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 *),
47 struct exofs_sb_info *sbi;
49 unsigned expected_pages;
50 struct exofs_io_state *ios;
56 loff_t pg_first; /* keep 64bit also in 32-arches */
57 bool read_4_write; /* This means two things: that the read is sync
58 * And the pages should not be unlocked.
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;
77 pcol->read_4_write = false;
80 static void _pcol_reset(struct page_collect *pcol)
82 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
85 pcol->alloc_pages = 0;
91 /* this is probably the end of the loop but in writes
92 * it might not end here. don't be left with nothing
94 if (!pcol->expected_pages)
95 pcol->expected_pages = MAX_PAGES_KMALLOC;
98 static int pcol_try_alloc(struct page_collect *pcol)
100 unsigned pages = min_t(unsigned, pcol->expected_pages,
103 if (!pcol->ios) { /* First time allocate io_state */
104 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
110 /* TODO: easily support bio chaining */
111 pages = min_t(unsigned, pages,
112 pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
114 for (; pages; pages >>= 1) {
115 pcol->pages = kmalloc(pages * sizeof(struct page *),
117 if (likely(pcol->pages)) {
118 pcol->alloc_pages = pages;
123 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
124 pcol->expected_pages);
128 static void pcol_free(struct page_collect *pcol)
134 exofs_put_io_state(pcol->ios);
139 static int pcol_add_page(struct page_collect *pcol, struct page *page,
142 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
145 pcol->pages[pcol->nr_pages++] = page;
150 static int update_read_page(struct page *page, int ret)
153 /* Everything is OK */
154 SetPageUptodate(page);
156 ClearPageError(page);
157 } else if (ret == -EFAULT) {
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 ret = 0; /* recovered error */
169 EXOFS_DBGMSG("recovered read error\n");
176 static void update_write_page(struct page *page, int ret)
179 mapping_set_error(page->mapping, ret);
182 end_page_writeback(page);
185 /* Called at the end of reads, to optionally unlock pages and update their
188 static int __readpages_done(struct page_collect *pcol)
194 int ret = exofs_check_io(pcol->ios, &resid);
197 good_bytes = pcol->length;
199 good_bytes = pcol->length - resid;
201 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
202 " length=0x%lx nr_pages=%u\n",
203 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
206 for (i = 0; i < pcol->nr_pages; i++) {
207 struct page *page = pcol->pages[i];
208 struct inode *inode = page->mapping->host;
211 if (inode != pcol->inode)
212 continue; /* osd might add more pages at end */
214 if (likely(length < good_bytes))
219 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
220 inode->i_ino, page->index,
221 page_stat ? "bad_bytes" : "good_bytes");
223 ret = update_read_page(page, page_stat);
224 if (!pcol->read_4_write)
230 EXOFS_DBGMSG2("readpages_done END\n");
234 /* callback of async reads */
235 static void readpages_done(struct exofs_io_state *ios, void *p)
237 struct page_collect *pcol = p;
239 __readpages_done(pcol);
240 atomic_dec(&pcol->sbi->s_curr_pending);
244 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
248 for (i = 0; i < pcol->nr_pages; i++) {
249 struct page *page = pcol->pages[i];
252 update_read_page(page, ret);
254 update_write_page(page, ret);
260 static int read_exec(struct page_collect *pcol)
262 struct exofs_i_info *oi = exofs_i(pcol->inode);
263 struct exofs_io_state *ios = pcol->ios;
264 struct page_collect *pcol_copy = NULL;
270 ios->pages = pcol->pages;
271 ios->nr_pages = pcol->nr_pages;
272 ios->length = pcol->length;
273 ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
275 if (pcol->read_4_write) {
276 exofs_oi_read(oi, pcol->ios);
277 return __readpages_done(pcol);
280 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
287 ios->done = readpages_done;
288 ios->private = pcol_copy;
289 ret = exofs_oi_read(oi, ios);
293 atomic_inc(&pcol->sbi->s_curr_pending);
295 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
296 ios->obj.id, _LLU(ios->offset), pcol->length);
298 /* pages ownership was passed to pcol_copy */
303 if (!pcol->read_4_write)
304 _unlock_pcol_pages(pcol, ret, READ);
312 /* readpage_strip is called either directly from readpage() or by the VFS from
313 * within read_cache_pages(), to add one more page to be read. It will try to
314 * collect as many contiguous pages as posible. If a discontinuity is
315 * encountered, or it runs out of resources, it will submit the previous segment
316 * and will start a new collection. Eventually caller must submit the last
317 * segment if present.
319 static int readpage_strip(void *data, struct page *page)
321 struct page_collect *pcol = data;
322 struct inode *inode = pcol->inode;
323 struct exofs_i_info *oi = exofs_i(inode);
324 loff_t i_size = i_size_read(inode);
325 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
329 /* FIXME: Just for debugging, will be removed */
330 if (PageUptodate(page))
331 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
334 if (page->index < end_index)
335 len = PAGE_CACHE_SIZE;
336 else if (page->index == end_index)
337 len = i_size & ~PAGE_CACHE_MASK;
341 if (!len || !obj_created(oi)) {
342 /* this will be out of bounds, or doesn't exist yet.
343 * Current page is cleared and the request is split
345 clear_highpage(page);
347 SetPageUptodate(page);
349 ClearPageError(page);
351 if (!pcol->read_4_write)
353 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
354 " splitting\n", inode->i_ino, page->index);
356 return read_exec(pcol);
361 if (unlikely(pcol->pg_first == -1)) {
362 pcol->pg_first = page->index;
363 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
365 /* Discontinuity detected, split the request */
366 ret = read_exec(pcol);
373 ret = pcol_try_alloc(pcol);
378 if (len != PAGE_CACHE_SIZE)
379 zero_user(page, len, PAGE_CACHE_SIZE - len);
381 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
382 inode->i_ino, page->index, len);
384 ret = pcol_add_page(pcol, page, len);
386 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
387 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
388 page, len, pcol->nr_pages, pcol->length);
390 /* split the request, and start again with current page */
391 ret = read_exec(pcol);
401 /* SetPageError(page); ??? */
406 static int exofs_readpages(struct file *file, struct address_space *mapping,
407 struct list_head *pages, unsigned nr_pages)
409 struct page_collect pcol;
412 _pcol_init(&pcol, nr_pages, mapping->host);
414 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
416 EXOFS_ERR("read_cache_pages => %d\n", ret);
420 return read_exec(&pcol);
423 static int _readpage(struct page *page, bool read_4_write)
425 struct page_collect pcol;
428 _pcol_init(&pcol, 1, page->mapping->host);
430 pcol.read_4_write = read_4_write;
431 ret = readpage_strip(&pcol, page);
433 EXOFS_ERR("_readpage => %d\n", ret);
437 return read_exec(&pcol);
441 * We don't need the file
443 static int exofs_readpage(struct file *file, struct page *page)
445 return _readpage(page, false);
448 /* Callback for osd_write. All writes are asynchronous */
449 static void writepages_done(struct exofs_io_state *ios, void *p)
451 struct page_collect *pcol = p;
456 int ret = exofs_check_io(ios, &resid);
458 atomic_dec(&pcol->sbi->s_curr_pending);
461 good_bytes = pcol->length;
463 good_bytes = pcol->length - resid;
465 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
466 " length=0x%lx nr_pages=%u\n",
467 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
470 for (i = 0; i < pcol->nr_pages; i++) {
471 struct page *page = pcol->pages[i];
472 struct inode *inode = page->mapping->host;
475 if (inode != pcol->inode)
476 continue; /* osd might add more pages to a bio */
478 if (likely(length < good_bytes))
483 update_write_page(page, page_stat);
485 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
486 inode->i_ino, page->index, page_stat);
493 EXOFS_DBGMSG2("writepages_done END\n");
496 static int write_exec(struct page_collect *pcol)
498 struct exofs_i_info *oi = exofs_i(pcol->inode);
499 struct exofs_io_state *ios = pcol->ios;
500 struct page_collect *pcol_copy = NULL;
506 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
508 EXOFS_ERR("write_exec: Failed to kmalloc(pcol)\n");
515 ios->pages = pcol_copy->pages;
516 ios->nr_pages = pcol_copy->nr_pages;
517 ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
518 ios->length = pcol_copy->length;
519 ios->done = writepages_done;
520 ios->private = pcol_copy;
522 ret = exofs_oi_write(oi, ios);
524 EXOFS_ERR("write_exec: exofs_oi_write() Failed\n");
528 atomic_inc(&pcol->sbi->s_curr_pending);
529 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
530 pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
532 /* pages ownership was passed to pcol_copy */
537 _unlock_pcol_pages(pcol, ret, WRITE);
544 /* writepage_strip is called either directly from writepage() or by the VFS from
545 * within write_cache_pages(), to add one more page to be written to storage.
546 * It will try to collect as many contiguous pages as possible. If a
547 * discontinuity is encountered or it runs out of resources it will submit the
548 * previous segment and will start a new collection.
549 * Eventually caller must submit the last segment if present.
551 static int writepage_strip(struct page *page,
552 struct writeback_control *wbc_unused, void *data)
554 struct page_collect *pcol = data;
555 struct inode *inode = pcol->inode;
556 struct exofs_i_info *oi = exofs_i(inode);
557 loff_t i_size = i_size_read(inode);
558 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
562 BUG_ON(!PageLocked(page));
564 ret = wait_obj_created(oi);
568 if (page->index < end_index)
569 /* in this case, the page is within the limits of the file */
570 len = PAGE_CACHE_SIZE;
572 len = i_size & ~PAGE_CACHE_MASK;
574 if (page->index > end_index || !len) {
575 /* in this case, the page is outside the limits
576 * (truncate in progress)
578 ret = write_exec(pcol);
582 ClearPageError(page);
584 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
585 "outside the limits\n",
586 inode->i_ino, page->index);
593 if (unlikely(pcol->pg_first == -1)) {
594 pcol->pg_first = page->index;
595 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
597 /* Discontinuity detected, split the request */
598 ret = write_exec(pcol);
602 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
603 inode->i_ino, page->index);
608 ret = pcol_try_alloc(pcol);
613 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
614 inode->i_ino, page->index, len);
616 ret = pcol_add_page(pcol, page, len);
618 EXOFS_DBGMSG2("Failed pcol_add_page "
619 "nr_pages=%u total_length=0x%lx\n",
620 pcol->nr_pages, pcol->length);
622 /* split the request, next loop will start again */
623 ret = write_exec(pcol);
625 EXOFS_DBGMSG("write_exec failed => %d", ret);
632 BUG_ON(PageWriteback(page));
633 set_page_writeback(page);
638 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
639 inode->i_ino, page->index, ret);
640 set_bit(AS_EIO, &page->mapping->flags);
645 static int exofs_writepages(struct address_space *mapping,
646 struct writeback_control *wbc)
648 struct page_collect pcol;
649 long start, end, expected_pages;
652 start = wbc->range_start >> PAGE_CACHE_SHIFT;
653 end = (wbc->range_end == LLONG_MAX) ?
654 start + mapping->nrpages :
655 wbc->range_end >> PAGE_CACHE_SHIFT;
658 expected_pages = end - start + 1;
660 expected_pages = mapping->nrpages;
662 if (expected_pages < 32L)
663 expected_pages = 32L;
665 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
666 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
667 mapping->host->i_ino, wbc->range_start, wbc->range_end,
668 mapping->nrpages, start, end, expected_pages);
670 _pcol_init(&pcol, expected_pages, mapping->host);
672 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
674 EXOFS_ERR("write_cache_pages => %d\n", ret);
678 return write_exec(&pcol);
681 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
683 struct page_collect pcol;
686 _pcol_init(&pcol, 1, page->mapping->host);
688 ret = writepage_strip(page, NULL, &pcol);
690 EXOFS_ERR("exofs_writepage => %d\n", ret);
694 return write_exec(&pcol);
697 /* i_mutex held using inode->i_size directly */
698 static void _write_failed(struct inode *inode, loff_t to)
700 if (to > inode->i_size)
701 truncate_pagecache(inode, to, inode->i_size);
704 int exofs_write_begin(struct file *file, struct address_space *mapping,
705 loff_t pos, unsigned len, unsigned flags,
706 struct page **pagep, void **fsdata)
713 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
716 EXOFS_DBGMSG("simple_write_begin failed\n");
723 /* read modify write */
724 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
725 ret = _readpage(page, true);
727 /*SetPageError was done by _readpage. Is it ok?*/
729 EXOFS_DBGMSG("__readpage_filler failed\n");
734 _write_failed(mapping->host, pos + len);
739 static int exofs_write_begin_export(struct file *file,
740 struct address_space *mapping,
741 loff_t pos, unsigned len, unsigned flags,
742 struct page **pagep, void **fsdata)
746 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
750 static int exofs_write_end(struct file *file, struct address_space *mapping,
751 loff_t pos, unsigned len, unsigned copied,
752 struct page *page, void *fsdata)
754 struct inode *inode = mapping->host;
755 /* According to comment in simple_write_end i_mutex is held */
756 loff_t i_size = inode->i_size;
759 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
761 _write_failed(inode, pos + len);
763 /* TODO: once simple_write_end marks inode dirty remove */
764 if (i_size != inode->i_size)
765 mark_inode_dirty(inode);
769 static int exofs_releasepage(struct page *page, gfp_t gfp)
771 EXOFS_DBGMSG("page 0x%lx\n", page->index);
776 static void exofs_invalidatepage(struct page *page, unsigned long offset)
778 EXOFS_DBGMSG("page 0x%lx offset 0x%lx\n", page->index, offset);
782 const struct address_space_operations exofs_aops = {
783 .readpage = exofs_readpage,
784 .readpages = exofs_readpages,
785 .writepage = exofs_writepage,
786 .writepages = exofs_writepages,
787 .write_begin = exofs_write_begin_export,
788 .write_end = exofs_write_end,
789 .releasepage = exofs_releasepage,
790 .set_page_dirty = __set_page_dirty_nobuffers,
791 .invalidatepage = exofs_invalidatepage,
793 /* Not implemented Yet */
794 .bmap = NULL, /* TODO: use osd's OSD_ACT_READ_MAP */
795 .direct_IO = NULL, /* TODO: Should be trivial to do */
797 /* With these NULL has special meaning or default is not exported */
800 .launder_page = NULL,
801 .is_partially_uptodate = NULL,
802 .error_remove_page = NULL,
805 /******************************************************************************
807 *****************************************************************************/
810 * Test whether an inode is a fast symlink.
812 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
814 struct exofs_i_info *oi = exofs_i(inode);
816 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
819 const struct osd_attr g_attr_logical_length = ATTR_DEF(
820 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
822 static int _do_truncate(struct inode *inode, loff_t newsize)
824 struct exofs_i_info *oi = exofs_i(inode);
827 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
829 ret = exofs_oi_truncate(oi, (u64)newsize);
831 truncate_setsize(inode, newsize);
833 EXOFS_DBGMSG("(0x%lx) size=0x%llx ret=>%d\n",
834 inode->i_ino, newsize, ret);
839 * Set inode attributes - update size attribute on OSD if needed,
840 * otherwise just call generic functions.
842 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
844 struct inode *inode = dentry->d_inode;
847 /* if we are about to modify an object, and it hasn't been
850 error = wait_obj_created(exofs_i(inode));
854 error = inode_change_ok(inode, iattr);
858 if ((iattr->ia_valid & ATTR_SIZE) &&
859 iattr->ia_size != i_size_read(inode)) {
860 error = _do_truncate(inode, iattr->ia_size);
865 setattr_copy(inode, iattr);
866 mark_inode_dirty(inode);
870 static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
872 EXOFS_ATTR_INODE_FILE_LAYOUT,
874 static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
876 EXOFS_ATTR_INODE_DIR_LAYOUT,
880 * Read the Linux inode info from the OSD, and return it as is. In exofs the
881 * inode info is in an application specific page/attribute of the osd-object.
883 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
884 struct exofs_fcb *inode)
886 struct exofs_sb_info *sbi = sb->s_fs_info;
887 struct osd_attr attrs[] = {
888 [0] = g_attr_inode_data,
889 [1] = g_attr_inode_file_layout,
890 [2] = g_attr_inode_dir_layout,
892 struct exofs_io_state *ios;
893 struct exofs_on_disk_inode_layout *layout;
896 ret = exofs_get_io_state(&sbi->layout, &ios);
898 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
902 ios->obj.id = exofs_oi_objno(oi);
903 exofs_make_credential(oi->i_cred, &ios->obj);
904 ios->cred = oi->i_cred;
906 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
907 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
909 ios->in_attr = attrs;
910 ios->in_attr_len = ARRAY_SIZE(attrs);
912 ret = exofs_sbi_read(ios);
914 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
915 _LLU(ios->obj.id), ret);
916 memset(inode, 0, sizeof(*inode));
917 inode->i_mode = 0040000 | (0777 & ~022);
918 /* If object is lost on target we might as well enable it's
921 if ((ret == -ENOENT) || (ret == -EINVAL))
926 ret = extract_attr_from_ios(ios, &attrs[0]);
928 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
931 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
932 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
934 ret = extract_attr_from_ios(ios, &attrs[1]);
936 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
940 layout = attrs[1].val_ptr;
941 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
942 EXOFS_ERR("%s: unsupported files layout %d\n",
943 __func__, layout->gen_func);
949 ret = extract_attr_from_ios(ios, &attrs[2]);
951 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
955 layout = attrs[2].val_ptr;
956 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
957 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
958 __func__, layout->gen_func);
965 exofs_put_io_state(ios);
969 static void __oi_init(struct exofs_i_info *oi)
971 init_waitqueue_head(&oi->i_wq);
975 * Fill in an inode read from the OSD and set it up for use
977 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
979 struct exofs_i_info *oi;
980 struct exofs_fcb fcb;
984 inode = iget_locked(sb, ino);
986 return ERR_PTR(-ENOMEM);
987 if (!(inode->i_state & I_NEW))
992 /* read the inode from the osd */
993 ret = exofs_get_inode(sb, oi, &fcb);
999 /* copy stuff from on-disk struct to in-memory struct */
1000 inode->i_mode = le16_to_cpu(fcb.i_mode);
1001 inode->i_uid = le32_to_cpu(fcb.i_uid);
1002 inode->i_gid = le32_to_cpu(fcb.i_gid);
1003 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
1004 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
1005 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
1006 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
1007 inode->i_ctime.tv_nsec =
1008 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
1009 oi->i_commit_size = le64_to_cpu(fcb.i_size);
1010 i_size_write(inode, oi->i_commit_size);
1011 inode->i_blkbits = EXOFS_BLKSHIFT;
1012 inode->i_generation = le32_to_cpu(fcb.i_generation);
1014 oi->i_dir_start_lookup = 0;
1016 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
1021 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1024 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
1027 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
1029 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
1032 if (S_ISREG(inode->i_mode)) {
1033 inode->i_op = &exofs_file_inode_operations;
1034 inode->i_fop = &exofs_file_operations;
1035 inode->i_mapping->a_ops = &exofs_aops;
1036 } else if (S_ISDIR(inode->i_mode)) {
1037 inode->i_op = &exofs_dir_inode_operations;
1038 inode->i_fop = &exofs_dir_operations;
1039 inode->i_mapping->a_ops = &exofs_aops;
1040 } else if (S_ISLNK(inode->i_mode)) {
1041 if (exofs_inode_is_fast_symlink(inode))
1042 inode->i_op = &exofs_fast_symlink_inode_operations;
1044 inode->i_op = &exofs_symlink_inode_operations;
1045 inode->i_mapping->a_ops = &exofs_aops;
1048 inode->i_op = &exofs_special_inode_operations;
1050 init_special_inode(inode, inode->i_mode,
1051 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
1053 init_special_inode(inode, inode->i_mode,
1054 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1057 unlock_new_inode(inode);
1062 return ERR_PTR(ret);
1065 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1067 if (!obj_created(oi)) {
1068 EXOFS_DBGMSG("!obj_created\n");
1069 BUG_ON(!obj_2bcreated(oi));
1070 wait_event(oi->i_wq, obj_created(oi));
1071 EXOFS_DBGMSG("wait_event done\n");
1073 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1076 * Callback function from exofs_new_inode(). The important thing is that we
1077 * set the obj_created flag so that other methods know that the object exists on
1080 static void create_done(struct exofs_io_state *ios, void *p)
1082 struct inode *inode = p;
1083 struct exofs_i_info *oi = exofs_i(inode);
1084 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1087 ret = exofs_check_io(ios, NULL);
1088 exofs_put_io_state(ios);
1090 atomic_dec(&sbi->s_curr_pending);
1092 if (unlikely(ret)) {
1093 EXOFS_ERR("object=0x%llx creation failed in pid=0x%llx",
1094 _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
1095 /*TODO: When FS is corrupted creation can fail, object already
1096 * exist. Get rid of this asynchronous creation, if exist
1097 * increment the obj counter and try the next object. Until we
1098 * succeed. All these dangling objects will be made into lost
1099 * files by chkfs.exofs
1103 set_obj_created(oi);
1109 * Set up a new inode and create an object for it on the OSD
1111 struct inode *exofs_new_inode(struct inode *dir, int mode)
1113 struct super_block *sb;
1114 struct inode *inode;
1115 struct exofs_i_info *oi;
1116 struct exofs_sb_info *sbi;
1117 struct exofs_io_state *ios;
1121 inode = new_inode(sb);
1123 return ERR_PTR(-ENOMEM);
1125 oi = exofs_i(inode);
1128 set_obj_2bcreated(oi);
1130 sbi = sb->s_fs_info;
1133 inode_init_owner(inode, dir, mode);
1134 inode->i_ino = sbi->s_nextid++;
1135 inode->i_blkbits = EXOFS_BLKSHIFT;
1136 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1137 oi->i_commit_size = inode->i_size = 0;
1138 spin_lock(&sbi->s_next_gen_lock);
1139 inode->i_generation = sbi->s_next_generation++;
1140 spin_unlock(&sbi->s_next_gen_lock);
1141 insert_inode_hash(inode);
1143 mark_inode_dirty(inode);
1145 ret = exofs_get_io_state(&sbi->layout, &ios);
1146 if (unlikely(ret)) {
1147 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1148 return ERR_PTR(ret);
1151 ios->obj.id = exofs_oi_objno(oi);
1152 exofs_make_credential(oi->i_cred, &ios->obj);
1154 ios->done = create_done;
1155 ios->private = inode;
1156 ios->cred = oi->i_cred;
1157 ret = exofs_sbi_create(ios);
1159 exofs_put_io_state(ios);
1160 return ERR_PTR(ret);
1162 atomic_inc(&sbi->s_curr_pending);
1168 * struct to pass two arguments to update_inode's callback
1170 struct updatei_args {
1171 struct exofs_sb_info *sbi;
1172 struct exofs_fcb fcb;
1176 * Callback function from exofs_update_inode().
1178 static void updatei_done(struct exofs_io_state *ios, void *p)
1180 struct updatei_args *args = p;
1182 exofs_put_io_state(ios);
1184 atomic_dec(&args->sbi->s_curr_pending);
1190 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1191 * synchronously or asynchronously depending on the do_sync flag.
1193 static int exofs_update_inode(struct inode *inode, int do_sync)
1195 struct exofs_i_info *oi = exofs_i(inode);
1196 struct super_block *sb = inode->i_sb;
1197 struct exofs_sb_info *sbi = sb->s_fs_info;
1198 struct exofs_io_state *ios;
1199 struct osd_attr attr;
1200 struct exofs_fcb *fcb;
1201 struct updatei_args *args;
1204 args = kzalloc(sizeof(*args), GFP_KERNEL);
1206 EXOFS_DBGMSG("Failed kzalloc of args\n");
1212 fcb->i_mode = cpu_to_le16(inode->i_mode);
1213 fcb->i_uid = cpu_to_le32(inode->i_uid);
1214 fcb->i_gid = cpu_to_le32(inode->i_gid);
1215 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1216 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1217 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1218 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1219 oi->i_commit_size = i_size_read(inode);
1220 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1221 fcb->i_generation = cpu_to_le32(inode->i_generation);
1223 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1224 if (old_valid_dev(inode->i_rdev)) {
1226 cpu_to_le32(old_encode_dev(inode->i_rdev));
1231 cpu_to_le32(new_encode_dev(inode->i_rdev));
1235 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1237 ret = exofs_get_io_state(&sbi->layout, &ios);
1238 if (unlikely(ret)) {
1239 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
1243 attr = g_attr_inode_data;
1245 ios->out_attr_len = 1;
1246 ios->out_attr = &attr;
1248 wait_obj_created(oi);
1252 ios->done = updatei_done;
1253 ios->private = args;
1256 ret = exofs_oi_write(oi, ios);
1257 if (!do_sync && !ret) {
1258 atomic_inc(&sbi->s_curr_pending);
1259 goto out; /* deallocation in updatei_done */
1262 exofs_put_io_state(ios);
1266 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1267 inode->i_ino, do_sync, ret);
1271 int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1273 return exofs_update_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1277 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1280 static void delete_done(struct exofs_io_state *ios, void *p)
1282 struct exofs_sb_info *sbi = p;
1284 exofs_put_io_state(ios);
1286 atomic_dec(&sbi->s_curr_pending);
1290 * Called when the refcount of an inode reaches zero. We remove the object
1291 * from the OSD here. We make sure the object was created before we try and
1294 void exofs_evict_inode(struct inode *inode)
1296 struct exofs_i_info *oi = exofs_i(inode);
1297 struct super_block *sb = inode->i_sb;
1298 struct exofs_sb_info *sbi = sb->s_fs_info;
1299 struct exofs_io_state *ios;
1302 truncate_inode_pages(&inode->i_data, 0);
1304 /* TODO: should do better here */
1305 if (inode->i_nlink || is_bad_inode(inode))
1309 end_writeback(inode);
1311 /* if we are deleting an obj that hasn't been created yet, wait.
1312 * This also makes sure that create_done cannot be called with an
1313 * already evicted inode.
1315 wait_obj_created(oi);
1316 /* ignore the error, attempt a remove anyway */
1318 /* Now Remove the OSD objects */
1319 ret = exofs_get_io_state(&sbi->layout, &ios);
1320 if (unlikely(ret)) {
1321 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
1325 ios->obj.id = exofs_oi_objno(oi);
1326 ios->done = delete_done;
1328 ios->cred = oi->i_cred;
1329 ret = exofs_sbi_remove(ios);
1331 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__);
1332 exofs_put_io_state(ios);
1335 atomic_inc(&sbi->s_curr_pending);
1340 end_writeback(inode);