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 { BIO_MAX_PAGES_KMALLOC =
41 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
43 PAGE_SIZE / sizeof(struct page *),
46 unsigned exofs_max_io_pages(struct ore_layout *layout,
47 unsigned expected_pages)
49 unsigned pages = min_t(unsigned, expected_pages, MAX_PAGES_KMALLOC);
51 /* TODO: easily support bio chaining */
52 pages = min_t(unsigned, pages,
53 layout->group_width * BIO_MAX_PAGES_KMALLOC);
58 struct exofs_sb_info *sbi;
60 unsigned expected_pages;
61 struct ore_io_state *ios;
67 loff_t pg_first; /* keep 64bit also in 32-arches */
68 bool read_4_write; /* This means two things: that the read is sync
69 * And the pages should not be unlocked.
73 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
76 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
80 pcol->expected_pages = expected_pages;
84 pcol->alloc_pages = 0;
88 pcol->read_4_write = false;
91 static void _pcol_reset(struct page_collect *pcol)
93 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
96 pcol->alloc_pages = 0;
102 /* this is probably the end of the loop but in writes
103 * it might not end here. don't be left with nothing
105 if (!pcol->expected_pages)
106 pcol->expected_pages = MAX_PAGES_KMALLOC;
109 static int pcol_try_alloc(struct page_collect *pcol)
113 /* TODO: easily support bio chaining */
114 pages = exofs_max_io_pages(&pcol->sbi->layout, pcol->expected_pages);
116 for (; pages; pages >>= 1) {
117 pcol->pages = kmalloc(pages * sizeof(struct page *),
119 if (likely(pcol->pages)) {
120 pcol->alloc_pages = pages;
125 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
126 pcol->expected_pages);
130 static void pcol_free(struct page_collect *pcol)
136 ore_put_io_state(pcol->ios);
141 static int pcol_add_page(struct page_collect *pcol, struct page *page,
144 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
147 pcol->pages[pcol->nr_pages++] = page;
152 static int update_read_page(struct page *page, int ret)
155 /* Everything is OK */
156 SetPageUptodate(page);
158 ClearPageError(page);
159 } else if (ret == -EFAULT) {
160 /* In this case we were trying to read something that wasn't on
161 * disk yet - return a page full of zeroes. This should be OK,
162 * because the object should be empty (if there was a write
163 * before this read, the read would be waiting with the page
165 clear_highpage(page);
167 SetPageUptodate(page);
169 ClearPageError(page);
170 ret = 0; /* recovered error */
171 EXOFS_DBGMSG("recovered read error\n");
178 static void update_write_page(struct page *page, int ret)
181 mapping_set_error(page->mapping, ret);
184 end_page_writeback(page);
187 /* Called at the end of reads, to optionally unlock pages and update their
190 static int __readpages_done(struct page_collect *pcol)
196 int ret = ore_check_io(pcol->ios, &resid);
199 good_bytes = pcol->length;
201 good_bytes = pcol->length - resid;
203 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
204 " length=0x%lx nr_pages=%u\n",
205 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
208 for (i = 0; i < pcol->nr_pages; i++) {
209 struct page *page = pcol->pages[i];
210 struct inode *inode = page->mapping->host;
213 if (inode != pcol->inode)
214 continue; /* osd might add more pages at end */
216 if (likely(length < good_bytes))
221 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
222 inode->i_ino, page->index,
223 page_stat ? "bad_bytes" : "good_bytes");
225 ret = update_read_page(page, page_stat);
226 if (!pcol->read_4_write)
232 EXOFS_DBGMSG2("readpages_done END\n");
236 /* callback of async reads */
237 static void readpages_done(struct ore_io_state *ios, void *p)
239 struct page_collect *pcol = p;
241 __readpages_done(pcol);
242 atomic_dec(&pcol->sbi->s_curr_pending);
246 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
250 for (i = 0; i < pcol->nr_pages; i++) {
251 struct page *page = pcol->pages[i];
254 update_read_page(page, ret);
256 update_write_page(page, ret);
262 static int _maybe_not_all_in_one_io(struct ore_io_state *ios,
263 struct page_collect *pcol_src, struct page_collect *pcol)
265 /* length was wrong or offset was not page aligned */
266 BUG_ON(pcol_src->nr_pages < ios->nr_pages);
268 if (pcol_src->nr_pages > ios->nr_pages) {
269 struct page **src_page;
270 unsigned pages_less = pcol_src->nr_pages - ios->nr_pages;
271 unsigned long len_less = pcol_src->length - ios->length;
275 /* This IO was trimmed */
276 pcol_src->nr_pages = ios->nr_pages;
277 pcol_src->length = ios->length;
279 /* Left over pages are passed to the next io */
280 pcol->expected_pages += pages_less;
281 pcol->nr_pages = pages_less;
282 pcol->length = len_less;
283 src_page = pcol_src->pages + pcol_src->nr_pages;
284 pcol->pg_first = (*src_page)->index;
286 ret = pcol_try_alloc(pcol);
290 for (i = 0; i < pages_less; ++i)
291 pcol->pages[i] = *src_page++;
293 EXOFS_DBGMSG("Length was adjusted nr_pages=0x%x "
294 "pages_less=0x%x expected_pages=0x%x "
295 "next_offset=0x%llx next_len=0x%lx\n",
296 pcol_src->nr_pages, pages_less, pcol->expected_pages,
297 pcol->pg_first * PAGE_SIZE, pcol->length);
302 static int read_exec(struct page_collect *pcol)
304 struct exofs_i_info *oi = exofs_i(pcol->inode);
305 struct ore_io_state *ios;
306 struct page_collect *pcol_copy = NULL;
313 int ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, true,
314 pcol->pg_first << PAGE_CACHE_SHIFT,
315 pcol->length, &pcol->ios);
322 ios->pages = pcol->pages;
324 if (pcol->read_4_write) {
326 return __readpages_done(pcol);
329 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
336 ios->done = readpages_done;
337 ios->private = pcol_copy;
339 /* pages ownership was passed to pcol_copy */
342 ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
346 EXOFS_DBGMSG2("read_exec(0x%lx) offset=0x%llx length=0x%llx\n",
347 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
353 atomic_inc(&pcol->sbi->s_curr_pending);
358 if (!pcol->read_4_write)
359 _unlock_pcol_pages(pcol, ret, READ);
367 /* readpage_strip is called either directly from readpage() or by the VFS from
368 * within read_cache_pages(), to add one more page to be read. It will try to
369 * collect as many contiguous pages as posible. If a discontinuity is
370 * encountered, or it runs out of resources, it will submit the previous segment
371 * and will start a new collection. Eventually caller must submit the last
372 * segment if present.
374 static int readpage_strip(void *data, struct page *page)
376 struct page_collect *pcol = data;
377 struct inode *inode = pcol->inode;
378 struct exofs_i_info *oi = exofs_i(inode);
379 loff_t i_size = i_size_read(inode);
380 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
384 /* FIXME: Just for debugging, will be removed */
385 if (PageUptodate(page))
386 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
389 if (page->index < end_index)
390 len = PAGE_CACHE_SIZE;
391 else if (page->index == end_index)
392 len = i_size & ~PAGE_CACHE_MASK;
396 if (!len || !obj_created(oi)) {
397 /* this will be out of bounds, or doesn't exist yet.
398 * Current page is cleared and the request is split
400 clear_highpage(page);
402 SetPageUptodate(page);
404 ClearPageError(page);
406 if (!pcol->read_4_write)
408 EXOFS_DBGMSG("readpage_strip(0x%lx) empty page len=%zx "
409 "read_4_write=%d index=0x%lx end_index=0x%lx "
410 "splitting\n", inode->i_ino, len,
411 pcol->read_4_write, page->index, end_index);
413 return read_exec(pcol);
418 if (unlikely(pcol->pg_first == -1)) {
419 pcol->pg_first = page->index;
420 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
422 /* Discontinuity detected, split the request */
423 ret = read_exec(pcol);
430 ret = pcol_try_alloc(pcol);
435 if (len != PAGE_CACHE_SIZE)
436 zero_user(page, len, PAGE_CACHE_SIZE - len);
438 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
439 inode->i_ino, page->index, len);
441 ret = pcol_add_page(pcol, page, len);
443 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
444 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
445 page, len, pcol->nr_pages, pcol->length);
447 /* split the request, and start again with current page */
448 ret = read_exec(pcol);
458 /* SetPageError(page); ??? */
463 static int exofs_readpages(struct file *file, struct address_space *mapping,
464 struct list_head *pages, unsigned nr_pages)
466 struct page_collect pcol;
469 _pcol_init(&pcol, nr_pages, mapping->host);
471 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
473 EXOFS_ERR("read_cache_pages => %d\n", ret);
477 ret = read_exec(&pcol);
481 return read_exec(&pcol);
484 static int _readpage(struct page *page, bool read_4_write)
486 struct page_collect pcol;
489 _pcol_init(&pcol, 1, page->mapping->host);
491 pcol.read_4_write = read_4_write;
492 ret = readpage_strip(&pcol, page);
494 EXOFS_ERR("_readpage => %d\n", ret);
498 return read_exec(&pcol);
502 * We don't need the file
504 static int exofs_readpage(struct file *file, struct page *page)
506 return _readpage(page, false);
509 /* Callback for osd_write. All writes are asynchronous */
510 static void writepages_done(struct ore_io_state *ios, void *p)
512 struct page_collect *pcol = p;
517 int ret = ore_check_io(ios, &resid);
519 atomic_dec(&pcol->sbi->s_curr_pending);
522 good_bytes = pcol->length;
524 good_bytes = pcol->length - resid;
526 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
527 " length=0x%lx nr_pages=%u\n",
528 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
531 for (i = 0; i < pcol->nr_pages; i++) {
532 struct page *page = pcol->pages[i];
533 struct inode *inode = page->mapping->host;
536 if (inode != pcol->inode)
537 continue; /* osd might add more pages to a bio */
539 if (likely(length < good_bytes))
544 update_write_page(page, page_stat);
546 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
547 inode->i_ino, page->index, page_stat);
554 EXOFS_DBGMSG2("writepages_done END\n");
557 static int write_exec(struct page_collect *pcol)
559 struct exofs_i_info *oi = exofs_i(pcol->inode);
560 struct ore_io_state *ios;
561 struct page_collect *pcol_copy = NULL;
568 ret = ore_get_rw_state(&pcol->sbi->layout, &oi->oc, false,
569 pcol->pg_first << PAGE_CACHE_SHIFT,
570 pcol->length, &pcol->ios);
574 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
576 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
584 ios->pages = pcol_copy->pages;
585 ios->done = writepages_done;
586 ios->private = pcol_copy;
588 /* pages ownership was passed to pcol_copy */
591 ret = _maybe_not_all_in_one_io(ios, pcol_copy, pcol);
595 EXOFS_DBGMSG2("write_exec(0x%lx) offset=0x%llx length=0x%llx\n",
596 pcol->inode->i_ino, _LLU(ios->offset), _LLU(ios->length));
598 ret = ore_write(ios);
600 EXOFS_ERR("write_exec: ore_write() Failed\n");
604 atomic_inc(&pcol->sbi->s_curr_pending);
608 _unlock_pcol_pages(pcol, ret, WRITE);
615 /* writepage_strip is called either directly from writepage() or by the VFS from
616 * within write_cache_pages(), to add one more page to be written to storage.
617 * It will try to collect as many contiguous pages as possible. If a
618 * discontinuity is encountered or it runs out of resources it will submit the
619 * previous segment and will start a new collection.
620 * Eventually caller must submit the last segment if present.
622 static int writepage_strip(struct page *page,
623 struct writeback_control *wbc_unused, void *data)
625 struct page_collect *pcol = data;
626 struct inode *inode = pcol->inode;
627 struct exofs_i_info *oi = exofs_i(inode);
628 loff_t i_size = i_size_read(inode);
629 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
633 BUG_ON(!PageLocked(page));
635 ret = wait_obj_created(oi);
639 if (page->index < end_index)
640 /* in this case, the page is within the limits of the file */
641 len = PAGE_CACHE_SIZE;
643 len = i_size & ~PAGE_CACHE_MASK;
645 if (page->index > end_index || !len) {
646 /* in this case, the page is outside the limits
647 * (truncate in progress)
649 ret = write_exec(pcol);
653 ClearPageError(page);
655 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
656 "outside the limits\n",
657 inode->i_ino, page->index);
664 if (unlikely(pcol->pg_first == -1)) {
665 pcol->pg_first = page->index;
666 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
668 /* Discontinuity detected, split the request */
669 ret = write_exec(pcol);
673 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
674 inode->i_ino, page->index);
679 ret = pcol_try_alloc(pcol);
684 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
685 inode->i_ino, page->index, len);
687 ret = pcol_add_page(pcol, page, len);
689 EXOFS_DBGMSG2("Failed pcol_add_page "
690 "nr_pages=%u total_length=0x%lx\n",
691 pcol->nr_pages, pcol->length);
693 /* split the request, next loop will start again */
694 ret = write_exec(pcol);
696 EXOFS_DBGMSG("write_exec failed => %d", ret);
703 BUG_ON(PageWriteback(page));
704 set_page_writeback(page);
709 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
710 inode->i_ino, page->index, ret);
711 set_bit(AS_EIO, &page->mapping->flags);
716 static int exofs_writepages(struct address_space *mapping,
717 struct writeback_control *wbc)
719 struct page_collect pcol;
720 long start, end, expected_pages;
723 start = wbc->range_start >> PAGE_CACHE_SHIFT;
724 end = (wbc->range_end == LLONG_MAX) ?
725 start + mapping->nrpages :
726 wbc->range_end >> PAGE_CACHE_SHIFT;
729 expected_pages = end - start + 1;
731 expected_pages = mapping->nrpages;
733 if (expected_pages < 32L)
734 expected_pages = 32L;
736 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
737 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
738 mapping->host->i_ino, wbc->range_start, wbc->range_end,
739 mapping->nrpages, start, end, expected_pages);
741 _pcol_init(&pcol, expected_pages, mapping->host);
743 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
745 EXOFS_ERR("write_cache_pages => %d\n", ret);
749 ret = write_exec(&pcol);
753 if (wbc->sync_mode == WB_SYNC_ALL) {
754 return write_exec(&pcol); /* pump the last reminder */
755 } else if (pcol.nr_pages) {
756 /* not SYNC let the reminder join the next writeout */
759 for (i = 0; i < pcol.nr_pages; i++) {
760 struct page *page = pcol.pages[i];
762 end_page_writeback(page);
763 set_page_dirty(page);
770 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
772 struct page_collect pcol;
775 _pcol_init(&pcol, 1, page->mapping->host);
777 ret = writepage_strip(page, NULL, &pcol);
779 EXOFS_ERR("exofs_writepage => %d\n", ret);
783 return write_exec(&pcol);
786 /* i_mutex held using inode->i_size directly */
787 static void _write_failed(struct inode *inode, loff_t to)
789 if (to > inode->i_size)
790 truncate_pagecache(inode, to, inode->i_size);
793 int exofs_write_begin(struct file *file, struct address_space *mapping,
794 loff_t pos, unsigned len, unsigned flags,
795 struct page **pagep, void **fsdata)
802 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
805 EXOFS_DBGMSG("simple_write_begin failed\n");
812 /* read modify write */
813 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
814 loff_t i_size = i_size_read(mapping->host);
815 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
818 if (page->index < end_index)
819 rlen = PAGE_CACHE_SIZE;
820 else if (page->index == end_index)
821 rlen = i_size & ~PAGE_CACHE_MASK;
826 clear_highpage(page);
827 SetPageUptodate(page);
831 ret = _readpage(page, true);
833 /*SetPageError was done by _readpage. Is it ok?*/
835 EXOFS_DBGMSG("__readpage failed\n");
840 _write_failed(mapping->host, pos + len);
845 static int exofs_write_begin_export(struct file *file,
846 struct address_space *mapping,
847 loff_t pos, unsigned len, unsigned flags,
848 struct page **pagep, void **fsdata)
852 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
856 static int exofs_write_end(struct file *file, struct address_space *mapping,
857 loff_t pos, unsigned len, unsigned copied,
858 struct page *page, void *fsdata)
860 struct inode *inode = mapping->host;
861 /* According to comment in simple_write_end i_mutex is held */
862 loff_t i_size = inode->i_size;
865 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
867 _write_failed(inode, pos + len);
869 /* TODO: once simple_write_end marks inode dirty remove */
870 if (i_size != inode->i_size)
871 mark_inode_dirty(inode);
875 static int exofs_releasepage(struct page *page, gfp_t gfp)
877 EXOFS_DBGMSG("page 0x%lx\n", page->index);
882 static void exofs_invalidatepage(struct page *page, unsigned long offset)
884 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset);
888 const struct address_space_operations exofs_aops = {
889 .readpage = exofs_readpage,
890 .readpages = exofs_readpages,
891 .writepage = exofs_writepage,
892 .writepages = exofs_writepages,
893 .write_begin = exofs_write_begin_export,
894 .write_end = exofs_write_end,
895 .releasepage = exofs_releasepage,
896 .set_page_dirty = __set_page_dirty_nobuffers,
897 .invalidatepage = exofs_invalidatepage,
899 /* Not implemented Yet */
900 .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
901 .direct_IO = NULL, /* TODO: Should be trivial to do */
903 /* With these NULL has special meaning or default is not exported */
906 .launder_page = NULL,
907 .is_partially_uptodate = NULL,
908 .error_remove_page = NULL,
911 /******************************************************************************
913 *****************************************************************************/
916 * Test whether an inode is a fast symlink.
918 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
920 struct exofs_i_info *oi = exofs_i(inode);
922 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
925 static int _do_truncate(struct inode *inode, loff_t newsize)
927 struct exofs_i_info *oi = exofs_i(inode);
928 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
931 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
933 ret = ore_truncate(&sbi->layout, &oi->oc, (u64)newsize);
935 truncate_setsize(inode, newsize);
937 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
938 inode->i_ino, newsize, ret);
943 * Set inode attributes - update size attribute on OSD if needed,
944 * otherwise just call generic functions.
946 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
948 struct inode *inode = dentry->d_inode;
951 /* if we are about to modify an object, and it hasn't been
954 error = wait_obj_created(exofs_i(inode));
958 error = inode_change_ok(inode, iattr);
962 if ((iattr->ia_valid & ATTR_SIZE) &&
963 iattr->ia_size != i_size_read(inode)) {
964 error = _do_truncate(inode, iattr->ia_size);
969 setattr_copy(inode, iattr);
970 mark_inode_dirty(inode);
974 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
976 EXOFS_ATTR_INODE_FILE_LAYOUT,
978 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
980 EXOFS_ATTR_INODE_DIR_LAYOUT,
984 * Read the Linux inode info from the OSD, and return it as is. In exofs the
985 * inode info is in an application specific page/attribute of the osd-object.
987 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
988 struct exofs_fcb *inode)
990 struct exofs_sb_info *sbi = sb->s_fs_info;
991 struct osd_attr attrs[] = {
992 [0] = g_attr_inode_data,
993 [1] = g_attr_inode_file_layout,
994 [2] = g_attr_inode_dir_layout,
996 struct ore_io_state *ios;
997 struct exofs_on_disk_inode_layout *layout;
1000 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1001 if (unlikely(ret)) {
1002 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1006 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1007 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->oc.numdevs);
1009 ios->in_attr = attrs;
1010 ios->in_attr_len = ARRAY_SIZE(attrs);
1012 ret = ore_read(ios);
1013 if (unlikely(ret)) {
1014 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
1015 _LLU(oi->one_comp.obj.id), ret);
1016 memset(inode, 0, sizeof(*inode));
1017 inode->i_mode = 0040000 | (0777 & ~022);
1018 /* If object is lost on target we might as well enable it's
1021 if ((ret == -ENOENT) || (ret == -EINVAL))
1026 ret = extract_attr_from_ios(ios, &attrs[0]);
1028 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1031 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
1032 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
1034 ret = extract_attr_from_ios(ios, &attrs[1]);
1036 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1040 layout = attrs[1].val_ptr;
1041 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1042 EXOFS_ERR("%s: unsupported files layout %d\n",
1043 __func__, layout->gen_func);
1049 ret = extract_attr_from_ios(ios, &attrs[2]);
1051 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
1055 layout = attrs[2].val_ptr;
1056 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
1057 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
1058 __func__, layout->gen_func);
1065 ore_put_io_state(ios);
1069 static void __oi_init(struct exofs_i_info *oi)
1071 init_waitqueue_head(&oi->i_wq);
1075 * Fill in an inode read from the OSD and set it up for use
1077 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1079 struct exofs_i_info *oi;
1080 struct exofs_fcb fcb;
1081 struct inode *inode;
1084 inode = iget_locked(sb, ino);
1086 return ERR_PTR(-ENOMEM);
1087 if (!(inode->i_state & I_NEW))
1089 oi = exofs_i(inode);
1091 exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1092 exofs_oi_objno(oi));
1094 /* read the inode from the osd */
1095 ret = exofs_get_inode(sb, oi, &fcb);
1099 set_obj_created(oi);
1101 /* copy stuff from on-disk struct to in-memory struct */
1102 inode->i_mode = le16_to_cpu(fcb.i_mode);
1103 inode->i_uid = le32_to_cpu(fcb.i_uid);
1104 inode->i_gid = le32_to_cpu(fcb.i_gid);
1105 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1106 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1107 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1108 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1109 inode->i_ctime.tv_nsec =
1110 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1111 oi->i_commit_size = le64_to_cpu(fcb.i_size);
1112 i_size_write(inode, oi->i_commit_size);
1113 inode->i_blkbits = EXOFS_BLKSHIFT;
1114 inode->i_generation = le32_to_cpu(fcb.i_generation);
1116 oi->i_dir_start_lookup = 0;
1118 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1123 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1126 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1129 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1131 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1134 inode->i_mapping->backing_dev_info = sb->s_bdi;
1135 if (S_ISREG(inode->i_mode)) {
1136 inode->i_op = &exofs_file_inode_operations;
1137 inode->i_fop = &exofs_file_operations;
1138 inode->i_mapping->a_ops = &exofs_aops;
1139 } else if (S_ISDIR(inode->i_mode)) {
1140 inode->i_op = &exofs_dir_inode_operations;
1141 inode->i_fop = &exofs_dir_operations;
1142 inode->i_mapping->a_ops = &exofs_aops;
1143 } else if (S_ISLNK(inode->i_mode)) {
1144 if (exofs_inode_is_fast_symlink(inode))
1145 inode->i_op = &exofs_fast_symlink_inode_operations;
1147 inode->i_op = &exofs_symlink_inode_operations;
1148 inode->i_mapping->a_ops = &exofs_aops;
1151 inode->i_op = &exofs_special_inode_operations;
1153 init_special_inode(inode, inode->i_mode,
1154 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1156 init_special_inode(inode, inode->i_mode,
1157 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1160 unlock_new_inode(inode);
1165 return ERR_PTR(ret);
1168 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1170 if (!obj_created(oi)) {
1171 EXOFS_DBGMSG("!obj_created\n");
1172 BUG_ON(!obj_2bcreated(oi));
1173 wait_event(oi->i_wq, obj_created(oi));
1174 EXOFS_DBGMSG("wait_event done\n");
1176 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1180 * Callback function from exofs_new_inode(). The important thing is that we
1181 * set the obj_created flag so that other methods know that the object exists on
1184 static void create_done(struct ore_io_state *ios, void *p)
1186 struct inode *inode = p;
1187 struct exofs_i_info *oi = exofs_i(inode);
1188 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1191 ret = ore_check_io(ios, NULL);
1192 ore_put_io_state(ios);
1194 atomic_dec(&sbi->s_curr_pending);
1196 if (unlikely(ret)) {
1197 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1198 _LLU(exofs_oi_objno(oi)),
1199 _LLU(oi->one_comp.obj.partition));
1200 /*TODO: When FS is corrupted creation can fail, object already
1201 * exist. Get rid of this asynchronous creation, if exist
1202 * increment the obj counter and try the next object. Until we
1203 * succeed. All these dangling objects will be made into lost
1204 * files by chkfs.exofs
1208 set_obj_created(oi);
1214 * Set up a new inode and create an object for it on the OSD
1216 struct inode *exofs_new_inode(struct inode *dir, int mode)
1218 struct super_block *sb = dir->i_sb;
1219 struct exofs_sb_info *sbi = sb->s_fs_info;
1220 struct inode *inode;
1221 struct exofs_i_info *oi;
1222 struct ore_io_state *ios;
1225 inode = new_inode(sb);
1227 return ERR_PTR(-ENOMEM);
1229 oi = exofs_i(inode);
1232 set_obj_2bcreated(oi);
1234 inode->i_mapping->backing_dev_info = sb->s_bdi;
1235 inode_init_owner(inode, dir, mode);
1236 inode->i_ino = sbi->s_nextid++;
1237 inode->i_blkbits = EXOFS_BLKSHIFT;
1238 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1239 oi->i_commit_size = inode->i_size = 0;
1240 spin_lock(&sbi->s_next_gen_lock);
1241 inode->i_generation = sbi->s_next_generation++;
1242 spin_unlock(&sbi->s_next_gen_lock);
1243 insert_inode_hash(inode);
1245 exofs_init_comps(&oi->oc, &oi->one_comp, sb->s_fs_info,
1246 exofs_oi_objno(oi));
1247 exofs_sbi_write_stats(sbi); /* Make sure new sbi->s_nextid is on disk */
1249 mark_inode_dirty(inode);
1251 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1252 if (unlikely(ret)) {
1253 EXOFS_ERR("exofs_new_inode: ore_get_io_state failed\n");
1254 return ERR_PTR(ret);
1257 ios->done = create_done;
1258 ios->private = inode;
1260 ret = ore_create(ios);
1262 ore_put_io_state(ios);
1263 return ERR_PTR(ret);
1265 atomic_inc(&sbi->s_curr_pending);
1271 * struct to pass two arguments to update_inode's callback
1273 struct updatei_args {
1274 struct exofs_sb_info *sbi;
1275 struct exofs_fcb fcb;
1279 * Callback function from exofs_update_inode().
1281 static void updatei_done(struct ore_io_state *ios, void *p)
1283 struct updatei_args *args = p;
1285 ore_put_io_state(ios);
1287 atomic_dec(&args->sbi->s_curr_pending);
1293 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1294 * synchronously or asynchronously depending on the do_sync flag.
1296 static int exofs_update_inode(struct inode *inode, int do_sync)
1298 struct exofs_i_info *oi = exofs_i(inode);
1299 struct super_block *sb = inode->i_sb;
1300 struct exofs_sb_info *sbi = sb->s_fs_info;
1301 struct ore_io_state *ios;
1302 struct osd_attr attr;
1303 struct exofs_fcb *fcb;
1304 struct updatei_args *args;
1307 args = kzalloc(sizeof(*args), GFP_KERNEL);
1309 EXOFS_DBGMSG("Failed kzalloc of args\n");
1315 fcb->i_mode = cpu_to_le16(inode->i_mode);
1316 fcb->i_uid = cpu_to_le32(inode->i_uid);
1317 fcb->i_gid = cpu_to_le32(inode->i_gid);
1318 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1319 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1320 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1321 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1322 oi->i_commit_size = i_size_read(inode);
1323 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1324 fcb->i_generation = cpu_to_le32(inode->i_generation);
1326 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1327 if (old_valid_dev(inode->i_rdev)) {
1329 cpu_to_le32(old_encode_dev(inode->i_rdev));
1334 cpu_to_le32(new_encode_dev(inode->i_rdev));
1338 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1340 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1341 if (unlikely(ret)) {
1342 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
1346 attr = g_attr_inode_data;
1348 ios->out_attr_len = 1;
1349 ios->out_attr = &attr;
1351 wait_obj_created(oi);
1355 ios->done = updatei_done;
1356 ios->private = args;
1359 ret = ore_write(ios);
1360 if (!do_sync && !ret) {
1361 atomic_inc(&sbi->s_curr_pending);
1362 goto out; /* deallocation in updatei_done */
1365 ore_put_io_state(ios);
1369 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1370 inode->i_ino, do_sync, ret);
1374 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1376 /* FIXME: fix fsync and use wbc->sync_mode == WB_SYNC_ALL */
1377 return exofs_update_inode(inode, 1);
1381 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1384 static void delete_done(struct ore_io_state *ios, void *p)
1386 struct exofs_sb_info *sbi = p;
1388 ore_put_io_state(ios);
1390 atomic_dec(&sbi->s_curr_pending);
1394 * Called when the refcount of an inode reaches zero. We remove the object
1395 * from the OSD here. We make sure the object was created before we try and
1398 void exofs_evict_inode(struct inode *inode)
1400 struct exofs_i_info *oi = exofs_i(inode);
1401 struct super_block *sb = inode->i_sb;
1402 struct exofs_sb_info *sbi = sb->s_fs_info;
1403 struct ore_io_state *ios;
1406 truncate_inode_pages(&inode->i_data, 0);
1408 /* TODO: should do better here */
1409 if (inode->i_nlink || is_bad_inode(inode))
1413 end_writeback(inode);
1415 /* if we are deleting an obj that hasn't been created yet, wait.
1416 * This also makes sure that create_done cannot be called with an
1417 * already evicted inode.
1419 wait_obj_created(oi);
1420 /* ignore the error, attempt a remove anyway */
1422 /* Now Remove the OSD objects */
1423 ret = ore_get_io_state(&sbi->layout, &oi->oc, &ios);
1424 if (unlikely(ret)) {
1425 EXOFS_ERR("%s: ore_get_io_state failed\n", __func__);
1429 ios->done = delete_done;
1432 ret = ore_remove(ios);
1434 EXOFS_ERR("%s: ore_remove failed\n", __func__);
1435 ore_put_io_state(ios);
1438 atomic_inc(&sbi->s_curr_pending);
1443 end_writeback(inode);