blob: 6c10f7476699c4da76b18453a844aefcaa6c9f09 [file] [log] [blame]
Boaz Harroshe8062712008-10-27 18:37:02 +02001/*
2 * Copyright (C) 2005, 2006
Boaz Harrosh27d2e142009-06-14 17:23:09 +03003 * Avishay Traeger (avishay@gmail.com)
Boaz Harroshe8062712008-10-27 18:37:02 +02004 * 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/writeback.h>
35#include <linux/buffer_head.h>
Boaz Harroshbeaec072008-10-27 19:31:34 +020036#include <scsi/scsi_device.h>
Boaz Harroshe8062712008-10-27 18:37:02 +020037
38#include "exofs.h"
39
40#ifdef CONFIG_EXOFS_DEBUG
41# define EXOFS_DEBUG_OBJ_ISIZE 1
42#endif
43
Boaz Harroshbeaec072008-10-27 19:31:34 +020044struct page_collect {
45 struct exofs_sb_info *sbi;
46 struct request_queue *req_q;
47 struct inode *inode;
48 unsigned expected_pages;
49
50 struct bio *bio;
51 unsigned nr_pages;
52 unsigned long length;
53 loff_t pg_first; /* keep 64bit also in 32-arches */
54};
55
56static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
57 struct inode *inode)
58{
59 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
Boaz Harroshbeaec072008-10-27 19:31:34 +020060
61 pcol->sbi = sbi;
Boaz Harroshfc2fac52009-05-24 20:04:43 +030062 pcol->req_q = osd_request_queue(sbi->s_dev);
Boaz Harroshbeaec072008-10-27 19:31:34 +020063 pcol->inode = inode;
64 pcol->expected_pages = expected_pages;
65
66 pcol->bio = NULL;
67 pcol->nr_pages = 0;
68 pcol->length = 0;
69 pcol->pg_first = -1;
70
71 EXOFS_DBGMSG("_pcol_init ino=0x%lx expected_pages=%u\n", inode->i_ino,
72 expected_pages);
73}
74
75static void _pcol_reset(struct page_collect *pcol)
76{
77 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
78
79 pcol->bio = NULL;
80 pcol->nr_pages = 0;
81 pcol->length = 0;
82 pcol->pg_first = -1;
83 EXOFS_DBGMSG("_pcol_reset ino=0x%lx expected_pages=%u\n",
84 pcol->inode->i_ino, pcol->expected_pages);
85
86 /* this is probably the end of the loop but in writes
87 * it might not end here. don't be left with nothing
88 */
89 if (!pcol->expected_pages)
90 pcol->expected_pages = 128;
91}
92
93static int pcol_try_alloc(struct page_collect *pcol)
94{
95 int pages = min_t(unsigned, pcol->expected_pages, BIO_MAX_PAGES);
96
97 for (; pages; pages >>= 1) {
98 pcol->bio = bio_alloc(GFP_KERNEL, pages);
99 if (likely(pcol->bio))
100 return 0;
101 }
102
103 EXOFS_ERR("Failed to kcalloc expected_pages=%u\n",
104 pcol->expected_pages);
105 return -ENOMEM;
106}
107
108static void pcol_free(struct page_collect *pcol)
109{
110 bio_put(pcol->bio);
111 pcol->bio = NULL;
112}
113
114static int pcol_add_page(struct page_collect *pcol, struct page *page,
115 unsigned len)
116{
117 int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0);
118 if (unlikely(len != added_len))
119 return -ENOMEM;
120
121 ++pcol->nr_pages;
122 pcol->length += len;
123 return 0;
124}
125
126static int update_read_page(struct page *page, int ret)
127{
128 if (ret == 0) {
129 /* Everything is OK */
130 SetPageUptodate(page);
131 if (PageError(page))
132 ClearPageError(page);
133 } else if (ret == -EFAULT) {
134 /* In this case we were trying to read something that wasn't on
135 * disk yet - return a page full of zeroes. This should be OK,
136 * because the object should be empty (if there was a write
137 * before this read, the read would be waiting with the page
138 * locked */
139 clear_highpage(page);
140
141 SetPageUptodate(page);
142 if (PageError(page))
143 ClearPageError(page);
144 ret = 0; /* recovered error */
145 EXOFS_DBGMSG("recovered read error\n");
146 } else /* Error */
147 SetPageError(page);
148
149 return ret;
150}
151
152static void update_write_page(struct page *page, int ret)
153{
154 if (ret) {
155 mapping_set_error(page->mapping, ret);
156 SetPageError(page);
157 }
158 end_page_writeback(page);
159}
160
161/* Called at the end of reads, to optionally unlock pages and update their
162 * status.
163 */
164static int __readpages_done(struct osd_request *or, struct page_collect *pcol,
165 bool do_unlock)
166{
167 struct bio_vec *bvec;
168 int i;
169 u64 resid;
170 u64 good_bytes;
171 u64 length = 0;
172 int ret = exofs_check_ok_resid(or, &resid, NULL);
173
174 osd_end_request(or);
175
176 if (likely(!ret))
177 good_bytes = pcol->length;
178 else if (!resid)
179 good_bytes = 0;
180 else
181 good_bytes = pcol->length - resid;
182
183 EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx"
184 " length=0x%lx nr_pages=%u\n",
185 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
186 pcol->nr_pages);
187
188 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
189 struct page *page = bvec->bv_page;
190 struct inode *inode = page->mapping->host;
191 int page_stat;
192
193 if (inode != pcol->inode)
194 continue; /* osd might add more pages at end */
195
196 if (likely(length < good_bytes))
197 page_stat = 0;
198 else
199 page_stat = ret;
200
201 EXOFS_DBGMSG(" readpages_done(0x%lx, 0x%lx) %s\n",
202 inode->i_ino, page->index,
203 page_stat ? "bad_bytes" : "good_bytes");
204
205 ret = update_read_page(page, page_stat);
206 if (do_unlock)
207 unlock_page(page);
208 length += bvec->bv_len;
209 }
210
211 pcol_free(pcol);
212 EXOFS_DBGMSG("readpages_done END\n");
213 return ret;
214}
215
216/* callback of async reads */
217static void readpages_done(struct osd_request *or, void *p)
218{
219 struct page_collect *pcol = p;
220
221 __readpages_done(or, pcol, true);
222 atomic_dec(&pcol->sbi->s_curr_pending);
223 kfree(p);
224}
225
226static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
227{
228 struct bio_vec *bvec;
229 int i;
230
231 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
232 struct page *page = bvec->bv_page;
233
234 if (rw == READ)
235 update_read_page(page, ret);
236 else
237 update_write_page(page, ret);
238
239 unlock_page(page);
240 }
241 pcol_free(pcol);
242}
243
244static int read_exec(struct page_collect *pcol, bool is_sync)
245{
246 struct exofs_i_info *oi = exofs_i(pcol->inode);
247 struct osd_obj_id obj = {pcol->sbi->s_pid,
248 pcol->inode->i_ino + EXOFS_OBJ_OFF};
249 struct osd_request *or = NULL;
250 struct page_collect *pcol_copy = NULL;
251 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
252 int ret;
253
254 if (!pcol->bio)
255 return 0;
256
257 /* see comment in _readpage() about sync reads */
258 WARN_ON(is_sync && (pcol->nr_pages != 1));
259
260 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
261 if (unlikely(!or)) {
262 ret = -ENOMEM;
263 goto err;
264 }
265
Boaz Harrosh62f469b2009-05-24 20:04:26 +0300266 osd_req_read(or, &obj, i_start, pcol->bio, pcol->length);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200267
268 if (is_sync) {
269 exofs_sync_op(or, pcol->sbi->s_timeout, oi->i_cred);
270 return __readpages_done(or, pcol, false);
271 }
272
273 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
274 if (!pcol_copy) {
275 ret = -ENOMEM;
276 goto err;
277 }
278
279 *pcol_copy = *pcol;
280 ret = exofs_async_op(or, readpages_done, pcol_copy, oi->i_cred);
281 if (unlikely(ret))
282 goto err;
283
284 atomic_inc(&pcol->sbi->s_curr_pending);
285
286 EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
287 obj.id, _LLU(i_start), pcol->length);
288
289 /* pages ownership was passed to pcol_copy */
290 _pcol_reset(pcol);
291 return 0;
292
293err:
294 if (!is_sync)
295 _unlock_pcol_pages(pcol, ret, READ);
Boaz Harroshb76a3f92009-06-08 19:28:41 +0300296 else /* Pages unlocked by caller in sync mode only free bio */
297 pcol_free(pcol);
298
Boaz Harroshbeaec072008-10-27 19:31:34 +0200299 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 */
312static 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
351try_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
392fail:
393 /* SetPageError(page); ??? */
394 unlock_page(page);
395 return ret;
396}
397
398static 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
415static 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 */
438static 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 */
444static 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
496static 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
Boaz Harrosh62f469b2009-05-24 20:04:26 +0300525 pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */
526 osd_req_write(or, &obj, i_start, pcol_copy->bio, pcol_copy->length);
Boaz Harroshbeaec072008-10-27 19:31:34 +0200527 ret = exofs_async_op(or, writepages_done, pcol_copy, oi->i_cred);
528 if (unlikely(ret)) {
529 EXOFS_ERR("write_exec: exofs_async_op() Faild\n");
530 goto err;
531 }
532
533 atomic_inc(&pcol->sbi->s_curr_pending);
534 EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
535 pcol->inode->i_ino, pcol->pg_first, _LLU(i_start),
536 pcol->length);
537 /* pages ownership was passed to pcol_copy */
538 _pcol_reset(pcol);
539 return 0;
540
541err:
542 _unlock_pcol_pages(pcol, ret, WRITE);
543 kfree(pcol_copy);
544 if (or)
545 osd_end_request(or);
546 return ret;
547}
548
549/* writepage_strip is called either directly from writepage() or by the VFS from
550 * within write_cache_pages(), to add one more page to be written to storage.
551 * It will try to collect as many contiguous pages as possible. If a
552 * discontinuity is encountered or it runs out of resources it will submit the
553 * previous segment and will start a new collection.
554 * Eventually caller must submit the last segment if present.
555 */
556static int writepage_strip(struct page *page,
557 struct writeback_control *wbc_unused, void *data)
558{
559 struct page_collect *pcol = data;
560 struct inode *inode = pcol->inode;
561 struct exofs_i_info *oi = exofs_i(inode);
562 loff_t i_size = i_size_read(inode);
563 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
564 size_t len;
565 int ret;
566
567 BUG_ON(!PageLocked(page));
568
569 ret = wait_obj_created(oi);
570 if (unlikely(ret))
571 goto fail;
572
573 if (page->index < end_index)
574 /* in this case, the page is within the limits of the file */
575 len = PAGE_CACHE_SIZE;
576 else {
577 len = i_size & ~PAGE_CACHE_MASK;
578
579 if (page->index > end_index || !len) {
580 /* in this case, the page is outside the limits
581 * (truncate in progress)
582 */
583 ret = write_exec(pcol);
584 if (unlikely(ret))
585 goto fail;
586 if (PageError(page))
587 ClearPageError(page);
588 unlock_page(page);
589 return 0;
590 }
591 }
592
593try_again:
594
595 if (unlikely(pcol->pg_first == -1)) {
596 pcol->pg_first = page->index;
597 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
598 page->index)) {
599 /* Discontinuity detected, split the request */
600 ret = write_exec(pcol);
601 if (unlikely(ret))
602 goto fail;
603 goto try_again;
604 }
605
606 if (!pcol->bio) {
607 ret = pcol_try_alloc(pcol);
608 if (unlikely(ret))
609 goto fail;
610 }
611
612 EXOFS_DBGMSG(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
613 inode->i_ino, page->index, len);
614
615 ret = pcol_add_page(pcol, page, len);
616 if (unlikely(ret)) {
617 EXOFS_DBGMSG("Failed pcol_add_page "
618 "nr_pages=%u total_length=0x%lx\n",
619 pcol->nr_pages, pcol->length);
620
621 /* split the request, next loop will start again */
622 ret = write_exec(pcol);
623 if (unlikely(ret)) {
624 EXOFS_DBGMSG("write_exec faild => %d", ret);
625 goto fail;
626 }
627
628 goto try_again;
629 }
630
631 BUG_ON(PageWriteback(page));
632 set_page_writeback(page);
633
634 return 0;
635
636fail:
637 set_bit(AS_EIO, &page->mapping->flags);
638 unlock_page(page);
639 return ret;
640}
641
642static int exofs_writepages(struct address_space *mapping,
643 struct writeback_control *wbc)
644{
645 struct page_collect pcol;
646 long start, end, expected_pages;
647 int ret;
648
649 start = wbc->range_start >> PAGE_CACHE_SHIFT;
650 end = (wbc->range_end == LLONG_MAX) ?
651 start + mapping->nrpages :
652 wbc->range_end >> PAGE_CACHE_SHIFT;
653
654 if (start || end)
655 expected_pages = min(end - start + 1, 32L);
656 else
657 expected_pages = mapping->nrpages;
658
659 EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx"
660 " m->nrpages=%lu start=0x%lx end=0x%lx\n",
661 mapping->host->i_ino, wbc->range_start, wbc->range_end,
662 mapping->nrpages, start, end);
663
664 _pcol_init(&pcol, expected_pages, mapping->host);
665
666 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
667 if (ret) {
668 EXOFS_ERR("write_cache_pages => %d\n", ret);
669 return ret;
670 }
671
672 return write_exec(&pcol);
673}
674
675static int exofs_writepage(struct page *page, struct writeback_control *wbc)
676{
677 struct page_collect pcol;
678 int ret;
679
680 _pcol_init(&pcol, 1, page->mapping->host);
681
682 ret = writepage_strip(page, NULL, &pcol);
683 if (ret) {
684 EXOFS_ERR("exofs_writepage => %d\n", ret);
685 return ret;
686 }
687
688 return write_exec(&pcol);
689}
690
691int exofs_write_begin(struct file *file, struct address_space *mapping,
692 loff_t pos, unsigned len, unsigned flags,
693 struct page **pagep, void **fsdata)
694{
695 int ret = 0;
696 struct page *page;
697
698 page = *pagep;
699 if (page == NULL) {
700 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
701 fsdata);
702 if (ret) {
703 EXOFS_DBGMSG("simple_write_begin faild\n");
704 return ret;
705 }
706
707 page = *pagep;
708 }
709
710 /* read modify write */
711 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
712 ret = _readpage(page, true);
713 if (ret) {
714 /*SetPageError was done by _readpage. Is it ok?*/
715 unlock_page(page);
716 EXOFS_DBGMSG("__readpage_filler faild\n");
717 }
718 }
719
720 return ret;
721}
722
723static int exofs_write_begin_export(struct file *file,
724 struct address_space *mapping,
725 loff_t pos, unsigned len, unsigned flags,
726 struct page **pagep, void **fsdata)
727{
728 *pagep = NULL;
729
730 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
731 fsdata);
732}
733
734const struct address_space_operations exofs_aops = {
735 .readpage = exofs_readpage,
736 .readpages = exofs_readpages,
737 .writepage = exofs_writepage,
738 .writepages = exofs_writepages,
739 .write_begin = exofs_write_begin_export,
740 .write_end = simple_write_end,
741};
742
Boaz Harroshe8062712008-10-27 18:37:02 +0200743/******************************************************************************
744 * INODE OPERATIONS
745 *****************************************************************************/
746
747/*
748 * Test whether an inode is a fast symlink.
749 */
750static inline int exofs_inode_is_fast_symlink(struct inode *inode)
751{
752 struct exofs_i_info *oi = exofs_i(inode);
753
754 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
755}
756
757/*
758 * get_block_t - Fill in a buffer_head
759 * An OSD takes care of block allocation so we just fake an allocation by
760 * putting in the inode's sector_t in the buffer_head.
761 * TODO: What about the case of create==0 and @iblock does not exist in the
762 * object?
763 */
764static int exofs_get_block(struct inode *inode, sector_t iblock,
765 struct buffer_head *bh_result, int create)
766{
767 map_bh(bh_result, inode->i_sb, iblock);
768 return 0;
769}
770
771const struct osd_attr g_attr_logical_length = ATTR_DEF(
772 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
773
774/*
775 * Truncate a file to the specified size - all we have to do is set the size
776 * attribute. We make sure the object exists first.
777 */
778void exofs_truncate(struct inode *inode)
779{
780 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
781 struct exofs_i_info *oi = exofs_i(inode);
782 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
783 struct osd_request *or;
784 struct osd_attr attr;
785 loff_t isize = i_size_read(inode);
786 __be64 newsize;
787 int ret;
788
789 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
790 || S_ISLNK(inode->i_mode)))
791 return;
792 if (exofs_inode_is_fast_symlink(inode))
793 return;
794 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
795 return;
796 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
797
798 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
799
800 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
801 if (unlikely(!or)) {
802 EXOFS_ERR("ERROR: exofs_truncate: osd_start_request failed\n");
803 goto fail;
804 }
805
806 osd_req_set_attributes(or, &obj);
807
808 newsize = cpu_to_be64((u64)isize);
809 attr = g_attr_logical_length;
810 attr.val_ptr = &newsize;
811 osd_req_add_set_attr_list(or, &attr, 1);
812
813 /* if we are about to truncate an object, and it hasn't been
814 * created yet, wait
815 */
816 if (unlikely(wait_obj_created(oi)))
817 goto fail;
818
819 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
820 osd_end_request(or);
821 if (ret)
822 goto fail;
823
824out:
825 mark_inode_dirty(inode);
826 return;
827fail:
828 make_bad_inode(inode);
829 goto out;
830}
831
832/*
833 * Set inode attributes - just call generic functions.
834 */
835int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
836{
837 struct inode *inode = dentry->d_inode;
838 int error;
839
840 error = inode_change_ok(inode, iattr);
841 if (error)
842 return error;
843
844 error = inode_setattr(inode, iattr);
845 return error;
846}
Boaz Harroshe6af00f2008-10-28 15:38:12 +0200847
848/*
849 * Read an inode from the OSD, and return it as is. We also return the size
850 * attribute in the 'sanity' argument if we got compiled with debugging turned
851 * on.
852 */
853static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
854 struct exofs_fcb *inode, uint64_t *sanity)
855{
856 struct exofs_sb_info *sbi = sb->s_fs_info;
857 struct osd_request *or;
858 struct osd_attr attr;
859 struct osd_obj_id obj = {sbi->s_pid,
860 oi->vfs_inode.i_ino + EXOFS_OBJ_OFF};
861 int ret;
862
863 exofs_make_credential(oi->i_cred, &obj);
864
865 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
866 if (unlikely(!or)) {
867 EXOFS_ERR("exofs_get_inode: osd_start_request failed.\n");
868 return -ENOMEM;
869 }
870 osd_req_get_attributes(or, &obj);
871
872 /* we need the inode attribute */
873 osd_req_add_get_attr_list(or, &g_attr_inode_data, 1);
874
875#ifdef EXOFS_DEBUG_OBJ_ISIZE
876 /* we get the size attributes to do a sanity check */
877 osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
878#endif
879
880 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
881 if (ret)
882 goto out;
883
884 attr = g_attr_inode_data;
885 ret = extract_attr_from_req(or, &attr);
886 if (ret) {
887 EXOFS_ERR("exofs_get_inode: extract_attr_from_req failed\n");
888 goto out;
889 }
890
891 WARN_ON(attr.len != EXOFS_INO_ATTR_SIZE);
892 memcpy(inode, attr.val_ptr, EXOFS_INO_ATTR_SIZE);
893
894#ifdef EXOFS_DEBUG_OBJ_ISIZE
895 attr = g_attr_logical_length;
896 ret = extract_attr_from_req(or, &attr);
897 if (ret) {
898 EXOFS_ERR("ERROR: extract attr from or failed\n");
899 goto out;
900 }
901 *sanity = get_unaligned_be64(attr.val_ptr);
902#endif
903
904out:
905 osd_end_request(or);
906 return ret;
907}
908
909/*
910 * Fill in an inode read from the OSD and set it up for use
911 */
912struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
913{
914 struct exofs_i_info *oi;
915 struct exofs_fcb fcb;
916 struct inode *inode;
917 uint64_t uninitialized_var(sanity);
918 int ret;
919
920 inode = iget_locked(sb, ino);
921 if (!inode)
922 return ERR_PTR(-ENOMEM);
923 if (!(inode->i_state & I_NEW))
924 return inode;
925 oi = exofs_i(inode);
926
927 /* read the inode from the osd */
928 ret = exofs_get_inode(sb, oi, &fcb, &sanity);
929 if (ret)
930 goto bad_inode;
931
932 init_waitqueue_head(&oi->i_wq);
933 set_obj_created(oi);
934
935 /* copy stuff from on-disk struct to in-memory struct */
936 inode->i_mode = le16_to_cpu(fcb.i_mode);
937 inode->i_uid = le32_to_cpu(fcb.i_uid);
938 inode->i_gid = le32_to_cpu(fcb.i_gid);
939 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
940 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
941 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
942 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
943 inode->i_ctime.tv_nsec =
944 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
945 oi->i_commit_size = le64_to_cpu(fcb.i_size);
946 i_size_write(inode, oi->i_commit_size);
947 inode->i_blkbits = EXOFS_BLKSHIFT;
948 inode->i_generation = le32_to_cpu(fcb.i_generation);
949
950#ifdef EXOFS_DEBUG_OBJ_ISIZE
951 if ((inode->i_size != sanity) &&
952 (!exofs_inode_is_fast_symlink(inode))) {
953 EXOFS_ERR("WARNING: Size of object from inode and "
954 "attributes differ (%lld != %llu)\n",
955 inode->i_size, _LLU(sanity));
956 }
957#endif
958
959 oi->i_dir_start_lookup = 0;
960
961 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
962 ret = -ESTALE;
963 goto bad_inode;
964 }
965
966 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
967 if (fcb.i_data[0])
968 inode->i_rdev =
969 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
970 else
971 inode->i_rdev =
972 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
973 } else {
974 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
975 }
976
977 if (S_ISREG(inode->i_mode)) {
978 inode->i_op = &exofs_file_inode_operations;
979 inode->i_fop = &exofs_file_operations;
980 inode->i_mapping->a_ops = &exofs_aops;
981 } else if (S_ISDIR(inode->i_mode)) {
982 inode->i_op = &exofs_dir_inode_operations;
983 inode->i_fop = &exofs_dir_operations;
984 inode->i_mapping->a_ops = &exofs_aops;
985 } else if (S_ISLNK(inode->i_mode)) {
986 if (exofs_inode_is_fast_symlink(inode))
987 inode->i_op = &exofs_fast_symlink_inode_operations;
988 else {
989 inode->i_op = &exofs_symlink_inode_operations;
990 inode->i_mapping->a_ops = &exofs_aops;
991 }
992 } else {
993 inode->i_op = &exofs_special_inode_operations;
994 if (fcb.i_data[0])
995 init_special_inode(inode, inode->i_mode,
996 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
997 else
998 init_special_inode(inode, inode->i_mode,
999 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1000 }
1001
1002 unlock_new_inode(inode);
1003 return inode;
1004
1005bad_inode:
1006 iget_failed(inode);
1007 return ERR_PTR(ret);
1008}
1009
1010int __exofs_wait_obj_created(struct exofs_i_info *oi)
1011{
1012 if (!obj_created(oi)) {
1013 BUG_ON(!obj_2bcreated(oi));
1014 wait_event(oi->i_wq, obj_created(oi));
1015 }
1016 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1017}
1018/*
1019 * Callback function from exofs_new_inode(). The important thing is that we
1020 * set the obj_created flag so that other methods know that the object exists on
1021 * the OSD.
1022 */
1023static void create_done(struct osd_request *or, void *p)
1024{
1025 struct inode *inode = p;
1026 struct exofs_i_info *oi = exofs_i(inode);
1027 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1028 int ret;
1029
1030 ret = exofs_check_ok(or);
1031 osd_end_request(or);
1032 atomic_dec(&sbi->s_curr_pending);
1033
1034 if (unlikely(ret)) {
1035 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1036 _LLU(sbi->s_pid), _LLU(inode->i_ino + EXOFS_OBJ_OFF));
1037 make_bad_inode(inode);
1038 } else
1039 set_obj_created(oi);
1040
1041 atomic_dec(&inode->i_count);
1042 wake_up(&oi->i_wq);
1043}
1044
1045/*
1046 * Set up a new inode and create an object for it on the OSD
1047 */
1048struct inode *exofs_new_inode(struct inode *dir, int mode)
1049{
1050 struct super_block *sb;
1051 struct inode *inode;
1052 struct exofs_i_info *oi;
1053 struct exofs_sb_info *sbi;
1054 struct osd_request *or;
1055 struct osd_obj_id obj;
1056 int ret;
1057
1058 sb = dir->i_sb;
1059 inode = new_inode(sb);
1060 if (!inode)
1061 return ERR_PTR(-ENOMEM);
1062
1063 oi = exofs_i(inode);
1064
1065 init_waitqueue_head(&oi->i_wq);
1066 set_obj_2bcreated(oi);
1067
1068 sbi = sb->s_fs_info;
1069
1070 sb->s_dirt = 1;
1071 inode->i_uid = current->cred->fsuid;
1072 if (dir->i_mode & S_ISGID) {
1073 inode->i_gid = dir->i_gid;
1074 if (S_ISDIR(mode))
1075 mode |= S_ISGID;
1076 } else {
1077 inode->i_gid = current->cred->fsgid;
1078 }
1079 inode->i_mode = mode;
1080
1081 inode->i_ino = sbi->s_nextid++;
1082 inode->i_blkbits = EXOFS_BLKSHIFT;
1083 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1084 oi->i_commit_size = inode->i_size = 0;
1085 spin_lock(&sbi->s_next_gen_lock);
1086 inode->i_generation = sbi->s_next_generation++;
1087 spin_unlock(&sbi->s_next_gen_lock);
1088 insert_inode_hash(inode);
1089
1090 mark_inode_dirty(inode);
1091
1092 obj.partition = sbi->s_pid;
1093 obj.id = inode->i_ino + EXOFS_OBJ_OFF;
1094 exofs_make_credential(oi->i_cred, &obj);
1095
1096 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1097 if (unlikely(!or)) {
1098 EXOFS_ERR("exofs_new_inode: osd_start_request failed\n");
1099 return ERR_PTR(-ENOMEM);
1100 }
1101
1102 osd_req_create_object(or, &obj);
1103
1104 /* increment the refcount so that the inode will still be around when we
1105 * reach the callback
1106 */
1107 atomic_inc(&inode->i_count);
1108
1109 ret = exofs_async_op(or, create_done, inode, oi->i_cred);
1110 if (ret) {
1111 atomic_dec(&inode->i_count);
1112 osd_end_request(or);
1113 return ERR_PTR(-EIO);
1114 }
1115 atomic_inc(&sbi->s_curr_pending);
1116
1117 return inode;
1118}
Boaz Harroshba9e5e92008-10-28 16:11:41 +02001119
1120/*
1121 * struct to pass two arguments to update_inode's callback
1122 */
1123struct updatei_args {
1124 struct exofs_sb_info *sbi;
1125 struct exofs_fcb fcb;
1126};
1127
1128/*
1129 * Callback function from exofs_update_inode().
1130 */
1131static void updatei_done(struct osd_request *or, void *p)
1132{
1133 struct updatei_args *args = p;
1134
1135 osd_end_request(or);
1136
1137 atomic_dec(&args->sbi->s_curr_pending);
1138
1139 kfree(args);
1140}
1141
1142/*
1143 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1144 * synchronously or asynchronously depending on the do_sync flag.
1145 */
1146static int exofs_update_inode(struct inode *inode, int do_sync)
1147{
1148 struct exofs_i_info *oi = exofs_i(inode);
1149 struct super_block *sb = inode->i_sb;
1150 struct exofs_sb_info *sbi = sb->s_fs_info;
1151 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1152 struct osd_request *or;
1153 struct osd_attr attr;
1154 struct exofs_fcb *fcb;
1155 struct updatei_args *args;
1156 int ret;
1157
1158 args = kzalloc(sizeof(*args), GFP_KERNEL);
1159 if (!args)
1160 return -ENOMEM;
1161
1162 fcb = &args->fcb;
1163
1164 fcb->i_mode = cpu_to_le16(inode->i_mode);
1165 fcb->i_uid = cpu_to_le32(inode->i_uid);
1166 fcb->i_gid = cpu_to_le32(inode->i_gid);
1167 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1168 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1169 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1170 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1171 oi->i_commit_size = i_size_read(inode);
1172 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1173 fcb->i_generation = cpu_to_le32(inode->i_generation);
1174
1175 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1176 if (old_valid_dev(inode->i_rdev)) {
1177 fcb->i_data[0] =
1178 cpu_to_le32(old_encode_dev(inode->i_rdev));
1179 fcb->i_data[1] = 0;
1180 } else {
1181 fcb->i_data[0] = 0;
1182 fcb->i_data[1] =
1183 cpu_to_le32(new_encode_dev(inode->i_rdev));
1184 fcb->i_data[2] = 0;
1185 }
1186 } else
1187 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1188
1189 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1190 if (unlikely(!or)) {
1191 EXOFS_ERR("exofs_update_inode: osd_start_request failed.\n");
1192 ret = -ENOMEM;
1193 goto free_args;
1194 }
1195
1196 osd_req_set_attributes(or, &obj);
1197
1198 attr = g_attr_inode_data;
1199 attr.val_ptr = fcb;
1200 osd_req_add_set_attr_list(or, &attr, 1);
1201
1202 if (!obj_created(oi)) {
1203 EXOFS_DBGMSG("!obj_created\n");
1204 BUG_ON(!obj_2bcreated(oi));
1205 wait_event(oi->i_wq, obj_created(oi));
1206 EXOFS_DBGMSG("wait_event done\n");
1207 }
1208
1209 if (do_sync) {
1210 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
1211 osd_end_request(or);
1212 goto free_args;
1213 } else {
1214 args->sbi = sbi;
1215
1216 ret = exofs_async_op(or, updatei_done, args, oi->i_cred);
1217 if (ret) {
1218 osd_end_request(or);
1219 goto free_args;
1220 }
1221 atomic_inc(&sbi->s_curr_pending);
1222 goto out; /* deallocation in updatei_done */
1223 }
1224
1225free_args:
1226 kfree(args);
1227out:
1228 EXOFS_DBGMSG("ret=>%d\n", ret);
1229 return ret;
1230}
1231
1232int exofs_write_inode(struct inode *inode, int wait)
1233{
1234 return exofs_update_inode(inode, wait);
1235}
1236
1237/*
1238 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1239 * do.
1240 */
1241static void delete_done(struct osd_request *or, void *p)
1242{
1243 struct exofs_sb_info *sbi;
1244 osd_end_request(or);
1245 sbi = p;
1246 atomic_dec(&sbi->s_curr_pending);
1247}
1248
1249/*
1250 * Called when the refcount of an inode reaches zero. We remove the object
1251 * from the OSD here. We make sure the object was created before we try and
1252 * delete it.
1253 */
1254void exofs_delete_inode(struct inode *inode)
1255{
1256 struct exofs_i_info *oi = exofs_i(inode);
1257 struct super_block *sb = inode->i_sb;
1258 struct exofs_sb_info *sbi = sb->s_fs_info;
1259 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1260 struct osd_request *or;
1261 int ret;
1262
1263 truncate_inode_pages(&inode->i_data, 0);
1264
1265 if (is_bad_inode(inode))
1266 goto no_delete;
1267
1268 mark_inode_dirty(inode);
1269 exofs_update_inode(inode, inode_needs_sync(inode));
1270
1271 inode->i_size = 0;
1272 if (inode->i_blocks)
1273 exofs_truncate(inode);
1274
1275 clear_inode(inode);
1276
1277 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1278 if (unlikely(!or)) {
1279 EXOFS_ERR("exofs_delete_inode: osd_start_request failed\n");
1280 return;
1281 }
1282
1283 osd_req_remove_object(or, &obj);
1284
1285 /* if we are deleting an obj that hasn't been created yet, wait */
1286 if (!obj_created(oi)) {
1287 BUG_ON(!obj_2bcreated(oi));
1288 wait_event(oi->i_wq, obj_created(oi));
1289 }
1290
1291 ret = exofs_async_op(or, delete_done, sbi, oi->i_cred);
1292 if (ret) {
1293 EXOFS_ERR(
1294 "ERROR: @exofs_delete_inode exofs_async_op failed\n");
1295 osd_end_request(or);
1296 return;
1297 }
1298 atomic_inc(&sbi->s_curr_pending);
1299
1300 return;
1301
1302no_delete:
1303 clear_inode(inode);
1304}