Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / fs / exofs / inode.c
blobea5e1f97806a7dc284e11cb1f1185981361d4e08
1 /*
2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
7 * Copyrights for code taken from ext2:
8 * Copyright (C) 1992, 1993, 1994, 1995
9 * Remy Card (card@masi.ibp.fr)
10 * Laboratoire MASI - Institut Blaise Pascal
11 * Universite Pierre et Marie Curie (Paris VI)
12 * from
13 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * This file is part of exofs.
18 * exofs is free software; you can redistribute it and/or modify
19 * it under the terms of the GNU General Public License as published by
20 * the Free Software Foundation. Since it is based on ext2, and the only
21 * valid version of GPL for the Linux kernel is version 2, the only valid
22 * version of GPL for exofs is version 2.
24 * exofs is distributed in the hope that it will be useful,
25 * but WITHOUT ANY WARRANTY; without even the implied warranty of
26 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27 * GNU General Public License for more details.
29 * You should have received a copy of the GNU General Public License
30 * along with exofs; if not, write to the Free Software
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
34 #include <linux/slab.h>
36 #include "exofs.h"
38 #define EXOFS_DBGMSG2(M...) do {} while (0)
40 enum {MAX_PAGES_KMALLOC = PAGE_SIZE / sizeof(struct page *), };
42 unsigned exofs_max_io_pages(struct ore_layout *layout,
43 unsigned expected_pages)
45 unsigned pages = min_t(unsigned, expected_pages, MAX_PAGES_KMALLOC);
47 /* TODO: easily support bio chaining */
48 pages = min_t(unsigned, pages, layout->max_io_length / PAGE_SIZE);
49 return pages;
52 struct page_collect {
53 struct exofs_sb_info *sbi;
54 struct inode *inode;
55 unsigned expected_pages;
56 struct ore_io_state *ios;
58 struct page **pages;
59 unsigned alloc_pages;
60 unsigned nr_pages;
61 unsigned long length;
62 loff_t pg_first; /* keep 64bit also in 32-arches */
63 bool read_4_write; /* This means two things: that the read is sync
64 * And the pages should not be unlocked.
66 struct page *that_locked_page;
69 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
70 struct inode *inode)
72 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
74 pcol->sbi = sbi;
75 pcol->inode = inode;
76 pcol->expected_pages = expected_pages;
78 pcol->ios = NULL;
79 pcol->pages = NULL;
80 pcol->alloc_pages = 0;
81 pcol->nr_pages = 0;
82 pcol->length = 0;
83 pcol->pg_first = -1;
84 pcol->read_4_write = false;
85 pcol->that_locked_page = NULL;
88 static void _pcol_reset(struct page_collect *pcol)
90 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
92 pcol->pages = NULL;
93 pcol->alloc_pages = 0;
94 pcol->nr_pages = 0;
95 pcol->length = 0;
96 pcol->pg_first = -1;
97 pcol->ios = NULL;
98 pcol->that_locked_page = NULL;
100 /* this is probably the end of the loop but in writes
101 * it might not end here. don't be left with nothing
103 if (!pcol->expected_pages)
104 pcol->expected_pages = MAX_PAGES_KMALLOC;
107 static int pcol_try_alloc(struct page_collect *pcol)
109 unsigned pages;
111 /* TODO: easily support bio chaining */
112 pages = exofs_max_io_pages(&pcol->sbi->layout, pcol->expected_pages);
114 for (; pages; pages >>= 1) {
115 pcol->pages = kmalloc(pages * sizeof(struct page *),
116 GFP_KERNEL);
117 if (likely(pcol->pages)) {
118 pcol->alloc_pages = pages;
119 return 0;
123 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
124 pcol->expected_pages);
125 return -ENOMEM;
128 static void pcol_free(struct page_collect *pcol)
130 kfree(pcol->pages);
131 pcol->pages = NULL;
133 if (pcol->ios) {
134 ore_put_io_state(pcol->ios);
135 pcol->ios = NULL;
139 static int pcol_add_page(struct page_collect *pcol, struct page *page,
140 unsigned len)
142 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
143 return -ENOMEM;
145 pcol->pages[pcol->nr_pages++] = page;
146 pcol->length += len;
147 return 0;
150 enum {PAGE_WAS_NOT_IN_IO = 17};
151 static int update_read_page(struct page *page, int ret)
153 switch (ret) {
154 case 0:
155 /* Everything is OK */
156 SetPageUptodate(page);
157 if (PageError(page))
158 ClearPageError(page);
159 break;
160 case -EFAULT:
161 /* In this case we were trying to read something that wasn't on
162 * disk yet - return a page full of zeroes. This should be OK,
163 * because the object should be empty (if there was a write
164 * before this read, the read would be waiting with the page
165 * locked */
166 clear_highpage(page);
168 SetPageUptodate(page);
169 if (PageError(page))
170 ClearPageError(page);
171 EXOFS_DBGMSG("recovered read error\n");
172 /* fall through */
173 case PAGE_WAS_NOT_IN_IO:
174 ret = 0; /* recovered error */
175 break;
176 default:
177 SetPageError(page);
179 return ret;
182 static void update_write_page(struct page *page, int ret)
184 if (unlikely(ret == PAGE_WAS_NOT_IN_IO))
185 return; /* don't pass start don't collect $200 */
187 if (ret) {
188 mapping_set_error(page->mapping, ret);
189 SetPageError(page);
191 end_page_writeback(page);
194 /* Called at the end of reads, to optionally unlock pages and update their
195 * status.
197 static int __readpages_done(struct page_collect *pcol)
199 int i;
200 u64 good_bytes;
201 u64 length = 0;
202 int ret = ore_check_io(pcol->ios, NULL);
204 if (likely(!ret)) {
205 good_bytes = pcol->length;
206 ret = PAGE_WAS_NOT_IN_IO;
207 } else {
208 good_bytes = 0;
211 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
212 " length=0x%lx nr_pages=%u\n",
213 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
214 pcol->nr_pages);
216 for (i = 0; i < pcol->nr_pages; i++) {
217 struct page *page = pcol->pages[i];
218 struct inode *inode = page->mapping->host;
219 int page_stat;
221 if (inode != pcol->inode)
222 continue; /* osd might add more pages at end */
224 if (likely(length < good_bytes))
225 page_stat = 0;
226 else
227 page_stat = ret;
229 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
230 inode->i_ino, page->index,
231 page_stat ? "bad_bytes" : "good_bytes");
233 ret = update_read_page(page, page_stat);
234 if (!pcol->read_4_write)
235 unlock_page(page);
236 length += PAGE_SIZE;
239 pcol_free(pcol);
240 EXOFS_DBGMSG2("readpages_done END\n");
241 return ret;
244 /* callback of async reads */
245 static void readpages_done(struct ore_io_state *ios, void *p)
247 struct page_collect *pcol = p;
249 __readpages_done(pcol);
250 atomic_dec(&pcol->sbi->s_curr_pending);
251 kfree(pcol);
254 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
256 int i;
258 for (i = 0; i < pcol->nr_pages; i++) {
259 struct page *page = pcol->pages[i];
261 if (rw == READ)
262 update_read_page(page, ret);
263 else
264 update_write_page(page, ret);
266 unlock_page(page);
270 static int _maybe_not_all_in_one_io(struct ore_io_state *ios,
271 struct page_collect *pcol_src, struct page_collect *pcol)
273 /* length was wrong or offset was not page aligned */
274 BUG_ON(pcol_src->nr_pages < ios->nr_pages);
276 if (pcol_src->nr_pages > ios->nr_pages) {
277 struct page **src_page;
278 unsigned pages_less = pcol_src->nr_pages - ios->nr_pages;
279 unsigned long len_less = pcol_src->length - ios->length;
280 unsigned i;
281 int ret;
283 /* This IO was trimmed */
284 pcol_src->nr_pages = ios->nr_pages;
285 pcol_src->length = ios->length;
287 /* Left over pages are passed to the next io */
288 pcol->expected_pages += pages_less;
289 pcol->nr_pages = pages_less;
290 pcol->length = len_less;
291 src_page = pcol_src->pages + pcol_src->nr_pages;
292 pcol->pg_first = (*src_page)->index;
294 ret = pcol_try_alloc(pcol);
295 if (unlikely(ret))
296 return ret;
298 for (i = 0; i < pages_less; ++i)
299 pcol->pages[i] = *src_page++;
301 EXOFS_DBGMSG("Length was adjusted nr_pages=0x%x "
302 "pages_less=0x%x expected_pages=0x%x "
303 "next_offset=0x%llx next_len=0x%lx\n",
304 pcol_src->nr_pages, pages_less, pcol->expected_pages,
305 pcol->pg_first * PAGE_SIZE, pcol->length);
307 return 0;
310 static int read_exec(struct page_collect *pcol)
312 struct exofs_i_info *oi = exofs_i(pcol->inode);
313 struct ore_io_state *ios;
314 struct page_collect *pcol_copy = NULL;
315 int ret;
317 if (!pcol->pages)
318 return 0;
320 if (!pcol->ios) {
321 int ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, true,
322 pcol->pg_first << PAGE_CACHE_SHIFT,
323 pcol->length, &pcol->ios);
325 if (ret)
326 return ret;
329 ios = pcol->ios;
330 ios->pages = pcol->pages;
332 if (pcol->read_4_write) {
333 ore_read(pcol->ios);
334 return __readpages_done(pcol);
337 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
338 if (!pcol_copy) {
339 ret = -ENOMEM;
340 goto err;
343 *pcol_copy = *pcol;
344 ios->done = readpages_done;
345 ios->private = pcol_copy;
347 /* pages ownership was passed to pcol_copy */
348 _pcol_reset(pcol);
350 ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
351 if (unlikely(ret))
352 goto err;
354 EXOFS_DBGMSG2("read_exec(0x%lx) offset=0x%llx length=0x%llx\n",
355 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
357 ret = ore_read(ios);
358 if (unlikely(ret))
359 goto err;
361 atomic_inc(&pcol->sbi->s_curr_pending);
363 return 0;
365 err:
366 if (!pcol->read_4_write)
367 _unlock_pcol_pages(pcol, ret, READ);
369 pcol_free(pcol);
371 kfree(pcol_copy);
372 return ret;
375 /* readpage_strip is called either directly from readpage() or by the VFS from
376 * within read_cache_pages(), to add one more page to be read. It will try to
377 * collect as many contiguous pages as posible. If a discontinuity is
378 * encountered, or it runs out of resources, it will submit the previous segment
379 * and will start a new collection. Eventually caller must submit the last
380 * segment if present.
382 static int readpage_strip(void *data, struct page *page)
384 struct page_collect *pcol = data;
385 struct inode *inode = pcol->inode;
386 struct exofs_i_info *oi = exofs_i(inode);
387 loff_t i_size = i_size_read(inode);
388 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
389 size_t len;
390 int ret;
392 /* FIXME: Just for debugging, will be removed */
393 if (PageUptodate(page))
394 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
395 page->index);
397 pcol->that_locked_page = page;
399 if (page->index < end_index)
400 len = PAGE_CACHE_SIZE;
401 else if (page->index == end_index)
402 len = i_size & ~PAGE_CACHE_MASK;
403 else
404 len = 0;
406 if (!len || !obj_created(oi)) {
407 /* this will be out of bounds, or doesn't exist yet.
408 * Current page is cleared and the request is split
410 clear_highpage(page);
412 SetPageUptodate(page);
413 if (PageError(page))
414 ClearPageError(page);
416 if (!pcol->read_4_write)
417 unlock_page(page);
418 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
419 "read_4_write=%d index=0x%lx end_index=0x%lx "
420 "splitting\n", inode->i_ino, len,
421 pcol->read_4_write, page->index, end_index);
423 return read_exec(pcol);
426 try_again:
428 if (unlikely(pcol->pg_first == -1)) {
429 pcol->pg_first = page->index;
430 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
431 page->index)) {
432 /* Discontinuity detected, split the request */
433 ret = read_exec(pcol);
434 if (unlikely(ret))
435 goto fail;
436 goto try_again;
439 if (!pcol->pages) {
440 ret = pcol_try_alloc(pcol);
441 if (unlikely(ret))
442 goto fail;
445 if (len != PAGE_CACHE_SIZE)
446 zero_user(page, len, PAGE_CACHE_SIZE - len);
448 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
449 inode->i_ino, page->index, len);
451 ret = pcol_add_page(pcol, page, len);
452 if (ret) {
453 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
454 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
455 page, len, pcol->nr_pages, pcol->length);
457 /* split the request, and start again with current page */
458 ret = read_exec(pcol);
459 if (unlikely(ret))
460 goto fail;
462 goto try_again;
465 return 0;
467 fail:
468 /* SetPageError(page); ??? */
469 unlock_page(page);
470 return ret;
473 static int exofs_readpages(struct file *file, struct address_space *mapping,
474 struct list_head *pages, unsigned nr_pages)
476 struct page_collect pcol;
477 int ret;
479 _pcol_init(&pcol, nr_pages, mapping->host);
481 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
482 if (ret) {
483 EXOFS_ERR("read_cache_pages => %d\n", ret);
484 return ret;
487 ret = read_exec(&pcol);
488 if (unlikely(ret))
489 return ret;
491 return read_exec(&pcol);
494 static int _readpage(struct page *page, bool read_4_write)
496 struct page_collect pcol;
497 int ret;
499 _pcol_init(&pcol, 1, page->mapping->host);
501 pcol.read_4_write = read_4_write;
502 ret = readpage_strip(&pcol, page);
503 if (ret) {
504 EXOFS_ERR("_readpage => %d\n", ret);
505 return ret;
508 return read_exec(&pcol);
512 * We don't need the file
514 static int exofs_readpage(struct file *file, struct page *page)
516 return _readpage(page, false);
519 /* Callback for osd_write. All writes are asynchronous */
520 static void writepages_done(struct ore_io_state *ios, void *p)
522 struct page_collect *pcol = p;
523 int i;
524 u64 good_bytes;
525 u64 length = 0;
526 int ret = ore_check_io(ios, NULL);
528 atomic_dec(&pcol->sbi->s_curr_pending);
530 if (likely(!ret)) {
531 good_bytes = pcol->length;
532 ret = PAGE_WAS_NOT_IN_IO;
533 } else {
534 good_bytes = 0;
537 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
538 " length=0x%lx nr_pages=%u\n",
539 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
540 pcol->nr_pages);
542 for (i = 0; i < pcol->nr_pages; i++) {
543 struct page *page = pcol->pages[i];
544 struct inode *inode = page->mapping->host;
545 int page_stat;
547 if (inode != pcol->inode)
548 continue; /* osd might add more pages to a bio */
550 if (likely(length < good_bytes))
551 page_stat = 0;
552 else
553 page_stat = ret;
555 update_write_page(page, page_stat);
556 unlock_page(page);
557 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
558 inode->i_ino, page->index, page_stat);
560 length += PAGE_SIZE;
563 pcol_free(pcol);
564 kfree(pcol);
565 EXOFS_DBGMSG2("writepages_done END\n");
568 static struct page *__r4w_get_page(void *priv, u64 offset, bool *uptodate)
570 struct page_collect *pcol = priv;
571 pgoff_t index = offset / PAGE_SIZE;
573 if (!pcol->that_locked_page ||
574 (pcol->that_locked_page->index != index)) {
575 struct page *page = find_get_page(pcol->inode->i_mapping, index);
577 if (!page) {
578 page = find_or_create_page(pcol->inode->i_mapping,
579 index, GFP_NOFS);
580 if (unlikely(!page)) {
581 EXOFS_DBGMSG("grab_cache_page Failed "
582 "index=0x%llx\n", _LLU(index));
583 return NULL;
585 unlock_page(page);
587 if (PageDirty(page) || PageWriteback(page))
588 *uptodate = true;
589 else
590 *uptodate = PageUptodate(page);
591 EXOFS_DBGMSG("index=0x%lx uptodate=%d\n", index, *uptodate);
592 return page;
593 } else {
594 EXOFS_DBGMSG("YES that_locked_page index=0x%lx\n",
595 pcol->that_locked_page->index);
596 *uptodate = true;
597 return pcol->that_locked_page;
601 static void __r4w_put_page(void *priv, struct page *page)
603 struct page_collect *pcol = priv;
605 if (pcol->that_locked_page != page) {
606 EXOFS_DBGMSG("index=0x%lx\n", page->index);
607 page_cache_release(page);
608 return;
610 EXOFS_DBGMSG("that_locked_page index=0x%lx\n", page->index);
613 static const struct _ore_r4w_op _r4w_op = {
614 .get_page = &__r4w_get_page,
615 .put_page = &__r4w_put_page,
618 static int write_exec(struct page_collect *pcol)
620 struct exofs_i_info *oi = exofs_i(pcol->inode);
621 struct ore_io_state *ios;
622 struct page_collect *pcol_copy = NULL;
623 int ret;
625 if (!pcol->pages)
626 return 0;
628 BUG_ON(pcol->ios);
629 ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, false,
630 pcol->pg_first << PAGE_CACHE_SHIFT,
631 pcol->length, &pcol->ios);
632 if (unlikely(ret))
633 goto err;
635 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
636 if (!pcol_copy) {
637 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
638 ret = -ENOMEM;
639 goto err;
642 *pcol_copy = *pcol;
644 ios = pcol->ios;
645 ios->pages = pcol_copy->pages;
646 ios->done = writepages_done;
647 ios->r4w = &_r4w_op;
648 ios->private = pcol_copy;
650 /* pages ownership was passed to pcol_copy */
651 _pcol_reset(pcol);
653 ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
654 if (unlikely(ret))
655 goto err;
657 EXOFS_DBGMSG2("write_exec(0x%lx) offset=0x%llx length=0x%llx\n",
658 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
660 ret = ore_write(ios);
661 if (unlikely(ret)) {
662 EXOFS_ERR("write_exec: ore_write() Failed\n");
663 goto err;
666 atomic_inc(&pcol->sbi->s_curr_pending);
667 return 0;
669 err:
670 _unlock_pcol_pages(pcol, ret, WRITE);
671 pcol_free(pcol);
672 kfree(pcol_copy);
674 return ret;
677 /* writepage_strip is called either directly from writepage() or by the VFS from
678 * within write_cache_pages(), to add one more page to be written to storage.
679 * It will try to collect as many contiguous pages as possible. If a
680 * discontinuity is encountered or it runs out of resources it will submit the
681 * previous segment and will start a new collection.
682 * Eventually caller must submit the last segment if present.
684 static int writepage_strip(struct page *page,
685 struct writeback_control *wbc_unused, void *data)
687 struct page_collect *pcol = data;
688 struct inode *inode = pcol->inode;
689 struct exofs_i_info *oi = exofs_i(inode);
690 loff_t i_size = i_size_read(inode);
691 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
692 size_t len;
693 int ret;
695 BUG_ON(!PageLocked(page));
697 ret = wait_obj_created(oi);
698 if (unlikely(ret))
699 goto fail;
701 if (page->index < end_index)
702 /* in this case, the page is within the limits of the file */
703 len = PAGE_CACHE_SIZE;
704 else {
705 len = i_size & ~PAGE_CACHE_MASK;
707 if (page->index > end_index || !len) {
708 /* in this case, the page is outside the limits
709 * (truncate in progress)
711 ret = write_exec(pcol);
712 if (unlikely(ret))
713 goto fail;
714 if (PageError(page))
715 ClearPageError(page);
716 unlock_page(page);
717 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
718 "outside the limits\n",
719 inode->i_ino, page->index);
720 return 0;
724 try_again:
726 if (unlikely(pcol->pg_first == -1)) {
727 pcol->pg_first = page->index;
728 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
729 page->index)) {
730 /* Discontinuity detected, split the request */
731 ret = write_exec(pcol);
732 if (unlikely(ret))
733 goto fail;
735 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
736 inode->i_ino, page->index);
737 goto try_again;
740 if (!pcol->pages) {
741 ret = pcol_try_alloc(pcol);
742 if (unlikely(ret))
743 goto fail;
746 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
747 inode->i_ino, page->index, len);
749 ret = pcol_add_page(pcol, page, len);
750 if (unlikely(ret)) {
751 EXOFS_DBGMSG2("Failed pcol_add_page "
752 "nr_pages=%u total_length=0x%lx\n",
753 pcol->nr_pages, pcol->length);
755 /* split the request, next loop will start again */
756 ret = write_exec(pcol);
757 if (unlikely(ret)) {
758 EXOFS_DBGMSG("write_exec failed => %d", ret);
759 goto fail;
762 goto try_again;
765 BUG_ON(PageWriteback(page));
766 set_page_writeback(page);
768 return 0;
770 fail:
771 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
772 inode->i_ino, page->index, ret);
773 set_bit(AS_EIO, &page->mapping->flags);
774 unlock_page(page);
775 return ret;
778 static int exofs_writepages(struct address_space *mapping,
779 struct writeback_control *wbc)
781 struct page_collect pcol;
782 long start, end, expected_pages;
783 int ret;
785 start = wbc->range_start >> PAGE_CACHE_SHIFT;
786 end = (wbc->range_end == LLONG_MAX) ?
787 start + mapping->nrpages :
788 wbc->range_end >> PAGE_CACHE_SHIFT;
790 if (start || end)
791 expected_pages = end - start + 1;
792 else
793 expected_pages = mapping->nrpages;
795 if (expected_pages < 32L)
796 expected_pages = 32L;
798 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
799 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
800 mapping->host->i_ino, wbc->range_start, wbc->range_end,
801 mapping->nrpages, start, end, expected_pages);
803 _pcol_init(&pcol, expected_pages, mapping->host);
805 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
806 if (unlikely(ret)) {
807 EXOFS_ERR("write_cache_pages => %d\n", ret);
808 return ret;
811 ret = write_exec(&pcol);
812 if (unlikely(ret))
813 return ret;
815 if (wbc->sync_mode == WB_SYNC_ALL) {
816 return write_exec(&pcol); /* pump the last reminder */
817 } else if (pcol.nr_pages) {
818 /* not SYNC let the reminder join the next writeout */
819 unsigned i;
821 for (i = 0; i < pcol.nr_pages; i++) {
822 struct page *page = pcol.pages[i];
824 end_page_writeback(page);
825 set_page_dirty(page);
826 unlock_page(page);
829 return 0;
833 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
835 struct page_collect pcol;
836 int ret;
838 _pcol_init(&pcol, 1, page->mapping->host);
840 ret = writepage_strip(page, NULL, &pcol);
841 if (ret) {
842 EXOFS_ERR("exofs_writepage => %d\n", ret);
843 return ret;
846 return write_exec(&pcol);
849 /* i_mutex held using inode->i_size directly */
850 static void _write_failed(struct inode *inode, loff_t to)
852 if (to > inode->i_size)
853 truncate_pagecache(inode, to, inode->i_size);
856 int exofs_write_begin(struct file *file, struct address_space *mapping,
857 loff_t pos, unsigned len, unsigned flags,
858 struct page **pagep, void **fsdata)
860 int ret = 0;
861 struct page *page;
863 page = *pagep;
864 if (page == NULL) {
865 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
866 fsdata);
867 if (ret) {
868 EXOFS_DBGMSG("simple_write_begin failed\n");
869 goto out;
872 page = *pagep;
875 /* read modify write */
876 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
877 loff_t i_size = i_size_read(mapping->host);
878 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
879 size_t rlen;
881 if (page->index < end_index)
882 rlen = PAGE_CACHE_SIZE;
883 else if (page->index == end_index)
884 rlen = i_size & ~PAGE_CACHE_MASK;
885 else
886 rlen = 0;
888 if (!rlen) {
889 clear_highpage(page);
890 SetPageUptodate(page);
891 goto out;
894 ret = _readpage(page, true);
895 if (ret) {
896 /*SetPageError was done by _readpage. Is it ok?*/
897 unlock_page(page);
898 EXOFS_DBGMSG("__readpage failed\n");
901 out:
902 if (unlikely(ret))
903 _write_failed(mapping->host, pos + len);
905 return ret;
908 static int exofs_write_begin_export(struct file *file,
909 struct address_space *mapping,
910 loff_t pos, unsigned len, unsigned flags,
911 struct page **pagep, void **fsdata)
913 *pagep = NULL;
915 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
916 fsdata);
919 static int exofs_write_end(struct file *file, struct address_space *mapping,
920 loff_t pos, unsigned len, unsigned copied,
921 struct page *page, void *fsdata)
923 struct inode *inode = mapping->host;
924 /* According to comment in simple_write_end i_mutex is held */
925 loff_t i_size = inode->i_size;
926 int ret;
928 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
929 if (unlikely(ret))
930 _write_failed(inode, pos + len);
932 /* TODO: once simple_write_end marks inode dirty remove */
933 if (i_size != inode->i_size)
934 mark_inode_dirty(inode);
935 return ret;
938 static int exofs_releasepage(struct page *page, gfp_t gfp)
940 EXOFS_DBGMSG("page 0x%lx\n", page->index);
941 WARN_ON(1);
942 return 0;
945 static void exofs_invalidatepage(struct page *page, unsigned long offset)
947 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset);
948 WARN_ON(1);
951 const struct address_space_operations exofs_aops = {
952 .readpage = exofs_readpage,
953 .readpages = exofs_readpages,
954 .writepage = NULL,
955 .writepages = exofs_writepages,
956 .write_begin = exofs_write_begin_export,
957 .write_end = exofs_write_end,
958 .releasepage = exofs_releasepage,
959 .set_page_dirty = __set_page_dirty_nobuffers,
960 .invalidatepage = exofs_invalidatepage,
962 /* Not implemented Yet */
963 .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
964 .direct_IO = NULL, /* TODO: Should be trivial to do */
966 /* With these NULL has special meaning or default is not exported */
967 .get_xip_mem = NULL,
968 .migratepage = NULL,
969 .launder_page = NULL,
970 .is_partially_uptodate = NULL,
971 .error_remove_page = NULL,
974 /******************************************************************************
975 * INODE OPERATIONS
976 *****************************************************************************/
979 * Test whether an inode is a fast symlink.
981 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
983 struct exofs_i_info *oi = exofs_i(inode);
985 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
988 static int _do_truncate(struct inode *inode, loff_t newsize)
990 struct exofs_i_info *oi = exofs_i(inode);
991 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
992 int ret;
994 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
996 ret = ore_truncate(&sbi->layout, &oi->oc, (u64)newsize);
997 if (likely(!ret))
998 truncate_setsize(inode, newsize);
1000 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
1001 inode->i_ino, newsize, ret);
1002 return ret;
1006 * Set inode attributes - update size attribute on OSD if needed,
1007 * otherwise just call generic functions.
1009 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
1011 struct inode *inode = dentry->d_inode;
1012 int error;
1014 /* if we are about to modify an object, and it hasn't been
1015 * created yet, wait
1017 error = wait_obj_created(exofs_i(inode));
1018 if (unlikely(error))
1019 return error;
1021 error = inode_change_ok(inode, iattr);
1022 if (unlikely(error))
1023 return error;
1025 if ((iattr->ia_valid & ATTR_SIZE) &&
1026 iattr->ia_size != i_size_read(inode)) {
1027 error = _do_truncate(inode, iattr->ia_size);
1028 if (unlikely(error))
1029 return error;
1032 setattr_copy(inode, iattr);
1033 mark_inode_dirty(inode);
1034 return 0;
1037 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
1038 EXOFS_APAGE_FS_DATA,
1039 EXOFS_ATTR_INODE_FILE_LAYOUT,
1041 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
1042 EXOFS_APAGE_FS_DATA,
1043 EXOFS_ATTR_INODE_DIR_LAYOUT,
1047 * Read the Linux inode info from the OSD, and return it as is. In exofs the
1048 * inode info is in an application specific page/attribute of the osd-object.
1050 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
1051 struct exofs_fcb *inode)
1053 struct exofs_sb_info *sbi = sb->s_fs_info;
1054 struct osd_attr attrs[] = {
1055 [0] = g_attr_inode_data,
1056 [1] = g_attr_inode_file_layout,
1057 [2] = g_attr_inode_dir_layout,
1059 struct ore_io_state *ios;
1060 struct exofs_on_disk_inode_layout *layout;
1061 int ret;
1063 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1064 if (unlikely(ret)) {
1065 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1066 return ret;
1069 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1070 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1072 ios->in_attr = attrs;
1073 ios->in_attr_len = ARRAY_SIZE(attrs);
1075 ret = ore_read(ios);
1076 if (unlikely(ret)) {
1077 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
1078 _LLU(oi->one_comp.obj.id), ret);
1079 memset(inode, 0, sizeof(*inode));
1080 inode->i_mode = 0040000 | (0777 & ~022);
1081 /* If object is lost on target we might as well enable it's
1082 * delete.
1084 if ((ret == -ENOENT) || (ret == -EINVAL))
1085 ret = 0;
1086 goto out;
1089 ret = extract_attr_from_ios(ios, &attrs[0]);
1090 if (ret) {
1091 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1092 goto out;
1094 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
1095 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
1097 ret = extract_attr_from_ios(ios, &attrs[1]);
1098 if (ret) {
1099 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1100 goto out;
1102 if (attrs[1].len) {
1103 layout = attrs[1].val_ptr;
1104 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1105 EXOFS_ERR("%s: unsupported files layout %d\n",
1106 __func__, layout->gen_func);
1107 ret = -ENOTSUPP;
1108 goto out;
1112 ret = extract_attr_from_ios(ios, &attrs[2]);
1113 if (ret) {
1114 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1115 goto out;
1117 if (attrs[2].len) {
1118 layout = attrs[2].val_ptr;
1119 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1120 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
1121 __func__, layout->gen_func);
1122 ret = -ENOTSUPP;
1123 goto out;
1127 out:
1128 ore_put_io_state(ios);
1129 return ret;
1132 static void __oi_init(struct exofs_i_info *oi)
1134 init_waitqueue_head(&oi->i_wq);
1135 oi->i_flags = 0;
1138 * Fill in an inode read from the OSD and set it up for use
1140 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1142 struct exofs_i_info *oi;
1143 struct exofs_fcb fcb;
1144 struct inode *inode;
1145 int ret;
1147 inode = iget_locked(sb, ino);
1148 if (!inode)
1149 return ERR_PTR(-ENOMEM);
1150 if (!(inode->i_state & I_NEW))
1151 return inode;
1152 oi = exofs_i(inode);
1153 __oi_init(oi);
1154 exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1155 exofs_oi_objno(oi));
1157 /* read the inode from the osd */
1158 ret = exofs_get_inode(sb, oi, &fcb);
1159 if (ret)
1160 goto bad_inode;
1162 set_obj_created(oi);
1164 /* copy stuff from on-disk struct to in-memory struct */
1165 inode->i_mode = le16_to_cpu(fcb.i_mode);
1166 inode->i_uid = le32_to_cpu(fcb.i_uid);
1167 inode->i_gid = le32_to_cpu(fcb.i_gid);
1168 set_nlink(inode, le16_to_cpu(fcb.i_links_count));
1169 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1170 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1171 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1172 inode->i_ctime.tv_nsec =
1173 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1174 oi->i_commit_size = le64_to_cpu(fcb.i_size);
1175 i_size_write(inode, oi->i_commit_size);
1176 inode->i_blkbits = EXOFS_BLKSHIFT;
1177 inode->i_generation = le32_to_cpu(fcb.i_generation);
1179 oi->i_dir_start_lookup = 0;
1181 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1182 ret = -ESTALE;
1183 goto bad_inode;
1186 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1187 if (fcb.i_data[0])
1188 inode->i_rdev =
1189 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1190 else
1191 inode->i_rdev =
1192 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1193 } else {
1194 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1197 inode->i_mapping->backing_dev_info = sb->s_bdi;
1198 if (S_ISREG(inode->i_mode)) {
1199 inode->i_op = &exofs_file_inode_operations;
1200 inode->i_fop = &exofs_file_operations;
1201 inode->i_mapping->a_ops = &exofs_aops;
1202 } else if (S_ISDIR(inode->i_mode)) {
1203 inode->i_op = &exofs_dir_inode_operations;
1204 inode->i_fop = &exofs_dir_operations;
1205 inode->i_mapping->a_ops = &exofs_aops;
1206 } else if (S_ISLNK(inode->i_mode)) {
1207 if (exofs_inode_is_fast_symlink(inode))
1208 inode->i_op = &exofs_fast_symlink_inode_operations;
1209 else {
1210 inode->i_op = &exofs_symlink_inode_operations;
1211 inode->i_mapping->a_ops = &exofs_aops;
1213 } else {
1214 inode->i_op = &exofs_special_inode_operations;
1215 if (fcb.i_data[0])
1216 init_special_inode(inode, inode->i_mode,
1217 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1218 else
1219 init_special_inode(inode, inode->i_mode,
1220 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1223 unlock_new_inode(inode);
1224 return inode;
1226 bad_inode:
1227 iget_failed(inode);
1228 return ERR_PTR(ret);
1231 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1233 if (!obj_created(oi)) {
1234 EXOFS_DBGMSG("!obj_created\n");
1235 BUG_ON(!obj_2bcreated(oi));
1236 wait_event(oi->i_wq, obj_created(oi));
1237 EXOFS_DBGMSG("wait_event done\n");
1239 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1243 * Callback function from exofs_new_inode(). The important thing is that we
1244 * set the obj_created flag so that other methods know that the object exists on
1245 * the OSD.
1247 static void create_done(struct ore_io_state *ios, void *p)
1249 struct inode *inode = p;
1250 struct exofs_i_info *oi = exofs_i(inode);
1251 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1252 int ret;
1254 ret = ore_check_io(ios, NULL);
1255 ore_put_io_state(ios);
1257 atomic_dec(&sbi->s_curr_pending);
1259 if (unlikely(ret)) {
1260 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1261 _LLU(exofs_oi_objno(oi)),
1262 _LLU(oi->one_comp.obj.partition));
1263 /*TODO: When FS is corrupted creation can fail, object already
1264 * exist. Get rid of this asynchronous creation, if exist
1265 * increment the obj counter and try the next object. Until we
1266 * succeed. All these dangling objects will be made into lost
1267 * files by chkfs.exofs
1271 set_obj_created(oi);
1273 wake_up(&oi->i_wq);
1277 * Set up a new inode and create an object for it on the OSD
1279 struct inode *exofs_new_inode(struct inode *dir, umode_t mode)
1281 struct super_block *sb = dir->i_sb;
1282 struct exofs_sb_info *sbi = sb->s_fs_info;
1283 struct inode *inode;
1284 struct exofs_i_info *oi;
1285 struct ore_io_state *ios;
1286 int ret;
1288 inode = new_inode(sb);
1289 if (!inode)
1290 return ERR_PTR(-ENOMEM);
1292 oi = exofs_i(inode);
1293 __oi_init(oi);
1295 set_obj_2bcreated(oi);
1297 inode->i_mapping->backing_dev_info = sb->s_bdi;
1298 inode_init_owner(inode, dir, mode);
1299 inode->i_ino = sbi->s_nextid++;
1300 inode->i_blkbits = EXOFS_BLKSHIFT;
1301 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1302 oi->i_commit_size = inode->i_size = 0;
1303 spin_lock(&sbi->s_next_gen_lock);
1304 inode->i_generation = sbi->s_next_generation++;
1305 spin_unlock(&sbi->s_next_gen_lock);
1306 insert_inode_hash(inode);
1308 exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1309 exofs_oi_objno(oi));
1310 exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */
1312 mark_inode_dirty(inode);
1314 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1315 if (unlikely(ret)) {
1316 EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n");
1317 return ERR_PTR(ret);
1320 ios->done = create_done;
1321 ios->private = inode;
1323 ret = ore_create(ios);
1324 if (ret) {
1325 ore_put_io_state(ios);
1326 return ERR_PTR(ret);
1328 atomic_inc(&sbi->s_curr_pending);
1330 return inode;
1334 * struct to pass two arguments to update_inode's callback
1336 struct updatei_args {
1337 struct exofs_sb_info *sbi;
1338 struct exofs_fcb fcb;
1342 * Callback function from exofs_update_inode().
1344 static void updatei_done(struct ore_io_state *ios, void *p)
1346 struct updatei_args *args = p;
1348 ore_put_io_state(ios);
1350 atomic_dec(&args->sbi->s_curr_pending);
1352 kfree(args);
1356 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1357 * synchronously or asynchronously depending on the do_sync flag.
1359 static int exofs_update_inode(struct inode *inode, int do_sync)
1361 struct exofs_i_info *oi = exofs_i(inode);
1362 struct super_block *sb = inode->i_sb;
1363 struct exofs_sb_info *sbi = sb->s_fs_info;
1364 struct ore_io_state *ios;
1365 struct osd_attr attr;
1366 struct exofs_fcb *fcb;
1367 struct updatei_args *args;
1368 int ret;
1370 args = kzalloc(sizeof(*args), GFP_KERNEL);
1371 if (!args) {
1372 EXOFS_DBGMSG("Failed kzalloc of args\n");
1373 return -ENOMEM;
1376 fcb = &args->fcb;
1378 fcb->i_mode = cpu_to_le16(inode->i_mode);
1379 fcb->i_uid = cpu_to_le32(inode->i_uid);
1380 fcb->i_gid = cpu_to_le32(inode->i_gid);
1381 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1382 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1383 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1384 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1385 oi->i_commit_size = i_size_read(inode);
1386 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1387 fcb->i_generation = cpu_to_le32(inode->i_generation);
1389 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1390 if (old_valid_dev(inode->i_rdev)) {
1391 fcb->i_data[0] =
1392 cpu_to_le32(old_encode_dev(inode->i_rdev));
1393 fcb->i_data[1] = 0;
1394 } else {
1395 fcb->i_data[0] = 0;
1396 fcb->i_data[1] =
1397 cpu_to_le32(new_encode_dev(inode->i_rdev));
1398 fcb->i_data[2] = 0;
1400 } else
1401 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1403 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1404 if (unlikely(ret)) {
1405 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1406 goto free_args;
1409 attr = g_attr_inode_data;
1410 attr.val_ptr = fcb;
1411 ios->out_attr_len = 1;
1412 ios->out_attr = &attr;
1414 wait_obj_created(oi);
1416 if (!do_sync) {
1417 args->sbi = sbi;
1418 ios->done = updatei_done;
1419 ios->private = args;
1422 ret = ore_write(ios);
1423 if (!do_sync && !ret) {
1424 atomic_inc(&sbi->s_curr_pending);
1425 goto out; /* deallocation in updatei_done */
1428 ore_put_io_state(ios);
1429 free_args:
1430 kfree(args);
1431 out:
1432 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1433 inode->i_ino, do_sync, ret);
1434 return ret;
1437 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1439 /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1440 return exofs_update_inode(inode, 1);
1444 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1445 * do.
1447 static void delete_done(struct ore_io_state *ios, void *p)
1449 struct exofs_sb_info *sbi = p;
1451 ore_put_io_state(ios);
1453 atomic_dec(&sbi->s_curr_pending);
1457 * Called when the refcount of an inode reaches zero. We remove the object
1458 * from the OSD here. We make sure the object was created before we try and
1459 * delete it.
1461 void exofs_evict_inode(struct inode *inode)
1463 struct exofs_i_info *oi = exofs_i(inode);
1464 struct super_block *sb = inode->i_sb;
1465 struct exofs_sb_info *sbi = sb->s_fs_info;
1466 struct ore_io_state *ios;
1467 int ret;
1469 truncate_inode_pages(&inode->i_data, 0);
1471 /* TODO: should do better here */
1472 if (inode->i_nlink || is_bad_inode(inode))
1473 goto no_delete;
1475 inode->i_size = 0;
1476 end_writeback(inode);
1478 /* if we are deleting an obj that hasn't been created yet, wait.
1479 * This also makes sure that create_done cannot be called with an
1480 * already evicted inode.
1482 wait_obj_created(oi);
1483 /* ignore the error, attempt a remove anyway */
1485 /* Now Remove the OSD objects */
1486 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1487 if (unlikely(ret)) {
1488 EXOFS_ERR("%s: ore_get_io_state failed\n", __func__);
1489 return;
1492 ios->done = delete_done;
1493 ios->private = sbi;
1495 ret = ore_remove(ios);
1496 if (ret) {
1497 EXOFS_ERR("%s: ore_remove failed\n", __func__);
1498 ore_put_io_state(ios);
1499 return;
1501 atomic_inc(&sbi->s_curr_pending);
1503 return;
1505 no_delete:
1506 end_writeback(inode);