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