio *folio,
399 size_t offset, size_t length)
400 {
401 struct orangefs_write_range *wr = folio_get_private(folio);
402
403 if (offset == 0 && length == PAGE_SIZE) {
404 kfree(folio_detach_private(folio));
405 return;
406 /* write range entirely within invalidate range (or equal) */
407 } else if (folio_pos(folio) + offset <= wr->pos &&
408 wr->pos + wr->len <= folio_pos(folio) + offset + length) {
409 kfree(folio_detach_private(folio));
410 /* XXX is this right? only caller in fs */
411 folio_cancel_dirty(folio);
412 return;
413 /* invalidate range chops off end of write range */
414 } else if (wr->pos < folio_pos(folio) + offset &&
415 wr->pos + wr->len <= folio_pos(folio) + offset + length &&
416 folio_pos(folio) + offset < wr->pos + wr->len) {
417 size_t x;
418 x = wr->pos + wr->len - (folio_pos(folio) + offset);
419 WARN_ON(x > wr->len);
420 wr->len -= x;
421 wr->uid = current_fsuid();
422 wr->gid = current_fsgid();
423 /* invalidate range chops off beginning of write range */
424 } else if (folio_pos(folio) + offset <= wr->pos &&
425 folio_pos(folio) + offset + length < wr->pos + wr->len &&
426 wr->pos < folio_pos(folio) + offset + length) {
427 size_t x;
428 x = folio_pos(folio) + offset + length - wr->pos;
429 WARN_ON(x > wr->len);
430 wr->pos += x;
431 wr->len -= x;
432 wr->uid = current_fsuid();
433 wr->gid = current_fsgid();
434 /* invalidate range entirely within write range (punch hole) */
435 } else if (wr->pos < folio_pos(folio) + offset &&
436 folio_pos(folio) + offset + length < wr->pos + wr->len) {
437 /* XXX what do we do here... should not WARN_ON */
438 WARN_ON(1);
439 /* punch hole */
440 /*
441 * should we just ignore this and write it out anyway?
442 * it hardly makes sense
443 */
444 return;
445 /* non-overlapping ranges */
446 } else {
447 /* WARN if they do overlap */
448 if (!((folio_pos(folio) + offset + length <= wr->pos) ^
449 (wr->pos + wr->len <= folio_pos(folio) + offset))) {
450 WARN_ON(1);
451 printk("invalidate range offset %llu length %zu\n",
452 folio_pos(folio) + offset, length);
453 printk("write range offset %llu length %zu\n",
454 wr->pos, wr->len);
455 }
456 return;
457 }
458
459 /*
460 * Above there are returns where wr is freed or where we WARN.
461 * Thus the following runs if wr was modified above.
462 */
463
464 orangefs_launder_folio(folio);
465 }
466
orangefs_release_folio(struct folio * folio,gfp_t foo)467 static bool orangefs_release_folio(struct folio *folio, gfp_t foo)
468 {
469 return !folio_test_private(folio);
470 }
471
orangefs_free_folio(struct folio * folio)472 static void orangefs_free_folio(struct folio *folio)
473 {
474 kfree(folio_detach_private(folio));
475 }
476
orangefs_launder_folio(struct folio * folio)477 static int orangefs_launder_folio(struct folio *folio)
478 {
479 int r = 0;
480 struct writeback_control wbc = {
481 .sync_mode = WB_SYNC_ALL,
482 .nr_to_write = 0,
483 };
484 folio_wait_writeback(folio);
485 if (folio_clear_dirty_for_io(folio)) {
486 r = orangefs_writepage_locked(&folio->page, &wbc);
487 folio_end_writeback(folio);
488 }
489 return r;
490 }
491
orangefs_direct_IO(struct kiocb * iocb,struct iov_iter * iter)492 static ssize_t orangefs_direct_IO(struct kiocb *iocb,
493 struct iov_iter *iter)
494 {
495 /*
496 * Comment from original do_readv_writev:
497 * Common entry point for read/write/readv/writev
498 * This function will dispatch it to either the direct I/O
499 * or buffered I/O path depending on the mount options and/or
500 * augmented/extended metadata attached to the file.
501 * Note: File extended attributes override any mount options.
502 */
503 struct file *file = iocb->ki_filp;
504 loff_t pos = iocb->ki_pos;
505 enum ORANGEFS_io_type type = iov_iter_rw(iter) == WRITE ?
506 ORANGEFS_IO_WRITE : ORANGEFS_IO_READ;
507 loff_t *offset = &pos;
508 struct inode *inode = file->f_mapping->host;
509 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
510 struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
511 size_t count = iov_iter_count(iter);
512 ssize_t total_count = 0;
513 ssize_t ret = -EINVAL;
514
515 gossip_debug(GOSSIP_FILE_DEBUG,
516 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
517 __func__,
518 handle,
519 (int)count);
520
521 if (type == ORANGEFS_IO_WRITE) {
522 gossip_debug(GOSSIP_FILE_DEBUG,
523 "%s(%pU): proceeding with offset : %llu, "
524 "size %d\n",
525 __func__,
526 handle,
527 llu(*offset),
528 (int)count);
529 }
530
531 if (count == 0) {
532 ret = 0;
533 goto out;
534 }
535
536 while (iov_iter_count(iter)) {
537 size_t each_count = iov_iter_count(iter);
538 size_t amt_complete;
539
540 /* how much to transfer in this loop iteration */
541 if (each_count > orangefs_bufmap_size_query())
542 each_count = orangefs_bufmap_size_query();
543
544 gossip_debug(GOSSIP_FILE_DEBUG,
545 "%s(%pU): size of each_count(%d)\n",
546 __func__,
547 handle,
548 (int)each_count);
549 gossip_debug(GOSSIP_FILE_DEBUG,
550 "%s(%pU): BEFORE wait_for_io: offset is %d\n",
551 __func__,
552 handle,
553 (int)*offset);
554
555 ret = wait_for_direct_io(type, inode, offset, iter,
556 each_count, 0, NULL, NULL, file);
557 gossip_debug(GOSSIP_FILE_DEBUG,
558 "%s(%pU): return from wait_for_io:%d\n",
559 __func__,
560 handle,
561 (int)ret);
562
563 if (ret < 0)
564 goto out;
565
566 *offset += ret;
567 total_count += ret;
568 amt_complete = ret;
569
570 gossip_debug(GOSSIP_FILE_DEBUG,
571 "%s(%pU): AFTER wait_for_io: offset is %d\n",
572 __func__,
573 handle,
574 (int)*offset);
575
576 /*
577 * if we got a short I/O operations,
578 * fall out and return what we got so far
579 */
580 if (amt_complete < each_count)
581 break;
582 } /*end while */
583
584 out:
585 if (total_count > 0)
586 ret = total_count;
587 if (ret > 0) {
588 if (type == ORANGEFS_IO_READ) {
589 file_accessed(file);
590 } else {
591 file_update_time(file);
592 if (*offset > i_size_read(inode))
593 i_size_write(inode, *offset);
594 }
595 }
596
597 gossip_debug(GOSSIP_FILE_DEBUG,
598 "%s(%pU): Value(%d) returned.\n",
599 __func__,
600 handle,
601 (int)ret);
602
603 return ret;
604 }
605
606 /** ORANGEFS2 implementation of address space operations */
607 static const struct address_space_operations orangefs_address_operations = {
608 .writepage = orangefs_writepage,
609 .readahead = orangefs_readahead,
610 .read_folio = orangefs_read_folio,
611 .writepages = orangefs_writepages,
612 .dirty_folio = filemap_dirty_folio,
613 .write_begin = orangefs_write_begin,
614 .write_end = orangefs_write_end,
615 .invalidate_folio = orangefs_invalidate_folio,
616 .release_folio = orangefs_release_folio,
617 .free_folio = orangefs_free_folio,
618 .launder_folio = orangefs_launder_folio,
619 .direct_IO = orangefs_direct_IO,
620 };
621
orangefs_page_mkwrite(struct vm_fault * vmf)622 vm_fault_t orangefs_page_mkwrite(struct vm_fault *vmf)
623 {
624 struct folio *folio = page_folio(vmf->page);
625 struct inode *inode = file_inode(vmf->vma->vm_file);
626 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
627 unsigned long *bitlock = &orangefs_inode->bitlock;
628 vm_fault_t ret;
629 struct orangefs_write_range *wr;
630
631 sb_start_pagefault(inode->i_sb);
632
633 if (wait_on_bit(bitlock, 1, TASK_KILLABLE)) {
634 ret = VM_FAULT_RETRY;
635 goto out;
636 }
637
638 folio_lock(folio);
639 if (folio_test_dirty(folio) && !folio_test_private(folio)) {
640 /*
641 * Should be impossible. If it happens, launder the folio
642 * since we don't know what's dirty. This will WARN in
643 * orangefs_writepage_locked.
644 */
645 if (orangefs_launder_folio(folio)) {
646 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
647 goto out;
648 }
649 }
650 if (folio_test_private(folio)) {
651 wr = folio_get_private(folio);
652 if (uid_eq(wr->uid, current_fsuid()) &&
653 gid_eq(wr->gid, current_fsgid())) {
654 wr->pos = page_offset(vmf->page);
655 wr->len = PAGE_SIZE;
656 goto okay;
657 } else {
658 if (orangefs_launder_folio(folio)) {
659 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
660 goto out;
661 }
662 }
663 }
664 wr = kmalloc(sizeof *wr, GFP_KERNEL);
665 if (!wr) {
666 ret = VM_FAULT_LOCKED|VM_FAULT_RETRY;
667 goto out;
668 }
669 wr->pos = page_offset(vmf->page);
670 wr->len = PAGE_SIZE;
671 wr->uid = current_fsuid();
672 wr->gid = current_fsgid();
673 folio_attach_private(folio, wr);
674 okay:
675
676 file_update_time(vmf->vma->vm_file);
677 if (folio->mapping != inode->i_mapping) {
678 folio_unlock(folio);
679 ret = VM_FAULT_LOCKED|VM_FAULT_NOPAGE;
680 goto out;
681 }
682
683 /*
684 * We mark the folio dirty already here so that when freeze is in
685 * progress, we are guaranteed that writeback during freezing will
686 * see the dirty folio and writeprotect it again.
687 */
688 folio_mark_dirty(folio);
689 folio_wait_stable(folio);
690 ret = VM_FAULT_LOCKED;
691 out:
692 sb_end_pagefault(inode->i_sb);
693 return ret;
694 }
695
orangefs_setattr_size(struct inode * inode,struct iattr * iattr)696 static int orangefs_setattr_size(struct inode *inode, struct iattr *iattr)
697 {
698 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
699 struct orangefs_kernel_op_s *new_op;
700 loff_t orig_size;
701 int ret = -EINVAL;
702
703 gossip_debug(GOSSIP_INODE_DEBUG,
704 "%s: %pU: Handle is %pU | fs_id %d | size is %llu\n",
705 __func__,
706 get_khandle_from_ino(inode),
707 &orangefs_inode->refn.khandle,
708 orangefs_inode->refn.fs_id,
709 iattr->ia_size);
710
711 /* Ensure that we have a up to date size, so we know if it changed. */
712 ret = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_SIZE);
713 if (ret == -ESTALE)
714 ret = -EIO;
715 if (ret) {
716 gossip_err("%s: orangefs_inode_getattr failed, ret:%d:.\n",
717 __func__, ret);
718 return ret;
719 }
720 orig_size = i_size_read(inode);
721
722 /* This is truncate_setsize in a different order. */
723 truncate_pagecache(inode, iattr->ia_size);
724 i_size_write(inode, iattr->ia_size);
725 if (iattr->ia_size > orig_size)
726 pagecache_isize_extended(inode, orig_size, iattr->ia_size);
727
728 new_op = op_alloc(ORANGEFS_VFS_OP_TRUNCATE);
729 if (!new_op)
730 return -ENOMEM;
731
732 new_op->upcall.req.truncate.refn = orangefs_inode->refn;
733 new_op->upcall.req.truncate.size = (__s64) iattr->ia_size;
734
735 ret = service_operation(new_op,
736 __func__,
737 get_interruptible_flag(inode));
738
739 /*
740 * the truncate has no downcall members to retrieve, but
741 * the status value tells us if it went through ok or not
742 */
743 gossip_debug(GOSSIP_INODE_DEBUG, "%s: ret:%d:\n", __func__, ret);
744
745 op_release(new_op);
746
747 if (ret != 0)
748 return ret;
749
750 if (orig_size != i_size_read(inode))
751 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME;
752
753 return ret;
754 }
755
__orangefs_setattr(struct inode * inode,struct iattr * iattr)756 int __orangefs_setattr(struct inode *inode, struct iattr *iattr)
757 {
758 int ret;
759
760 if (iattr->ia_valid & ATTR_MODE) {
761 if (iattr->ia_mode & (S_ISVTX)) {
762 if (is_root_handle(inode)) {
763 /*
764 * allow sticky bit to be set on root (since
765 * it shows up that way by default anyhow),
766 * but don't show it to the server
767 */
768 iattr->ia_mode -= S_ISVTX;
769 } else {
770 gossip_debug(GOSSIP_UTILS_DEBUG,
771 "User attempted to set sticky bit on non-root directory; returning EINVAL.\n");
772 ret = -EINVAL;
773 goto out;
774 }
775 }
776 if (iattr->ia_mode & (S_ISUID)) {
777 gossip_debug(GOSSIP_UTILS_DEBUG,
778 "Attempting to set setuid bit (not supported); returning EINVAL.\n");
779 ret = -EINVAL;
780 goto out;
781 }
782 }
783
784 if (iattr->ia_valid & ATTR_SIZE) {
785 ret = orangefs_setattr_size(inode, iattr);
786 if (ret)
787 goto out;
788 }
789
790 again:
791 spin_lock(&inode->i_lock);
792 if (ORANGEFS_I(inode)->attr_valid) {
793 if (uid_eq(ORANGEFS_I(inode)->attr_uid, current_fsuid()) &&
794 gid_eq(ORANGEFS_I(inode)->attr_gid, current_fsgid())) {
795 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
796 } else {
797 spin_unlock(&inode->i_lock);
798 write_inode_now(inode, 1);
799 goto again;
800 }
801 } else {
802 ORANGEFS_I(inode)->attr_valid = iattr->ia_valid;
803 ORANGEFS_I(inode)->attr_uid = current_fsuid();
804 ORANGEFS_I(inode)->attr_gid = current_fsgid();
805 }
806 setattr_copy(&nop_mnt_idmap, inode, iattr);
807 spin_unlock(&inode->i_lock);
808 mark_inode_dirty(inode);
809
810 ret = 0;
811 out:
812 return ret;
813 }
814
__orangefs_setattr_mode(struct dentry * dentry,struct iattr * iattr)815 int __orangefs_setattr_mode(struct dentry *dentry, struct iattr *iattr)
816 {
817 int ret;
818 struct inode *inode = d_inode(dentry);
819
820 ret = __orangefs_setattr(inode, iattr);
821 /* change mode on a file that has ACLs */
822 if (!ret && (iattr->ia_valid & ATTR_MODE))
823 ret = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
824 return ret;
825 }
826
827 /*
828 * Change attributes of an object referenced by dentry.
829 */
orangefs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * iattr)830 int orangefs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
831 struct iattr *iattr)
832 {
833 int ret;
834 gossip_debug(GOSSIP_INODE_DEBUG, "__orangefs_setattr: called on %pd\n",
835 dentry);
836 ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
837 if (ret)
838 goto out;
839 ret = __orangefs_setattr_mode(dentry, iattr);
840 sync_inode_metadata(d_inode(dentry), 1);
841 out:
842 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_setattr: returning %d\n",
843 ret);
844 return ret;
845 }
846
847 /*
848 * Obtain attributes of an object given a dentry
849 */
orangefs_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int flags)850 int orangefs_getattr(struct mnt_idmap *idmap, const struct path *path,
851 struct kstat *stat, u32 request_mask, unsigned int flags)
852 {
853 int ret;
854 struct inode *inode = path->dentry->d_inode;
855
856 gossip_debug(GOSSIP_INODE_DEBUG,
857 "orangefs_getattr: called on %pd mask %u\n",
858 path->dentry, request_mask);
859
860 ret = orangefs_inode_getattr(inode,
861 request_mask & STATX_SIZE ? ORANGEFS_GETATTR_SIZE : 0);
862 if (ret == 0) {
863 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
864
865 /* override block size reported to stat */
866 if (!(request_mask & STATX_SIZE))
867 stat->result_mask &= ~STATX_SIZE;
868
869 generic_fill_statx_attr(inode, stat);
870 }
871 return ret;
872 }
873
orangefs_permission(struct mnt_idmap * idmap,struct inode * inode,int mask)874 int orangefs_permission(struct mnt_idmap *idmap,
875 struct inode *inode, int mask)
876 {
877 int ret;
878
879 if (mask & MAY_NOT_BLOCK)
880 return -ECHILD;
881
882 gossip_debug(GOSSIP_INODE_DEBUG, "%s: refreshing\n", __func__);
883
884 /* Make sure the permission (and other common attrs) are up to date. */
885 ret = orangefs_inode_getattr(inode, 0);
886 if (ret < 0)
887 return ret;
888
889 return generic_permission(&nop_mnt_idmap, inode, mask);
890 }
891
orangefs_update_time(struct inode * inode,int flags)892 int orangefs_update_time(struct inode *inode, int flags)
893 {
894 struct iattr iattr;
895
896 gossip_debug(GOSSIP_INODE_DEBUG, "orangefs_update_time: %pU\n",
897 get_khandle_from_ino(inode));
898 flags = generic_update_time(inode, flags);
899 memset(&iattr, 0, sizeof iattr);
900 if (flags & S_ATIME)
901 iattr.ia_valid |= ATTR_ATIME;
902 if (flags & S_CTIME)
903 iattr.ia_valid |= ATTR_CTIME;
904 if (flags & S_MTIME)
905 iattr.ia_valid |= ATTR_MTIME;
906 return __orangefs_setattr(inode, &iattr);
907 }
908
orangefs_fileattr_get(struct dentry * dentry,struct fileattr * fa)909 static int orangefs_fileattr_get(struct dentry *dentry, struct fileattr *fa)
910 {
911 u64 val = 0;
912 int ret;
913
914 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
915 dentry);
916
917 ret = orangefs_inode_getxattr(d_inode(dentry),
918 "user.pvfs2.meta_hint",
919 &val, sizeof(val));
920 if (ret < 0 && ret != -ENODATA)
921 return ret;
922
923 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
924
925 fileattr_fill_flags(fa, val);
926 return 0;
927 }
928
orangefs_fileattr_set(struct mnt_idmap * idmap,struct dentry * dentry,struct fileattr * fa)929 static int orangefs_fileattr_set(struct mnt_idmap *idmap,
930 struct dentry *dentry, struct fileattr *fa)
931 {
932 u64 val = 0;
933
934 gossip_debug(GOSSIP_FILE_DEBUG, "%s: called on %pd\n", __func__,
935 dentry);
936 /*
937 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode is
938 * turned on for a file. The user is not allowed to turn on this bit,
939 * but the bit is present if the user first gets the flags and then
940 * updates the flags with some new settings. So, we ignore it in the
941 * following edit. bligon.
942 */
943 if (fileattr_has_fsx(fa) ||
944 (fa->flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL | ORANGEFS_MIRROR_FL))) {
945 gossip_err("%s: only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n",
946 __func__);
947 return -EOPNOTSUPP;
948 }
949 val = fa->flags;
950 gossip_debug(GOSSIP_FILE_DEBUG, "%s: flags=%u\n", __func__, (u32) val);
951 return orangefs_inode_setxattr(d_inode(dentry),
952 "user.pvfs2.meta_hint",
953 &val, sizeof(val), 0);
954 }
955
956 /* ORANGEFS2 implementation of VFS inode operations for files */
957 static const struct inode_operations orangefs_file_inode_operations = {
958 .get_inode_acl = orangefs_get_acl,
959 .set_acl = orangefs_set_acl,
960 .setattr = orangefs_setattr,
961 .getattr = orangefs_getattr,
962 .listxattr = orangefs_listxattr,
963 .permission = orangefs_permission,
964 .update_time = orangefs_update_time,
965 .fileattr_get = orangefs_fileattr_get,
966 .fileattr_set = orangefs_fileattr_set,
967 };
968
orangefs_init_iops(struct inode * inode)969 static int orangefs_init_iops(struct inode *inode)
970 {
971 inode->i_mapping->a_ops = &orangefs_address_operations;
972
973 switch (inode->i_mode & S_IFMT) {
974 case S_IFREG:
975 inode->i_op = &orangefs_file_inode_operations;
976 inode->i_fop = &orangefs_file_operations;
977 break;
978 case S_IFLNK:
979 inode->i_op = &orangefs_symlink_inode_operations;
980 break;
981 case S_IFDIR:
982 inode->i_op = &orangefs_dir_inode_operations;
983 inode->i_fop = &orangefs_dir_operations;
984 break;
985 default:
986 gossip_debug(GOSSIP_INODE_DEBUG,
987 "%s: unsupported mode\n",
988 __func__);
989 return -EINVAL;
990 }
991
992 return 0;
993 }
994
995 /*
996 * Given an ORANGEFS object identifier (fsid, handle), convert it into
997 * a ino_t type that will be used as a hash-index from where the handle will
998 * be searched for in the VFS hash table of inodes.
999 */
orangefs_handle_hash(struct orangefs_object_kref * ref)1000 static inline ino_t orangefs_handle_hash(struct orangefs_object_kref *ref)
1001 {
1002 if (!ref)
1003 return 0;
1004 return orangefs_khandle_to_ino(&(ref->khandle));
1005 }
1006
1007 /*
1008 * Called to set up an inode from iget5_locked.
1009 */
orangefs_set_inode(struct inode * inode,void * data)1010 static int orangefs_set_inode(struct inode *inode, void *data)
1011 {
1012 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1013 ORANGEFS_I(inode)->refn.fs_id = ref->fs_id;
1014 ORANGEFS_I(inode)->refn.khandle = ref->khandle;
1015 ORANGEFS_I(inode)->attr_valid = 0;
1016 hash_init(ORANGEFS_I(inode)->xattr_cache);
1017 ORANGEFS_I(inode)->mapping_time = jiffies - 1;
1018 ORANGEFS_I(inode)->bitlock = 0;
1019 return 0;
1020 }
1021
1022 /*
1023 * Called to determine if handles match.
1024 */
orangefs_test_inode(struct inode * inode,void * data)1025 static int orangefs_test_inode(struct inode *inode, void *data)
1026 {
1027 struct orangefs_object_kref *ref = (struct orangefs_object_kref *) data;
1028 struct orangefs_inode_s *orangefs_inode = NULL;
1029
1030 orangefs_inode = ORANGEFS_I(inode);
1031 /* test handles and fs_ids... */
1032 return (!ORANGEFS_khandle_cmp(&(orangefs_inode->refn.khandle),
1033 &(ref->khandle)) &&
1034 orangefs_inode->refn.fs_id == ref->fs_id);
1035 }
1036
1037 /*
1038 * Front-end to lookup the inode-cache maintained by the VFS using the ORANGEFS
1039 * file handle.
1040 *
1041 * @sb: the file system super block instance.
1042 * @ref: The ORANGEFS object for which we are trying to locate an inode.
1043 */
orangefs_iget(struct super_block * sb,struct orangefs_object_kref * ref)1044 struct inode *orangefs_iget(struct super_block *sb,
1045 struct orangefs_object_kref *ref)
1046 {
1047 struct inode *inode = NULL;
1048 unsigned long hash;
1049 int error;
1050
1051 hash = orangefs_handle_hash(ref);
1052 inode = iget5_locked(sb,
1053 hash,
1054 orangefs_test_inode,
1055 orangefs_set_inode,
1056 ref);
1057
1058 if (!inode)
1059 return ERR_PTR(-ENOMEM);
1060
1061 if (!(inode->i_state & I_NEW))
1062 return inode;
1063
1064 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1065 if (error) {
1066 iget_failed(inode);
1067 return ERR_PTR(error);
1068 }
1069
1070 inode->i_ino = hash; /* needed for stat etc */
1071 orangefs_init_iops(inode);
1072 unlock_new_inode(inode);
1073
1074 gossip_debug(GOSSIP_INODE_DEBUG,
1075 "iget handle %pU, fsid %d hash %ld i_ino %lu\n",
1076 &ref->khandle,
1077 ref->fs_id,
1078 hash,
1079 inode->i_ino);
1080
1081 return inode;
1082 }
1083
1084 /*
1085 * Allocate an inode for a newly created file and insert it into the inode hash.
1086 */
orangefs_new_inode(struct super_block * sb,struct inode * dir,umode_t mode,dev_t dev,struct orangefs_object_kref * ref)1087 struct inode *orangefs_new_inode(struct super_block *sb, struct inode *dir,
1088 umode_t mode, dev_t dev, struct orangefs_object_kref *ref)
1089 {
1090 struct posix_acl *acl = NULL, *default_acl = NULL;
1091 unsigned long hash = orangefs_handle_hash(ref);
1092 struct inode *inode;
1093 int error;
1094
1095 gossip_debug(GOSSIP_INODE_DEBUG,
1096 "%s:(sb is %p | MAJOR(dev)=%u | MINOR(dev)=%u mode=%o)\n",
1097 __func__,
1098 sb,
1099 MAJOR(dev),
1100 MINOR(dev),
1101 mode);
1102
1103 inode = new_inode(sb);
1104 if (!inode)
1105 return ERR_PTR(-ENOMEM);
1106
1107 error = posix_acl_create(dir, &mode, &default_acl, &acl);
1108 if (error)
1109 goto out_iput;
1110
1111 orangefs_set_inode(inode, ref);
1112 inode->i_ino = hash; /* needed for stat etc */
1113
1114 error = orangefs_inode_getattr(inode, ORANGEFS_GETATTR_NEW);
1115 if (error)
1116 goto out_iput;
1117
1118 orangefs_init_iops(inode);
1119 inode->i_rdev = dev;
1120
1121 if (default_acl) {
1122 error = __orangefs_set_acl(inode, default_acl,
1123 ACL_TYPE_DEFAULT);
1124 if (error)
1125 goto out_iput;
1126 }
1127
1128 if (acl) {
1129 error = __orangefs_set_acl(inode, acl, ACL_TYPE_ACCESS);
1130 if (error)
1131 goto out_iput;
1132 }
1133
1134 error = insert_inode_locked4(inode, hash, orangefs_test_inode, ref);
1135 if (error < 0)
1136 goto out_iput;
1137
1138 gossip_debug(GOSSIP_INODE_DEBUG,
1139 "Initializing ACL's for inode %pU\n",
1140 get_khandle_from_ino(inode));
1141 if (mode != inode->i_mode) {
1142 struct iattr iattr = {
1143 .ia_mode = mode,
1144 .ia_valid = ATTR_MODE,
1145 };
1146 inode->i_mode = mode;
1147 __orangefs_setattr(inode, &iattr);
1148 __posix_acl_chmod(&acl, GFP_KERNEL, inode->i_mode);
1149 }
1150 posix_acl_release(acl);
1151 posix_acl_release(default_acl);
1152 return inode;
1153
1154 out_iput:
1155 iput(inode);
1156 posix_acl_release(acl);
1157 posix_acl_release(default_acl);
1158 return ERR_PTR(error);
1159 }
1160