Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / fs / exofs / inode.c
1 /*
2  * Copyright (C) 2005, 2006
3  * Avishay Traeger (avishay@gmail.com) (avishay@il.ibm.com)
4  * Copyright (C) 2005, 2006
5  * International Business Machines
6  * Copyright (C) 2008, 2009
7  * Boaz Harrosh <bharrosh@panasas.com>
8  *
9  * Copyrights for code taken from ext2:
10  *     Copyright (C) 1992, 1993, 1994, 1995
11  *     Remy Card (card@masi.ibp.fr)
12  *     Laboratoire MASI - Institut Blaise Pascal
13  *     Universite Pierre et Marie Curie (Paris VI)
14  *     from
15  *     linux/fs/minix/inode.c
16  *     Copyright (C) 1991, 1992  Linus Torvalds
17  *
18  * This file is part of exofs.
19  *
20  * exofs is free software; you can redistribute it and/or modify
21  * it under the terms of the GNU General Public License as published by
22  * the Free Software Foundation.  Since it is based on ext2, and the only
23  * valid version of GPL for the Linux kernel is version 2, the only valid
24  * version of GPL for exofs is version 2.
25  *
26  * exofs is distributed in the hope that it will be useful,
27  * but WITHOUT ANY WARRANTY; without even the implied warranty of
28  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29  * GNU General Public License for more details.
30  *
31  * You should have received a copy of the GNU General Public License
32  * along with exofs; if not, write to the Free Software
33  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
34  */
35
36 #include <linux/writeback.h>
37 #include <linux/buffer_head.h>
38 #include <scsi/scsi_device.h>
39
40 #include "exofs.h"
41
42 #ifdef CONFIG_EXOFS_DEBUG
43 #  define EXOFS_DEBUG_OBJ_ISIZE 1
44 #endif
45
46 struct page_collect {
47         struct exofs_sb_info *sbi;
48         struct request_queue *req_q;
49         struct inode *inode;
50         unsigned expected_pages;
51
52         struct bio *bio;
53         unsigned nr_pages;
54         unsigned long length;
55         loff_t pg_first; /* keep 64bit also in 32-arches */
56 };
57
58 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
59                 struct inode *inode)
60 {
61         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
62         struct request_queue *req_q = sbi->s_dev->scsi_device->request_queue;
63
64         pcol->sbi = sbi;
65         pcol->req_q = req_q;
66         pcol->inode = inode;
67         pcol->expected_pages = expected_pages;
68
69         pcol->bio = NULL;
70         pcol->nr_pages = 0;
71         pcol->length = 0;
72         pcol->pg_first = -1;
73
74         EXOFS_DBGMSG("_pcol_init ino=0x%lx expected_pages=%u\n", inode->i_ino,
75                      expected_pages);
76 }
77
78 static void _pcol_reset(struct page_collect *pcol)
79 {
80         pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
81
82         pcol->bio = NULL;
83         pcol->nr_pages = 0;
84         pcol->length = 0;
85         pcol->pg_first = -1;
86         EXOFS_DBGMSG("_pcol_reset ino=0x%lx expected_pages=%u\n",
87                      pcol->inode->i_ino, pcol->expected_pages);
88
89         /* this is probably the end of the loop but in writes
90          * it might not end here. don't be left with nothing
91          */
92         if (!pcol->expected_pages)
93                 pcol->expected_pages = 128;
94 }
95
96 static int pcol_try_alloc(struct page_collect *pcol)
97 {
98         int pages = min_t(unsigned, pcol->expected_pages, BIO_MAX_PAGES);
99
100         for (; pages; pages >>= 1) {
101                 pcol->bio = bio_alloc(GFP_KERNEL, pages);
102                 if (likely(pcol->bio))
103                         return 0;
104         }
105
106         EXOFS_ERR("Failed to kcalloc expected_pages=%u\n",
107                   pcol->expected_pages);
108         return -ENOMEM;
109 }
110
111 static void pcol_free(struct page_collect *pcol)
112 {
113         bio_put(pcol->bio);
114         pcol->bio = NULL;
115 }
116
117 static int pcol_add_page(struct page_collect *pcol, struct page *page,
118                          unsigned len)
119 {
120         int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0);
121         if (unlikely(len != added_len))
122                 return -ENOMEM;
123
124         ++pcol->nr_pages;
125         pcol->length += len;
126         return 0;
127 }
128
129 static int update_read_page(struct page *page, int ret)
130 {
131         if (ret == 0) {
132                 /* Everything is OK */
133                 SetPageUptodate(page);
134                 if (PageError(page))
135                         ClearPageError(page);
136         } else if (ret == -EFAULT) {
137                 /* In this case we were trying to read something that wasn't on
138                  * disk yet - return a page full of zeroes.  This should be OK,
139                  * because the object should be empty (if there was a write
140                  * before this read, the read would be waiting with the page
141                  * locked */
142                 clear_highpage(page);
143
144                 SetPageUptodate(page);
145                 if (PageError(page))
146                         ClearPageError(page);
147                 ret = 0; /* recovered error */
148                 EXOFS_DBGMSG("recovered read error\n");
149         } else /* Error */
150                 SetPageError(page);
151
152         return ret;
153 }
154
155 static void update_write_page(struct page *page, int ret)
156 {
157         if (ret) {
158                 mapping_set_error(page->mapping, ret);
159                 SetPageError(page);
160         }
161         end_page_writeback(page);
162 }
163
164 /* Called at the end of reads, to optionally unlock pages and update their
165  * status.
166  */
167 static int __readpages_done(struct osd_request *or, struct page_collect *pcol,
168                             bool do_unlock)
169 {
170         struct bio_vec *bvec;
171         int i;
172         u64 resid;
173         u64 good_bytes;
174         u64 length = 0;
175         int ret = exofs_check_ok_resid(or, &resid, NULL);
176
177         osd_end_request(or);
178
179         if (likely(!ret))
180                 good_bytes = pcol->length;
181         else if (!resid)
182                 good_bytes = 0;
183         else
184                 good_bytes = pcol->length - resid;
185
186         EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx"
187                      " length=0x%lx nr_pages=%u\n",
188                      pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
189                      pcol->nr_pages);
190
191         __bio_for_each_segment(bvec, pcol->bio, i, 0) {
192                 struct page *page = bvec->bv_page;
193                 struct inode *inode = page->mapping->host;
194                 int page_stat;
195
196                 if (inode != pcol->inode)
197                         continue; /* osd might add more pages at end */
198
199                 if (likely(length < good_bytes))
200                         page_stat = 0;
201                 else
202                         page_stat = ret;
203
204                 EXOFS_DBGMSG("    readpages_done(0x%lx, 0x%lx) %s\n",
205                           inode->i_ino, page->index,
206                           page_stat ? "bad_bytes" : "good_bytes");
207
208                 ret = update_read_page(page, page_stat);
209                 if (do_unlock)
210                         unlock_page(page);
211                 length += bvec->bv_len;
212         }
213
214         pcol_free(pcol);
215         EXOFS_DBGMSG("readpages_done END\n");
216         return ret;
217 }
218
219 /* callback of async reads */
220 static void readpages_done(struct osd_request *or, void *p)
221 {
222         struct page_collect *pcol = p;
223
224         __readpages_done(or, pcol, true);
225         atomic_dec(&pcol->sbi->s_curr_pending);
226         kfree(p);
227 }
228
229 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
230 {
231         struct bio_vec *bvec;
232         int i;
233
234         __bio_for_each_segment(bvec, pcol->bio, i, 0) {
235                 struct page *page = bvec->bv_page;
236
237                 if (rw == READ)
238                         update_read_page(page, ret);
239                 else
240                         update_write_page(page, ret);
241
242                 unlock_page(page);
243         }
244         pcol_free(pcol);
245 }
246
247 static int read_exec(struct page_collect *pcol, bool is_sync)
248 {
249         struct exofs_i_info *oi = exofs_i(pcol->inode);
250         struct osd_obj_id obj = {pcol->sbi->s_pid,
251                                         pcol->inode->i_ino + EXOFS_OBJ_OFF};
252         struct osd_request *or = NULL;
253         struct page_collect *pcol_copy = NULL;
254         loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
255         int ret;
256
257         if (!pcol->bio)
258                 return 0;
259
260         /* see comment in _readpage() about sync reads */
261         WARN_ON(is_sync && (pcol->nr_pages != 1));
262
263         or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
264         if (unlikely(!or)) {
265                 ret = -ENOMEM;
266                 goto err;
267         }
268
269         osd_req_read(or, &obj, pcol->bio, i_start);
270
271         if (is_sync) {
272                 exofs_sync_op(or, pcol->sbi->s_timeout, oi->i_cred);
273                 return __readpages_done(or, pcol, false);
274         }
275
276         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
277         if (!pcol_copy) {
278                 ret = -ENOMEM;
279                 goto err;
280         }
281
282         *pcol_copy = *pcol;
283         ret = exofs_async_op(or, readpages_done, pcol_copy, oi->i_cred);
284         if (unlikely(ret))
285                 goto err;
286
287         atomic_inc(&pcol->sbi->s_curr_pending);
288
289         EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
290                   obj.id, _LLU(i_start), pcol->length);
291
292         /* pages ownership was passed to pcol_copy */
293         _pcol_reset(pcol);
294         return 0;
295
296 err:
297         if (!is_sync)
298                 _unlock_pcol_pages(pcol, ret, READ);
299         kfree(pcol_copy);
300         if (or)
301                 osd_end_request(or);
302         return ret;
303 }
304
305 /* readpage_strip is called either directly from readpage() or by the VFS from
306  * within read_cache_pages(), to add one more page to be read. It will try to
307  * collect as many contiguous pages as posible. If a discontinuity is
308  * encountered, or it runs out of resources, it will submit the previous segment
309  * and will start a new collection. Eventually caller must submit the last
310  * segment if present.
311  */
312 static int readpage_strip(void *data, struct page *page)
313 {
314         struct page_collect *pcol = data;
315         struct inode *inode = pcol->inode;
316         struct exofs_i_info *oi = exofs_i(inode);
317         loff_t i_size = i_size_read(inode);
318         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
319         size_t len;
320         int ret;
321
322         /* FIXME: Just for debugging, will be removed */
323         if (PageUptodate(page))
324                 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
325                           page->index);
326
327         if (page->index < end_index)
328                 len = PAGE_CACHE_SIZE;
329         else if (page->index == end_index)
330                 len = i_size & ~PAGE_CACHE_MASK;
331         else
332                 len = 0;
333
334         if (!len || !obj_created(oi)) {
335                 /* this will be out of bounds, or doesn't exist yet.
336                  * Current page is cleared and the request is split
337                  */
338                 clear_highpage(page);
339
340                 SetPageUptodate(page);
341                 if (PageError(page))
342                         ClearPageError(page);
343
344                 unlock_page(page);
345                 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
346                              " splitting\n", inode->i_ino, page->index);
347
348                 return read_exec(pcol, false);
349         }
350
351 try_again:
352
353         if (unlikely(pcol->pg_first == -1)) {
354                 pcol->pg_first = page->index;
355         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
356                    page->index)) {
357                 /* Discontinuity detected, split the request */
358                 ret = read_exec(pcol, false);
359                 if (unlikely(ret))
360                         goto fail;
361                 goto try_again;
362         }
363
364         if (!pcol->bio) {
365                 ret = pcol_try_alloc(pcol);
366                 if (unlikely(ret))
367                         goto fail;
368         }
369
370         if (len != PAGE_CACHE_SIZE)
371                 zero_user(page, len, PAGE_CACHE_SIZE - len);
372
373         EXOFS_DBGMSG("    readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
374                      inode->i_ino, page->index, len);
375
376         ret = pcol_add_page(pcol, page, len);
377         if (ret) {
378                 EXOFS_DBGMSG("Failed pcol_add_page pages[i]=%p "
379                           "this_len=0x%zx nr_pages=%u length=0x%lx\n",
380                           page, len, pcol->nr_pages, pcol->length);
381
382                 /* split the request, and start again with current page */
383                 ret = read_exec(pcol, false);
384                 if (unlikely(ret))
385                         goto fail;
386
387                 goto try_again;
388         }
389
390         return 0;
391
392 fail:
393         /* SetPageError(page); ??? */
394         unlock_page(page);
395         return ret;
396 }
397
398 static int exofs_readpages(struct file *file, struct address_space *mapping,
399                            struct list_head *pages, unsigned nr_pages)
400 {
401         struct page_collect pcol;
402         int ret;
403
404         _pcol_init(&pcol, nr_pages, mapping->host);
405
406         ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
407         if (ret) {
408                 EXOFS_ERR("read_cache_pages => %d\n", ret);
409                 return ret;
410         }
411
412         return read_exec(&pcol, false);
413 }
414
415 static int _readpage(struct page *page, bool is_sync)
416 {
417         struct page_collect pcol;
418         int ret;
419
420         _pcol_init(&pcol, 1, page->mapping->host);
421
422         /* readpage_strip might call read_exec(,async) inside at several places
423          * but this is safe for is_async=0 since read_exec will not do anything
424          * when we have a single page.
425          */
426         ret = readpage_strip(&pcol, page);
427         if (ret) {
428                 EXOFS_ERR("_readpage => %d\n", ret);
429                 return ret;
430         }
431
432         return read_exec(&pcol, is_sync);
433 }
434
435 /*
436  * We don't need the file
437  */
438 static int exofs_readpage(struct file *file, struct page *page)
439 {
440         return _readpage(page, false);
441 }
442
443 /* Callback for osd_write. All writes are asynchronouse */
444 static void writepages_done(struct osd_request *or, void *p)
445 {
446         struct page_collect *pcol = p;
447         struct bio_vec *bvec;
448         int i;
449         u64 resid;
450         u64  good_bytes;
451         u64  length = 0;
452
453         int ret = exofs_check_ok_resid(or, NULL, &resid);
454
455         osd_end_request(or);
456         atomic_dec(&pcol->sbi->s_curr_pending);
457
458         if (likely(!ret))
459                 good_bytes = pcol->length;
460         else if (!resid)
461                 good_bytes = 0;
462         else
463                 good_bytes = pcol->length - resid;
464
465         EXOFS_DBGMSG("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         __bio_for_each_segment(bvec, pcol->bio, i, 0) {
471                 struct page *page = bvec->bv_page;
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_DBGMSG("    writepages_done(0x%lx, 0x%lx) status=%d\n",
486                              inode->i_ino, page->index, page_stat);
487
488                 length += bvec->bv_len;
489         }
490
491         pcol_free(pcol);
492         kfree(pcol);
493         EXOFS_DBGMSG("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 osd_obj_id obj = {pcol->sbi->s_pid,
500                                         pcol->inode->i_ino + EXOFS_OBJ_OFF};
501         struct osd_request *or = NULL;
502         struct page_collect *pcol_copy = NULL;
503         loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
504         int ret;
505
506         if (!pcol->bio)
507                 return 0;
508
509         or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
510         if (unlikely(!or)) {
511                 EXOFS_ERR("write_exec: Faild to osd_start_request()\n");
512                 ret = -ENOMEM;
513                 goto err;
514         }
515
516         pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
517         if (!pcol_copy) {
518                 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
519                 ret = -ENOMEM;
520                 goto err;
521         }
522
523         *pcol_copy = *pcol;
524
525         osd_req_write(or, &obj, pcol_copy->bio, i_start);
526         ret = exofs_async_op(or, writepages_done, pcol_copy, oi->i_cred);
527         if (unlikely(ret)) {
528                 EXOFS_ERR("write_exec: exofs_async_op() Faild\n");
529                 goto err;
530         }
531
532         atomic_inc(&pcol->sbi->s_curr_pending);
533         EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
534                   pcol->inode->i_ino, pcol->pg_first, _LLU(i_start),
535                   pcol->length);
536         /* pages ownership was passed to pcol_copy */
537         _pcol_reset(pcol);
538         return 0;
539
540 err:
541         _unlock_pcol_pages(pcol, ret, WRITE);
542         kfree(pcol_copy);
543         if (or)
544                 osd_end_request(or);
545         return ret;
546 }
547
548 /* writepage_strip is called either directly from writepage() or by the VFS from
549  * within write_cache_pages(), to add one more page to be written to storage.
550  * It will try to collect as many contiguous pages as possible. If a
551  * discontinuity is encountered or it runs out of resources it will submit the
552  * previous segment and will start a new collection.
553  * Eventually caller must submit the last segment if present.
554  */
555 static int writepage_strip(struct page *page,
556                            struct writeback_control *wbc_unused, void *data)
557 {
558         struct page_collect *pcol = data;
559         struct inode *inode = pcol->inode;
560         struct exofs_i_info *oi = exofs_i(inode);
561         loff_t i_size = i_size_read(inode);
562         pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
563         size_t len;
564         int ret;
565
566         BUG_ON(!PageLocked(page));
567
568         ret = wait_obj_created(oi);
569         if (unlikely(ret))
570                 goto fail;
571
572         if (page->index < end_index)
573                 /* in this case, the page is within the limits of the file */
574                 len = PAGE_CACHE_SIZE;
575         else {
576                 len = i_size & ~PAGE_CACHE_MASK;
577
578                 if (page->index > end_index || !len) {
579                         /* in this case, the page is outside the limits
580                          * (truncate in progress)
581                          */
582                         ret = write_exec(pcol);
583                         if (unlikely(ret))
584                                 goto fail;
585                         if (PageError(page))
586                                 ClearPageError(page);
587                         unlock_page(page);
588                         return 0;
589                 }
590         }
591
592 try_again:
593
594         if (unlikely(pcol->pg_first == -1)) {
595                 pcol->pg_first = page->index;
596         } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
597                    page->index)) {
598                 /* Discontinuity detected, split the request */
599                 ret = write_exec(pcol);
600                 if (unlikely(ret))
601                         goto fail;
602                 goto try_again;
603         }
604
605         if (!pcol->bio) {
606                 ret = pcol_try_alloc(pcol);
607                 if (unlikely(ret))
608                         goto fail;
609         }
610
611         EXOFS_DBGMSG("    writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
612                      inode->i_ino, page->index, len);
613
614         ret = pcol_add_page(pcol, page, len);
615         if (unlikely(ret)) {
616                 EXOFS_DBGMSG("Failed pcol_add_page "
617                              "nr_pages=%u total_length=0x%lx\n",
618                              pcol->nr_pages, pcol->length);
619
620                 /* split the request, next loop will start again */
621                 ret = write_exec(pcol);
622                 if (unlikely(ret)) {
623                         EXOFS_DBGMSG("write_exec faild => %d", ret);
624                         goto fail;
625                 }
626
627                 goto try_again;
628         }
629
630         BUG_ON(PageWriteback(page));
631         set_page_writeback(page);
632
633         return 0;
634
635 fail:
636         set_bit(AS_EIO, &page->mapping->flags);
637         unlock_page(page);
638         return ret;
639 }
640
641 static int exofs_writepages(struct address_space *mapping,
642                        struct writeback_control *wbc)
643 {
644         struct page_collect pcol;
645         long start, end, expected_pages;
646         int ret;
647
648         start = wbc->range_start >> PAGE_CACHE_SHIFT;
649         end = (wbc->range_end == LLONG_MAX) ?
650                         start + mapping->nrpages :
651                         wbc->range_end >> PAGE_CACHE_SHIFT;
652
653         if (start || end)
654                 expected_pages = min(end - start + 1, 32L);
655         else
656                 expected_pages = mapping->nrpages;
657
658         EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx"
659                      " m->nrpages=%lu start=0x%lx end=0x%lx\n",
660                      mapping->host->i_ino, wbc->range_start, wbc->range_end,
661                      mapping->nrpages, start, end);
662
663         _pcol_init(&pcol, expected_pages, mapping->host);
664
665         ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
666         if (ret) {
667                 EXOFS_ERR("write_cache_pages => %d\n", ret);
668                 return ret;
669         }
670
671         return write_exec(&pcol);
672 }
673
674 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
675 {
676         struct page_collect pcol;
677         int ret;
678
679         _pcol_init(&pcol, 1, page->mapping->host);
680
681         ret = writepage_strip(page, NULL, &pcol);
682         if (ret) {
683                 EXOFS_ERR("exofs_writepage => %d\n", ret);
684                 return ret;
685         }
686
687         return write_exec(&pcol);
688 }
689
690 int exofs_write_begin(struct file *file, struct address_space *mapping,
691                 loff_t pos, unsigned len, unsigned flags,
692                 struct page **pagep, void **fsdata)
693 {
694         int ret = 0;
695         struct page *page;
696
697         page = *pagep;
698         if (page == NULL) {
699                 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
700                                          fsdata);
701                 if (ret) {
702                         EXOFS_DBGMSG("simple_write_begin faild\n");
703                         return ret;
704                 }
705
706                 page = *pagep;
707         }
708
709          /* read modify write */
710         if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
711                 ret = _readpage(page, true);
712                 if (ret) {
713                         /*SetPageError was done by _readpage. Is it ok?*/
714                         unlock_page(page);
715                         EXOFS_DBGMSG("__readpage_filler faild\n");
716                 }
717         }
718
719         return ret;
720 }
721
722 static int exofs_write_begin_export(struct file *file,
723                 struct address_space *mapping,
724                 loff_t pos, unsigned len, unsigned flags,
725                 struct page **pagep, void **fsdata)
726 {
727         *pagep = NULL;
728
729         return exofs_write_begin(file, mapping, pos, len, flags, pagep,
730                                         fsdata);
731 }
732
733 const struct address_space_operations exofs_aops = {
734         .readpage       = exofs_readpage,
735         .readpages      = exofs_readpages,
736         .writepage      = exofs_writepage,
737         .writepages     = exofs_writepages,
738         .write_begin    = exofs_write_begin_export,
739         .write_end      = simple_write_end,
740 };
741
742 /******************************************************************************
743  * INODE OPERATIONS
744  *****************************************************************************/
745
746 /*
747  * Test whether an inode is a fast symlink.
748  */
749 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
750 {
751         struct exofs_i_info *oi = exofs_i(inode);
752
753         return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
754 }
755
756 /*
757  * get_block_t - Fill in a buffer_head
758  * An OSD takes care of block allocation so we just fake an allocation by
759  * putting in the inode's sector_t in the buffer_head.
760  * TODO: What about the case of create==0 and @iblock does not exist in the
761  * object?
762  */
763 static int exofs_get_block(struct inode *inode, sector_t iblock,
764                     struct buffer_head *bh_result, int create)
765 {
766         map_bh(bh_result, inode->i_sb, iblock);
767         return 0;
768 }
769
770 const struct osd_attr g_attr_logical_length = ATTR_DEF(
771         OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
772
773 /*
774  * Truncate a file to the specified size - all we have to do is set the size
775  * attribute.  We make sure the object exists first.
776  */
777 void exofs_truncate(struct inode *inode)
778 {
779         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
780         struct exofs_i_info *oi = exofs_i(inode);
781         struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
782         struct osd_request *or;
783         struct osd_attr attr;
784         loff_t isize = i_size_read(inode);
785         __be64 newsize;
786         int ret;
787
788         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
789              || S_ISLNK(inode->i_mode)))
790                 return;
791         if (exofs_inode_is_fast_symlink(inode))
792                 return;
793         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
794                 return;
795         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
796
797         nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
798
799         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
800         if (unlikely(!or)) {
801                 EXOFS_ERR("ERROR: exofs_truncate: osd_start_request failed\n");
802                 goto fail;
803         }
804
805         osd_req_set_attributes(or, &obj);
806
807         newsize = cpu_to_be64((u64)isize);
808         attr = g_attr_logical_length;
809         attr.val_ptr = &newsize;
810         osd_req_add_set_attr_list(or, &attr, 1);
811
812         /* if we are about to truncate an object, and it hasn't been
813          * created yet, wait
814          */
815         if (unlikely(wait_obj_created(oi)))
816                 goto fail;
817
818         ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
819         osd_end_request(or);
820         if (ret)
821                 goto fail;
822
823 out:
824         mark_inode_dirty(inode);
825         return;
826 fail:
827         make_bad_inode(inode);
828         goto out;
829 }
830
831 /*
832  * Set inode attributes - just call generic functions.
833  */
834 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
835 {
836         struct inode *inode = dentry->d_inode;
837         int error;
838
839         error = inode_change_ok(inode, iattr);
840         if (error)
841                 return error;
842
843         error = inode_setattr(inode, iattr);
844         return error;
845 }
846
847 /*
848  * Read an inode from the OSD, and return it as is.  We also return the size
849  * attribute in the 'sanity' argument if we got compiled with debugging turned
850  * on.
851  */
852 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
853                     struct exofs_fcb *inode, uint64_t *sanity)
854 {
855         struct exofs_sb_info *sbi = sb->s_fs_info;
856         struct osd_request *or;
857         struct osd_attr attr;
858         struct osd_obj_id obj = {sbi->s_pid,
859                                  oi->vfs_inode.i_ino + EXOFS_OBJ_OFF};
860         int ret;
861
862         exofs_make_credential(oi->i_cred, &obj);
863
864         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
865         if (unlikely(!or)) {
866                 EXOFS_ERR("exofs_get_inode: osd_start_request failed.\n");
867                 return -ENOMEM;
868         }
869         osd_req_get_attributes(or, &obj);
870
871         /* we need the inode attribute */
872         osd_req_add_get_attr_list(or, &g_attr_inode_data, 1);
873
874 #ifdef EXOFS_DEBUG_OBJ_ISIZE
875         /* we get the size attributes to do a sanity check */
876         osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
877 #endif
878
879         ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
880         if (ret)
881                 goto out;
882
883         attr = g_attr_inode_data;
884         ret = extract_attr_from_req(or, &attr);
885         if (ret) {
886                 EXOFS_ERR("exofs_get_inode: extract_attr_from_req failed\n");
887                 goto out;
888         }
889
890         WARN_ON(attr.len != EXOFS_INO_ATTR_SIZE);
891         memcpy(inode, attr.val_ptr, EXOFS_INO_ATTR_SIZE);
892
893 #ifdef EXOFS_DEBUG_OBJ_ISIZE
894         attr = g_attr_logical_length;
895         ret = extract_attr_from_req(or, &attr);
896         if (ret) {
897                 EXOFS_ERR("ERROR: extract attr from or failed\n");
898                 goto out;
899         }
900         *sanity = get_unaligned_be64(attr.val_ptr);
901 #endif
902
903 out:
904         osd_end_request(or);
905         return ret;
906 }
907
908 /*
909  * Fill in an inode read from the OSD and set it up for use
910  */
911 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
912 {
913         struct exofs_i_info *oi;
914         struct exofs_fcb fcb;
915         struct inode *inode;
916         uint64_t uninitialized_var(sanity);
917         int ret;
918
919         inode = iget_locked(sb, ino);
920         if (!inode)
921                 return ERR_PTR(-ENOMEM);
922         if (!(inode->i_state & I_NEW))
923                 return inode;
924         oi = exofs_i(inode);
925
926         /* read the inode from the osd */
927         ret = exofs_get_inode(sb, oi, &fcb, &sanity);
928         if (ret)
929                 goto bad_inode;
930
931         init_waitqueue_head(&oi->i_wq);
932         set_obj_created(oi);
933
934         /* copy stuff from on-disk struct to in-memory struct */
935         inode->i_mode = le16_to_cpu(fcb.i_mode);
936         inode->i_uid = le32_to_cpu(fcb.i_uid);
937         inode->i_gid = le32_to_cpu(fcb.i_gid);
938         inode->i_nlink = le16_to_cpu(fcb.i_links_count);
939         inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
940         inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
941         inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
942         inode->i_ctime.tv_nsec =
943                 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
944         oi->i_commit_size = le64_to_cpu(fcb.i_size);
945         i_size_write(inode, oi->i_commit_size);
946         inode->i_blkbits = EXOFS_BLKSHIFT;
947         inode->i_generation = le32_to_cpu(fcb.i_generation);
948
949 #ifdef EXOFS_DEBUG_OBJ_ISIZE
950         if ((inode->i_size != sanity) &&
951                 (!exofs_inode_is_fast_symlink(inode))) {
952                 EXOFS_ERR("WARNING: Size of object from inode and "
953                           "attributes differ (%lld != %llu)\n",
954                           inode->i_size, _LLU(sanity));
955         }
956 #endif
957
958         oi->i_dir_start_lookup = 0;
959
960         if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
961                 ret = -ESTALE;
962                 goto bad_inode;
963         }
964
965         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
966                 if (fcb.i_data[0])
967                         inode->i_rdev =
968                                 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
969                 else
970                         inode->i_rdev =
971                                 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
972         } else {
973                 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
974         }
975
976         if (S_ISREG(inode->i_mode)) {
977                 inode->i_op = &exofs_file_inode_operations;
978                 inode->i_fop = &exofs_file_operations;
979                 inode->i_mapping->a_ops = &exofs_aops;
980         } else if (S_ISDIR(inode->i_mode)) {
981                 inode->i_op = &exofs_dir_inode_operations;
982                 inode->i_fop = &exofs_dir_operations;
983                 inode->i_mapping->a_ops = &exofs_aops;
984         } else if (S_ISLNK(inode->i_mode)) {
985                 if (exofs_inode_is_fast_symlink(inode))
986                         inode->i_op = &exofs_fast_symlink_inode_operations;
987                 else {
988                         inode->i_op = &exofs_symlink_inode_operations;
989                         inode->i_mapping->a_ops = &exofs_aops;
990                 }
991         } else {
992                 inode->i_op = &exofs_special_inode_operations;
993                 if (fcb.i_data[0])
994                         init_special_inode(inode, inode->i_mode,
995                            old_decode_dev(le32_to_cpu(fcb.i_data[0])));
996                 else
997                         init_special_inode(inode, inode->i_mode,
998                            new_decode_dev(le32_to_cpu(fcb.i_data[1])));
999         }
1000
1001         unlock_new_inode(inode);
1002         return inode;
1003
1004 bad_inode:
1005         iget_failed(inode);
1006         return ERR_PTR(ret);
1007 }
1008
1009 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1010 {
1011         if (!obj_created(oi)) {
1012                 BUG_ON(!obj_2bcreated(oi));
1013                 wait_event(oi->i_wq, obj_created(oi));
1014         }
1015         return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1016 }
1017 /*
1018  * Callback function from exofs_new_inode().  The important thing is that we
1019  * set the obj_created flag so that other methods know that the object exists on
1020  * the OSD.
1021  */
1022 static void create_done(struct osd_request *or, void *p)
1023 {
1024         struct inode *inode = p;
1025         struct exofs_i_info *oi = exofs_i(inode);
1026         struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1027         int ret;
1028
1029         ret = exofs_check_ok(or);
1030         osd_end_request(or);
1031         atomic_dec(&sbi->s_curr_pending);
1032
1033         if (unlikely(ret)) {
1034                 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1035                           _LLU(sbi->s_pid), _LLU(inode->i_ino + EXOFS_OBJ_OFF));
1036                 make_bad_inode(inode);
1037         } else
1038                 set_obj_created(oi);
1039
1040         atomic_dec(&inode->i_count);
1041         wake_up(&oi->i_wq);
1042 }
1043
1044 /*
1045  * Set up a new inode and create an object for it on the OSD
1046  */
1047 struct inode *exofs_new_inode(struct inode *dir, int mode)
1048 {
1049         struct super_block *sb;
1050         struct inode *inode;
1051         struct exofs_i_info *oi;
1052         struct exofs_sb_info *sbi;
1053         struct osd_request *or;
1054         struct osd_obj_id obj;
1055         int ret;
1056
1057         sb = dir->i_sb;
1058         inode = new_inode(sb);
1059         if (!inode)
1060                 return ERR_PTR(-ENOMEM);
1061
1062         oi = exofs_i(inode);
1063
1064         init_waitqueue_head(&oi->i_wq);
1065         set_obj_2bcreated(oi);
1066
1067         sbi = sb->s_fs_info;
1068
1069         sb->s_dirt = 1;
1070         inode->i_uid = current->cred->fsuid;
1071         if (dir->i_mode & S_ISGID) {
1072                 inode->i_gid = dir->i_gid;
1073                 if (S_ISDIR(mode))
1074                         mode |= S_ISGID;
1075         } else {
1076                 inode->i_gid = current->cred->fsgid;
1077         }
1078         inode->i_mode = mode;
1079
1080         inode->i_ino = sbi->s_nextid++;
1081         inode->i_blkbits = EXOFS_BLKSHIFT;
1082         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1083         oi->i_commit_size = inode->i_size = 0;
1084         spin_lock(&sbi->s_next_gen_lock);
1085         inode->i_generation = sbi->s_next_generation++;
1086         spin_unlock(&sbi->s_next_gen_lock);
1087         insert_inode_hash(inode);
1088
1089         mark_inode_dirty(inode);
1090
1091         obj.partition = sbi->s_pid;
1092         obj.id = inode->i_ino + EXOFS_OBJ_OFF;
1093         exofs_make_credential(oi->i_cred, &obj);
1094
1095         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1096         if (unlikely(!or)) {
1097                 EXOFS_ERR("exofs_new_inode: osd_start_request failed\n");
1098                 return ERR_PTR(-ENOMEM);
1099         }
1100
1101         osd_req_create_object(or, &obj);
1102
1103         /* increment the refcount so that the inode will still be around when we
1104          * reach the callback
1105          */
1106         atomic_inc(&inode->i_count);
1107
1108         ret = exofs_async_op(or, create_done, inode, oi->i_cred);
1109         if (ret) {
1110                 atomic_dec(&inode->i_count);
1111                 osd_end_request(or);
1112                 return ERR_PTR(-EIO);
1113         }
1114         atomic_inc(&sbi->s_curr_pending);
1115
1116         return inode;
1117 }
1118
1119 /*
1120  * struct to pass two arguments to update_inode's callback
1121  */
1122 struct updatei_args {
1123         struct exofs_sb_info    *sbi;
1124         struct exofs_fcb        fcb;
1125 };
1126
1127 /*
1128  * Callback function from exofs_update_inode().
1129  */
1130 static void updatei_done(struct osd_request *or, void *p)
1131 {
1132         struct updatei_args *args = p;
1133
1134         osd_end_request(or);
1135
1136         atomic_dec(&args->sbi->s_curr_pending);
1137
1138         kfree(args);
1139 }
1140
1141 /*
1142  * Write the inode to the OSD.  Just fill up the struct, and set the attribute
1143  * synchronously or asynchronously depending on the do_sync flag.
1144  */
1145 static int exofs_update_inode(struct inode *inode, int do_sync)
1146 {
1147         struct exofs_i_info *oi = exofs_i(inode);
1148         struct super_block *sb = inode->i_sb;
1149         struct exofs_sb_info *sbi = sb->s_fs_info;
1150         struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1151         struct osd_request *or;
1152         struct osd_attr attr;
1153         struct exofs_fcb *fcb;
1154         struct updatei_args *args;
1155         int ret;
1156
1157         args = kzalloc(sizeof(*args), GFP_KERNEL);
1158         if (!args)
1159                 return -ENOMEM;
1160
1161         fcb = &args->fcb;
1162
1163         fcb->i_mode = cpu_to_le16(inode->i_mode);
1164         fcb->i_uid = cpu_to_le32(inode->i_uid);
1165         fcb->i_gid = cpu_to_le32(inode->i_gid);
1166         fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1167         fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1168         fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1169         fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1170         oi->i_commit_size = i_size_read(inode);
1171         fcb->i_size = cpu_to_le64(oi->i_commit_size);
1172         fcb->i_generation = cpu_to_le32(inode->i_generation);
1173
1174         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1175                 if (old_valid_dev(inode->i_rdev)) {
1176                         fcb->i_data[0] =
1177                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1178                         fcb->i_data[1] = 0;
1179                 } else {
1180                         fcb->i_data[0] = 0;
1181                         fcb->i_data[1] =
1182                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1183                         fcb->i_data[2] = 0;
1184                 }
1185         } else
1186                 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1187
1188         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1189         if (unlikely(!or)) {
1190                 EXOFS_ERR("exofs_update_inode: osd_start_request failed.\n");
1191                 ret = -ENOMEM;
1192                 goto free_args;
1193         }
1194
1195         osd_req_set_attributes(or, &obj);
1196
1197         attr = g_attr_inode_data;
1198         attr.val_ptr = fcb;
1199         osd_req_add_set_attr_list(or, &attr, 1);
1200
1201         if (!obj_created(oi)) {
1202                 EXOFS_DBGMSG("!obj_created\n");
1203                 BUG_ON(!obj_2bcreated(oi));
1204                 wait_event(oi->i_wq, obj_created(oi));
1205                 EXOFS_DBGMSG("wait_event done\n");
1206         }
1207
1208         if (do_sync) {
1209                 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
1210                 osd_end_request(or);
1211                 goto free_args;
1212         } else {
1213                 args->sbi = sbi;
1214
1215                 ret = exofs_async_op(or, updatei_done, args, oi->i_cred);
1216                 if (ret) {
1217                         osd_end_request(or);
1218                         goto free_args;
1219                 }
1220                 atomic_inc(&sbi->s_curr_pending);
1221                 goto out; /* deallocation in updatei_done */
1222         }
1223
1224 free_args:
1225         kfree(args);
1226 out:
1227         EXOFS_DBGMSG("ret=>%d\n", ret);
1228         return ret;
1229 }
1230
1231 int exofs_write_inode(struct inode *inode, int wait)
1232 {
1233         return exofs_update_inode(inode, wait);
1234 }
1235
1236 /*
1237  * Callback function from exofs_delete_inode() - don't have much cleaning up to
1238  * do.
1239  */
1240 static void delete_done(struct osd_request *or, void *p)
1241 {
1242         struct exofs_sb_info *sbi;
1243         osd_end_request(or);
1244         sbi = p;
1245         atomic_dec(&sbi->s_curr_pending);
1246 }
1247
1248 /*
1249  * Called when the refcount of an inode reaches zero.  We remove the object
1250  * from the OSD here.  We make sure the object was created before we try and
1251  * delete it.
1252  */
1253 void exofs_delete_inode(struct inode *inode)
1254 {
1255         struct exofs_i_info *oi = exofs_i(inode);
1256         struct super_block *sb = inode->i_sb;
1257         struct exofs_sb_info *sbi = sb->s_fs_info;
1258         struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1259         struct osd_request *or;
1260         int ret;
1261
1262         truncate_inode_pages(&inode->i_data, 0);
1263
1264         if (is_bad_inode(inode))
1265                 goto no_delete;
1266
1267         mark_inode_dirty(inode);
1268         exofs_update_inode(inode, inode_needs_sync(inode));
1269
1270         inode->i_size = 0;
1271         if (inode->i_blocks)
1272                 exofs_truncate(inode);
1273
1274         clear_inode(inode);
1275
1276         or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1277         if (unlikely(!or)) {
1278                 EXOFS_ERR("exofs_delete_inode: osd_start_request failed\n");
1279                 return;
1280         }
1281
1282         osd_req_remove_object(or, &obj);
1283
1284         /* if we are deleting an obj that hasn't been created yet, wait */
1285         if (!obj_created(oi)) {
1286                 BUG_ON(!obj_2bcreated(oi));
1287                 wait_event(oi->i_wq, obj_created(oi));
1288         }
1289
1290         ret = exofs_async_op(or, delete_done, sbi, oi->i_cred);
1291         if (ret) {
1292                 EXOFS_ERR(
1293                        "ERROR: @exofs_delete_inode exofs_async_op failed\n");
1294                 osd_end_request(or);
1295                 return;
1296         }
1297         atomic_inc(&sbi->s_curr_pending);
1298
1299         return;
1300
1301 no_delete:
1302         clear_inode(inode);
1303 }