Merge branch 'master' into for-next
[linux-2.6.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 #include <linux/writeback.h>
36 #include <linux/buffer_head.h>
37 #include <scsi/scsi_device.h>
38
39 #include "exofs.h"
40
41 #define EXOFS_DBGMSG2(M...) do {} while (0)
42
43 enum { BIO_MAX_PAGES_KMALLOC =
44                 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
45         MAX_PAGES_KMALLOC =
46                 PAGE_SIZE / sizeof(struct page *),
47 };
48
49 struct page_collect {
50         struct exofs_sb_info *sbi;
51         struct inode *inode;
52         unsigned expected_pages;
53         struct exofs_io_state *ios;
54
55         struct page **pages;
56         unsigned alloc_pages;
57         unsigned nr_pages;
58         unsigned long length;
59         loff_t pg_first; /* keep 64bit also in 32-arches */
60 };
61
62 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
63                        struct inode *inode)
64 {
65         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
66
67         pcol->sbi = sbi;
68         pcol->inode = inode;
69         pcol->expected_pages = expected_pages;
70
71         pcol->ios = NULL;
72         pcol->pages = NULL;
73         pcol->alloc_pages = 0;
74         pcol->nr_pages = 0;
75         pcol->length = 0;
76         pcol->pg_first = -1;
77 }
78
79 static void _pcol_reset(struct page_collect *pcol)
80 {
81         pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
82
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->ios = NULL;
89
90         /* this is probably the end of the loop but in writes
91          * it might not end here. don't be left with nothing
92          */
93         if (!pcol->expected_pages)
94                 pcol->expected_pages = MAX_PAGES_KMALLOC;
95 }
96
97 static int pcol_try_alloc(struct page_collect *pcol)
98 {
99         unsigned pages = min_t(unsigned, pcol->expected_pages,
100                           MAX_PAGES_KMALLOC);
101
102         if (!pcol->ios) { /* First time allocate io_state */
103                 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
104
105                 if (ret)
106                         return ret;
107         }
108
109         /* TODO: easily support bio chaining */
110         pages =  min_t(unsigned, pages,
111                        pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
112
113         for (; pages; pages >>= 1) {
114                 pcol->pages = kmalloc(pages * sizeof(struct page *),
115                                       GFP_KERNEL);
116                 if (likely(pcol->pages)) {
117                         pcol->alloc_pages = pages;
118                         return 0;
119                 }
120         }
121
122         EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
123                   pcol->expected_pages);
124         return -ENOMEM;
125 }
126
127 static void pcol_free(struct page_collect *pcol)
128 {
129         kfree(pcol->pages);
130         pcol->pages = NULL;
131
132         if (pcol->ios) {
133                 exofs_put_io_state(pcol->ios);
134                 pcol->ios = NULL;
135         }
136 }
137
138 static int pcol_add_page(struct page_collect *pcol, struct page *page,
139                          unsigned len)
140 {
141         if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
142                 return -ENOMEM;
143
144         pcol->pages[pcol->nr_pages++] = page;
145         pcol->length += len;
146         return 0;
147 }
148
149 static int update_read_page(struct page *page, int ret)
150 {
151         if (ret == 0) {
152                 /* Everything is OK */
153                 SetPageUptodate(page);
154                 if (PageError(page))
155                         ClearPageError(page);
156         } else if (ret == -EFAULT) {
157                 /* In this case we were trying to read something that wasn't on
158                  * disk yet - return a page full of zeroes.  This should be OK,
159                  * because the object should be empty (if there was a write
160                  * before this read, the read would be waiting with the page
161                  * locked */
162                 clear_highpage(page);
163
164                 SetPageUptodate(page);
165                 if (PageError(page))
166                         ClearPageError(page);
167                 ret = 0; /* recovered error */
168                 EXOFS_DBGMSG("recovered read error\n");
169         } else /* Error */
170                 SetPageError(page);
171
172         return ret;
173 }
174
175 static void update_write_page(struct page *page, int ret)
176 {
177         if (ret) {
178                 mapping_set_error(page->mapping, ret);
179                 SetPageError(page);
180         }
181         end_page_writeback(page);
182 }
183
184 /* Called at the end of reads, to optionally unlock pages and update their
185  * status.
186  */
187 static int __readpages_done(struct page_collect *pcol, bool do_unlock)
188 {
189         int i;
190         u64 resid;
191         u64 good_bytes;
192         u64 length = 0;
193         int ret = exofs_check_io(pcol->ios, &resid);
194
195         if (likely(!ret))
196                 good_bytes = pcol->length;
197         else
198                 good_bytes = pcol->length - resid;
199
200         EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
201                      " length=0x%lx nr_pages=%u\n",
202                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
203                      pcol->nr_pages);
204
205         for (i = 0; i < pcol->nr_pages; i++) {
206                 struct page *page = pcol->pages[i];
207                 struct inode *inode = page->mapping->host;
208                 int page_stat;
209
210                 if (inode != pcol->inode)
211                         continue; /* osd might add more pages at end */
212
213                 if (likely(length < good_bytes))
214                         page_stat = 0;
215                 else
216                         page_stat = ret;
217
218                 EXOFS_DBGMSG2("    readpages_done(0x%lx, 0x%lx) %s\n",
219                           inode->i_ino, page->index,
220                           page_stat ? "bad_bytes" : "good_bytes");
221
222                 ret = update_read_page(page, page_stat);
223                 if (do_unlock)
224                         unlock_page(page);
225                 length += PAGE_SIZE;
226         }
227
228         pcol_free(pcol);
229         EXOFS_DBGMSG2("readpages_done END\n");
230         return ret;
231 }
232
233 /* callback of async reads */
234 static void readpages_done(struct exofs_io_state *ios, void *p)
235 {
236         struct page_collect *pcol = p;
237
238         __readpages_done(pcol, true);
239         atomic_dec(&pcol->sbi->s_curr_pending);
240         kfree(pcol);
241 }
242
243 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
244 {
245         int i;
246
247         for (i = 0; i < pcol->nr_pages; i++) {
248                 struct page *page = pcol->pages[i];
249
250                 if (rw == READ)
251                         update_read_page(page, ret);
252                 else
253                         update_write_page(page, ret);
254
255                 unlock_page(page);
256         }
257 }
258
259 static int read_exec(struct page_collect *pcol, bool is_sync)
260 {
261         struct exofs_i_info *oi = exofs_i(pcol->inode);
262         struct exofs_io_state *ios = pcol->ios;
263         struct page_collect *pcol_copy = NULL;
264         int ret;
265
266         if (!pcol->pages)
267                 return 0;
268
269         /* see comment in _readpage() about sync reads */
270         WARN_ON(is_sync && (pcol->nr_pages != 1));
271
272         ios->pages = pcol->pages;
273         ios->nr_pages = pcol->nr_pages;
274         ios->length = pcol->length;
275         ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
276
277         if (is_sync) {
278                 exofs_oi_read(oi, pcol->ios);
279                 return __readpages_done(pcol, false);
280         }
281
282         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
283         if (!pcol_copy) {
284                 ret = -ENOMEM;
285                 goto err;
286         }
287
288         *pcol_copy = *pcol;
289         ios->done = readpages_done;
290         ios->private = pcol_copy;
291         ret = exofs_oi_read(oi, ios);
292         if (unlikely(ret))
293                 goto err;
294
295         atomic_inc(&pcol->sbi->s_curr_pending);
296
297         EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
298                   ios->obj.id, _LLU(ios->offset), pcol->length);
299
300         /* pages ownership was passed to pcol_copy */
301         _pcol_reset(pcol);
302         return 0;
303
304 err:
305         if (!is_sync)
306                 _unlock_pcol_pages(pcol, ret, READ);
307
308         pcol_free(pcol);
309
310         kfree(pcol_copy);
311         return ret;
312 }
313
314 /* readpage_strip is called either directly from readpage() or by the VFS from
315  * within read_cache_pages(), to add one more page to be read. It will try to
316  * collect as many contiguous pages as posible. If a discontinuity is
317  * encountered, or it runs out of resources, it will submit the previous segment
318  * and will start a new collection. Eventually caller must submit the last
319  * segment if present.
320  */
321 static int readpage_strip(void *data, struct page *page)
322 {
323         struct page_collect *pcol = data;
324         struct inode *inode = pcol->inode;
325         struct exofs_i_info *oi = exofs_i(inode);
326         loff_t i_size = i_size_read(inode);
327         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
328         size_t len;
329         int ret;
330
331         /* FIXME: Just for debugging, will be removed */
332         if (PageUptodate(page))
333                 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
334                           page->index);
335
336         if (page->index < end_index)
337                 len = PAGE_CACHE_SIZE;
338         else if (page->index == end_index)
339                 len = i_size & ~PAGE_CACHE_MASK;
340         else
341                 len = 0;
342
343         if (!len || !obj_created(oi)) {
344                 /* this will be out of bounds, or doesn't exist yet.
345                  * Current page is cleared and the request is split
346                  */
347                 clear_highpage(page);
348
349                 SetPageUptodate(page);
350                 if (PageError(page))
351                         ClearPageError(page);
352
353                 unlock_page(page);
354                 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
355                              " splitting\n", inode->i_ino, page->index);
356
357                 return read_exec(pcol, false);
358         }
359
360 try_again:
361
362         if (unlikely(pcol->pg_first == -1)) {
363                 pcol->pg_first = page->index;
364         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
365                    page->index)) {
366                 /* Discontinuity detected, split the request */
367                 ret = read_exec(pcol, false);
368                 if (unlikely(ret))
369                         goto fail;
370                 goto try_again;
371         }
372
373         if (!pcol->pages) {
374                 ret = pcol_try_alloc(pcol);
375                 if (unlikely(ret))
376                         goto fail;
377         }
378
379         if (len != PAGE_CACHE_SIZE)
380                 zero_user(page, len, PAGE_CACHE_SIZE - len);
381
382         EXOFS_DBGMSG2("    readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
383                      inode->i_ino, page->index, len);
384
385         ret = pcol_add_page(pcol, page, len);
386         if (ret) {
387                 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
388                           "this_len=0x%zx nr_pages=%u length=0x%lx\n",
389                           page, len, pcol->nr_pages, pcol->length);
390
391                 /* split the request, and start again with current page */
392                 ret = read_exec(pcol, false);
393                 if (unlikely(ret))
394                         goto fail;
395
396                 goto try_again;
397         }
398
399         return 0;
400
401 fail:
402         /* SetPageError(page); ??? */
403         unlock_page(page);
404         return ret;
405 }
406
407 static int exofs_readpages(struct file *file, struct address_space *mapping,
408                            struct list_head *pages, unsigned nr_pages)
409 {
410         struct page_collect pcol;
411         int ret;
412
413         _pcol_init(&pcol, nr_pages, mapping->host);
414
415         ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
416         if (ret) {
417                 EXOFS_ERR("read_cache_pages => %d\n", ret);
418                 return ret;
419         }
420
421         return read_exec(&pcol, false);
422 }
423
424 static int _readpage(struct page *page, bool is_sync)
425 {
426         struct page_collect pcol;
427         int ret;
428
429         _pcol_init(&pcol, 1, page->mapping->host);
430
431         /* readpage_strip might call read_exec(,is_sync==false) at several
432          * places but not if we have a single page.
433          */
434         ret = readpage_strip(&pcol, page);
435         if (ret) {
436                 EXOFS_ERR("_readpage => %d\n", ret);
437                 return ret;
438         }
439
440         return read_exec(&pcol, is_sync);
441 }
442
443 /*
444  * We don't need the file
445  */
446 static int exofs_readpage(struct file *file, struct page *page)
447 {
448         return _readpage(page, false);
449 }
450
451 /* Callback for osd_write. All writes are asynchronous */
452 static void writepages_done(struct exofs_io_state *ios, void *p)
453 {
454         struct page_collect *pcol = p;
455         int i;
456         u64 resid;
457         u64  good_bytes;
458         u64  length = 0;
459         int ret = exofs_check_io(ios, &resid);
460
461         atomic_dec(&pcol->sbi->s_curr_pending);
462
463         if (likely(!ret))
464                 good_bytes = pcol->length;
465         else
466                 good_bytes = pcol->length - resid;
467
468         EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
469                      " length=0x%lx nr_pages=%u\n",
470                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
471                      pcol->nr_pages);
472
473         for (i = 0; i < pcol->nr_pages; i++) {
474                 struct page *page = pcol->pages[i];
475                 struct inode *inode = page->mapping->host;
476                 int page_stat;
477
478                 if (inode != pcol->inode)
479                         continue; /* osd might add more pages to a bio */
480
481                 if (likely(length < good_bytes))
482                         page_stat = 0;
483                 else
484                         page_stat = ret;
485
486                 update_write_page(page, page_stat);
487                 unlock_page(page);
488                 EXOFS_DBGMSG2("    writepages_done(0x%lx, 0x%lx) status=%d\n",
489                              inode->i_ino, page->index, page_stat);
490
491                 length += PAGE_SIZE;
492         }
493
494         pcol_free(pcol);
495         kfree(pcol);
496         EXOFS_DBGMSG2("writepages_done END\n");
497 }
498
499 static int write_exec(struct page_collect *pcol)
500 {
501         struct exofs_i_info *oi = exofs_i(pcol->inode);
502         struct exofs_io_state *ios = pcol->ios;
503         struct page_collect *pcol_copy = NULL;
504         int ret;
505
506         if (!pcol->pages)
507                 return 0;
508
509         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
510         if (!pcol_copy) {
511                 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
512                 ret = -ENOMEM;
513                 goto err;
514         }
515
516         *pcol_copy = *pcol;
517
518         ios->pages = pcol_copy->pages;
519         ios->nr_pages = pcol_copy->nr_pages;
520         ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
521         ios->length = pcol_copy->length;
522         ios->done = writepages_done;
523         ios->private = pcol_copy;
524
525         ret = exofs_oi_write(oi, ios);
526         if (unlikely(ret)) {
527                 EXOFS_ERR("write_exec: exofs_oi_write() Failed\n");
528                 goto err;
529         }
530
531         atomic_inc(&pcol->sbi->s_curr_pending);
532         EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
533                   pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
534                   pcol->length);
535         /* pages ownership was passed to pcol_copy */
536         _pcol_reset(pcol);
537         return 0;
538
539 err:
540         _unlock_pcol_pages(pcol, ret, WRITE);
541         pcol_free(pcol);
542         kfree(pcol_copy);
543
544         return ret;
545 }
546
547 /* writepage_strip is called either directly from writepage() or by the VFS from
548  * within write_cache_pages(), to add one more page to be written to storage.
549  * It will try to collect as many contiguous pages as possible. If a
550  * discontinuity is encountered or it runs out of resources it will submit the
551  * previous segment and will start a new collection.
552  * Eventually caller must submit the last segment if present.
553  */
554 static int writepage_strip(struct page *page,
555                            struct writeback_control *wbc_unused, void *data)
556 {
557         struct page_collect *pcol = data;
558         struct inode *inode = pcol->inode;
559         struct exofs_i_info *oi = exofs_i(inode);
560         loff_t i_size = i_size_read(inode);
561         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
562         size_t len;
563         int ret;
564
565         BUG_ON(!PageLocked(page));
566
567         ret = wait_obj_created(oi);
568         if (unlikely(ret))
569                 goto fail;
570
571         if (page->index < end_index)
572                 /* in this case, the page is within the limits of the file */
573                 len = PAGE_CACHE_SIZE;
574         else {
575                 len = i_size & ~PAGE_CACHE_MASK;
576
577                 if (page->index > end_index || !len) {
578                         /* in this case, the page is outside the limits
579                          * (truncate in progress)
580                          */
581                         ret = write_exec(pcol);
582                         if (unlikely(ret))
583                                 goto fail;
584                         if (PageError(page))
585                                 ClearPageError(page);
586                         unlock_page(page);
587                         EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
588                                      "outside the limits\n",
589                                      inode->i_ino, page->index);
590                         return 0;
591                 }
592         }
593
594 try_again:
595
596         if (unlikely(pcol->pg_first == -1)) {
597                 pcol->pg_first = page->index;
598         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
599                    page->index)) {
600                 /* Discontinuity detected, split the request */
601                 ret = write_exec(pcol);
602                 if (unlikely(ret))
603                         goto fail;
604
605                 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
606                              inode->i_ino, page->index);
607                 goto try_again;
608         }
609
610         if (!pcol->pages) {
611                 ret = pcol_try_alloc(pcol);
612                 if (unlikely(ret))
613                         goto fail;
614         }
615
616         EXOFS_DBGMSG2("    writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
617                      inode->i_ino, page->index, len);
618
619         ret = pcol_add_page(pcol, page, len);
620         if (unlikely(ret)) {
621                 EXOFS_DBGMSG2("Failed pcol_add_page "
622                              "nr_pages=%u total_length=0x%lx\n",
623                              pcol->nr_pages, pcol->length);
624
625                 /* split the request, next loop will start again */
626                 ret = write_exec(pcol);
627                 if (unlikely(ret)) {
628                         EXOFS_DBGMSG("write_exec failed => %d", ret);
629                         goto fail;
630                 }
631
632                 goto try_again;
633         }
634
635         BUG_ON(PageWriteback(page));
636         set_page_writeback(page);
637
638         return 0;
639
640 fail:
641         EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
642                      inode->i_ino, page->index, ret);
643         set_bit(AS_EIO, &page->mapping->flags);
644         unlock_page(page);
645         return ret;
646 }
647
648 static int exofs_writepages(struct address_space *mapping,
649                        struct writeback_control *wbc)
650 {
651         struct page_collect pcol;
652         long start, end, expected_pages;
653         int ret;
654
655         start = wbc->range_start >> PAGE_CACHE_SHIFT;
656         end = (wbc->range_end == LLONG_MAX) ?
657                         start + mapping->nrpages :
658                         wbc->range_end >> PAGE_CACHE_SHIFT;
659
660         if (start || end)
661                 expected_pages = end - start + 1;
662         else
663                 expected_pages = mapping->nrpages;
664
665         if (expected_pages < 32L)
666                 expected_pages = 32L;
667
668         EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
669                      "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
670                      mapping->host->i_ino, wbc->range_start, wbc->range_end,
671                      mapping->nrpages, start, end, expected_pages);
672
673         _pcol_init(&pcol, expected_pages, mapping->host);
674
675         ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
676         if (ret) {
677                 EXOFS_ERR("write_cache_pages => %d\n", ret);
678                 return ret;
679         }
680
681         return write_exec(&pcol);
682 }
683
684 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
685 {
686         struct page_collect pcol;
687         int ret;
688
689         _pcol_init(&pcol, 1, page->mapping->host);
690
691         ret = writepage_strip(page, NULL, &pcol);
692         if (ret) {
693                 EXOFS_ERR("exofs_writepage => %d\n", ret);
694                 return ret;
695         }
696
697         return write_exec(&pcol);
698 }
699
700 /* i_mutex held using inode->i_size directly */
701 static void _write_failed(struct inode *inode, loff_t to)
702 {
703         if (to > inode->i_size)
704                 truncate_pagecache(inode, to, inode->i_size);
705 }
706
707 int exofs_write_begin(struct file *file, struct address_space *mapping,
708                 loff_t pos, unsigned len, unsigned flags,
709                 struct page **pagep, void **fsdata)
710 {
711         int ret = 0;
712         struct page *page;
713
714         page = *pagep;
715         if (page == NULL) {
716                 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
717                                          fsdata);
718                 if (ret) {
719                         EXOFS_DBGMSG("simple_write_begin failed\n");
720                         goto out;
721                 }
722
723                 page = *pagep;
724         }
725
726          /* read modify write */
727         if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
728                 ret = _readpage(page, true);
729                 if (ret) {
730                         /*SetPageError was done by _readpage. Is it ok?*/
731                         unlock_page(page);
732                         EXOFS_DBGMSG("__readpage_filler failed\n");
733                 }
734         }
735 out:
736         if (unlikely(ret))
737                 _write_failed(mapping->host, pos + len);
738
739         return ret;
740 }
741
742 static int exofs_write_begin_export(struct file *file,
743                 struct address_space *mapping,
744                 loff_t pos, unsigned len, unsigned flags,
745                 struct page **pagep, void **fsdata)
746 {
747         *pagep = NULL;
748
749         return exofs_write_begin(file, mapping, pos, len, flags, pagep,
750                                         fsdata);
751 }
752
753 static int exofs_write_end(struct file *file, struct address_space *mapping,
754                         loff_t pos, unsigned len, unsigned copied,
755                         struct page *page, void *fsdata)
756 {
757         struct inode *inode = mapping->host;
758         /* According to comment in simple_write_end i_mutex is held */
759         loff_t i_size = inode->i_size;
760         int ret;
761
762         ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
763         if (unlikely(ret))
764                 _write_failed(inode, pos + len);
765
766         /* TODO: once simple_write_end marks inode dirty remove */
767         if (i_size != inode->i_size)
768                 mark_inode_dirty(inode);
769         return ret;
770 }
771
772 static int exofs_releasepage(struct page *page, gfp_t gfp)
773 {
774         EXOFS_DBGMSG("page 0x%lx\n", page->index);
775         WARN_ON(1);
776         return try_to_free_buffers(page);
777 }
778
779 static void exofs_invalidatepage(struct page *page, unsigned long offset)
780 {
781         EXOFS_DBGMSG("page_has_buffers=>%d\n", page_has_buffers(page));
782         WARN_ON(1);
783
784         block_invalidatepage(page, offset);
785 }
786
787 const struct address_space_operations exofs_aops = {
788         .readpage       = exofs_readpage,
789         .readpages      = exofs_readpages,
790         .writepage      = exofs_writepage,
791         .writepages     = exofs_writepages,
792         .write_begin    = exofs_write_begin_export,
793         .write_end      = exofs_write_end,
794         .releasepage    = exofs_releasepage,
795         .set_page_dirty = __set_page_dirty_nobuffers,
796         .invalidatepage = exofs_invalidatepage,
797
798         /* Not implemented Yet */
799         .bmap           = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
800         .direct_IO      = NULL, /* TODO: Should be trivial to do */
801
802         /* With these NULL has special meaning or default is not exported */
803         .sync_page      = NULL,
804         .get_xip_mem    = NULL,
805         .migratepage    = NULL,
806         .launder_page   = NULL,
807         .is_partially_uptodate = NULL,
808         .error_remove_page = NULL,
809 };
810
811 /******************************************************************************
812  * INODE OPERATIONS
813  *****************************************************************************/
814
815 /*
816  * Test whether an inode is a fast symlink.
817  */
818 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
819 {
820         struct exofs_i_info *oi = exofs_i(inode);
821
822         return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
823 }
824
825 const struct osd_attr g_attr_logical_length = ATTR_DEF(
826         OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
827
828 static int _do_truncate(struct inode *inode, loff_t newsize)
829 {
830         struct exofs_i_info *oi = exofs_i(inode);
831         int ret;
832
833         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
834
835         ret = exofs_oi_truncate(oi, (u64)newsize);
836         if (likely(!ret))
837                 truncate_setsize(inode, newsize);
838
839         EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
840                      inode->i_ino, newsize, ret);
841         return ret;
842 }
843
844 /*
845  * Set inode attributes - update size attribute on OSD if needed,
846  *                        otherwise just call generic functions.
847  */
848 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
849 {
850         struct inode *inode = dentry->d_inode;
851         int error;
852
853         /* if we are about to modify an object, and it hasn't been
854          * created yet, wait
855          */
856         error = wait_obj_created(exofs_i(inode));
857         if (unlikely(error))
858                 return error;
859
860         error = inode_change_ok(inode, iattr);
861         if (unlikely(error))
862                 return error;
863
864         if ((iattr->ia_valid & ATTR_SIZE) &&
865             iattr->ia_size != i_size_read(inode)) {
866                 error = _do_truncate(inode, iattr->ia_size);
867                 if (unlikely(error))
868                         return error;
869         }
870
871         setattr_copy(inode, iattr);
872         mark_inode_dirty(inode);
873         return 0;
874 }
875
876 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
877         EXOFS_APAGE_FS_DATA,
878         EXOFS_ATTR_INODE_FILE_LAYOUT,
879         0);
880 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
881         EXOFS_APAGE_FS_DATA,
882         EXOFS_ATTR_INODE_DIR_LAYOUT,
883         0);
884
885 /*
886  * Read the Linux inode info from the OSD, and return it as is. In exofs the
887  * inode info is in an application specific page/attribute of the osd-object.
888  */
889 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
890                     struct exofs_fcb *inode)
891 {
892         struct exofs_sb_info *sbi = sb->s_fs_info;
893         struct osd_attr attrs[] = {
894                 [0] = g_attr_inode_data,
895                 [1] = g_attr_inode_file_layout,
896                 [2] = g_attr_inode_dir_layout,
897         };
898         struct exofs_io_state *ios;
899         struct exofs_on_disk_inode_layout *layout;
900         int ret;
901
902         ret = exofs_get_io_state(&sbi->layout, &ios);
903         if (unlikely(ret)) {
904                 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
905                 return ret;
906         }
907
908         ios->obj.id = exofs_oi_objno(oi);
909         exofs_make_credential(oi->i_cred, &ios->obj);
910         ios->cred = oi->i_cred;
911
912         attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
913         attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
914
915         ios->in_attr = attrs;
916         ios->in_attr_len = ARRAY_SIZE(attrs);
917
918         ret = exofs_sbi_read(ios);
919         if (unlikely(ret)) {
920                 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
921                           _LLU(ios->obj.id), ret);
922                 memset(inode, 0, sizeof(*inode));
923                 inode->i_mode = 0040000 | (0777 & ~022);
924                 /* If object is lost on target we might as well enable it's
925                  * delete.
926                  */
927                 if ((ret == -ENOENT) || (ret == -EINVAL))
928                         ret = 0;
929                 goto out;
930         }
931
932         ret = extract_attr_from_ios(ios, &attrs[0]);
933         if (ret) {
934                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
935                 goto out;
936         }
937         WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
938         memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
939
940         ret = extract_attr_from_ios(ios, &attrs[1]);
941         if (ret) {
942                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
943                 goto out;
944         }
945         if (attrs[1].len) {
946                 layout = attrs[1].val_ptr;
947                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
948                         EXOFS_ERR("%s: unsupported files layout %d\n",
949                                 __func__, layout->gen_func);
950                         ret = -ENOTSUPP;
951                         goto out;
952                 }
953         }
954
955         ret = extract_attr_from_ios(ios, &attrs[2]);
956         if (ret) {
957                 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
958                 goto out;
959         }
960         if (attrs[2].len) {
961                 layout = attrs[2].val_ptr;
962                 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
963                         EXOFS_ERR("%s: unsupported meta-data layout %d\n",
964                                 __func__, layout->gen_func);
965                         ret = -ENOTSUPP;
966                         goto out;
967                 }
968         }
969
970 out:
971         exofs_put_io_state(ios);
972         return ret;
973 }
974
975 static void __oi_init(struct exofs_i_info *oi)
976 {
977         init_waitqueue_head(&oi->i_wq);
978         oi->i_flags = 0;
979 }
980 /*
981  * Fill in an inode read from the OSD and set it up for use
982  */
983 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
984 {
985         struct exofs_i_info *oi;
986         struct exofs_fcb fcb;
987         struct inode *inode;
988         int ret;
989
990         inode = iget_locked(sb, ino);
991         if (!inode)
992                 return ERR_PTR(-ENOMEM);
993         if (!(inode->i_state & I_NEW))
994                 return inode;
995         oi = exofs_i(inode);
996         __oi_init(oi);
997
998         /* read the inode from the osd */
999         ret = exofs_get_inode(sb, oi, &fcb);
1000         if (ret)
1001                 goto bad_inode;
1002
1003         set_obj_created(oi);
1004
1005         /* copy stuff from on-disk struct to in-memory struct */
1006         inode->i_mode = le16_to_cpu(fcb.i_mode);
1007         inode->i_uid = le32_to_cpu(fcb.i_uid);
1008         inode->i_gid = le32_to_cpu(fcb.i_gid);
1009         inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1010         inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1011         inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1012         inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1013         inode->i_ctime.tv_nsec =
1014                 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1015         oi->i_commit_size = le64_to_cpu(fcb.i_size);
1016         i_size_write(inode, oi->i_commit_size);
1017         inode->i_blkbits = EXOFS_BLKSHIFT;
1018         inode->i_generation = le32_to_cpu(fcb.i_generation);
1019
1020         oi->i_dir_start_lookup = 0;
1021
1022         if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1023                 ret = -ESTALE;
1024                 goto bad_inode;
1025         }
1026
1027         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1028                 if (fcb.i_data[0])
1029                         inode->i_rdev =
1030                                 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1031                 else
1032                         inode->i_rdev =
1033                                 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1034         } else {
1035                 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1036         }
1037
1038         if (S_ISREG(inode->i_mode)) {
1039                 inode->i_op = &exofs_file_inode_operations;
1040                 inode->i_fop = &exofs_file_operations;
1041                 inode->i_mapping->a_ops = &exofs_aops;
1042         } else if (S_ISDIR(inode->i_mode)) {
1043                 inode->i_op = &exofs_dir_inode_operations;
1044                 inode->i_fop = &exofs_dir_operations;
1045                 inode->i_mapping->a_ops = &exofs_aops;
1046         } else if (S_ISLNK(inode->i_mode)) {
1047                 if (exofs_inode_is_fast_symlink(inode))
1048                         inode->i_op = &exofs_fast_symlink_inode_operations;
1049                 else {
1050                         inode->i_op = &exofs_symlink_inode_operations;
1051                         inode->i_mapping->a_ops = &exofs_aops;
1052                 }
1053         } else {
1054                 inode->i_op = &exofs_special_inode_operations;
1055                 if (fcb.i_data[0])
1056                         init_special_inode(inode, inode->i_mode,
1057                            old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1058                 else
1059                         init_special_inode(inode, inode->i_mode,
1060                            new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1061         }
1062
1063         unlock_new_inode(inode);
1064         return inode;
1065
1066 bad_inode:
1067         iget_failed(inode);
1068         return ERR_PTR(ret);
1069 }
1070
1071 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1072 {
1073         if (!obj_created(oi)) {
1074                 BUG_ON(!obj_2bcreated(oi));
1075                 wait_event(oi->i_wq, obj_created(oi));
1076         }
1077         return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1078 }
1079 /*
1080  * Callback function from exofs_new_inode().  The important thing is that we
1081  * set the obj_created flag so that other methods know that the object exists on
1082  * the OSD.
1083  */
1084 static void create_done(struct exofs_io_state *ios, void *p)
1085 {
1086         struct inode *inode = p;
1087         struct exofs_i_info *oi = exofs_i(inode);
1088         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1089         int ret;
1090
1091         ret = exofs_check_io(ios, NULL);
1092         exofs_put_io_state(ios);
1093
1094         atomic_dec(&sbi->s_curr_pending);
1095
1096         if (unlikely(ret)) {
1097                 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1098                           _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
1099                 /*TODO: When FS is corrupted creation can fail, object already
1100                  * exist. Get rid of this asynchronous creation, if exist
1101                  * increment the obj counter and try the next object. Until we
1102                  * succeed. All these dangling objects will be made into lost
1103                  * files by chkfs.exofs
1104                  */
1105         }
1106
1107         set_obj_created(oi);
1108
1109         atomic_dec(&inode->i_count);
1110         wake_up(&oi->i_wq);
1111 }
1112
1113 /*
1114  * Set up a new inode and create an object for it on the OSD
1115  */
1116 struct inode *exofs_new_inode(struct inode *dir, int mode)
1117 {
1118         struct super_block *sb;
1119         struct inode *inode;
1120         struct exofs_i_info *oi;
1121         struct exofs_sb_info *sbi;
1122         struct exofs_io_state *ios;
1123         int ret;
1124
1125         sb = dir->i_sb;
1126         inode = new_inode(sb);
1127         if (!inode)
1128                 return ERR_PTR(-ENOMEM);
1129
1130         oi = exofs_i(inode);
1131         __oi_init(oi);
1132
1133         set_obj_2bcreated(oi);
1134
1135         sbi = sb->s_fs_info;
1136
1137         sb->s_dirt = 1;
1138         inode_init_owner(inode, dir, mode);
1139         inode->i_ino = sbi->s_nextid++;
1140         inode->i_blkbits = EXOFS_BLKSHIFT;
1141         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1142         oi->i_commit_size = inode->i_size = 0;
1143         spin_lock(&sbi->s_next_gen_lock);
1144         inode->i_generation = sbi->s_next_generation++;
1145         spin_unlock(&sbi->s_next_gen_lock);
1146         insert_inode_hash(inode);
1147
1148         mark_inode_dirty(inode);
1149
1150         ret = exofs_get_io_state(&sbi->layout, &ios);
1151         if (unlikely(ret)) {
1152                 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1153                 return ERR_PTR(ret);
1154         }
1155
1156         ios->obj.id = exofs_oi_objno(oi);
1157         exofs_make_credential(oi->i_cred, &ios->obj);
1158
1159         /* increment the refcount so that the inode will still be around when we
1160          * reach the callback
1161          */
1162         atomic_inc(&inode->i_count);
1163
1164         ios->done = create_done;
1165         ios->private = inode;
1166         ios->cred = oi->i_cred;
1167         ret = exofs_sbi_create(ios);
1168         if (ret) {
1169                 atomic_dec(&inode->i_count);
1170                 exofs_put_io_state(ios);
1171                 return ERR_PTR(ret);
1172         }
1173         atomic_inc(&sbi->s_curr_pending);
1174
1175         return inode;
1176 }
1177
1178 /*
1179  * struct to pass two arguments to update_inode's callback
1180  */
1181 struct updatei_args {
1182         struct exofs_sb_info    *sbi;
1183         struct exofs_fcb        fcb;
1184 };
1185
1186 /*
1187  * Callback function from exofs_update_inode().
1188  */
1189 static void updatei_done(struct exofs_io_state *ios, void *p)
1190 {
1191         struct updatei_args *args = p;
1192
1193         exofs_put_io_state(ios);
1194
1195         atomic_dec(&args->sbi->s_curr_pending);
1196
1197         kfree(args);
1198 }
1199
1200 /*
1201  * Write the inode to the OSD.  Just fill up the struct, and set the attribute
1202  * synchronously or asynchronously depending on the do_sync flag.
1203  */
1204 static int exofs_update_inode(struct inode *inode, int do_sync)
1205 {
1206         struct exofs_i_info *oi = exofs_i(inode);
1207         struct super_block *sb = inode->i_sb;
1208         struct exofs_sb_info *sbi = sb->s_fs_info;
1209         struct exofs_io_state *ios;
1210         struct osd_attr attr;
1211         struct exofs_fcb *fcb;
1212         struct updatei_args *args;
1213         int ret;
1214
1215         args = kzalloc(sizeof(*args), GFP_KERNEL);
1216         if (!args) {
1217                 EXOFS_DBGMSG("Failed kzalloc of args\n");
1218                 return -ENOMEM;
1219         }
1220
1221         fcb = &args->fcb;
1222
1223         fcb->i_mode = cpu_to_le16(inode->i_mode);
1224         fcb->i_uid = cpu_to_le32(inode->i_uid);
1225         fcb->i_gid = cpu_to_le32(inode->i_gid);
1226         fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1227         fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1228         fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1229         fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1230         oi->i_commit_size = i_size_read(inode);
1231         fcb->i_size = cpu_to_le64(oi->i_commit_size);
1232         fcb->i_generation = cpu_to_le32(inode->i_generation);
1233
1234         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1235                 if (old_valid_dev(inode->i_rdev)) {
1236                         fcb->i_data[0] =
1237                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1238                         fcb->i_data[1] = 0;
1239                 } else {
1240                         fcb->i_data[0] = 0;
1241                         fcb->i_data[1] =
1242                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1243                         fcb->i_data[2] = 0;
1244                 }
1245         } else
1246                 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1247
1248         ret = exofs_get_io_state(&sbi->layout, &ios);
1249         if (unlikely(ret)) {
1250                 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
1251                 goto free_args;
1252         }
1253
1254         attr = g_attr_inode_data;
1255         attr.val_ptr = fcb;
1256         ios->out_attr_len = 1;
1257         ios->out_attr = &attr;
1258
1259         if (!obj_created(oi)) {
1260                 EXOFS_DBGMSG("!obj_created\n");
1261                 BUG_ON(!obj_2bcreated(oi));
1262                 wait_event(oi->i_wq, obj_created(oi));
1263                 EXOFS_DBGMSG("wait_event done\n");
1264         }
1265
1266         if (!do_sync) {
1267                 args->sbi = sbi;
1268                 ios->done = updatei_done;
1269                 ios->private = args;
1270         }
1271
1272         ret = exofs_oi_write(oi, ios);
1273         if (!do_sync && !ret) {
1274                 atomic_inc(&sbi->s_curr_pending);
1275                 goto out; /* deallocation in updatei_done */
1276         }
1277
1278         exofs_put_io_state(ios);
1279 free_args:
1280         kfree(args);
1281 out:
1282         EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1283                      inode->i_ino, do_sync, ret);
1284         return ret;
1285 }
1286
1287 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1288 {
1289         return exofs_update_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1290 }
1291
1292 /*
1293  * Callback function from exofs_delete_inode() - don't have much cleaning up to
1294  * do.
1295  */
1296 static void delete_done(struct exofs_io_state *ios, void *p)
1297 {
1298         struct exofs_sb_info *sbi = p;
1299
1300         exofs_put_io_state(ios);
1301
1302         atomic_dec(&sbi->s_curr_pending);
1303 }
1304
1305 /*
1306  * Called when the refcount of an inode reaches zero.  We remove the object
1307  * from the OSD here.  We make sure the object was created before we try and
1308  * delete it.
1309  */
1310 void exofs_evict_inode(struct inode *inode)
1311 {
1312         struct exofs_i_info *oi = exofs_i(inode);
1313         struct super_block *sb = inode->i_sb;
1314         struct exofs_sb_info *sbi = sb->s_fs_info;
1315         struct exofs_io_state *ios;
1316         int ret;
1317
1318         truncate_inode_pages(&inode->i_data, 0);
1319
1320         /* TODO: should do better here */
1321         if (inode->i_nlink || is_bad_inode(inode))
1322                 goto no_delete;
1323
1324         inode->i_size = 0;
1325         end_writeback(inode);
1326
1327         /* if we are deleting an obj that hasn't been created yet, wait */
1328         if (!obj_created(oi)) {
1329                 BUG_ON(!obj_2bcreated(oi));
1330                 wait_event(oi->i_wq, obj_created(oi));
1331                 /* ignore the error attempt a remove anyway */
1332         }
1333
1334         /* Now Remove the OSD objects */
1335         ret = exofs_get_io_state(&sbi->layout, &ios);
1336         if (unlikely(ret)) {
1337                 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
1338                 return;
1339         }
1340
1341         ios->obj.id = exofs_oi_objno(oi);
1342         ios->done = delete_done;
1343         ios->private = sbi;
1344         ios->cred = oi->i_cred;
1345         ret = exofs_sbi_remove(ios);
1346         if (ret) {
1347                 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__);
1348                 exofs_put_io_state(ios);
1349                 return;
1350         }
1351         atomic_inc(&sbi->s_curr_pending);
1352
1353         return;
1354
1355 no_delete:
1356         end_writeback(inode);
1357 }