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>
35 #include <linux/writeback.h>
36 #include <linux/buffer_head.h>
37 #include <scsi/scsi_device.h>
41 #define EXOFS_DBGMSG2(M...) do {} while (0)
43 enum { BIO_MAX_PAGES_KMALLOC =
44 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
46 PAGE_SIZE / sizeof(struct page *),
50 struct exofs_sb_info *sbi;
52 unsigned expected_pages;
53 struct exofs_io_state *ios;
59 loff_t pg_first; /* keep 64bit also in 32-arches */
62 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
65 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
69 pcol->expected_pages = expected_pages;
73 pcol->alloc_pages = 0;
79 static void _pcol_reset(struct page_collect *pcol)
81 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
84 pcol->alloc_pages = 0;
90 /* this is probably the end of the loop but in writes
91 * it might not end here. don't be left with nothing
93 if (!pcol->expected_pages)
94 pcol->expected_pages = MAX_PAGES_KMALLOC;
97 static int pcol_try_alloc(struct page_collect *pcol)
99 unsigned pages = min_t(unsigned, pcol->expected_pages,
102 if (!pcol->ios) { /* First time allocate io_state */
103 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
109 /* TODO: easily support bio chaining */
110 pages = min_t(unsigned, pages,
111 pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
113 for (; pages; pages >>= 1) {
114 pcol->pages = kmalloc(pages * sizeof(struct page *),
116 if (likely(pcol->pages)) {
117 pcol->alloc_pages = pages;
122 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
123 pcol->expected_pages);
127 static void pcol_free(struct page_collect *pcol)
133 exofs_put_io_state(pcol->ios);
138 static int pcol_add_page(struct page_collect *pcol, struct page *page,
141 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
144 pcol->pages[pcol->nr_pages++] = page;
149 static int update_read_page(struct page *page, int ret)
152 /* Everything is OK */
153 SetPageUptodate(page);
155 ClearPageError(page);
156 } else if (ret == -EFAULT) {
157 /* In this case we were trying to read something that wasn't on
158 * disk yet - return a page full of zeroes. This should be OK,
159 * because the object should be empty (if there was a write
160 * before this read, the read would be waiting with the page
162 clear_highpage(page);
164 SetPageUptodate(page);
166 ClearPageError(page);
167 ret = 0; /* recovered error */
168 EXOFS_DBGMSG("recovered read error\n");
175 static void update_write_page(struct page *page, int ret)
178 mapping_set_error(page->mapping, ret);
181 end_page_writeback(page);
184 /* Called at the end of reads, to optionally unlock pages and update their
187 static int __readpages_done(struct page_collect *pcol, bool do_unlock)
193 int ret = exofs_check_io(pcol->ios, &resid);
196 good_bytes = pcol->length;
198 good_bytes = pcol->length - resid;
200 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
201 " length=0x%lx nr_pages=%u\n",
202 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
205 for (i = 0; i < pcol->nr_pages; i++) {
206 struct page *page = pcol->pages[i];
207 struct inode *inode = page->mapping->host;
210 if (inode != pcol->inode)
211 continue; /* osd might add more pages at end */
213 if (likely(length < good_bytes))
218 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
219 inode->i_ino, page->index,
220 page_stat ? "bad_bytes" : "good_bytes");
222 ret = update_read_page(page, page_stat);
229 EXOFS_DBGMSG2("readpages_done END\n");
233 /* callback of async reads */
234 static void readpages_done(struct exofs_io_state *ios, void *p)
236 struct page_collect *pcol = p;
238 __readpages_done(pcol, true);
239 atomic_dec(&pcol->sbi->s_curr_pending);
243 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
247 for (i = 0; i < pcol->nr_pages; i++) {
248 struct page *page = pcol->pages[i];
251 update_read_page(page, ret);
253 update_write_page(page, ret);
259 static int read_exec(struct page_collect *pcol, bool is_sync)
261 struct exofs_i_info *oi = exofs_i(pcol->inode);
262 struct exofs_io_state *ios = pcol->ios;
263 struct page_collect *pcol_copy = NULL;
269 /* see comment in _readpage() about sync reads */
270 WARN_ON(is_sync && (pcol->nr_pages != 1));
272 ios->pages = pcol->pages;
273 ios->nr_pages = pcol->nr_pages;
274 ios->length = pcol->length;
275 ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
278 exofs_oi_read(oi, pcol->ios);
279 return __readpages_done(pcol, false);
282 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
289 ios->done = readpages_done;
290 ios->private = pcol_copy;
291 ret = exofs_oi_read(oi, ios);
295 atomic_inc(&pcol->sbi->s_curr_pending);
297 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
298 ios->obj.id, _LLU(ios->offset), pcol->length);
300 /* pages ownership was passed to pcol_copy */
306 _unlock_pcol_pages(pcol, ret, READ);
314 /* readpage_strip is called either directly from readpage() or by the VFS from
315 * within read_cache_pages(), to add one more page to be read. It will try to
316 * collect as many contiguous pages as posible. If a discontinuity is
317 * encountered, or it runs out of resources, it will submit the previous segment
318 * and will start a new collection. Eventually caller must submit the last
319 * segment if present.
321 static int readpage_strip(void *data, struct page *page)
323 struct page_collect *pcol = data;
324 struct inode *inode = pcol->inode;
325 struct exofs_i_info *oi = exofs_i(inode);
326 loff_t i_size = i_size_read(inode);
327 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
331 /* FIXME: Just for debugging, will be removed */
332 if (PageUptodate(page))
333 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
336 if (page->index < end_index)
337 len = PAGE_CACHE_SIZE;
338 else if (page->index == end_index)
339 len = i_size & ~PAGE_CACHE_MASK;
343 if (!len || !obj_created(oi)) {
344 /* this will be out of bounds, or doesn't exist yet.
345 * Current page is cleared and the request is split
347 clear_highpage(page);
349 SetPageUptodate(page);
351 ClearPageError(page);
354 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
355 " splitting\n", inode->i_ino, page->index);
357 return read_exec(pcol, false);
362 if (unlikely(pcol->pg_first == -1)) {
363 pcol->pg_first = page->index;
364 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
366 /* Discontinuity detected, split the request */
367 ret = read_exec(pcol, false);
374 ret = pcol_try_alloc(pcol);
379 if (len != PAGE_CACHE_SIZE)
380 zero_user(page, len, PAGE_CACHE_SIZE - len);
382 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
383 inode->i_ino, page->index, len);
385 ret = pcol_add_page(pcol, page, len);
387 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
388 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
389 page, len, pcol->nr_pages, pcol->length);
391 /* split the request, and start again with current page */
392 ret = read_exec(pcol, false);
402 /* SetPageError(page); ??? */
407 static int exofs_readpages(struct file *file, struct address_space *mapping,
408 struct list_head *pages, unsigned nr_pages)
410 struct page_collect pcol;
413 _pcol_init(&pcol, nr_pages, mapping->host);
415 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
417 EXOFS_ERR("read_cache_pages => %d\n", ret);
421 return read_exec(&pcol, false);
424 static int _readpage(struct page *page, bool is_sync)
426 struct page_collect pcol;
429 _pcol_init(&pcol, 1, page->mapping->host);
431 /* readpage_strip might call read_exec(,is_sync==false) at several
432 * places but not if we have a single page.
434 ret = readpage_strip(&pcol, page);
436 EXOFS_ERR("_readpage => %d\n", ret);
440 return read_exec(&pcol, is_sync);
444 * We don't need the file
446 static int exofs_readpage(struct file *file, struct page *page)
448 return _readpage(page, false);
451 /* Callback for osd_write. All writes are asynchronous */
452 static void writepages_done(struct exofs_io_state *ios, void *p)
454 struct page_collect *pcol = p;
459 int ret = exofs_check_io(ios, &resid);
461 atomic_dec(&pcol->sbi->s_curr_pending);
464 good_bytes = pcol->length;
466 good_bytes = pcol->length - resid;
468 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
469 " length=0x%lx nr_pages=%u\n",
470 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
473 for (i = 0; i < pcol->nr_pages; i++) {
474 struct page *page = pcol->pages[i];
475 struct inode *inode = page->mapping->host;
478 if (inode != pcol->inode)
479 continue; /* osd might add more pages to a bio */
481 if (likely(length < good_bytes))
486 update_write_page(page, page_stat);
488 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
489 inode->i_ino, page->index, page_stat);
496 EXOFS_DBGMSG2("writepages_done END\n");
499 static int write_exec(struct page_collect *pcol)
501 struct exofs_i_info *oi = exofs_i(pcol->inode);
502 struct exofs_io_state *ios = pcol->ios;
503 struct page_collect *pcol_copy = NULL;
509 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
511 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
518 ios->pages = pcol_copy->pages;
519 ios->nr_pages = pcol_copy->nr_pages;
520 ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
521 ios->length = pcol_copy->length;
522 ios->done = writepages_done;
523 ios->private = pcol_copy;
525 ret = exofs_oi_write(oi, ios);
527 EXOFS_ERR("write_exec: exofs_oi_write() Faild\n");
531 atomic_inc(&pcol->sbi->s_curr_pending);
532 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
533 pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
535 /* pages ownership was passed to pcol_copy */
540 _unlock_pcol_pages(pcol, ret, WRITE);
547 /* writepage_strip is called either directly from writepage() or by the VFS from
548 * within write_cache_pages(), to add one more page to be written to storage.
549 * It will try to collect as many contiguous pages as possible. If a
550 * discontinuity is encountered or it runs out of resources it will submit the
551 * previous segment and will start a new collection.
552 * Eventually caller must submit the last segment if present.
554 static int writepage_strip(struct page *page,
555 struct writeback_control *wbc_unused, void *data)
557 struct page_collect *pcol = data;
558 struct inode *inode = pcol->inode;
559 struct exofs_i_info *oi = exofs_i(inode);
560 loff_t i_size = i_size_read(inode);
561 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
565 BUG_ON(!PageLocked(page));
567 ret = wait_obj_created(oi);
571 if (page->index < end_index)
572 /* in this case, the page is within the limits of the file */
573 len = PAGE_CACHE_SIZE;
575 len = i_size & ~PAGE_CACHE_MASK;
577 if (page->index > end_index || !len) {
578 /* in this case, the page is outside the limits
579 * (truncate in progress)
581 ret = write_exec(pcol);
585 ClearPageError(page);
587 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
588 "outside the limits\n",
589 inode->i_ino, page->index);
596 if (unlikely(pcol->pg_first == -1)) {
597 pcol->pg_first = page->index;
598 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
600 /* Discontinuity detected, split the request */
601 ret = write_exec(pcol);
605 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
606 inode->i_ino, page->index);
611 ret = pcol_try_alloc(pcol);
616 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
617 inode->i_ino, page->index, len);
619 ret = pcol_add_page(pcol, page, len);
621 EXOFS_DBGMSG2("Failed pcol_add_page "
622 "nr_pages=%u total_length=0x%lx\n",
623 pcol->nr_pages, pcol->length);
625 /* split the request, next loop will start again */
626 ret = write_exec(pcol);
628 EXOFS_DBGMSG("write_exec faild => %d", ret);
635 BUG_ON(PageWriteback(page));
636 set_page_writeback(page);
641 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
642 inode->i_ino, page->index, ret);
643 set_bit(AS_EIO, &page->mapping->flags);
648 static int exofs_writepages(struct address_space *mapping,
649 struct writeback_control *wbc)
651 struct page_collect pcol;
652 long start, end, expected_pages;
655 start = wbc->range_start >> PAGE_CACHE_SHIFT;
656 end = (wbc->range_end == LLONG_MAX) ?
657 start + mapping->nrpages :
658 wbc->range_end >> PAGE_CACHE_SHIFT;
661 expected_pages = end - start + 1;
663 expected_pages = mapping->nrpages;
665 if (expected_pages < 32L)
666 expected_pages = 32L;
668 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
669 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
670 mapping->host->i_ino, wbc->range_start, wbc->range_end,
671 mapping->nrpages, start, end, expected_pages);
673 _pcol_init(&pcol, expected_pages, mapping->host);
675 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
677 EXOFS_ERR("write_cache_pages => %d\n", ret);
681 return write_exec(&pcol);
684 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
686 struct page_collect pcol;
689 _pcol_init(&pcol, 1, page->mapping->host);
691 ret = writepage_strip(page, NULL, &pcol);
693 EXOFS_ERR("exofs_writepage => %d\n", ret);
697 return write_exec(&pcol);
700 int exofs_write_begin(struct file *file, struct address_space *mapping,
701 loff_t pos, unsigned len, unsigned flags,
702 struct page **pagep, void **fsdata)
709 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
712 EXOFS_DBGMSG("simple_write_begin faild\n");
719 /* read modify write */
720 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
721 ret = _readpage(page, true);
723 /*SetPageError was done by _readpage. Is it ok?*/
725 EXOFS_DBGMSG("__readpage_filler faild\n");
732 static int exofs_write_begin_export(struct file *file,
733 struct address_space *mapping,
734 loff_t pos, unsigned len, unsigned flags,
735 struct page **pagep, void **fsdata)
739 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
743 static int exofs_write_end(struct file *file, struct address_space *mapping,
744 loff_t pos, unsigned len, unsigned copied,
745 struct page *page, void *fsdata)
747 struct inode *inode = mapping->host;
748 /* According to comment in simple_write_end i_mutex is held */
749 loff_t i_size = inode->i_size;
752 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
753 if (i_size != inode->i_size)
754 mark_inode_dirty(inode);
758 static int exofs_releasepage(struct page *page, gfp_t gfp)
760 EXOFS_DBGMSG("page 0x%lx\n", page->index);
762 return try_to_free_buffers(page);
765 static void exofs_invalidatepage(struct page *page, unsigned long offset)
767 EXOFS_DBGMSG("page_has_buffers=>%d\n", page_has_buffers(page));
770 block_invalidatepage(page, offset);
773 const struct address_space_operations exofs_aops = {
774 .readpage = exofs_readpage,
775 .readpages = exofs_readpages,
776 .writepage = exofs_writepage,
777 .writepages = exofs_writepages,
778 .write_begin = exofs_write_begin_export,
779 .write_end = exofs_write_end,
780 .releasepage = exofs_releasepage,
781 .set_page_dirty = __set_page_dirty_nobuffers,
782 .invalidatepage = exofs_invalidatepage,
784 /* Not implemented Yet */
785 .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
786 .direct_IO = NULL, /* TODO: Should be trivial to do */
788 /* With these NULL has special meaning or default is not exported */
792 .launder_page = NULL,
793 .is_partially_uptodate = NULL,
794 .error_remove_page = NULL,
797 /******************************************************************************
799 *****************************************************************************/
802 * Test whether an inode is a fast symlink.
804 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
806 struct exofs_i_info *oi = exofs_i(inode);
808 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
812 * get_block_t - Fill in a buffer_head
813 * An OSD takes care of block allocation so we just fake an allocation by
814 * putting in the inode's sector_t in the buffer_head.
815 * TODO: What about the case of create==0 and @iblock does not exist in the
818 static int exofs_get_block(struct inode *inode, sector_t iblock,
819 struct buffer_head *bh_result, int create)
821 map_bh(bh_result, inode->i_sb, iblock);
825 const struct osd_attr g_attr_logical_length = ATTR_DEF(
826 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
828 static int _do_truncate(struct inode *inode)
830 struct exofs_i_info *oi = exofs_i(inode);
831 loff_t isize = i_size_read(inode);
834 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
836 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
838 ret = exofs_oi_truncate(oi, (u64)isize);
839 EXOFS_DBGMSG("(0x%lx) size=0x%llx\n", inode->i_ino, isize);
844 * Truncate a file to the specified size - all we have to do is set the size
845 * attribute. We make sure the object exists first.
847 void exofs_truncate(struct inode *inode)
849 struct exofs_i_info *oi = exofs_i(inode);
852 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
853 || S_ISLNK(inode->i_mode)))
855 if (exofs_inode_is_fast_symlink(inode))
857 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
860 /* if we are about to truncate an object, and it hasn't been
863 if (unlikely(wait_obj_created(oi)))
866 ret = _do_truncate(inode);
871 mark_inode_dirty(inode);
874 make_bad_inode(inode);
879 * Set inode attributes - just call generic functions.
881 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
883 struct inode *inode = dentry->d_inode;
886 error = inode_change_ok(inode, iattr);
890 if ((iattr->ia_valid & ATTR_SIZE) &&
891 iattr->ia_size != i_size_read(inode)) {
894 error = vmtruncate(inode, iattr->ia_size);
899 setattr_copy(inode, iattr);
900 mark_inode_dirty(inode);
904 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
906 EXOFS_ATTR_INODE_FILE_LAYOUT,
908 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
910 EXOFS_ATTR_INODE_DIR_LAYOUT,
914 * Read the Linux inode info from the OSD, and return it as is. In exofs the
915 * inode info is in an application specific page/attribute of the osd-object.
917 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
918 struct exofs_fcb *inode)
920 struct exofs_sb_info *sbi = sb->s_fs_info;
921 struct osd_attr attrs[] = {
922 [0] = g_attr_inode_data,
923 [1] = g_attr_inode_file_layout,
924 [2] = g_attr_inode_dir_layout,
926 struct exofs_io_state *ios;
927 struct exofs_on_disk_inode_layout *layout;
930 ret = exofs_get_io_state(&sbi->layout, &ios);
932 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
936 ios->obj.id = exofs_oi_objno(oi);
937 exofs_make_credential(oi->i_cred, &ios->obj);
938 ios->cred = oi->i_cred;
940 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
941 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
943 ios->in_attr = attrs;
944 ios->in_attr_len = ARRAY_SIZE(attrs);
946 ret = exofs_sbi_read(ios);
948 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
949 _LLU(ios->obj.id), ret);
950 memset(inode, 0, sizeof(*inode));
951 inode->i_mode = 0040000 | (0777 & ~022);
952 /* If object is lost on target we might as well enable it's
955 if ((ret == -ENOENT) || (ret == -EINVAL))
960 ret = extract_attr_from_ios(ios, &attrs[0]);
962 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
965 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
966 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
968 ret = extract_attr_from_ios(ios, &attrs[1]);
970 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
974 layout = attrs[1].val_ptr;
975 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
976 EXOFS_ERR("%s: unsupported files layout %d\n",
977 __func__, layout->gen_func);
983 ret = extract_attr_from_ios(ios, &attrs[2]);
985 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
989 layout = attrs[2].val_ptr;
990 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
991 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
992 __func__, layout->gen_func);
999 exofs_put_io_state(ios);
1003 static void __oi_init(struct exofs_i_info *oi)
1005 init_waitqueue_head(&oi->i_wq);
1009 * Fill in an inode read from the OSD and set it up for use
1011 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
1013 struct exofs_i_info *oi;
1014 struct exofs_fcb fcb;
1015 struct inode *inode;
1018 inode = iget_locked(sb, ino);
1020 return ERR_PTR(-ENOMEM);
1021 if (!(inode->i_state & I_NEW))
1023 oi = exofs_i(inode);
1026 /* read the inode from the osd */
1027 ret = exofs_get_inode(sb, oi, &fcb);
1031 set_obj_created(oi);
1033 /* copy stuff from on-disk struct to in-memory struct */
1034 inode->i_mode = le16_to_cpu(fcb.i_mode);
1035 inode->i_uid = le32_to_cpu(fcb.i_uid);
1036 inode->i_gid = le32_to_cpu(fcb.i_gid);
1037 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1038 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1039 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1040 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1041 inode->i_ctime.tv_nsec =
1042 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1043 oi->i_commit_size = le64_to_cpu(fcb.i_size);
1044 i_size_write(inode, oi->i_commit_size);
1045 inode->i_blkbits = EXOFS_BLKSHIFT;
1046 inode->i_generation = le32_to_cpu(fcb.i_generation);
1048 oi->i_dir_start_lookup = 0;
1050 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1055 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1058 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1061 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1063 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1066 if (S_ISREG(inode->i_mode)) {
1067 inode->i_op = &exofs_file_inode_operations;
1068 inode->i_fop = &exofs_file_operations;
1069 inode->i_mapping->a_ops = &exofs_aops;
1070 } else if (S_ISDIR(inode->i_mode)) {
1071 inode->i_op = &exofs_dir_inode_operations;
1072 inode->i_fop = &exofs_dir_operations;
1073 inode->i_mapping->a_ops = &exofs_aops;
1074 } else if (S_ISLNK(inode->i_mode)) {
1075 if (exofs_inode_is_fast_symlink(inode))
1076 inode->i_op = &exofs_fast_symlink_inode_operations;
1078 inode->i_op = &exofs_symlink_inode_operations;
1079 inode->i_mapping->a_ops = &exofs_aops;
1082 inode->i_op = &exofs_special_inode_operations;
1084 init_special_inode(inode, inode->i_mode,
1085 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1087 init_special_inode(inode, inode->i_mode,
1088 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1091 unlock_new_inode(inode);
1096 return ERR_PTR(ret);
1099 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1101 if (!obj_created(oi)) {
1102 BUG_ON(!obj_2bcreated(oi));
1103 wait_event(oi->i_wq, obj_created(oi));
1105 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1108 * Callback function from exofs_new_inode(). The important thing is that we
1109 * set the obj_created flag so that other methods know that the object exists on
1112 static void create_done(struct exofs_io_state *ios, void *p)
1114 struct inode *inode = p;
1115 struct exofs_i_info *oi = exofs_i(inode);
1116 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1119 ret = exofs_check_io(ios, NULL);
1120 exofs_put_io_state(ios);
1122 atomic_dec(&sbi->s_curr_pending);
1124 if (unlikely(ret)) {
1125 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1126 _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
1127 /*TODO: When FS is corrupted creation can fail, object already
1128 * exist. Get rid of this asynchronous creation, if exist
1129 * increment the obj counter and try the next object. Until we
1130 * succeed. All these dangling objects will be made into lost
1131 * files by chkfs.exofs
1135 set_obj_created(oi);
1137 atomic_dec(&inode->i_count);
1142 * Set up a new inode and create an object for it on the OSD
1144 struct inode *exofs_new_inode(struct inode *dir, int mode)
1146 struct super_block *sb;
1147 struct inode *inode;
1148 struct exofs_i_info *oi;
1149 struct exofs_sb_info *sbi;
1150 struct exofs_io_state *ios;
1154 inode = new_inode(sb);
1156 return ERR_PTR(-ENOMEM);
1158 oi = exofs_i(inode);
1161 set_obj_2bcreated(oi);
1163 sbi = sb->s_fs_info;
1166 inode_init_owner(inode, dir, mode);
1167 inode->i_ino = sbi->s_nextid++;
1168 inode->i_blkbits = EXOFS_BLKSHIFT;
1169 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1170 oi->i_commit_size = inode->i_size = 0;
1171 spin_lock(&sbi->s_next_gen_lock);
1172 inode->i_generation = sbi->s_next_generation++;
1173 spin_unlock(&sbi->s_next_gen_lock);
1174 insert_inode_hash(inode);
1176 mark_inode_dirty(inode);
1178 ret = exofs_get_io_state(&sbi->layout, &ios);
1179 if (unlikely(ret)) {
1180 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1181 return ERR_PTR(ret);
1184 ios->obj.id = exofs_oi_objno(oi);
1185 exofs_make_credential(oi->i_cred, &ios->obj);
1187 /* increment the refcount so that the inode will still be around when we
1188 * reach the callback
1190 atomic_inc(&inode->i_count);
1192 ios->done = create_done;
1193 ios->private = inode;
1194 ios->cred = oi->i_cred;
1195 ret = exofs_sbi_create(ios);
1197 atomic_dec(&inode->i_count);
1198 exofs_put_io_state(ios);
1199 return ERR_PTR(ret);
1201 atomic_inc(&sbi->s_curr_pending);
1207 * struct to pass two arguments to update_inode's callback
1209 struct updatei_args {
1210 struct exofs_sb_info *sbi;
1211 struct exofs_fcb fcb;
1215 * Callback function from exofs_update_inode().
1217 static void updatei_done(struct exofs_io_state *ios, void *p)
1219 struct updatei_args *args = p;
1221 exofs_put_io_state(ios);
1223 atomic_dec(&args->sbi->s_curr_pending);
1229 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1230 * synchronously or asynchronously depending on the do_sync flag.
1232 static int exofs_update_inode(struct inode *inode, int do_sync)
1234 struct exofs_i_info *oi = exofs_i(inode);
1235 struct super_block *sb = inode->i_sb;
1236 struct exofs_sb_info *sbi = sb->s_fs_info;
1237 struct exofs_io_state *ios;
1238 struct osd_attr attr;
1239 struct exofs_fcb *fcb;
1240 struct updatei_args *args;
1243 args = kzalloc(sizeof(*args), GFP_KERNEL);
1245 EXOFS_DBGMSG("Faild kzalloc of args\n");
1251 fcb->i_mode = cpu_to_le16(inode->i_mode);
1252 fcb->i_uid = cpu_to_le32(inode->i_uid);
1253 fcb->i_gid = cpu_to_le32(inode->i_gid);
1254 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1255 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1256 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1257 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1258 oi->i_commit_size = i_size_read(inode);
1259 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1260 fcb->i_generation = cpu_to_le32(inode->i_generation);
1262 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1263 if (old_valid_dev(inode->i_rdev)) {
1265 cpu_to_le32(old_encode_dev(inode->i_rdev));
1270 cpu_to_le32(new_encode_dev(inode->i_rdev));
1274 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1276 ret = exofs_get_io_state(&sbi->layout, &ios);
1277 if (unlikely(ret)) {
1278 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
1282 attr = g_attr_inode_data;
1284 ios->out_attr_len = 1;
1285 ios->out_attr = &attr;
1287 if (!obj_created(oi)) {
1288 EXOFS_DBGMSG("!obj_created\n");
1289 BUG_ON(!obj_2bcreated(oi));
1290 wait_event(oi->i_wq, obj_created(oi));
1291 EXOFS_DBGMSG("wait_event done\n");
1296 ios->done = updatei_done;
1297 ios->private = args;
1300 ret = exofs_oi_write(oi, ios);
1301 if (!do_sync && !ret) {
1302 atomic_inc(&sbi->s_curr_pending);
1303 goto out; /* deallocation in updatei_done */
1306 exofs_put_io_state(ios);
1310 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1311 inode->i_ino, do_sync, ret);
1315 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1317 return exofs_update_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1321 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1324 static void delete_done(struct exofs_io_state *ios, void *p)
1326 struct exofs_sb_info *sbi = p;
1328 exofs_put_io_state(ios);
1330 atomic_dec(&sbi->s_curr_pending);
1334 * Called when the refcount of an inode reaches zero. We remove the object
1335 * from the OSD here. We make sure the object was created before we try and
1338 void exofs_delete_inode(struct inode *inode)
1340 struct exofs_i_info *oi = exofs_i(inode);
1341 struct super_block *sb = inode->i_sb;
1342 struct exofs_sb_info *sbi = sb->s_fs_info;
1343 struct exofs_io_state *ios;
1346 truncate_inode_pages(&inode->i_data, 0);
1348 if (is_bad_inode(inode))
1351 mark_inode_dirty(inode);
1352 exofs_update_inode(inode, inode_needs_sync(inode));
1355 if (inode->i_blocks)
1356 exofs_truncate(inode);
1360 ret = exofs_get_io_state(&sbi->layout, &ios);
1361 if (unlikely(ret)) {
1362 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
1366 /* if we are deleting an obj that hasn't been created yet, wait */
1367 if (!obj_created(oi)) {
1368 BUG_ON(!obj_2bcreated(oi));
1369 wait_event(oi->i_wq, obj_created(oi));
1372 ios->obj.id = exofs_oi_objno(oi);
1373 ios->done = delete_done;
1375 ios->cred = oi->i_cred;
1376 ret = exofs_sbi_remove(ios);
1378 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__);
1379 exofs_put_io_state(ios);
1382 atomic_inc(&sbi->s_curr_pending);