aboutsummaryrefslogtreecommitdiffstats
path: root/fs/exofs/inode.c
diff options
context:
space:
mode:
authorBoaz Harrosh <bharrosh@panasas.com>2008-10-27 13:31:34 -0400
committerBoaz Harrosh <bharrosh@panasas.com>2009-03-31 12:44:29 -0400
commitbeaec07ba6af35d387643b76a2920a7a6e22207b (patch)
tree74ffd4738198424f698ae238e4d3164938ef5af7 /fs/exofs/inode.c
parent982980d753223fda3864038236b7b94e246895cb (diff)
exofs: address_space_operations
OK Now we start to read and write from osd-objects. We try to collect at most contiguous pages as possible in a single write/read. The first page index is the object's offset. TODO: In 64-bit a single bio can carry at most 128 pages. Add support of chaining multiple bios Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Diffstat (limited to 'fs/exofs/inode.c')
-rw-r--r--fs/exofs/inode.c697
1 files changed, 697 insertions, 0 deletions
diff --git a/fs/exofs/inode.c b/fs/exofs/inode.c
index b0bda1e91225..a3691d8bfb98 100644
--- a/fs/exofs/inode.c
+++ b/fs/exofs/inode.c
@@ -35,6 +35,7 @@
35 35
36#include <linux/writeback.h> 36#include <linux/writeback.h>
37#include <linux/buffer_head.h> 37#include <linux/buffer_head.h>
38#include <scsi/scsi_device.h>
38 39
39#include "exofs.h" 40#include "exofs.h"
40 41
@@ -42,6 +43,702 @@
42# define EXOFS_DEBUG_OBJ_ISIZE 1 43# define EXOFS_DEBUG_OBJ_ISIZE 1
43#endif 44#endif
44 45
46struct 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
58static 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
78static 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
96static 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
111static void pcol_free(struct page_collect *pcol)
112{
113 bio_put(pcol->bio);
114 pcol->bio = NULL;
115}
116
117static 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
129static 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
155static 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 */
167static 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 */
220static 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
229static 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
247static 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
296err:
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 */
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
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
540err:
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 */
555static 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
592try_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
635fail:
636 set_bit(AS_EIO, &page->mapping->flags);
637 unlock_page(page);
638 return ret;
639}
640
641static 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
674static 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
690int 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
722static 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
733const 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
45/****************************************************************************** 742/******************************************************************************
46 * INODE OPERATIONS 743 * INODE OPERATIONS
47 *****************************************************************************/ 744 *****************************************************************************/