2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
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)
13 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * This file is part of exofs.
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.
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.
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
34 #include <linux/slab.h>
38 #define EXOFS_DBGMSG2(M...) do {} while (0)
40 enum {MAX_PAGES_KMALLOC = PAGE_SIZE / sizeof(struct page *), };
42 unsigned exofs_max_io_pages(struct ore_layout *layout,
43 unsigned expected_pages)
45 unsigned pages = min_t(unsigned, expected_pages, MAX_PAGES_KMALLOC);
47 /* TODO: easily support bio chaining */
48 pages = min_t(unsigned, pages, layout->max_io_length / PAGE_SIZE);
53 struct exofs_sb_info *sbi;
55 unsigned expected_pages;
56 struct ore_io_state *ios;
62 loff_t pg_first; /* keep 64bit also in 32-arches */
63 bool read_4_write; /* This means two things: that the read is sync
64 * And the pages should not be unlocked.
68 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
71 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
75 pcol->expected_pages = expected_pages;
79 pcol->alloc_pages = 0;
83 pcol->read_4_write = false;
86 static void _pcol_reset(struct page_collect *pcol)
88 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
91 pcol->alloc_pages = 0;
97 /* this is probably the end of the loop but in writes
98 * it might not end here. don't be left with nothing
100 if (!pcol->expected_pages)
101 pcol->expected_pages = MAX_PAGES_KMALLOC;
104 static int pcol_try_alloc(struct page_collect *pcol)
108 /* TODO: easily support bio chaining */
109 pages = exofs_max_io_pages(&pcol->sbi->layout, pcol->expected_pages);
111 for (; pages; pages >>= 1) {
112 pcol->pages = kmalloc(pages * sizeof(struct page *),
114 if (likely(pcol->pages)) {
115 pcol->alloc_pages = pages;
120 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
121 pcol->expected_pages);
125 static void pcol_free(struct page_collect *pcol)
131 ore_put_io_state(pcol->ios);
136 static int pcol_add_page(struct page_collect *pcol, struct page *page,
139 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
142 pcol->pages[pcol->nr_pages++] = page;
147 enum {PAGE_WAS_NOT_IN_IO = 17};
148 static int update_read_page(struct page *page, int ret)
152 /* Everything is OK */
153 SetPageUptodate(page);
155 ClearPageError(page);
158 /* In this case we were trying to read something that wasn't on
159 * disk yet - return a page full of zeroes. This should be OK,
160 * because the object should be empty (if there was a write
161 * before this read, the read would be waiting with the page
163 clear_highpage(page);
165 SetPageUptodate(page);
167 ClearPageError(page);
168 EXOFS_DBGMSG("recovered read error\n");
170 case PAGE_WAS_NOT_IN_IO:
171 ret = 0; /* recovered error */
179 static void update_write_page(struct page *page, int ret)
181 if (unlikely(ret == PAGE_WAS_NOT_IN_IO))
182 return; /* don't pass start don't collect $200 */
185 mapping_set_error(page->mapping, ret);
188 end_page_writeback(page);
191 /* Called at the end of reads, to optionally unlock pages and update their
194 static int __readpages_done(struct page_collect *pcol)
200 int ret = ore_check_io(pcol->ios, &resid);
203 good_bytes = pcol->length;
204 ret = PAGE_WAS_NOT_IN_IO;
206 good_bytes = pcol->length - resid;
208 if (good_bytes > pcol->ios->length)
209 good_bytes = pcol->ios->length;
211 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
212 " length=0x%lx nr_pages=%u\n",
213 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
216 for (i = 0; i < pcol->nr_pages; i++) {
217 struct page *page = pcol->pages[i];
218 struct inode *inode = page->mapping->host;
221 if (inode != pcol->inode)
222 continue; /* osd might add more pages at end */
224 if (likely(length < good_bytes))
229 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
230 inode->i_ino, page->index,
231 page_stat ? "bad_bytes" : "good_bytes");
233 ret = update_read_page(page, page_stat);
234 if (!pcol->read_4_write)
240 EXOFS_DBGMSG2("readpages_done END\n");
244 /* callback of async reads */
245 static void readpages_done(struct ore_io_state *ios, void *p)
247 struct page_collect *pcol = p;
249 __readpages_done(pcol);
250 atomic_dec(&pcol->sbi->s_curr_pending);
254 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
258 for (i = 0; i < pcol->nr_pages; i++) {
259 struct page *page = pcol->pages[i];
262 update_read_page(page, ret);
264 update_write_page(page, ret);
270 static int _maybe_not_all_in_one_io(struct ore_io_state *ios,
271 struct page_collect *pcol_src, struct page_collect *pcol)
273 /* length was wrong or offset was not page aligned */
274 BUG_ON(pcol_src->nr_pages < ios->nr_pages);
276 if (pcol_src->nr_pages > ios->nr_pages) {
277 struct page **src_page;
278 unsigned pages_less = pcol_src->nr_pages - ios->nr_pages;
279 unsigned long len_less = pcol_src->length - ios->length;
283 /* This IO was trimmed */
284 pcol_src->nr_pages = ios->nr_pages;
285 pcol_src->length = ios->length;
287 /* Left over pages are passed to the next io */
288 pcol->expected_pages += pages_less;
289 pcol->nr_pages = pages_less;
290 pcol->length = len_less;
291 src_page = pcol_src->pages + pcol_src->nr_pages;
292 pcol->pg_first = (*src_page)->index;
294 ret = pcol_try_alloc(pcol);
298 for (i = 0; i < pages_less; ++i)
299 pcol->pages[i] = *src_page++;
301 EXOFS_DBGMSG("Length was adjusted nr_pages=0x%x "
302 "pages_less=0x%x expected_pages=0x%x "
303 "next_offset=0x%llx next_len=0x%lx\n",
304 pcol_src->nr_pages, pages_less, pcol->expected_pages,
305 pcol->pg_first * PAGE_SIZE, pcol->length);
310 static int read_exec(struct page_collect *pcol)
312 struct exofs_i_info *oi = exofs_i(pcol->inode);
313 struct ore_io_state *ios;
314 struct page_collect *pcol_copy = NULL;
321 int ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, true,
322 pcol->pg_first << PAGE_CACHE_SHIFT,
323 pcol->length, &pcol->ios);
330 ios->pages = pcol->pages;
332 if (pcol->read_4_write) {
334 return __readpages_done(pcol);
337 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
344 ios->done = readpages_done;
345 ios->private = pcol_copy;
347 /* pages ownership was passed to pcol_copy */
350 ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
354 EXOFS_DBGMSG2("read_exec(0x%lx) offset=0x%llx length=0x%llx\n",
355 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
361 atomic_inc(&pcol->sbi->s_curr_pending);
366 if (!pcol->read_4_write)
367 _unlock_pcol_pages(pcol, ret, READ);
375 /* readpage_strip is called either directly from readpage() or by the VFS from
376 * within read_cache_pages(), to add one more page to be read. It will try to
377 * collect as many contiguous pages as posible. If a discontinuity is
378 * encountered, or it runs out of resources, it will submit the previous segment
379 * and will start a new collection. Eventually caller must submit the last
380 * segment if present.
382 static int readpage_strip(void *data, struct page *page)
384 struct page_collect *pcol = data;
385 struct inode *inode = pcol->inode;
386 struct exofs_i_info *oi = exofs_i(inode);
387 loff_t i_size = i_size_read(inode);
388 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
392 /* FIXME: Just for debugging, will be removed */
393 if (PageUptodate(page))
394 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
397 if (page->index < end_index)
398 len = PAGE_CACHE_SIZE;
399 else if (page->index == end_index)
400 len = i_size & ~PAGE_CACHE_MASK;
404 if (!len || !obj_created(oi)) {
405 /* this will be out of bounds, or doesn't exist yet.
406 * Current page is cleared and the request is split
408 clear_highpage(page);
410 SetPageUptodate(page);
412 ClearPageError(page);
414 if (!pcol->read_4_write)
416 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
417 "read_4_write=%d index=0x%lx end_index=0x%lx "
418 "splitting\n", inode->i_ino, len,
419 pcol->read_4_write, page->index, end_index);
421 return read_exec(pcol);
426 if (unlikely(pcol->pg_first == -1)) {
427 pcol->pg_first = page->index;
428 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
430 /* Discontinuity detected, split the request */
431 ret = read_exec(pcol);
438 ret = pcol_try_alloc(pcol);
443 if (len != PAGE_CACHE_SIZE)
444 zero_user(page, len, PAGE_CACHE_SIZE - len);
446 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
447 inode->i_ino, page->index, len);
449 ret = pcol_add_page(pcol, page, len);
451 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
452 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
453 page, len, pcol->nr_pages, pcol->length);
455 /* split the request, and start again with current page */
456 ret = read_exec(pcol);
466 /* SetPageError(page); ??? */
471 static int exofs_readpages(struct file *file, struct address_space *mapping,
472 struct list_head *pages, unsigned nr_pages)
474 struct page_collect pcol;
477 _pcol_init(&pcol, nr_pages, mapping->host);
479 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
481 EXOFS_ERR("read_cache_pages => %d\n", ret);
485 ret = read_exec(&pcol);
489 return read_exec(&pcol);
492 static int _readpage(struct page *page, bool read_4_write)
494 struct page_collect pcol;
497 _pcol_init(&pcol, 1, page->mapping->host);
499 pcol.read_4_write = read_4_write;
500 ret = readpage_strip(&pcol, page);
502 EXOFS_ERR("_readpage => %d\n", ret);
506 return read_exec(&pcol);
510 * We don't need the file
512 static int exofs_readpage(struct file *file, struct page *page)
514 return _readpage(page, false);
517 /* Callback for osd_write. All writes are asynchronous */
518 static void writepages_done(struct ore_io_state *ios, void *p)
520 struct page_collect *pcol = p;
525 int ret = ore_check_io(ios, &resid);
527 atomic_dec(&pcol->sbi->s_curr_pending);
530 good_bytes = pcol->length;
531 ret = PAGE_WAS_NOT_IN_IO;
533 good_bytes = pcol->length - resid;
535 if (good_bytes > pcol->ios->length)
536 good_bytes = pcol->ios->length;
538 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
539 " length=0x%lx nr_pages=%u\n",
540 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
543 for (i = 0; i < pcol->nr_pages; i++) {
544 struct page *page = pcol->pages[i];
545 struct inode *inode = page->mapping->host;
548 if (inode != pcol->inode)
549 continue; /* osd might add more pages to a bio */
551 if (likely(length < good_bytes))
556 update_write_page(page, page_stat);
558 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
559 inode->i_ino, page->index, page_stat);
566 EXOFS_DBGMSG2("writepages_done END\n");
569 static int write_exec(struct page_collect *pcol)
571 struct exofs_i_info *oi = exofs_i(pcol->inode);
572 struct ore_io_state *ios;
573 struct page_collect *pcol_copy = NULL;
580 ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, false,
581 pcol->pg_first << PAGE_CACHE_SHIFT,
582 pcol->length, &pcol->ios);
586 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
588 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
596 ios->pages = pcol_copy->pages;
597 ios->done = writepages_done;
598 ios->private = pcol_copy;
600 /* pages ownership was passed to pcol_copy */
603 ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
607 EXOFS_DBGMSG2("write_exec(0x%lx) offset=0x%llx length=0x%llx\n",
608 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
610 ret = ore_write(ios);
612 EXOFS_ERR("write_exec: ore_write() Failed\n");
616 atomic_inc(&pcol->sbi->s_curr_pending);
620 _unlock_pcol_pages(pcol, ret, WRITE);
627 /* writepage_strip is called either directly from writepage() or by the VFS from
628 * within write_cache_pages(), to add one more page to be written to storage.
629 * It will try to collect as many contiguous pages as possible. If a
630 * discontinuity is encountered or it runs out of resources it will submit the
631 * previous segment and will start a new collection.
632 * Eventually caller must submit the last segment if present.
634 static int writepage_strip(struct page *page,
635 struct writeback_control *wbc_unused, void *data)
637 struct page_collect *pcol = data;
638 struct inode *inode = pcol->inode;
639 struct exofs_i_info *oi = exofs_i(inode);
640 loff_t i_size = i_size_read(inode);
641 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
645 BUG_ON(!PageLocked(page));
647 ret = wait_obj_created(oi);
651 if (page->index < end_index)
652 /* in this case, the page is within the limits of the file */
653 len = PAGE_CACHE_SIZE;
655 len = i_size & ~PAGE_CACHE_MASK;
657 if (page->index > end_index || !len) {
658 /* in this case, the page is outside the limits
659 * (truncate in progress)
661 ret = write_exec(pcol);
665 ClearPageError(page);
667 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
668 "outside the limits\n",
669 inode->i_ino, page->index);
676 if (unlikely(pcol->pg_first == -1)) {
677 pcol->pg_first = page->index;
678 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
680 /* Discontinuity detected, split the request */
681 ret = write_exec(pcol);
685 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
686 inode->i_ino, page->index);
691 ret = pcol_try_alloc(pcol);
696 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
697 inode->i_ino, page->index, len);
699 ret = pcol_add_page(pcol, page, len);
701 EXOFS_DBGMSG2("Failed pcol_add_page "
702 "nr_pages=%u total_length=0x%lx\n",
703 pcol->nr_pages, pcol->length);
705 /* split the request, next loop will start again */
706 ret = write_exec(pcol);
708 EXOFS_DBGMSG("write_exec failed => %d", ret);
715 BUG_ON(PageWriteback(page));
716 set_page_writeback(page);
721 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
722 inode->i_ino, page->index, ret);
723 set_bit(AS_EIO, &page->mapping->flags);
728 static int exofs_writepages(struct address_space *mapping,
729 struct writeback_control *wbc)
731 struct page_collect pcol;
732 long start, end, expected_pages;
735 start = wbc->range_start >> PAGE_CACHE_SHIFT;
736 end = (wbc->range_end == LLONG_MAX) ?
737 start + mapping->nrpages :
738 wbc->range_end >> PAGE_CACHE_SHIFT;
741 expected_pages = end - start + 1;
743 expected_pages = mapping->nrpages;
745 if (expected_pages < 32L)
746 expected_pages = 32L;
748 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
749 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
750 mapping->host->i_ino, wbc->range_start, wbc->range_end,
751 mapping->nrpages, start, end, expected_pages);
753 _pcol_init(&pcol, expected_pages, mapping->host);
755 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
757 EXOFS_ERR("write_cache_pages => %d\n", ret);
761 ret = write_exec(&pcol);
765 if (wbc->sync_mode == WB_SYNC_ALL) {
766 return write_exec(&pcol); /* pump the last reminder */
767 } else if (pcol.nr_pages) {
768 /* not SYNC let the reminder join the next writeout */
771 for (i = 0; i < pcol.nr_pages; i++) {
772 struct page *page = pcol.pages[i];
774 end_page_writeback(page);
775 set_page_dirty(page);
782 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
784 struct page_collect pcol;
787 _pcol_init(&pcol, 1, page->mapping->host);
789 ret = writepage_strip(page, NULL, &pcol);
791 EXOFS_ERR("exofs_writepage => %d\n", ret);
795 return write_exec(&pcol);
798 /* i_mutex held using inode->i_size directly */
799 static void _write_failed(struct inode *inode, loff_t to)
801 if (to > inode->i_size)
802 truncate_pagecache(inode, to, inode->i_size);
805 int exofs_write_begin(struct file *file, struct address_space *mapping,
806 loff_t pos, unsigned len, unsigned flags,
807 struct page **pagep, void **fsdata)
814 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
817 EXOFS_DBGMSG("simple_write_begin failed\n");
824 /* read modify write */
825 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
826 loff_t i_size = i_size_read(mapping->host);
827 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
830 if (page->index < end_index)
831 rlen = PAGE_CACHE_SIZE;
832 else if (page->index == end_index)
833 rlen = i_size & ~PAGE_CACHE_MASK;
838 clear_highpage(page);
839 SetPageUptodate(page);
843 ret = _readpage(page, true);
845 /*SetPageError was done by _readpage. Is it ok?*/
847 EXOFS_DBGMSG("__readpage failed\n");
852 _write_failed(mapping->host, pos + len);
857 static int exofs_write_begin_export(struct file *file,
858 struct address_space *mapping,
859 loff_t pos, unsigned len, unsigned flags,
860 struct page **pagep, void **fsdata)
864 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
868 static int exofs_write_end(struct file *file, struct address_space *mapping,
869 loff_t pos, unsigned len, unsigned copied,
870 struct page *page, void *fsdata)
872 struct inode *inode = mapping->host;
873 /* According to comment in simple_write_end i_mutex is held */
874 loff_t i_size = inode->i_size;
877 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
879 _write_failed(inode, pos + len);
881 /* TODO: once simple_write_end marks inode dirty remove */
882 if (i_size != inode->i_size)
883 mark_inode_dirty(inode);
887 static int exofs_releasepage(struct page *page, gfp_t gfp)
889 EXOFS_DBGMSG("page 0x%lx\n", page->index);
894 static void exofs_invalidatepage(struct page *page, unsigned long offset)
896 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset);
900 const struct address_space_operations exofs_aops = {
901 .readpage = exofs_readpage,
902 .readpages = exofs_readpages,
903 .writepage = exofs_writepage,
904 .writepages = exofs_writepages,
905 .write_begin = exofs_write_begin_export,
906 .write_end = exofs_write_end,
907 .releasepage = exofs_releasepage,
908 .set_page_dirty = __set_page_dirty_nobuffers,
909 .invalidatepage = exofs_invalidatepage,
911 /* Not implemented Yet */
912 .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
913 .direct_IO = NULL, /* TODO: Should be trivial to do */
915 /* With these NULL has special meaning or default is not exported */
918 .launder_page = NULL,
919 .is_partially_uptodate = NULL,
920 .error_remove_page = NULL,
923 /******************************************************************************
925 *****************************************************************************/
928 * Test whether an inode is a fast symlink.
930 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
932 struct exofs_i_info *oi = exofs_i(inode);
934 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
937 static int _do_truncate(struct inode *inode, loff_t newsize)
939 struct exofs_i_info *oi = exofs_i(inode);
940 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
943 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
945 ret = ore_truncate(&sbi->layout, &oi->oc, (u64)newsize);
947 truncate_setsize(inode, newsize);
949 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
950 inode->i_ino, newsize, ret);
955 * Set inode attributes - update size attribute on OSD if needed,
956 * otherwise just call generic functions.
958 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
960 struct inode *inode = dentry->d_inode;
963 /* if we are about to modify an object, and it hasn't been
966 error = wait_obj_created(exofs_i(inode));
970 error = inode_change_ok(inode, iattr);
974 if ((iattr->ia_valid & ATTR_SIZE) &&
975 iattr->ia_size != i_size_read(inode)) {
976 error = _do_truncate(inode, iattr->ia_size);
981 setattr_copy(inode, iattr);
982 mark_inode_dirty(inode);
986 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
988 EXOFS_ATTR_INODE_FILE_LAYOUT,
990 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
992 EXOFS_ATTR_INODE_DIR_LAYOUT,
996 * Read the Linux inode info from the OSD, and return it as is. In exofs the
997 * inode info is in an application specific page/attribute of the osd-object.
999 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
1000 struct exofs_fcb *inode)
1002 struct exofs_sb_info *sbi = sb->s_fs_info;
1003 struct osd_attr attrs[] = {
1004 [0] = g_attr_inode_data,
1005 [1] = g_attr_inode_file_layout,
1006 [2] = g_attr_inode_dir_layout,
1008 struct ore_io_state *ios;
1009 struct exofs_on_disk_inode_layout *layout;
1012 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1013 if (unlikely(ret)) {
1014 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1018 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1019 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1021 ios->in_attr = attrs;
1022 ios->in_attr_len = ARRAY_SIZE(attrs);
1024 ret = ore_read(ios);
1025 if (unlikely(ret)) {
1026 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
1027 _LLU(oi->one_comp.obj.id), ret);
1028 memset(inode, 0, sizeof(*inode));
1029 inode->i_mode = 0040000 | (0777 & ~022);
1030 /* If object is lost on target we might as well enable it's
1033 if ((ret == -ENOENT) || (ret == -EINVAL))
1038 ret = extract_attr_from_ios(ios, &attrs[0]);
1040 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1043 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
1044 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
1046 ret = extract_attr_from_ios(ios, &attrs[1]);
1048 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1052 layout = attrs[1].val_ptr;
1053 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1054 EXOFS_ERR("%s: unsupported files layout %d\n",
1055 __func__, layout->gen_func);
1061 ret = extract_attr_from_ios(ios, &attrs[2]);
1063 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1067 layout = attrs[2].val_ptr;
1068 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1069 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
1070 __func__, layout->gen_func);
1077 ore_put_io_state(ios);
1081 static void __oi_init(struct exofs_i_info *oi)
1083 init_waitqueue_head(&oi->i_wq);
1087 * Fill in an inode read from the OSD and set it up for use
1089 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1091 struct exofs_i_info *oi;
1092 struct exofs_fcb fcb;
1093 struct inode *inode;
1096 inode = iget_locked(sb, ino);
1098 return ERR_PTR(-ENOMEM);
1099 if (!(inode->i_state & I_NEW))
1101 oi = exofs_i(inode);
1103 exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1104 exofs_oi_objno(oi));
1106 /* read the inode from the osd */
1107 ret = exofs_get_inode(sb, oi, &fcb);
1111 set_obj_created(oi);
1113 /* copy stuff from on-disk struct to in-memory struct */
1114 inode->i_mode = le16_to_cpu(fcb.i_mode);
1115 inode->i_uid = le32_to_cpu(fcb.i_uid);
1116 inode->i_gid = le32_to_cpu(fcb.i_gid);
1117 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1118 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1119 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1120 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1121 inode->i_ctime.tv_nsec =
1122 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1123 oi->i_commit_size = le64_to_cpu(fcb.i_size);
1124 i_size_write(inode, oi->i_commit_size);
1125 inode->i_blkbits = EXOFS_BLKSHIFT;
1126 inode->i_generation = le32_to_cpu(fcb.i_generation);
1128 oi->i_dir_start_lookup = 0;
1130 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1135 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1138 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1141 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1143 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1146 inode->i_mapping->backing_dev_info = sb->s_bdi;
1147 if (S_ISREG(inode->i_mode)) {
1148 inode->i_op = &exofs_file_inode_operations;
1149 inode->i_fop = &exofs_file_operations;
1150 inode->i_mapping->a_ops = &exofs_aops;
1151 } else if (S_ISDIR(inode->i_mode)) {
1152 inode->i_op = &exofs_dir_inode_operations;
1153 inode->i_fop = &exofs_dir_operations;
1154 inode->i_mapping->a_ops = &exofs_aops;
1155 } else if (S_ISLNK(inode->i_mode)) {
1156 if (exofs_inode_is_fast_symlink(inode))
1157 inode->i_op = &exofs_fast_symlink_inode_operations;
1159 inode->i_op = &exofs_symlink_inode_operations;
1160 inode->i_mapping->a_ops = &exofs_aops;
1163 inode->i_op = &exofs_special_inode_operations;
1165 init_special_inode(inode, inode->i_mode,
1166 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1168 init_special_inode(inode, inode->i_mode,
1169 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1172 unlock_new_inode(inode);
1177 return ERR_PTR(ret);
1180 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1182 if (!obj_created(oi)) {
1183 EXOFS_DBGMSG("!obj_created\n");
1184 BUG_ON(!obj_2bcreated(oi));
1185 wait_event(oi->i_wq, obj_created(oi));
1186 EXOFS_DBGMSG("wait_event done\n");
1188 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1192 * Callback function from exofs_new_inode(). The important thing is that we
1193 * set the obj_created flag so that other methods know that the object exists on
1196 static void create_done(struct ore_io_state *ios, void *p)
1198 struct inode *inode = p;
1199 struct exofs_i_info *oi = exofs_i(inode);
1200 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1203 ret = ore_check_io(ios, NULL);
1204 ore_put_io_state(ios);
1206 atomic_dec(&sbi->s_curr_pending);
1208 if (unlikely(ret)) {
1209 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1210 _LLU(exofs_oi_objno(oi)),
1211 _LLU(oi->one_comp.obj.partition));
1212 /*TODO: When FS is corrupted creation can fail, object already
1213 * exist. Get rid of this asynchronous creation, if exist
1214 * increment the obj counter and try the next object. Until we
1215 * succeed. All these dangling objects will be made into lost
1216 * files by chkfs.exofs
1220 set_obj_created(oi);
1226 * Set up a new inode and create an object for it on the OSD
1228 struct inode *exofs_new_inode(struct inode *dir, int mode)
1230 struct super_block *sb = dir->i_sb;
1231 struct exofs_sb_info *sbi = sb->s_fs_info;
1232 struct inode *inode;
1233 struct exofs_i_info *oi;
1234 struct ore_io_state *ios;
1237 inode = new_inode(sb);
1239 return ERR_PTR(-ENOMEM);
1241 oi = exofs_i(inode);
1244 set_obj_2bcreated(oi);
1246 inode->i_mapping->backing_dev_info = sb->s_bdi;
1247 inode_init_owner(inode, dir, mode);
1248 inode->i_ino = sbi->s_nextid++;
1249 inode->i_blkbits = EXOFS_BLKSHIFT;
1250 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1251 oi->i_commit_size = inode->i_size = 0;
1252 spin_lock(&sbi->s_next_gen_lock);
1253 inode->i_generation = sbi->s_next_generation++;
1254 spin_unlock(&sbi->s_next_gen_lock);
1255 insert_inode_hash(inode);
1257 exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1258 exofs_oi_objno(oi));
1259 exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */
1261 mark_inode_dirty(inode);
1263 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1264 if (unlikely(ret)) {
1265 EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n");
1266 return ERR_PTR(ret);
1269 ios->done = create_done;
1270 ios->private = inode;
1272 ret = ore_create(ios);
1274 ore_put_io_state(ios);
1275 return ERR_PTR(ret);
1277 atomic_inc(&sbi->s_curr_pending);
1283 * struct to pass two arguments to update_inode's callback
1285 struct updatei_args {
1286 struct exofs_sb_info *sbi;
1287 struct exofs_fcb fcb;
1291 * Callback function from exofs_update_inode().
1293 static void updatei_done(struct ore_io_state *ios, void *p)
1295 struct updatei_args *args = p;
1297 ore_put_io_state(ios);
1299 atomic_dec(&args->sbi->s_curr_pending);
1305 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1306 * synchronously or asynchronously depending on the do_sync flag.
1308 static int exofs_update_inode(struct inode *inode, int do_sync)
1310 struct exofs_i_info *oi = exofs_i(inode);
1311 struct super_block *sb = inode->i_sb;
1312 struct exofs_sb_info *sbi = sb->s_fs_info;
1313 struct ore_io_state *ios;
1314 struct osd_attr attr;
1315 struct exofs_fcb *fcb;
1316 struct updatei_args *args;
1319 args = kzalloc(sizeof(*args), GFP_KERNEL);
1321 EXOFS_DBGMSG("Failed kzalloc of args\n");
1327 fcb->i_mode = cpu_to_le16(inode->i_mode);
1328 fcb->i_uid = cpu_to_le32(inode->i_uid);
1329 fcb->i_gid = cpu_to_le32(inode->i_gid);
1330 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1331 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1332 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1333 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1334 oi->i_commit_size = i_size_read(inode);
1335 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1336 fcb->i_generation = cpu_to_le32(inode->i_generation);
1338 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1339 if (old_valid_dev(inode->i_rdev)) {
1341 cpu_to_le32(old_encode_dev(inode->i_rdev));
1346 cpu_to_le32(new_encode_dev(inode->i_rdev));
1350 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1352 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1353 if (unlikely(ret)) {
1354 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1358 attr = g_attr_inode_data;
1360 ios->out_attr_len = 1;
1361 ios->out_attr = &attr;
1363 wait_obj_created(oi);
1367 ios->done = updatei_done;
1368 ios->private = args;
1371 ret = ore_write(ios);
1372 if (!do_sync && !ret) {
1373 atomic_inc(&sbi->s_curr_pending);
1374 goto out; /* deallocation in updatei_done */
1377 ore_put_io_state(ios);
1381 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1382 inode->i_ino, do_sync, ret);
1386 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1388 /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1389 return exofs_update_inode(inode, 1);
1393 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1396 static void delete_done(struct ore_io_state *ios, void *p)
1398 struct exofs_sb_info *sbi = p;
1400 ore_put_io_state(ios);
1402 atomic_dec(&sbi->s_curr_pending);
1406 * Called when the refcount of an inode reaches zero. We remove the object
1407 * from the OSD here. We make sure the object was created before we try and
1410 void exofs_evict_inode(struct inode *inode)
1412 struct exofs_i_info *oi = exofs_i(inode);
1413 struct super_block *sb = inode->i_sb;
1414 struct exofs_sb_info *sbi = sb->s_fs_info;
1415 struct ore_io_state *ios;
1418 truncate_inode_pages(&inode->i_data, 0);
1420 /* TODO: should do better here */
1421 if (inode->i_nlink || is_bad_inode(inode))
1425 end_writeback(inode);
1427 /* if we are deleting an obj that hasn't been created yet, wait.
1428 * This also makes sure that create_done cannot be called with an
1429 * already evicted inode.
1431 wait_obj_created(oi);
1432 /* ignore the error, attempt a remove anyway */
1434 /* Now Remove the OSD objects */
1435 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1436 if (unlikely(ret)) {
1437 EXOFS_ERR("%s: ore_get_io_state failed\n", __func__);
1441 ios->done = delete_done;
1444 ret = ore_remove(ios);
1446 EXOFS_ERR("%s: ore_remove failed\n", __func__);
1447 ore_put_io_state(ios);
1450 atomic_inc(&sbi->s_curr_pending);
1455 end_writeback(inode);