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