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/writeback.h>
35 #include <linux/buffer_head.h>
36 #include <scsi/scsi_device.h>
40 #ifdef CONFIG_EXOFS_DEBUG
41 # define EXOFS_DEBUG_OBJ_ISIZE 1
44 #define EXOFS_DBGMSG2(M...) do {} while (0)
47 struct exofs_sb_info *sbi;
48 struct request_queue *req_q;
50 unsigned expected_pages;
55 loff_t pg_first; /* keep 64bit also in 32-arches */
58 static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
61 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
64 pcol->req_q = osd_request_queue(sbi->s_dev);
66 pcol->expected_pages = expected_pages;
73 EXOFS_DBGMSG("_pcol_init ino=0x%lx expected_pages=%u\n", inode->i_ino,
77 static void _pcol_reset(struct page_collect *pcol)
79 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
85 EXOFS_DBGMSG("_pcol_reset ino=0x%lx expected_pages=%u\n",
86 pcol->inode->i_ino, pcol->expected_pages);
88 /* this is probably the end of the loop but in writes
89 * it might not end here. don't be left with nothing
91 if (!pcol->expected_pages)
92 pcol->expected_pages = 128;
95 static int pcol_try_alloc(struct page_collect *pcol)
97 int pages = min_t(unsigned, pcol->expected_pages, BIO_MAX_PAGES);
99 for (; pages; pages >>= 1) {
100 pcol->bio = bio_alloc(GFP_KERNEL, pages);
101 if (likely(pcol->bio))
105 EXOFS_ERR("Failed to kcalloc expected_pages=%u\n",
106 pcol->expected_pages);
110 static void pcol_free(struct page_collect *pcol)
116 static int pcol_add_page(struct page_collect *pcol, struct page *page,
119 int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0);
120 if (unlikely(len != added_len))
128 static int update_read_page(struct page *page, int ret)
131 /* Everything is OK */
132 SetPageUptodate(page);
134 ClearPageError(page);
135 } else if (ret == -EFAULT) {
136 /* In this case we were trying to read something that wasn't on
137 * disk yet - return a page full of zeroes. This should be OK,
138 * because the object should be empty (if there was a write
139 * before this read, the read would be waiting with the page
141 clear_highpage(page);
143 SetPageUptodate(page);
145 ClearPageError(page);
146 ret = 0; /* recovered error */
147 EXOFS_DBGMSG("recovered read error\n");
154 static void update_write_page(struct page *page, int ret)
157 mapping_set_error(page->mapping, ret);
160 end_page_writeback(page);
163 /* Called at the end of reads, to optionally unlock pages and update their
166 static int __readpages_done(struct osd_request *or, struct page_collect *pcol,
169 struct bio_vec *bvec;
174 int ret = exofs_check_ok_resid(or, &resid, NULL);
179 good_bytes = pcol->length;
183 good_bytes = pcol->length - resid;
185 EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx"
186 " length=0x%lx nr_pages=%u\n",
187 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
190 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
191 struct page *page = bvec->bv_page;
192 struct inode *inode = page->mapping->host;
195 if (inode != pcol->inode)
196 continue; /* osd might add more pages at end */
198 if (likely(length < good_bytes))
203 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
204 inode->i_ino, page->index,
205 page_stat ? "bad_bytes" : "good_bytes");
207 ret = update_read_page(page, page_stat);
210 length += bvec->bv_len;
214 EXOFS_DBGMSG("readpages_done END\n");
218 /* callback of async reads */
219 static void readpages_done(struct osd_request *or, void *p)
221 struct page_collect *pcol = p;
223 __readpages_done(or, pcol, true);
224 atomic_dec(&pcol->sbi->s_curr_pending);
228 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
230 struct bio_vec *bvec;
233 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
234 struct page *page = bvec->bv_page;
237 update_read_page(page, ret);
239 update_write_page(page, ret);
246 static int read_exec(struct page_collect *pcol, bool is_sync)
248 struct exofs_i_info *oi = exofs_i(pcol->inode);
249 struct osd_obj_id obj = {pcol->sbi->s_pid,
250 pcol->inode->i_ino + EXOFS_OBJ_OFF};
251 struct osd_request *or = NULL;
252 struct page_collect *pcol_copy = NULL;
253 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
259 /* see comment in _readpage() about sync reads */
260 WARN_ON(is_sync && (pcol->nr_pages != 1));
262 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
268 osd_req_read(or, &obj, i_start, pcol->bio, pcol->length);
271 exofs_sync_op(or, pcol->sbi->s_timeout, oi->i_cred);
272 return __readpages_done(or, pcol, false);
275 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
282 ret = exofs_async_op(or, readpages_done, pcol_copy, oi->i_cred);
286 atomic_inc(&pcol->sbi->s_curr_pending);
288 EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
289 obj.id, _LLU(i_start), pcol->length);
291 /* pages ownership was passed to pcol_copy */
297 _unlock_pcol_pages(pcol, ret, READ);
298 else /* Pages unlocked by caller in sync mode only free bio */
307 /* readpage_strip is called either directly from readpage() or by the VFS from
308 * within read_cache_pages(), to add one more page to be read. It will try to
309 * collect as many contiguous pages as posible. If a discontinuity is
310 * encountered, or it runs out of resources, it will submit the previous segment
311 * and will start a new collection. Eventually caller must submit the last
312 * segment if present.
314 static int readpage_strip(void *data, struct page *page)
316 struct page_collect *pcol = data;
317 struct inode *inode = pcol->inode;
318 struct exofs_i_info *oi = exofs_i(inode);
319 loff_t i_size = i_size_read(inode);
320 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
324 /* FIXME: Just for debugging, will be removed */
325 if (PageUptodate(page))
326 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
329 if (page->index < end_index)
330 len = PAGE_CACHE_SIZE;
331 else if (page->index == end_index)
332 len = i_size & ~PAGE_CACHE_MASK;
336 if (!len || !obj_created(oi)) {
337 /* this will be out of bounds, or doesn't exist yet.
338 * Current page is cleared and the request is split
340 clear_highpage(page);
342 SetPageUptodate(page);
344 ClearPageError(page);
347 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
348 " splitting\n", inode->i_ino, page->index);
350 return read_exec(pcol, false);
355 if (unlikely(pcol->pg_first == -1)) {
356 pcol->pg_first = page->index;
357 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
359 /* Discontinuity detected, split the request */
360 ret = read_exec(pcol, false);
367 ret = pcol_try_alloc(pcol);
372 if (len != PAGE_CACHE_SIZE)
373 zero_user(page, len, PAGE_CACHE_SIZE - len);
375 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
376 inode->i_ino, page->index, len);
378 ret = pcol_add_page(pcol, page, len);
380 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
381 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
382 page, len, pcol->nr_pages, pcol->length);
384 /* split the request, and start again with current page */
385 ret = read_exec(pcol, false);
395 /* SetPageError(page); ??? */
400 static int exofs_readpages(struct file *file, struct address_space *mapping,
401 struct list_head *pages, unsigned nr_pages)
403 struct page_collect pcol;
406 _pcol_init(&pcol, nr_pages, mapping->host);
408 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
410 EXOFS_ERR("read_cache_pages => %d\n", ret);
414 return read_exec(&pcol, false);
417 static int _readpage(struct page *page, bool is_sync)
419 struct page_collect pcol;
422 _pcol_init(&pcol, 1, page->mapping->host);
424 /* readpage_strip might call read_exec(,async) inside at several places
425 * but this is safe for is_async=0 since read_exec will not do anything
426 * when we have a single page.
428 ret = readpage_strip(&pcol, page);
430 EXOFS_ERR("_readpage => %d\n", ret);
434 return read_exec(&pcol, is_sync);
438 * We don't need the file
440 static int exofs_readpage(struct file *file, struct page *page)
442 return _readpage(page, false);
445 /* Callback for osd_write. All writes are asynchronouse */
446 static void writepages_done(struct osd_request *or, void *p)
448 struct page_collect *pcol = p;
449 struct bio_vec *bvec;
455 int ret = exofs_check_ok_resid(or, NULL, &resid);
458 atomic_dec(&pcol->sbi->s_curr_pending);
461 good_bytes = pcol->length;
465 good_bytes = pcol->length - resid;
467 EXOFS_DBGMSG("writepages_done(0x%lx) good_bytes=0x%llx"
468 " length=0x%lx nr_pages=%u\n",
469 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
472 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
473 struct page *page = bvec->bv_page;
474 struct inode *inode = page->mapping->host;
477 if (inode != pcol->inode)
478 continue; /* osd might add more pages to a bio */
480 if (likely(length < good_bytes))
485 update_write_page(page, page_stat);
487 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
488 inode->i_ino, page->index, page_stat);
490 length += bvec->bv_len;
495 EXOFS_DBGMSG("writepages_done END\n");
498 static int write_exec(struct page_collect *pcol)
500 struct exofs_i_info *oi = exofs_i(pcol->inode);
501 struct osd_obj_id obj = {pcol->sbi->s_pid,
502 pcol->inode->i_ino + EXOFS_OBJ_OFF};
503 struct osd_request *or = NULL;
504 struct page_collect *pcol_copy = NULL;
505 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
511 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
513 EXOFS_ERR("write_exec: Faild to osd_start_request()\n");
518 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
520 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
527 pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */
528 osd_req_write(or, &obj, i_start, pcol_copy->bio, pcol_copy->length);
529 ret = exofs_async_op(or, writepages_done, pcol_copy, oi->i_cred);
531 EXOFS_ERR("write_exec: exofs_async_op() Faild\n");
535 atomic_inc(&pcol->sbi->s_curr_pending);
536 EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
537 pcol->inode->i_ino, pcol->pg_first, _LLU(i_start),
539 /* pages ownership was passed to pcol_copy */
544 _unlock_pcol_pages(pcol, ret, WRITE);
551 /* writepage_strip is called either directly from writepage() or by the VFS from
552 * within write_cache_pages(), to add one more page to be written to storage.
553 * It will try to collect as many contiguous pages as possible. If a
554 * discontinuity is encountered or it runs out of resources it will submit the
555 * previous segment and will start a new collection.
556 * Eventually caller must submit the last segment if present.
558 static int writepage_strip(struct page *page,
559 struct writeback_control *wbc_unused, void *data)
561 struct page_collect *pcol = data;
562 struct inode *inode = pcol->inode;
563 struct exofs_i_info *oi = exofs_i(inode);
564 loff_t i_size = i_size_read(inode);
565 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
569 BUG_ON(!PageLocked(page));
571 ret = wait_obj_created(oi);
575 if (page->index < end_index)
576 /* in this case, the page is within the limits of the file */
577 len = PAGE_CACHE_SIZE;
579 len = i_size & ~PAGE_CACHE_MASK;
581 if (page->index > end_index || !len) {
582 /* in this case, the page is outside the limits
583 * (truncate in progress)
585 ret = write_exec(pcol);
589 ClearPageError(page);
597 if (unlikely(pcol->pg_first == -1)) {
598 pcol->pg_first = page->index;
599 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
601 /* Discontinuity detected, split the request */
602 ret = write_exec(pcol);
609 ret = pcol_try_alloc(pcol);
614 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
615 inode->i_ino, page->index, len);
617 ret = pcol_add_page(pcol, page, len);
619 EXOFS_DBGMSG("Failed pcol_add_page "
620 "nr_pages=%u total_length=0x%lx\n",
621 pcol->nr_pages, pcol->length);
623 /* split the request, next loop will start again */
624 ret = write_exec(pcol);
626 EXOFS_DBGMSG("write_exec faild => %d", ret);
633 BUG_ON(PageWriteback(page));
634 set_page_writeback(page);
639 set_bit(AS_EIO, &page->mapping->flags);
644 static int exofs_writepages(struct address_space *mapping,
645 struct writeback_control *wbc)
647 struct page_collect pcol;
648 long start, end, expected_pages;
651 start = wbc->range_start >> PAGE_CACHE_SHIFT;
652 end = (wbc->range_end == LLONG_MAX) ?
653 start + mapping->nrpages :
654 wbc->range_end >> PAGE_CACHE_SHIFT;
657 expected_pages = min(end - start + 1, 32L);
659 expected_pages = mapping->nrpages;
661 EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx"
662 " m->nrpages=%lu start=0x%lx end=0x%lx\n",
663 mapping->host->i_ino, wbc->range_start, wbc->range_end,
664 mapping->nrpages, start, end);
666 _pcol_init(&pcol, expected_pages, mapping->host);
668 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
670 EXOFS_ERR("write_cache_pages => %d\n", ret);
674 return write_exec(&pcol);
677 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
679 struct page_collect pcol;
682 _pcol_init(&pcol, 1, page->mapping->host);
684 ret = writepage_strip(page, NULL, &pcol);
686 EXOFS_ERR("exofs_writepage => %d\n", ret);
690 return write_exec(&pcol);
693 int exofs_write_begin(struct file *file, struct address_space *mapping,
694 loff_t pos, unsigned len, unsigned flags,
695 struct page **pagep, void **fsdata)
702 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
705 EXOFS_DBGMSG("simple_write_begin faild\n");
712 /* read modify write */
713 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
714 ret = _readpage(page, true);
716 /*SetPageError was done by _readpage. Is it ok?*/
718 EXOFS_DBGMSG("__readpage_filler faild\n");
725 static int exofs_write_begin_export(struct file *file,
726 struct address_space *mapping,
727 loff_t pos, unsigned len, unsigned flags,
728 struct page **pagep, void **fsdata)
732 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
736 const struct address_space_operations exofs_aops = {
737 .readpage = exofs_readpage,
738 .readpages = exofs_readpages,
739 .writepage = exofs_writepage,
740 .writepages = exofs_writepages,
741 .write_begin = exofs_write_begin_export,
742 .write_end = simple_write_end,
745 /******************************************************************************
747 *****************************************************************************/
750 * Test whether an inode is a fast symlink.
752 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
754 struct exofs_i_info *oi = exofs_i(inode);
756 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
760 * get_block_t - Fill in a buffer_head
761 * An OSD takes care of block allocation so we just fake an allocation by
762 * putting in the inode's sector_t in the buffer_head.
763 * TODO: What about the case of create==0 and @iblock does not exist in the
766 static int exofs_get_block(struct inode *inode, sector_t iblock,
767 struct buffer_head *bh_result, int create)
769 map_bh(bh_result, inode->i_sb, iblock);
773 const struct osd_attr g_attr_logical_length = ATTR_DEF(
774 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
777 * Truncate a file to the specified size - all we have to do is set the size
778 * attribute. We make sure the object exists first.
780 void exofs_truncate(struct inode *inode)
782 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
783 struct exofs_i_info *oi = exofs_i(inode);
784 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
785 struct osd_request *or;
786 struct osd_attr attr;
787 loff_t isize = i_size_read(inode);
791 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
792 || S_ISLNK(inode->i_mode)))
794 if (exofs_inode_is_fast_symlink(inode))
796 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
798 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
800 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
802 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
804 EXOFS_ERR("ERROR: exofs_truncate: osd_start_request failed\n");
808 osd_req_set_attributes(or, &obj);
810 newsize = cpu_to_be64((u64)isize);
811 attr = g_attr_logical_length;
812 attr.val_ptr = &newsize;
813 osd_req_add_set_attr_list(or, &attr, 1);
815 /* if we are about to truncate an object, and it hasn't been
818 if (unlikely(wait_obj_created(oi)))
821 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
827 mark_inode_dirty(inode);
830 make_bad_inode(inode);
835 * Set inode attributes - just call generic functions.
837 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
839 struct inode *inode = dentry->d_inode;
842 error = inode_change_ok(inode, iattr);
846 error = inode_setattr(inode, iattr);
851 * Read an inode from the OSD, and return it as is. We also return the size
852 * attribute in the 'sanity' argument if we got compiled with debugging turned
855 static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
856 struct exofs_fcb *inode, uint64_t *sanity)
858 struct exofs_sb_info *sbi = sb->s_fs_info;
859 struct osd_request *or;
860 struct osd_attr attr;
861 struct osd_obj_id obj = {sbi->s_pid,
862 oi->vfs_inode.i_ino + EXOFS_OBJ_OFF};
865 exofs_make_credential(oi->i_cred, &obj);
867 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
869 EXOFS_ERR("exofs_get_inode: osd_start_request failed.\n");
872 osd_req_get_attributes(or, &obj);
874 /* we need the inode attribute */
875 osd_req_add_get_attr_list(or, &g_attr_inode_data, 1);
877 #ifdef EXOFS_DEBUG_OBJ_ISIZE
878 /* we get the size attributes to do a sanity check */
879 osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
882 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
886 attr = g_attr_inode_data;
887 ret = extract_attr_from_req(or, &attr);
889 EXOFS_ERR("exofs_get_inode: extract_attr_from_req failed\n");
893 WARN_ON(attr.len != EXOFS_INO_ATTR_SIZE);
894 memcpy(inode, attr.val_ptr, EXOFS_INO_ATTR_SIZE);
896 #ifdef EXOFS_DEBUG_OBJ_ISIZE
897 attr = g_attr_logical_length;
898 ret = extract_attr_from_req(or, &attr);
900 EXOFS_ERR("ERROR: extract attr from or failed\n");
903 *sanity = get_unaligned_be64(attr.val_ptr);
912 * Fill in an inode read from the OSD and set it up for use
914 struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
916 struct exofs_i_info *oi;
917 struct exofs_fcb fcb;
919 uint64_t uninitialized_var(sanity);
922 inode = iget_locked(sb, ino);
924 return ERR_PTR(-ENOMEM);
925 if (!(inode->i_state & I_NEW))
929 /* read the inode from the osd */
930 ret = exofs_get_inode(sb, oi, &fcb, &sanity);
934 init_waitqueue_head(&oi->i_wq);
937 /* copy stuff from on-disk struct to in-memory struct */
938 inode->i_mode = le16_to_cpu(fcb.i_mode);
939 inode->i_uid = le32_to_cpu(fcb.i_uid);
940 inode->i_gid = le32_to_cpu(fcb.i_gid);
941 inode->i_nlink = le16_to_cpu(fcb.i_links_count);
942 inode->i_ctime.tv_sec = (signed)le32_to_cpu(fcb.i_ctime);
943 inode->i_atime.tv_sec = (signed)le32_to_cpu(fcb.i_atime);
944 inode->i_mtime.tv_sec = (signed)le32_to_cpu(fcb.i_mtime);
945 inode->i_ctime.tv_nsec =
946 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = 0;
947 oi->i_commit_size = le64_to_cpu(fcb.i_size);
948 i_size_write(inode, oi->i_commit_size);
949 inode->i_blkbits = EXOFS_BLKSHIFT;
950 inode->i_generation = le32_to_cpu(fcb.i_generation);
952 #ifdef EXOFS_DEBUG_OBJ_ISIZE
953 if ((inode->i_size != sanity) &&
954 (!exofs_inode_is_fast_symlink(inode))) {
955 EXOFS_ERR("WARNING: Size of inode=%llu != object=%llu\n",
956 inode->i_size, _LLU(sanity));
960 oi->i_dir_start_lookup = 0;
962 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
967 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
970 old_decode_dev(le32_to_cpu(fcb.i_data[0]));
973 new_decode_dev(le32_to_cpu(fcb.i_data[1]));
975 memcpy(oi->i_data, fcb.i_data, sizeof(fcb.i_data));
978 if (S_ISREG(inode->i_mode)) {
979 inode->i_op = &exofs_file_inode_operations;
980 inode->i_fop = &exofs_file_operations;
981 inode->i_mapping->a_ops = &exofs_aops;
982 } else if (S_ISDIR(inode->i_mode)) {
983 inode->i_op = &exofs_dir_inode_operations;
984 inode->i_fop = &exofs_dir_operations;
985 inode->i_mapping->a_ops = &exofs_aops;
986 } else if (S_ISLNK(inode->i_mode)) {
987 if (exofs_inode_is_fast_symlink(inode))
988 inode->i_op = &exofs_fast_symlink_inode_operations;
990 inode->i_op = &exofs_symlink_inode_operations;
991 inode->i_mapping->a_ops = &exofs_aops;
994 inode->i_op = &exofs_special_inode_operations;
996 init_special_inode(inode, inode->i_mode,
997 old_decode_dev(le32_to_cpu(fcb.i_data[0])));
999 init_special_inode(inode, inode->i_mode,
1000 new_decode_dev(le32_to_cpu(fcb.i_data[1])));
1003 unlock_new_inode(inode);
1008 return ERR_PTR(ret);
1011 int __exofs_wait_obj_created(struct exofs_i_info *oi)
1013 if (!obj_created(oi)) {
1014 BUG_ON(!obj_2bcreated(oi));
1015 wait_event(oi->i_wq, obj_created(oi));
1017 return unlikely(is_bad_inode(&oi->vfs_inode)) ? -EIO : 0;
1020 * Callback function from exofs_new_inode(). The important thing is that we
1021 * set the obj_created flag so that other methods know that the object exists on
1024 static void create_done(struct osd_request *or, void *p)
1026 struct inode *inode = p;
1027 struct exofs_i_info *oi = exofs_i(inode);
1028 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1031 ret = exofs_check_ok(or);
1032 osd_end_request(or);
1033 atomic_dec(&sbi->s_curr_pending);
1035 if (unlikely(ret)) {
1036 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1037 _LLU(sbi->s_pid), _LLU(inode->i_ino + EXOFS_OBJ_OFF));
1038 make_bad_inode(inode);
1040 set_obj_created(oi);
1042 atomic_dec(&inode->i_count);
1047 * Set up a new inode and create an object for it on the OSD
1049 struct inode *exofs_new_inode(struct inode *dir, int mode)
1051 struct super_block *sb;
1052 struct inode *inode;
1053 struct exofs_i_info *oi;
1054 struct exofs_sb_info *sbi;
1055 struct osd_request *or;
1056 struct osd_obj_id obj;
1060 inode = new_inode(sb);
1062 return ERR_PTR(-ENOMEM);
1064 oi = exofs_i(inode);
1066 init_waitqueue_head(&oi->i_wq);
1067 set_obj_2bcreated(oi);
1069 sbi = sb->s_fs_info;
1072 inode->i_uid = current->cred->fsuid;
1073 if (dir->i_mode & S_ISGID) {
1074 inode->i_gid = dir->i_gid;
1078 inode->i_gid = current->cred->fsgid;
1080 inode->i_mode = mode;
1082 inode->i_ino = sbi->s_nextid++;
1083 inode->i_blkbits = EXOFS_BLKSHIFT;
1084 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1085 oi->i_commit_size = inode->i_size = 0;
1086 spin_lock(&sbi->s_next_gen_lock);
1087 inode->i_generation = sbi->s_next_generation++;
1088 spin_unlock(&sbi->s_next_gen_lock);
1089 insert_inode_hash(inode);
1091 mark_inode_dirty(inode);
1093 obj.partition = sbi->s_pid;
1094 obj.id = inode->i_ino + EXOFS_OBJ_OFF;
1095 exofs_make_credential(oi->i_cred, &obj);
1097 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1098 if (unlikely(!or)) {
1099 EXOFS_ERR("exofs_new_inode: osd_start_request failed\n");
1100 return ERR_PTR(-ENOMEM);
1103 osd_req_create_object(or, &obj);
1105 /* increment the refcount so that the inode will still be around when we
1106 * reach the callback
1108 atomic_inc(&inode->i_count);
1110 ret = exofs_async_op(or, create_done, inode, oi->i_cred);
1112 atomic_dec(&inode->i_count);
1113 osd_end_request(or);
1114 return ERR_PTR(-EIO);
1116 atomic_inc(&sbi->s_curr_pending);
1122 * struct to pass two arguments to update_inode's callback
1124 struct updatei_args {
1125 struct exofs_sb_info *sbi;
1126 struct exofs_fcb fcb;
1130 * Callback function from exofs_update_inode().
1132 static void updatei_done(struct osd_request *or, void *p)
1134 struct updatei_args *args = p;
1136 osd_end_request(or);
1138 atomic_dec(&args->sbi->s_curr_pending);
1144 * Write the inode to the OSD. Just fill up the struct, and set the attribute
1145 * synchronously or asynchronously depending on the do_sync flag.
1147 static int exofs_update_inode(struct inode *inode, int do_sync)
1149 struct exofs_i_info *oi = exofs_i(inode);
1150 struct super_block *sb = inode->i_sb;
1151 struct exofs_sb_info *sbi = sb->s_fs_info;
1152 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1153 struct osd_request *or;
1154 struct osd_attr attr;
1155 struct exofs_fcb *fcb;
1156 struct updatei_args *args;
1159 args = kzalloc(sizeof(*args), GFP_KERNEL);
1165 fcb->i_mode = cpu_to_le16(inode->i_mode);
1166 fcb->i_uid = cpu_to_le32(inode->i_uid);
1167 fcb->i_gid = cpu_to_le32(inode->i_gid);
1168 fcb->i_links_count = cpu_to_le16(inode->i_nlink);
1169 fcb->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1170 fcb->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1171 fcb->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1172 oi->i_commit_size = i_size_read(inode);
1173 fcb->i_size = cpu_to_le64(oi->i_commit_size);
1174 fcb->i_generation = cpu_to_le32(inode->i_generation);
1176 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1177 if (old_valid_dev(inode->i_rdev)) {
1179 cpu_to_le32(old_encode_dev(inode->i_rdev));
1184 cpu_to_le32(new_encode_dev(inode->i_rdev));
1188 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1190 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1191 if (unlikely(!or)) {
1192 EXOFS_ERR("exofs_update_inode: osd_start_request failed.\n");
1197 osd_req_set_attributes(or, &obj);
1199 attr = g_attr_inode_data;
1201 osd_req_add_set_attr_list(or, &attr, 1);
1203 if (!obj_created(oi)) {
1204 EXOFS_DBGMSG("!obj_created\n");
1205 BUG_ON(!obj_2bcreated(oi));
1206 wait_event(oi->i_wq, obj_created(oi));
1207 EXOFS_DBGMSG("wait_event done\n");
1211 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
1212 osd_end_request(or);
1217 ret = exofs_async_op(or, updatei_done, args, oi->i_cred);
1219 osd_end_request(or);
1222 atomic_inc(&sbi->s_curr_pending);
1223 goto out; /* deallocation in updatei_done */
1229 EXOFS_DBGMSG("ret=>%d\n", ret);
1233 int exofs_write_inode(struct inode *inode, int wait)
1235 return exofs_update_inode(inode, wait);
1239 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1242 static void delete_done(struct osd_request *or, void *p)
1244 struct exofs_sb_info *sbi;
1245 osd_end_request(or);
1247 atomic_dec(&sbi->s_curr_pending);
1251 * Called when the refcount of an inode reaches zero. We remove the object
1252 * from the OSD here. We make sure the object was created before we try and
1255 void exofs_delete_inode(struct inode *inode)
1257 struct exofs_i_info *oi = exofs_i(inode);
1258 struct super_block *sb = inode->i_sb;
1259 struct exofs_sb_info *sbi = sb->s_fs_info;
1260 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
1261 struct osd_request *or;
1264 truncate_inode_pages(&inode->i_data, 0);
1266 if (is_bad_inode(inode))
1269 mark_inode_dirty(inode);
1270 exofs_update_inode(inode, inode_needs_sync(inode));
1273 if (inode->i_blocks)
1274 exofs_truncate(inode);
1278 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1279 if (unlikely(!or)) {
1280 EXOFS_ERR("exofs_delete_inode: osd_start_request failed\n");
1284 osd_req_remove_object(or, &obj);
1286 /* if we are deleting an obj that hasn't been created yet, wait */
1287 if (!obj_created(oi)) {
1288 BUG_ON(!obj_2bcreated(oi));
1289 wait_event(oi->i_wq, obj_created(oi));
1292 ret = exofs_async_op(or, delete_done, sbi, oi->i_cred);
1295 "ERROR: @exofs_delete_inode exofs_async_op failed\n");
1296 osd_end_request(or);
1299 atomic_inc(&sbi->s_curr_pending);