sparc: Fix build with DEBUG_PAGEALLOC enabled.
[~shefty/rdma-dev.git] / fs / exofs / inode.c
1 /*
2  * Copyright (C) 2005, 2006
3  * Avishay Traeger (avishay@gmail.com)
4  * Copyright (C) 2008, 2009
5  * Boaz Harrosh <bharrosh@panasas.com>
6  *
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
15  *
16  * This file is part of exofs.
17  *
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.
23  *
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.
28  *
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
32  */
33
34 #include <linux/slab.h>
35
36 #include "exofs.h"
37
38 #define EXOFS_DBGMSG2(M...) do {} while (0)
39
40 enum { BIO_MAX_PAGES_KMALLOC =
41                 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
42         MAX_PAGES_KMALLOC =
43                 PAGE_SIZE / sizeof(struct page *),
44 };
45
46 unsigned exofs_max_io_pages(struct exofs_layout *layout,
47                             unsigned expected_pages)
48 {
49         unsigned pages = min_t(unsigned, expected_pages, MAX_PAGES_KMALLOC);
50
51         /* TODO: easily support bio chaining */
52         pages =  min_t(unsigned, pages,
53                        layout->group_width * BIO_MAX_PAGES_KMALLOC);
54         return pages;
55 }
56
57 struct page_collect {
58         struct exofs_sb_info *sbi;
59         struct inode *inode;
60         unsigned expected_pages;
61         struct exofs_io_state *ios;
62
63         struct page **pages;
64         unsigned alloc_pages;
65         unsigned nr_pages;
66         unsigned long length;
67         loff_t pg_first; /* keep 64bit also in 32-arches */
68         bool read_4_write; /* This means two things: that the read is sync
69                             * And the pages should not be unlocked.
70                             */
71 };
72
73 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
74                        struct inode *inode)
75 {
76         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
77
78         pcol->sbi = sbi;
79         pcol->inode = inode;
80         pcol->expected_pages = expected_pages;
81
82         pcol->ios = NULL;
83         pcol->pages = NULL;
84         pcol->alloc_pages = 0;
85         pcol->nr_pages = 0;
86         pcol->length = 0;
87         pcol->pg_first = -1;
88         pcol->read_4_write = false;
89 }
90
91 static void _pcol_reset(struct page_collect *pcol)
92 {
93         pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
94
95         pcol->pages = NULL;
96         pcol->alloc_pages = 0;
97         pcol->nr_pages = 0;
98         pcol->length = 0;
99         pcol->pg_first = -1;
100         pcol->ios = NULL;
101
102         /* this is probably the end of the loop but in writes
103          * it might not end here. don't be left with nothing
104          */
105         if (!pcol->expected_pages)
106                 pcol->expected_pages = MAX_PAGES_KMALLOC;
107 }
108
109 static int pcol_try_alloc(struct page_collect *pcol)
110 {
111         unsigned pages;
112
113         if (!pcol->ios) { /* First time allocate io_state */
114                 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
115
116                 if (ret)
117                         return ret;
118         }
119
120         /* TODO: easily support bio chaining */
121         pages =  exofs_max_io_pages(&pcol->sbi->layout, pcol->expected_pages);
122
123         for (; pages; pages >>= 1) {
124                 pcol->pages = kmalloc(pages * sizeof(struct page *),
125                                       GFP_KERNEL);
126                 if (likely(pcol->pages)) {
127                         pcol->alloc_pages = pages;
128                         return 0;
129                 }
130         }
131
132         EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
133                   pcol->expected_pages);
134         return -ENOMEM;
135 }
136
137 static void pcol_free(struct page_collect *pcol)
138 {
139         kfree(pcol->pages);
140         pcol->pages = NULL;
141
142         if (pcol->ios) {
143                 exofs_put_io_state(pcol->ios);
144                 pcol->ios = NULL;
145         }
146 }
147
148 static int pcol_add_page(struct page_collect *pcol, struct page *page,
149                          unsigned len)
150 {
151         if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
152                 return -ENOMEM;
153
154         pcol->pages[pcol->nr_pages++] = page;
155         pcol->length += len;
156         return 0;
157 }
158
159 static int update_read_page(struct page *page, int ret)
160 {
161         if (ret == 0) {
162                 /* Everything is OK */
163                 SetPageUptodate(page);
164                 if (PageError(page))
165                         ClearPageError(page);
166         } else if (ret == -EFAULT) {
167                 /* In this case we were trying to read something that wasn't on
168                  * disk yet - return a page full of zeroes.  This should be OK,
169                  * because the object should be empty (if there was a write
170                  * before this read, the read would be waiting with the page
171                  * locked */
172                 clear_highpage(page);
173
174                 SetPageUptodate(page);
175                 if (PageError(page))
176                         ClearPageError(page);
177                 ret = 0; /* recovered error */
178                 EXOFS_DBGMSG("recovered read error\n");
179         } else /* Error */
180                 SetPageError(page);
181
182         return ret;
183 }
184
185 static void update_write_page(struct page *page, int ret)
186 {
187         if (ret) {
188                 mapping_set_error(page->mapping, ret);
189                 SetPageError(page);
190         }
191         end_page_writeback(page);
192 }
193
194 /* Called at the end of reads, to optionally unlock pages and update their
195  * status.
196  */
197 static int __readpages_done(struct page_collect *pcol)
198 {
199         int i;
200         u64 resid;
201         u64 good_bytes;
202         u64 length = 0;
203         int ret = exofs_check_io(pcol->ios, &resid);
204
205         if (likely(!ret))
206                 good_bytes = pcol->length;
207         else
208                 good_bytes = pcol->length - resid;
209
210         EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
211                      " length=0x%lx nr_pages=%u\n",
212                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
213                      pcol->nr_pages);
214
215         for (i = 0; i < pcol->nr_pages; i++) {
216                 struct page *page = pcol->pages[i];
217                 struct inode *inode = page->mapping->host;
218                 int page_stat;
219
220                 if (inode != pcol->inode)
221                         continue; /* osd might add more pages at end */
222
223                 if (likely(length < good_bytes))
224                         page_stat = 0;
225                 else
226                         page_stat = ret;
227
228                 EXOFS_DBGMSG2("    readpages_done(0x%lx, 0x%lx) %s\n",
229                           inode->i_ino, page->index,
230                           page_stat ? "bad_bytes" : "good_bytes");
231
232                 ret = update_read_page(page, page_stat);
233                 if (!pcol->read_4_write)
234                         unlock_page(page);
235                 length += PAGE_SIZE;
236         }
237
238         pcol_free(pcol);
239         EXOFS_DBGMSG2("readpages_done END\n");
240         return ret;
241 }
242
243 /* callback of async reads */
244 static void readpages_done(struct exofs_io_state *ios, void *p)
245 {
246         struct page_collect *pcol = p;
247
248         __readpages_done(pcol);
249         atomic_dec(&pcol->sbi->s_curr_pending);
250         kfree(pcol);
251 }
252
253 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
254 {
255         int i;
256
257         for (i = 0; i < pcol->nr_pages; i++) {
258                 struct page *page = pcol->pages[i];
259
260                 if (rw == READ)
261                         update_read_page(page, ret);
262                 else
263                         update_write_page(page, ret);
264
265                 unlock_page(page);
266         }
267 }
268
269 static int read_exec(struct page_collect *pcol)
270 {
271         struct exofs_i_info *oi = exofs_i(pcol->inode);
272         struct exofs_io_state *ios = pcol->ios;
273         struct page_collect *pcol_copy = NULL;
274         int ret;
275
276         if (!pcol->pages)
277                 return 0;
278
279         ios->pages = pcol->pages;
280         ios->nr_pages = pcol->nr_pages;
281         ios->length = pcol->length;
282         ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
283
284         if (pcol->read_4_write) {
285                 exofs_oi_read(oi, pcol->ios);
286                 return __readpages_done(pcol);
287         }
288
289         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
290         if (!pcol_copy) {
291                 ret = -ENOMEM;
292                 goto err;
293         }
294
295         *pcol_copy = *pcol;
296         ios->done = readpages_done;
297         ios->private = pcol_copy;
298         ret = exofs_oi_read(oi, ios);
299         if (unlikely(ret))
300                 goto err;
301
302         atomic_inc(&pcol->sbi->s_curr_pending);
303
304         EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
305                   ios->obj.id, _LLU(ios->offset), pcol->length);
306
307         /* pages ownership was passed to pcol_copy */
308         _pcol_reset(pcol);
309         return 0;
310
311 err:
312         if (!pcol->read_4_write)
313                 _unlock_pcol_pages(pcol, ret, READ);
314
315         pcol_free(pcol);
316
317         kfree(pcol_copy);
318         return ret;
319 }
320
321 /* readpage_strip is called either directly from readpage() or by the VFS from
322  * within read_cache_pages(), to add one more page to be read. It will try to
323  * collect as many contiguous pages as posible. If a discontinuity is
324  * encountered, or it runs out of resources, it will submit the previous segment
325  * and will start a new collection. Eventually caller must submit the last
326  * segment if present.
327  */
328 static int readpage_strip(void *data, struct page *page)
329 {
330         struct page_collect *pcol = data;
331         struct inode *inode = pcol->inode;
332         struct exofs_i_info *oi = exofs_i(inode);
333         loff_t i_size = i_size_read(inode);
334         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
335         size_t len;
336         int ret;
337
338         /* FIXME: Just for debugging, will be removed */
339         if (PageUptodate(page))
340                 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
341                           page->index);
342
343         if (page->index < end_index)
344                 len = PAGE_CACHE_SIZE;
345         else if (page->index == end_index)
346                 len = i_size & ~PAGE_CACHE_MASK;
347         else
348                 len = 0;
349
350         if (!len || !obj_created(oi)) {
351                 /* this will be out of bounds, or doesn't exist yet.
352                  * Current page is cleared and the request is split
353                  */
354                 clear_highpage(page);
355
356                 SetPageUptodate(page);
357                 if (PageError(page))
358                         ClearPageError(page);
359
360                 if (!pcol->read_4_write)
361                         unlock_page(page);
362                 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
363                              "read_4_write=%d index=0x%lx end_index=0x%lx "
364                              "splitting\n", inode->i_ino, len,
365                              pcol->read_4_write, page->index, end_index);
366
367                 return read_exec(pcol);
368         }
369
370 try_again:
371
372         if (unlikely(pcol->pg_first == -1)) {
373                 pcol->pg_first = page->index;
374         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
375                    page->index)) {
376                 /* Discontinuity detected, split the request */
377                 ret = read_exec(pcol);
378                 if (unlikely(ret))
379                         goto fail;
380                 goto try_again;
381         }
382
383         if (!pcol->pages) {
384                 ret = pcol_try_alloc(pcol);
385                 if (unlikely(ret))
386                         goto fail;
387         }
388
389         if (len != PAGE_CACHE_SIZE)
390                 zero_user(page, len, PAGE_CACHE_SIZE - len);
391
392         EXOFS_DBGMSG2("    readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
393                      inode->i_ino, page->index, len);
394
395         ret = pcol_add_page(pcol, page, len);
396         if (ret) {
397                 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
398                           "this_len=0x%zx nr_pages=%u length=0x%lx\n",
399                           page, len, pcol->nr_pages, pcol->length);
400
401                 /* split the request, and start again with current page */
402                 ret = read_exec(pcol);
403                 if (unlikely(ret))
404                         goto fail;
405
406                 goto try_again;
407         }
408
409         return 0;
410
411 fail:
412         /* SetPageError(page); ??? */
413         unlock_page(page);
414         return ret;
415 }
416
417 static int exofs_readpages(struct file *file, struct address_space *mapping,
418                            struct list_head *pages, unsigned nr_pages)
419 {
420         struct page_collect pcol;
421         int ret;
422
423         _pcol_init(&pcol, nr_pages, mapping->host);
424
425         ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
426         if (ret) {
427                 EXOFS_ERR("read_cache_pages => %d\n", ret);
428                 return ret;
429         }
430
431         return read_exec(&pcol);
432 }
433
434 static int _readpage(struct page *page, bool read_4_write)
435 {
436         struct page_collect pcol;
437         int ret;
438
439         _pcol_init(&pcol, 1, page->mapping->host);
440
441         pcol.read_4_write = read_4_write;
442         ret = readpage_strip(&pcol, page);
443         if (ret) {
444                 EXOFS_ERR("_readpage => %d\n", ret);
445                 return ret;
446         }
447
448         return read_exec(&pcol);
449 }
450
451 /*
452  * We don't need the file
453  */
454 static int exofs_readpage(struct file *file, struct page *page)
455 {
456         return _readpage(page, false);
457 }
458
459 /* Callback for osd_write. All writes are asynchronous */
460 static void writepages_done(struct exofs_io_state *ios, void *p)
461 {
462         struct page_collect *pcol = p;
463         int i;
464         u64 resid;
465         u64  good_bytes;
466         u64  length = 0;
467         int ret = exofs_check_io(ios, &resid);
468
469         atomic_dec(&pcol->sbi->s_curr_pending);
470
471         if (likely(!ret))
472                 good_bytes = pcol->length;
473         else
474                 good_bytes = pcol->length - resid;
475
476         EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
477                      " length=0x%lx nr_pages=%u\n",
478                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
479                      pcol->nr_pages);
480
481         for (i = 0; i < pcol->nr_pages; i++) {
482                 struct page *page = pcol->pages[i];
483                 struct inode *inode = page->mapping->host;
484                 int page_stat;
485
486                 if (inode != pcol->inode)
487                         continue; /* osd might add more pages to a bio */
488
489                 if (likely(length < good_bytes))
490                         page_stat = 0;
491                 else
492                         page_stat = ret;
493
494                 update_write_page(page, page_stat);
495                 unlock_page(page);
496                 EXOFS_DBGMSG2("    writepages_done(0x%lx, 0x%lx) status=%d\n",
497                              inode->i_ino, page->index, page_stat);
498
499                 length += PAGE_SIZE;
500         }
501
502         pcol_free(pcol);
503         kfree(pcol);
504         EXOFS_DBGMSG2("writepages_done END\n");
505 }
506
507 static int write_exec(struct page_collect *pcol)
508 {
509         struct exofs_i_info *oi = exofs_i(pcol->inode);
510         struct exofs_io_state *ios = pcol->ios;
511         struct page_collect *pcol_copy = NULL;
512         int ret;
513
514         if (!pcol->pages)
515                 return 0;
516
517         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
518         if (!pcol_copy) {
519                 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
520                 ret = -ENOMEM;
521                 goto err;
522         }
523
524         *pcol_copy = *pcol;
525
526         ios->pages = pcol_copy->pages;
527         ios->nr_pages = pcol_copy->nr_pages;
528         ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
529         ios->length = pcol_copy->length;
530         ios->done = writepages_done;
531         ios->private = pcol_copy;
532
533         ret = exofs_oi_write(oi, ios);
534         if (unlikely(ret)) {
535                 EXOFS_ERR("write_exec: exofs_oi_write() Failed\n");
536                 goto err;
537         }
538
539         atomic_inc(&pcol->sbi->s_curr_pending);
540         EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
541                   pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
542                   pcol->length);
543         /* pages ownership was passed to pcol_copy */
544         _pcol_reset(pcol);
545         return 0;
546
547 err:
548         _unlock_pcol_pages(pcol, ret, WRITE);
549         pcol_free(pcol);
550         kfree(pcol_copy);
551
552         return ret;
553 }
554
555 /* writepage_strip is called either directly from writepage() or by the VFS from
556  * within write_cache_pages(), to add one more page to be written to storage.
557  * It will try to collect as many contiguous pages as possible. If a
558  * discontinuity is encountered or it runs out of resources it will submit the
559  * previous segment and will start a new collection.
560  * Eventually caller must submit the last segment if present.
561  */
562 static int writepage_strip(struct page *page,
563                            struct writeback_control *wbc_unused, void *data)
564 {
565         struct page_collect *pcol = data;
566         struct inode *inode = pcol->inode;
567         struct exofs_i_info *oi = exofs_i(inode);
568         loff_t i_size = i_size_read(inode);
569         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
570         size_t len;
571         int ret;
572
573         BUG_ON(!PageLocked(page));
574
575         ret = wait_obj_created(oi);
576         if (unlikely(ret))
577                 goto fail;
578
579         if (page->index < end_index)
580                 /* in this case, the page is within the limits of the file */
581                 len = PAGE_CACHE_SIZE;
582         else {
583                 len = i_size & ~PAGE_CACHE_MASK;
584
585                 if (page->index > end_index || !len) {
586                         /* in this case, the page is outside the limits
587                          * (truncate in progress)
588                          */
589                         ret = write_exec(pcol);
590                         if (unlikely(ret))
591                                 goto fail;
592                         if (PageError(page))
593                                 ClearPageError(page);
594                         unlock_page(page);
595                         EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
596                                      "outside the limits\n",
597                                      inode->i_ino, page->index);
598                         return 0;
599                 }
600         }
601
602 try_again:
603
604         if (unlikely(pcol->pg_first == -1)) {
605                 pcol->pg_first = page->index;
606         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
607                    page->index)) {
608                 /* Discontinuity detected, split the request */
609                 ret = write_exec(pcol);
610                 if (unlikely(ret))
611                         goto fail;
612
613                 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
614                              inode->i_ino, page->index);
615                 goto try_again;
616         }
617
618         if (!pcol->pages) {
619                 ret = pcol_try_alloc(pcol);
620                 if (unlikely(ret))
621                         goto fail;
622         }
623
624         EXOFS_DBGMSG2("    writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
625                      inode->i_ino, page->index, len);
626
627         ret = pcol_add_page(pcol, page, len);
628         if (unlikely(ret)) {
629                 EXOFS_DBGMSG2("Failed pcol_add_page "
630                              "nr_pages=%u total_length=0x%lx\n",
631                              pcol->nr_pages, pcol->length);
632
633                 /* split the request, next loop will start again */
634                 ret = write_exec(pcol);
635                 if (unlikely(ret)) {
636                         EXOFS_DBGMSG("write_exec failed => %d", ret);
637                         goto fail;
638                 }
639
640                 goto try_again;
641         }
642
643         BUG_ON(PageWriteback(page));
644         set_page_writeback(page);
645
646         return 0;
647
648 fail:
649         EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
650                      inode->i_ino, page->index, ret);
651         set_bit(AS_EIO, &page->mapping->flags);
652         unlock_page(page);
653         return ret;
654 }
655
656 static int exofs_writepages(struct address_space *mapping,
657                        struct writeback_control *wbc)
658 {
659         struct page_collect pcol;
660         long start, end, expected_pages;
661         int ret;
662
663         start = wbc->range_start >> PAGE_CACHE_SHIFT;
664         end = (wbc->range_end == LLONG_MAX) ?
665                         start + mapping->nrpages :
666                         wbc->range_end >> PAGE_CACHE_SHIFT;
667
668         if (start || end)
669                 expected_pages = end - start + 1;
670         else
671                 expected_pages = mapping->nrpages;
672
673         if (expected_pages < 32L)
674                 expected_pages = 32L;
675
676         EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
677                      "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
678                      mapping->host->i_ino, wbc->range_start, wbc->range_end,
679                      mapping->nrpages, start, end, expected_pages);
680
681         _pcol_init(&pcol, expected_pages, mapping->host);
682
683         ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
684         if (ret) {
685                 EXOFS_ERR("write_cache_pages => %d\n", ret);
686                 return ret;
687         }
688
689         return write_exec(&pcol);
690 }
691
692 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
693 {
694         struct page_collect pcol;
695         int ret;
696
697         _pcol_init(&pcol, 1, page->mapping->host);
698
699         ret = writepage_strip(page, NULL, &pcol);
700         if (ret) {
701                 EXOFS_ERR("exofs_writepage => %d\n", ret);
702                 return ret;
703         }
704
705         return write_exec(&pcol);
706 }
707
708 /* i_mutex held using inode->i_size directly */
709 static void _write_failed(struct inode *inode, loff_t to)
710 {
711         if (to > inode->i_size)
712                 truncate_pagecache(inode, to, inode->i_size);
713 }
714
715 int exofs_write_begin(struct file *file, struct address_space *mapping,
716                 loff_t pos, unsigned len, unsigned flags,
717                 struct page **pagep, void **fsdata)
718 {
719         int ret = 0;
720         struct page *page;
721
722         page = *pagep;
723         if (page == NULL) {
724                 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
725                                          fsdata);
726                 if (ret) {
727                         EXOFS_DBGMSG("simple_write_begin failed\n");
728                         goto out;
729                 }
730
731                 page = *pagep;
732         }
733
734          /* read modify write */
735         if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
736                 loff_t i_size = i_size_read(mapping->host);
737                 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
738                 size_t rlen;
739
740                 if (page->index < end_index)
741                         rlen = PAGE_CACHE_SIZE;
742                 else if (page->index == end_index)
743                         rlen = i_size & ~PAGE_CACHE_MASK;
744                 else
745                         rlen = 0;
746
747                 if (!rlen) {
748                         clear_highpage(page);
749                         SetPageUptodate(page);
750                         goto out;
751                 }
752
753                 ret = _readpage(page, true);
754                 if (ret) {
755                         /*SetPageError was done by _readpage. Is it ok?*/
756                         unlock_page(page);
757                         EXOFS_DBGMSG("__readpage failed\n");
758                 }
759         }
760 out:
761         if (unlikely(ret))
762                 _write_failed(mapping->host, pos + len);
763
764         return ret;
765 }
766
767 static int exofs_write_begin_export(struct file *file,
768                 struct address_space *mapping,
769                 loff_t pos, unsigned len, unsigned flags,
770                 struct page **pagep, void **fsdata)
771 {
772         *pagep = NULL;
773
774         return exofs_write_begin(file, mapping, pos, len, flags, pagep,
775                                         fsdata);
776 }
777
778 static int exofs_write_end(struct file *file, struct address_space *mapping,
779                         loff_t pos, unsigned len, unsigned copied,
780                         struct page *page, void *fsdata)
781 {
782         struct inode *inode = mapping->host;
783         /* According to comment in simple_write_end i_mutex is held */
784         loff_t i_size = inode->i_size;
785         int ret;
786
787         ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
788         if (unlikely(ret))
789                 _write_failed(inode, pos + len);
790
791         /* TODO: once simple_write_end marks inode dirty remove */
792         if (i_size != inode->i_size)
793                 mark_inode_dirty(inode);
794         return ret;
795 }
796
797 static int exofs_releasepage(struct page *page, gfp_t gfp)
798 {
799         EXOFS_DBGMSG("page 0x%lx\n", page->index);
800         WARN_ON(1);
801         return 0;
802 }
803
804 static void exofs_invalidatepage(struct page *page, unsigned long offset)
805 {
806         EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset);
807         WARN_ON(1);
808 }
809
810 const struct address_space_operations exofs_aops = {
811         .readpage       = exofs_readpage,
812         .readpages      = exofs_readpages,
813         .writepage      = exofs_writepage,
814         .writepages     = exofs_writepages,
815         .write_begin    = exofs_write_begin_export,
816         .write_end      = exofs_write_end,
817         .releasepage    = exofs_releasepage,
818         .set_page_dirty = __set_page_dirty_nobuffers,
819         .invalidatepage = exofs_invalidatepage,
820
821         /* Not implemented Yet */
822         .bmap           = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
823         .direct_IO      = NULL, /* TODO: Should be trivial to do */
824
825         /* With these NULL has special meaning or default is not exported */
826         .get_xip_mem    = NULL,
827         .migratepage    = NULL,
828         .launder_page   = NULL,
829         .is_partially_uptodate = NULL,
830         .error_remove_page = NULL,
831 };
832
833 /******************************************************************************
834  * INODE OPERATIONS
835  *****************************************************************************/
836
837 /*
838  * Test whether an inode is a fast symlink.
839  */
840 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
841 {
842         struct exofs_i_info *oi = exofs_i(inode);
843
844         return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
845 }
846
847 const struct osd_attr g_attr_logical_length = ATTR_DEF(
848         OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
849
850 static int _do_truncate(struct inode *inode, loff_t newsize)
851 {
852         struct exofs_i_info *oi = exofs_i(inode);
853         int ret;
854
855         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
856
857         ret = exofs_oi_truncate(oi, (u64)newsize);
858         if (likely(!ret))
859                 truncate_setsize(inode, newsize);
860
861         EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
862                      inode->i_ino, newsize, ret);
863         return ret;
864 }
865
866 /*
867  * Set inode attributes - update size attribute on OSD if needed,
868  *                        otherwise just call generic functions.
869  */
870 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
871 {
872         struct inode *inode = dentry->d_inode;
873         int error;
874
875         /* if we are about to modify an object, and it hasn't been
876          * created yet, wait
877          */
878         error = wait_obj_created(exofs_i(inode));
879         if (unlikely(error))
880                 return error;
881
882         error = inode_change_ok(inode, iattr);
883         if (unlikely(error))
884                 return error;
885
886         if ((iattr->ia_valid & ATTR_SIZE) &&
887             iattr->ia_size != i_size_read(inode)) {
888                 error = _do_truncate(inode, iattr->ia_size);
889                 if (unlikely(error))
890                         return error;
891         }
892
893         setattr_copy(inode, iattr);
894         mark_inode_dirty(inode);
895         return 0;
896 }
897
898 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
899         EXOFS_APAGE_FS_DATA,
900         EXOFS_ATTR_INODE_FILE_LAYOUT,
901         0);
902 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
903         EXOFS_APAGE_FS_DATA,
904         EXOFS_ATTR_INODE_DIR_LAYOUT,
905         0);
906
907 /*
908  * Read the Linux inode info from the OSD, and return it as is. In exofs the
909  * inode info is in an application specific page/attribute of the osd-object.
910  */
911 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
912                     struct exofs_fcb *inode)
913 {
914         struct exofs_sb_info *sbi = sb->s_fs_info;
915         struct osd_attr attrs[] = {
916                 [0] = g_attr_inode_data,
917                 [1] = g_attr_inode_file_layout,
918                 [2] = g_attr_inode_dir_layout,
919         };
920         struct exofs_io_state *ios;
921         struct exofs_on_disk_inode_layout *layout;
922         int ret;
923
924         ret = exofs_get_io_state(&sbi->layout, &ios);
925         if (unlikely(ret)) {
926                 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
927                 return ret;
928         }
929
930         ios->obj.id = exofs_oi_objno(oi);
931         exofs_make_credential(oi->i_cred, &ios->obj);
932         ios->cred = oi->i_cred;
933
934         attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
935         attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
936
937         ios->in_attr = attrs;
938         ios->in_attr_len = ARRAY_SIZE(attrs);
939
940         ret = exofs_sbi_read(ios);
941         if (unlikely(ret)) {
942                 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
943                           _LLU(ios->obj.id), ret);
944                 memset(inode, 0, sizeof(*inode));
945                 inode->i_mode = 0040000 | (0777 & ~022);
946                 /* If object is lost on target we might as well enable it's
947                  * delete.
948                  */
949                 if ((ret == -ENOENT) || (ret == -EINVAL))
950                         ret = 0;
951                 goto out;
952         }
953
954         ret = extract_attr_from_ios(ios, &attrs[0]);
955         if (ret) {
956                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
957                 goto out;
958         }
959         WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
960         memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
961
962         ret = extract_attr_from_ios(ios, &attrs[1]);
963         if (ret) {
964                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
965                 goto out;
966         }
967         if (attrs[1].len) {
968                 layout = attrs[1].val_ptr;
969                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
970                         EXOFS_ERR("%s: unsupported files layout %d\n",
971                                 __func__, layout->gen_func);
972                         ret = -ENOTSUPP;
973                         goto out;
974                 }
975         }
976
977         ret = extract_attr_from_ios(ios, &attrs[2]);
978         if (ret) {
979                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
980                 goto out;
981         }
982         if (attrs[2].len) {
983                 layout = attrs[2].val_ptr;
984                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
985                         EXOFS_ERR("%s: unsupported meta-data layout %d\n",
986                                 __func__, layout->gen_func);
987                         ret = -ENOTSUPP;
988                         goto out;
989                 }
990         }
991
992 out:
993         exofs_put_io_state(ios);
994         return ret;
995 }
996
997 static void __oi_init(struct exofs_i_info *oi)
998 {
999         init_waitqueue_head(&oi->i_wq);
1000         oi->i_flags = 0;
1001 }
1002 /*
1003  * Fill in an inode read from the OSD and set it up for use
1004  */
1005 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1006 {
1007         struct exofs_i_info *oi;
1008         struct exofs_fcb fcb;
1009         struct inode *inode;
1010         int ret;
1011
1012         inode = iget_locked(sb, ino);
1013         if (!inode)
1014                 return ERR_PTR(-ENOMEM);
1015         if (!(inode->i_state & I_NEW))
1016                 return inode;
1017         oi = exofs_i(inode);
1018         __oi_init(oi);
1019
1020         /* read the inode from the osd */
1021         ret = exofs_get_inode(sb, oi, &fcb);
1022         if (ret)
1023                 goto bad_inode;
1024
1025         set_obj_created(oi);
1026
1027         /* copy stuff from on-disk struct to in-memory struct */
1028         inode->i_mode = le16_to_cpu(fcb.i_mode);
1029         inode->i_uid = le32_to_cpu(fcb.i_uid);
1030         inode->i_gid = le32_to_cpu(fcb.i_gid);
1031         inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1032         inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1033         inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1034         inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1035         inode->i_ctime.tv_nsec =
1036                 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1037         oi->i_commit_size = le64_to_cpu(fcb.i_size);
1038         i_size_write(inode, oi->i_commit_size);
1039         inode->i_blkbits = EXOFS_BLKSHIFT;
1040         inode->i_generation = le32_to_cpu(fcb.i_generation);
1041
1042         oi->i_dir_start_lookup = 0;
1043
1044         if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1045                 ret = -ESTALE;
1046                 goto bad_inode;
1047         }
1048
1049         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1050                 if (fcb.i_data[0])
1051                         inode->i_rdev =
1052                                 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1053                 else
1054                         inode->i_rdev =
1055                                 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1056         } else {
1057                 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1058         }
1059
1060         inode->i_mapping->backing_dev_info = sb->s_bdi;
1061         if (S_ISREG(inode->i_mode)) {
1062                 inode->i_op = &exofs_file_inode_operations;
1063                 inode->i_fop = &exofs_file_operations;
1064                 inode->i_mapping->a_ops = &exofs_aops;
1065         } else if (S_ISDIR(inode->i_mode)) {
1066                 inode->i_op = &exofs_dir_inode_operations;
1067                 inode->i_fop = &exofs_dir_operations;
1068                 inode->i_mapping->a_ops = &exofs_aops;
1069         } else if (S_ISLNK(inode->i_mode)) {
1070                 if (exofs_inode_is_fast_symlink(inode))
1071                         inode->i_op = &exofs_fast_symlink_inode_operations;
1072                 else {
1073                         inode->i_op = &exofs_symlink_inode_operations;
1074                         inode->i_mapping->a_ops = &exofs_aops;
1075                 }
1076         } else {
1077                 inode->i_op = &exofs_special_inode_operations;
1078                 if (fcb.i_data[0])
1079                         init_special_inode(inode, inode->i_mode,
1080                            old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1081                 else
1082                         init_special_inode(inode, inode->i_mode,
1083                            new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1084         }
1085
1086         unlock_new_inode(inode);
1087         return inode;
1088
1089 bad_inode:
1090         iget_failed(inode);
1091         return ERR_PTR(ret);
1092 }
1093
1094 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1095 {
1096         if (!obj_created(oi)) {
1097                 EXOFS_DBGMSG("!obj_created\n");
1098                 BUG_ON(!obj_2bcreated(oi));
1099                 wait_event(oi->i_wq, obj_created(oi));
1100                 EXOFS_DBGMSG("wait_event done\n");
1101         }
1102         return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1103 }
1104
1105 /*
1106  * Callback function from exofs_new_inode().  The important thing is that we
1107  * set the obj_created flag so that other methods know that the object exists on
1108  * the OSD.
1109  */
1110 static void create_done(struct exofs_io_state *ios, void *p)
1111 {
1112         struct inode *inode = p;
1113         struct exofs_i_info *oi = exofs_i(inode);
1114         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1115         int ret;
1116
1117         ret = exofs_check_io(ios, NULL);
1118         exofs_put_io_state(ios);
1119
1120         atomic_dec(&sbi->s_curr_pending);
1121
1122         if (unlikely(ret)) {
1123                 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1124                           _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
1125                 /*TODO: When FS is corrupted creation can fail, object already
1126                  * exist. Get rid of this asynchronous creation, if exist
1127                  * increment the obj counter and try the next object. Until we
1128                  * succeed. All these dangling objects will be made into lost
1129                  * files by chkfs.exofs
1130                  */
1131         }
1132
1133         set_obj_created(oi);
1134
1135         wake_up(&oi->i_wq);
1136 }
1137
1138 /*
1139  * Set up a new inode and create an object for it on the OSD
1140  */
1141 struct inode *exofs_new_inode(struct inode *dir, int mode)
1142 {
1143         struct super_block *sb;
1144         struct inode *inode;
1145         struct exofs_i_info *oi;
1146         struct exofs_sb_info *sbi;
1147         struct exofs_io_state *ios;
1148         int ret;
1149
1150         sb = dir->i_sb;
1151         inode = new_inode(sb);
1152         if (!inode)
1153                 return ERR_PTR(-ENOMEM);
1154
1155         oi = exofs_i(inode);
1156         __oi_init(oi);
1157
1158         set_obj_2bcreated(oi);
1159
1160         sbi = sb->s_fs_info;
1161
1162         inode->i_mapping->backing_dev_info = sb->s_bdi;
1163         inode_init_owner(inode, dir, mode);
1164         inode->i_ino = sbi->s_nextid++;
1165         inode->i_blkbits = EXOFS_BLKSHIFT;
1166         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1167         oi->i_commit_size = inode->i_size = 0;
1168         spin_lock(&sbi->s_next_gen_lock);
1169         inode->i_generation = sbi->s_next_generation++;
1170         spin_unlock(&sbi->s_next_gen_lock);
1171         insert_inode_hash(inode);
1172
1173         exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */
1174
1175         mark_inode_dirty(inode);
1176
1177         ret = exofs_get_io_state(&sbi->layout, &ios);
1178         if (unlikely(ret)) {
1179                 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1180                 return ERR_PTR(ret);
1181         }
1182
1183         ios->obj.id = exofs_oi_objno(oi);
1184         exofs_make_credential(oi->i_cred, &ios->obj);
1185
1186         ios->done = create_done;
1187         ios->private = inode;
1188         ios->cred = oi->i_cred;
1189         ret = exofs_sbi_create(ios);
1190         if (ret) {
1191                 exofs_put_io_state(ios);
1192                 return ERR_PTR(ret);
1193         }
1194         atomic_inc(&sbi->s_curr_pending);
1195
1196         return inode;
1197 }
1198
1199 /*
1200  * struct to pass two arguments to update_inode's callback
1201  */
1202 struct updatei_args {
1203         struct exofs_sb_info    *sbi;
1204         struct exofs_fcb        fcb;
1205 };
1206
1207 /*
1208  * Callback function from exofs_update_inode().
1209  */
1210 static void updatei_done(struct exofs_io_state *ios, void *p)
1211 {
1212         struct updatei_args *args = p;
1213
1214         exofs_put_io_state(ios);
1215
1216         atomic_dec(&args->sbi->s_curr_pending);
1217
1218         kfree(args);
1219 }
1220
1221 /*
1222  * Write the inode to the OSD.  Just fill up the struct, and set the attribute
1223  * synchronously or asynchronously depending on the do_sync flag.
1224  */
1225 static int exofs_update_inode(struct inode *inode, int do_sync)
1226 {
1227         struct exofs_i_info *oi = exofs_i(inode);
1228         struct super_block *sb = inode->i_sb;
1229         struct exofs_sb_info *sbi = sb->s_fs_info;
1230         struct exofs_io_state *ios;
1231         struct osd_attr attr;
1232         struct exofs_fcb *fcb;
1233         struct updatei_args *args;
1234         int ret;
1235
1236         args = kzalloc(sizeof(*args), GFP_KERNEL);
1237         if (!args) {
1238                 EXOFS_DBGMSG("Failed kzalloc of args\n");
1239                 return -ENOMEM;
1240         }
1241
1242         fcb = &args->fcb;
1243
1244         fcb->i_mode = cpu_to_le16(inode->i_mode);
1245         fcb->i_uid = cpu_to_le32(inode->i_uid);
1246         fcb->i_gid = cpu_to_le32(inode->i_gid);
1247         fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1248         fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1249         fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1250         fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1251         oi->i_commit_size = i_size_read(inode);
1252         fcb->i_size = cpu_to_le64(oi->i_commit_size);
1253         fcb->i_generation = cpu_to_le32(inode->i_generation);
1254
1255         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1256                 if (old_valid_dev(inode->i_rdev)) {
1257                         fcb->i_data[0] =
1258                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1259                         fcb->i_data[1] = 0;
1260                 } else {
1261                         fcb->i_data[0] = 0;
1262                         fcb->i_data[1] =
1263                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1264                         fcb->i_data[2] = 0;
1265                 }
1266         } else
1267                 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1268
1269         ret = exofs_get_io_state(&sbi->layout, &ios);
1270         if (unlikely(ret)) {
1271                 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
1272                 goto free_args;
1273         }
1274
1275         attr = g_attr_inode_data;
1276         attr.val_ptr = fcb;
1277         ios->out_attr_len = 1;
1278         ios->out_attr = &attr;
1279
1280         wait_obj_created(oi);
1281
1282         if (!do_sync) {
1283                 args->sbi = sbi;
1284                 ios->done = updatei_done;
1285                 ios->private = args;
1286         }
1287
1288         ret = exofs_oi_write(oi, ios);
1289         if (!do_sync && !ret) {
1290                 atomic_inc(&sbi->s_curr_pending);
1291                 goto out; /* deallocation in updatei_done */
1292         }
1293
1294         exofs_put_io_state(ios);
1295 free_args:
1296         kfree(args);
1297 out:
1298         EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1299                      inode->i_ino, do_sync, ret);
1300         return ret;
1301 }
1302
1303 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1304 {
1305         /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1306         return exofs_update_inode(inode, 1);
1307 }
1308
1309 /*
1310  * Callback function from exofs_delete_inode() - don't have much cleaning up to
1311  * do.
1312  */
1313 static void delete_done(struct exofs_io_state *ios, void *p)
1314 {
1315         struct exofs_sb_info *sbi = p;
1316
1317         exofs_put_io_state(ios);
1318
1319         atomic_dec(&sbi->s_curr_pending);
1320 }
1321
1322 /*
1323  * Called when the refcount of an inode reaches zero.  We remove the object
1324  * from the OSD here.  We make sure the object was created before we try and
1325  * delete it.
1326  */
1327 void exofs_evict_inode(struct inode *inode)
1328 {
1329         struct exofs_i_info *oi = exofs_i(inode);
1330         struct super_block *sb = inode->i_sb;
1331         struct exofs_sb_info *sbi = sb->s_fs_info;
1332         struct exofs_io_state *ios;
1333         int ret;
1334
1335         truncate_inode_pages(&inode->i_data, 0);
1336
1337         /* TODO: should do better here */
1338         if (inode->i_nlink || is_bad_inode(inode))
1339                 goto no_delete;
1340
1341         inode->i_size = 0;
1342         end_writeback(inode);
1343
1344         /* if we are deleting an obj that hasn't been created yet, wait.
1345          * This also makes sure that create_done cannot be called with an
1346          * already evicted inode.
1347          */
1348         wait_obj_created(oi);
1349         /* ignore the error, attempt a remove anyway */
1350
1351         /* Now Remove the OSD objects */
1352         ret = exofs_get_io_state(&sbi->layout, &ios);
1353         if (unlikely(ret)) {
1354                 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
1355                 return;
1356         }
1357
1358         ios->obj.id = exofs_oi_objno(oi);
1359         ios->done = delete_done;
1360         ios->private = sbi;
1361         ios->cred = oi->i_cred;
1362         ret = exofs_sbi_remove(ios);
1363         if (ret) {
1364                 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__);
1365                 exofs_put_io_state(ios);
1366                 return;
1367         }
1368         atomic_inc(&sbi->s_curr_pending);
1369
1370         return;
1371
1372 no_delete:
1373         end_writeback(inode);
1374 }