2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com) (avishay@il.ibm.com)
4 * Copyright (C) 2005, 2006
5 * International Business Machines
6 * Copyright (C) 2008, 2009
7 * Boaz Harrosh <bharrosh@panasas.com>
9 * Copyrights for code taken from ext2:
10 * Copyright (C) 1992, 1993, 1994, 1995
11 * Remy Card (card@masi.ibp.fr)
12 * Laboratoire MASI - Institut Blaise Pascal
13 * Universite Pierre et Marie Curie (Paris VI)
15 * linux/fs/minix/inode.c
16 * Copyright (C) 1991, 1992 Linus Torvalds
18 * This file is part of exofs.
20 * exofs is free software; you can redistribute it and/or modify
21 * it under the terms of the GNU General Public License as published by
22 * the Free Software Foundation. Since it is based on ext2, and the only
23 * valid version of GPL for the Linux kernel is version 2, the only valid
24 * version of GPL for exofs is version 2.
26 * exofs is distributed in the hope that it will be useful,
27 * but WITHOUT ANY WARRANTY; without even the implied warranty of
28 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
29 * GNU General Public License for more details.
31 * You should have received a copy of the GNU General Public License
32 * along with exofs; if not, write to the Free Software
33 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
36 #include <linux/writeback.h>
37 #include <linux/buffer_head.h>
38 #include <scsi/scsi_device.h>
42 #ifdef CONFIG_EXOFS_DEBUG
43 # define EXOFS_DEBUG_OBJ_ISIZE 1
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;
62 struct request_queue *req_q = sbi->s_dev->scsi_device->request_queue;
67 pcol->expected_pages = expected_pages;
74 EXOFS_DBGMSG("_pcol_init ino=0x%lx expected_pages=%u\n", inode->i_ino,
78 static void _pcol_reset(struct page_collect *pcol)
80 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
86 EXOFS_DBGMSG("_pcol_reset ino=0x%lx expected_pages=%u\n",
87 pcol->inode->i_ino, pcol->expected_pages);
89 /* this is probably the end of the loop but in writes
90 * it might not end here. don't be left with nothing
92 if (!pcol->expected_pages)
93 pcol->expected_pages = 128;
96 static int pcol_try_alloc(struct page_collect *pcol)
98 int pages = min_t(unsigned, pcol->expected_pages, BIO_MAX_PAGES);
100 for (; pages; pages >>= 1) {
101 pcol->bio = bio_alloc(GFP_KERNEL, pages);
102 if (likely(pcol->bio))
106 EXOFS_ERR("Failed to kcalloc expected_pages=%u\n",
107 pcol->expected_pages);
111 static void pcol_free(struct page_collect *pcol)
117 static int pcol_add_page(struct page_collect *pcol, struct page *page,
120 int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0);
121 if (unlikely(len != added_len))
129 static int update_read_page(struct page *page, int ret)
132 /* Everything is OK */
133 SetPageUptodate(page);
135 ClearPageError(page);
136 } else if (ret == -EFAULT) {
137 /* In this case we were trying to read something that wasn't on
138 * disk yet - return a page full of zeroes. This should be OK,
139 * because the object should be empty (if there was a write
140 * before this read, the read would be waiting with the page
142 clear_highpage(page);
144 SetPageUptodate(page);
146 ClearPageError(page);
147 ret = 0; /* recovered error */
148 EXOFS_DBGMSG("recovered read error\n");
155 static void update_write_page(struct page *page, int ret)
158 mapping_set_error(page->mapping, ret);
161 end_page_writeback(page);
164 /* Called at the end of reads, to optionally unlock pages and update their
167 static int __readpages_done(struct osd_request *or, struct page_collect *pcol,
170 struct bio_vec *bvec;
175 int ret = exofs_check_ok_resid(or, &resid, NULL);
180 good_bytes = pcol->length;
184 good_bytes = pcol->length - resid;
186 EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx"
187 " length=0x%lx nr_pages=%u\n",
188 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
191 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
192 struct page *page = bvec->bv_page;
193 struct inode *inode = page->mapping->host;
196 if (inode != pcol->inode)
197 continue; /* osd might add more pages at end */
199 if (likely(length < good_bytes))
204 EXOFS_DBGMSG(" readpages_done(0x%lx, 0x%lx) %s\n",
205 inode->i_ino, page->index,
206 page_stat ? "bad_bytes" : "good_bytes");
208 ret = update_read_page(page, page_stat);
211 length += bvec->bv_len;
215 EXOFS_DBGMSG("readpages_done END\n");
219 /* callback of async reads */
220 static void readpages_done(struct osd_request *or, void *p)
222 struct page_collect *pcol = p;
224 __readpages_done(or, pcol, true);
225 atomic_dec(&pcol->sbi->s_curr_pending);
229 static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
231 struct bio_vec *bvec;
234 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
235 struct page *page = bvec->bv_page;
238 update_read_page(page, ret);
240 update_write_page(page, ret);
247 static int read_exec(struct page_collect *pcol, bool is_sync)
249 struct exofs_i_info *oi = exofs_i(pcol->inode);
250 struct osd_obj_id obj = {pcol->sbi->s_pid,
251 pcol->inode->i_ino + EXOFS_OBJ_OFF};
252 struct osd_request *or = NULL;
253 struct page_collect *pcol_copy = NULL;
254 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
260 /* see comment in _readpage() about sync reads */
261 WARN_ON(is_sync && (pcol->nr_pages != 1));
263 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
269 osd_req_read(or, &obj, pcol->bio, i_start);
272 exofs_sync_op(or, pcol->sbi->s_timeout, oi->i_cred);
273 return __readpages_done(or, pcol, false);
276 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
283 ret = exofs_async_op(or, readpages_done, pcol_copy, oi->i_cred);
287 atomic_inc(&pcol->sbi->s_curr_pending);
289 EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
290 obj.id, _LLU(i_start), pcol->length);
292 /* pages ownership was passed to pcol_copy */
298 _unlock_pcol_pages(pcol, ret, READ);
305 /* readpage_strip is called either directly from readpage() or by the VFS from
306 * within read_cache_pages(), to add one more page to be read. It will try to
307 * collect as many contiguous pages as posible. If a discontinuity is
308 * encountered, or it runs out of resources, it will submit the previous segment
309 * and will start a new collection. Eventually caller must submit the last
310 * segment if present.
312 static int readpage_strip(void *data, struct page *page)
314 struct page_collect *pcol = data;
315 struct inode *inode = pcol->inode;
316 struct exofs_i_info *oi = exofs_i(inode);
317 loff_t i_size = i_size_read(inode);
318 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
322 /* FIXME: Just for debugging, will be removed */
323 if (PageUptodate(page))
324 EXOFS_ERR("PageUptodate(0x%lx, 0x%lx)\n", pcol->inode->i_ino,
327 if (page->index < end_index)
328 len = PAGE_CACHE_SIZE;
329 else if (page->index == end_index)
330 len = i_size & ~PAGE_CACHE_MASK;
334 if (!len || !obj_created(oi)) {
335 /* this will be out of bounds, or doesn't exist yet.
336 * Current page is cleared and the request is split
338 clear_highpage(page);
340 SetPageUptodate(page);
342 ClearPageError(page);
345 EXOFS_DBGMSG("readpage_strip(0x%lx, 0x%lx) empty page,"
346 " splitting\n", inode->i_ino, page->index);
348 return read_exec(pcol, false);
353 if (unlikely(pcol->pg_first == -1)) {
354 pcol->pg_first = page->index;
355 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
357 /* Discontinuity detected, split the request */
358 ret = read_exec(pcol, false);
365 ret = pcol_try_alloc(pcol);
370 if (len != PAGE_CACHE_SIZE)
371 zero_user(page, len, PAGE_CACHE_SIZE - len);
373 EXOFS_DBGMSG(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
374 inode->i_ino, page->index, len);
376 ret = pcol_add_page(pcol, page, len);
378 EXOFS_DBGMSG("Failed pcol_add_page pages[i]=%p "
379 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
380 page, len, pcol->nr_pages, pcol->length);
382 /* split the request, and start again with current page */
383 ret = read_exec(pcol, false);
393 /* SetPageError(page); ??? */
398 static int exofs_readpages(struct file *file, struct address_space *mapping,
399 struct list_head *pages, unsigned nr_pages)
401 struct page_collect pcol;
404 _pcol_init(&pcol, nr_pages, mapping->host);
406 ret = read_cache_pages(mapping, pages, readpage_strip, &pcol);
408 EXOFS_ERR("read_cache_pages => %d\n", ret);
412 return read_exec(&pcol, false);
415 static int _readpage(struct page *page, bool is_sync)
417 struct page_collect pcol;
420 _pcol_init(&pcol, 1, page->mapping->host);
422 /* readpage_strip might call read_exec(,async) inside at several places
423 * but this is safe for is_async=0 since read_exec will not do anything
424 * when we have a single page.
426 ret = readpage_strip(&pcol, page);
428 EXOFS_ERR("_readpage => %d\n", ret);
432 return read_exec(&pcol, is_sync);
436 * We don't need the file
438 static int exofs_readpage(struct file *file, struct page *page)
440 return _readpage(page, false);
443 /* Callback for osd_write. All writes are asynchronouse */
444 static void writepages_done(struct osd_request *or, void *p)
446 struct page_collect *pcol = p;
447 struct bio_vec *bvec;
453 int ret = exofs_check_ok_resid(or, NULL, &resid);
456 atomic_dec(&pcol->sbi->s_curr_pending);
459 good_bytes = pcol->length;
463 good_bytes = pcol->length - resid;
465 EXOFS_DBGMSG("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 __bio_for_each_segment(bvec, pcol->bio, i, 0) {
471 struct page *page = bvec->bv_page;
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_DBGMSG(" writepages_done(0x%lx, 0x%lx) status=%d\n",
486 inode->i_ino, page->index, page_stat);
488 length += bvec->bv_len;
493 EXOFS_DBGMSG("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 osd_obj_id obj = {pcol->sbi->s_pid,
500 pcol->inode->i_ino + EXOFS_OBJ_OFF};
501 struct osd_request *or = NULL;
502 struct page_collect *pcol_copy = NULL;
503 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
509 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
511 EXOFS_ERR("write_exec: Faild to osd_start_request()\n");
516 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
518 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
525 osd_req_write(or, &obj, pcol_copy->bio, i_start);
526 ret = exofs_async_op(or, writepages_done, pcol_copy, oi->i_cred);
528 EXOFS_ERR("write_exec: exofs_async_op() Faild\n");
532 atomic_inc(&pcol->sbi->s_curr_pending);
533 EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
534 pcol->inode->i_ino, pcol->pg_first, _LLU(i_start),
536 /* pages ownership was passed to pcol_copy */
541 _unlock_pcol_pages(pcol, ret, WRITE);
548 /* writepage_strip is called either directly from writepage() or by the VFS from
549 * within write_cache_pages(), to add one more page to be written to storage.
550 * It will try to collect as many contiguous pages as possible. If a
551 * discontinuity is encountered or it runs out of resources it will submit the
552 * previous segment and will start a new collection.
553 * Eventually caller must submit the last segment if present.
555 static int writepage_strip(struct page *page,
556 struct writeback_control *wbc_unused, void *data)
558 struct page_collect *pcol = data;
559 struct inode *inode = pcol->inode;
560 struct exofs_i_info *oi = exofs_i(inode);
561 loff_t i_size = i_size_read(inode);
562 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
566 BUG_ON(!PageLocked(page));
568 ret = wait_obj_created(oi);
572 if (page->index < end_index)
573 /* in this case, the page is within the limits of the file */
574 len = PAGE_CACHE_SIZE;
576 len = i_size & ~PAGE_CACHE_MASK;
578 if (page->index > end_index || !len) {
579 /* in this case, the page is outside the limits
580 * (truncate in progress)
582 ret = write_exec(pcol);
586 ClearPageError(page);
594 if (unlikely(pcol->pg_first == -1)) {
595 pcol->pg_first = page->index;
596 } else if (unlikely((pcol->pg_first + pcol->nr_pages) !=
598 /* Discontinuity detected, split the request */
599 ret = write_exec(pcol);
606 ret = pcol_try_alloc(pcol);
611 EXOFS_DBGMSG(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
612 inode->i_ino, page->index, len);
614 ret = pcol_add_page(pcol, page, len);
616 EXOFS_DBGMSG("Failed pcol_add_page "
617 "nr_pages=%u total_length=0x%lx\n",
618 pcol->nr_pages, pcol->length);
620 /* split the request, next loop will start again */
621 ret = write_exec(pcol);
623 EXOFS_DBGMSG("write_exec faild => %d", ret);
630 BUG_ON(PageWriteback(page));
631 set_page_writeback(page);
636 set_bit(AS_EIO, &page->mapping->flags);
641 static int exofs_writepages(struct address_space *mapping,
642 struct writeback_control *wbc)
644 struct page_collect pcol;
645 long start, end, expected_pages;
648 start = wbc->range_start >> PAGE_CACHE_SHIFT;
649 end = (wbc->range_end == LLONG_MAX) ?
650 start + mapping->nrpages :
651 wbc->range_end >> PAGE_CACHE_SHIFT;
654 expected_pages = min(end - start + 1, 32L);
656 expected_pages = mapping->nrpages;
658 EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx"
659 " m->nrpages=%lu start=0x%lx end=0x%lx\n",
660 mapping->host->i_ino, wbc->range_start, wbc->range_end,
661 mapping->nrpages, start, end);
663 _pcol_init(&pcol, expected_pages, mapping->host);
665 ret = write_cache_pages(mapping, wbc, writepage_strip, &pcol);
667 EXOFS_ERR("write_cache_pages => %d\n", ret);
671 return write_exec(&pcol);
674 static int exofs_writepage(struct page *page, struct writeback_control *wbc)
676 struct page_collect pcol;
679 _pcol_init(&pcol, 1, page->mapping->host);
681 ret = writepage_strip(page, NULL, &pcol);
683 EXOFS_ERR("exofs_writepage => %d\n", ret);
687 return write_exec(&pcol);
690 int exofs_write_begin(struct file *file, struct address_space *mapping,
691 loff_t pos, unsigned len, unsigned flags,
692 struct page **pagep, void **fsdata)
699 ret = simple_write_begin(file, mapping, pos, len, flags, pagep,
702 EXOFS_DBGMSG("simple_write_begin faild\n");
709 /* read modify write */
710 if (!PageUptodate(page) && (len != PAGE_CACHE_SIZE)) {
711 ret = _readpage(page, true);
713 /*SetPageError was done by _readpage. Is it ok?*/
715 EXOFS_DBGMSG("__readpage_filler faild\n");
722 static int exofs_write_begin_export(struct file *file,
723 struct address_space *mapping,
724 loff_t pos, unsigned len, unsigned flags,
725 struct page **pagep, void **fsdata)
729 return exofs_write_begin(file, mapping, pos, len, flags, pagep,
733 const struct address_space_operations exofs_aops = {
734 .readpage = exofs_readpage,
735 .readpages = exofs_readpages,
736 .writepage = exofs_writepage,
737 .writepages = exofs_writepages,
738 .write_begin = exofs_write_begin_export,
739 .write_end = simple_write_end,
742 /******************************************************************************
744 *****************************************************************************/
747 * Test whether an inode is a fast symlink.
749 static inline int exofs_inode_is_fast_symlink(struct inode *inode)
751 struct exofs_i_info *oi = exofs_i(inode);
753 return S_ISLNK(inode->i_mode) && (oi->i_data[0] != 0);
757 * get_block_t - Fill in a buffer_head
758 * An OSD takes care of block allocation so we just fake an allocation by
759 * putting in the inode's sector_t in the buffer_head.
760 * TODO: What about the case of create==0 and @iblock does not exist in the
763 static int exofs_get_block(struct inode *inode, sector_t iblock,
764 struct buffer_head *bh_result, int create)
766 map_bh(bh_result, inode->i_sb, iblock);
770 const struct osd_attr g_attr_logical_length = ATTR_DEF(
771 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
774 * Truncate a file to the specified size - all we have to do is set the size
775 * attribute. We make sure the object exists first.
777 void exofs_truncate(struct inode *inode)
779 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
780 struct exofs_i_info *oi = exofs_i(inode);
781 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
782 struct osd_request *or;
783 struct osd_attr attr;
784 loff_t isize = i_size_read(inode);
788 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
789 || S_ISLNK(inode->i_mode)))
791 if (exofs_inode_is_fast_symlink(inode))
793 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
795 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
797 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
799 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
801 EXOFS_ERR("ERROR: exofs_truncate: osd_start_request failed\n");
805 osd_req_set_attributes(or, &obj);
807 newsize = cpu_to_be64((u64)isize);
808 attr = g_attr_logical_length;
809 attr.val_ptr = &newsize;
810 osd_req_add_set_attr_list(or, &attr, 1);
812 /* if we are about to truncate an object, and it hasn't been
815 if (unlikely(wait_obj_created(oi)))
818 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
824 mark_inode_dirty(inode);
827 make_bad_inode(inode);
832 * Set inode attributes - just call generic functions.
834 int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
836 struct inode *inode = dentry->d_inode;
839 error = inode_change_ok(inode, iattr);
843 error = inode_setattr(inode, iattr);