2 * linux/fs/nfs/blocklayout/blocklayout.c
4 * Module for the NFSv4.1 pNFS block layout driver.
6 * Copyright (c) 2006 The Regents of the University of Michigan.
9 * Andy Adamson <andros@citi.umich.edu>
10 * Fred Isaman <iisaman@umich.edu>
12 * permission is granted to use, copy, create derivative works and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the university of michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization. if
17 * the above copyright notice or any other identification of the
18 * university of michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
21 * this software is provided as is, without representation from the
22 * university of michigan as to its fitness for any purpose, and without
23 * warranty by the university of michigan of any kind, either express
24 * or implied, including without limitation the implied warranties of
25 * merchantability and fitness for a particular purpose. the regents
26 * of the university of michigan shall not be liable for any damages,
27 * including special, indirect, incidental, or consequential damages,
28 * with respect to any claim arising out or in connection with the use
29 * of the software, even if it has been or is hereafter advised of the
30 * possibility of such damages.
33 #include <linux/module.h>
34 #include <linux/init.h>
35 #include <linux/mount.h>
36 #include <linux/namei.h>
37 #include <linux/bio.h> /* struct bio */
38 #include <linux/buffer_head.h> /* various write calls */
39 #include <linux/prefetch.h>
40 #include <linux/pagevec.h>
43 #include "../nfs4session.h"
44 #include "../internal.h"
45 #include "blocklayout.h"
47 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
49 MODULE_LICENSE("GPL");
50 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
51 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
53 static void print_page(struct page *page)
55 dprintk("PRINTPAGE page %p\n", page);
56 dprintk(" PagePrivate %d\n", PagePrivate(page));
57 dprintk(" PageUptodate %d\n", PageUptodate(page));
58 dprintk(" PageError %d\n", PageError(page));
59 dprintk(" PageDirty %d\n", PageDirty(page));
60 dprintk(" PageReferenced %d\n", PageReferenced(page));
61 dprintk(" PageLocked %d\n", PageLocked(page));
62 dprintk(" PageWriteback %d\n", PageWriteback(page));
63 dprintk(" PageMappedToDisk %d\n", PageMappedToDisk(page));
67 /* Given the be associated with isect, determine if page data needs to be
70 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
72 if (be->be_state == PNFS_BLOCK_NONE_DATA)
74 else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
77 return !bl_is_sector_init(be->be_inval, isect);
80 /* Given the be associated with isect, determine if page data can be
83 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
85 return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
86 be->be_state == PNFS_BLOCK_INVALID_DATA);
89 /* The data we are handed might be spread across several bios. We need
90 * to track when the last one is finished.
94 void (*pnfs_callback) (void *data, int num_se);
99 static inline struct parallel_io *alloc_parallel(void *data)
101 struct parallel_io *rv;
103 rv = kmalloc(sizeof(*rv), GFP_NOFS);
106 kref_init(&rv->refcnt);
112 static inline void get_parallel(struct parallel_io *p)
114 kref_get(&p->refcnt);
117 static void destroy_parallel(struct kref *kref)
119 struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
121 dprintk("%s enter\n", __func__);
122 p->pnfs_callback(p->data, p->bse_count);
126 static inline void put_parallel(struct parallel_io *p)
128 kref_put(&p->refcnt, destroy_parallel);
132 bl_submit_bio(int rw, struct bio *bio)
135 get_parallel(bio->bi_private);
136 dprintk("%s submitting %s bio %u@%llu\n", __func__,
137 rw == READ ? "read" : "write",
138 bio->bi_size, (unsigned long long)bio->bi_sector);
144 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
145 struct pnfs_block_extent *be,
146 void (*end_io)(struct bio *, int err),
147 struct parallel_io *par)
151 npg = min(npg, BIO_MAX_PAGES);
152 bio = bio_alloc(GFP_NOIO, npg);
153 if (!bio && (current->flags & PF_MEMALLOC)) {
154 while (!bio && (npg /= 2))
155 bio = bio_alloc(GFP_NOIO, npg);
159 bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
160 bio->bi_bdev = be->be_mdev;
161 bio->bi_end_io = end_io;
162 bio->bi_private = par;
167 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
168 sector_t isect, struct page *page,
169 struct pnfs_block_extent *be,
170 void (*end_io)(struct bio *, int err),
171 struct parallel_io *par,
172 unsigned int offset, int len)
174 isect = isect + (offset >> SECTOR_SHIFT);
175 dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
176 npg, rw, (unsigned long long)isect, offset, len);
179 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
181 return ERR_PTR(-ENOMEM);
183 if (bio_add_page(bio, page, len, offset) < len) {
184 bio = bl_submit_bio(rw, bio);
190 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
191 sector_t isect, struct page *page,
192 struct pnfs_block_extent *be,
193 void (*end_io)(struct bio *, int err),
194 struct parallel_io *par)
196 return do_add_page_to_bio(bio, npg, rw, isect, page, be,
197 end_io, par, 0, PAGE_CACHE_SIZE);
200 /* This is basically copied from mpage_end_io_read */
201 static void bl_end_io_read(struct bio *bio, int err)
203 struct parallel_io *par = bio->bi_private;
204 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
205 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
208 struct page *page = bvec->bv_page;
210 if (--bvec >= bio->bi_io_vec)
211 prefetchw(&bvec->bv_page->flags);
213 SetPageUptodate(page);
214 } while (bvec >= bio->bi_io_vec);
216 struct nfs_read_data *rdata = par->data;
217 struct nfs_pgio_header *header = rdata->header;
219 if (!header->pnfs_error)
220 header->pnfs_error = -EIO;
221 pnfs_set_lo_fail(header->lseg);
227 static void bl_read_cleanup(struct work_struct *work)
229 struct rpc_task *task;
230 struct nfs_read_data *rdata;
231 dprintk("%s enter\n", __func__);
232 task = container_of(work, struct rpc_task, u.tk_work);
233 rdata = container_of(task, struct nfs_read_data, task);
234 pnfs_ld_read_done(rdata);
238 bl_end_par_io_read(void *data, int unused)
240 struct nfs_read_data *rdata = data;
242 rdata->task.tk_status = rdata->header->pnfs_error;
243 INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
244 schedule_work(&rdata->task.u.tk_work);
247 static enum pnfs_try_status
248 bl_read_pagelist(struct nfs_read_data *rdata)
250 struct nfs_pgio_header *header = rdata->header;
252 struct bio *bio = NULL;
253 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
254 sector_t isect, extent_length = 0;
255 struct parallel_io *par;
256 loff_t f_offset = rdata->args.offset;
257 size_t bytes_left = rdata->args.count;
258 unsigned int pg_offset, pg_len;
259 struct page **pages = rdata->args.pages;
260 int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
261 const bool is_dio = (header->dreq != NULL);
263 dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
264 rdata->pages.npages, f_offset, (unsigned int)rdata->args.count);
266 par = alloc_parallel(rdata);
269 par->pnfs_callback = bl_end_par_io_read;
270 /* At this point, we can no longer jump to use_mds */
272 isect = (sector_t) (f_offset >> SECTOR_SHIFT);
273 /* Code assumes extents are page-aligned */
274 for (i = pg_index; i < rdata->pages.npages; i++) {
275 if (!extent_length) {
276 /* We've used up the previous extent */
278 bl_put_extent(cow_read);
279 bio = bl_submit_bio(READ, bio);
280 /* Get the next one */
281 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
284 header->pnfs_error = -EIO;
287 extent_length = be->be_length -
288 (isect - be->be_f_offset);
290 sector_t cow_length = cow_read->be_length -
291 (isect - cow_read->be_f_offset);
292 extent_length = min(extent_length, cow_length);
297 pg_offset = f_offset & ~PAGE_CACHE_MASK;
298 if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
299 pg_len = PAGE_CACHE_SIZE - pg_offset;
304 bytes_left -= pg_len;
305 isect += (pg_offset >> SECTOR_SHIFT);
308 pg_len = PAGE_CACHE_SIZE;
311 hole = is_hole(be, isect);
312 if (hole && !cow_read) {
313 bio = bl_submit_bio(READ, bio);
314 /* Fill hole w/ zeroes w/o accessing device */
315 dprintk("%s Zeroing page for hole\n", __func__);
316 zero_user_segment(pages[i], pg_offset, pg_len);
317 print_page(pages[i]);
318 SetPageUptodate(pages[i]);
320 struct pnfs_block_extent *be_read;
322 be_read = (hole && cow_read) ? cow_read : be;
323 bio = do_add_page_to_bio(bio, rdata->pages.npages - i,
325 isect, pages[i], be_read,
329 header->pnfs_error = PTR_ERR(bio);
334 isect += (pg_len >> SECTOR_SHIFT);
335 extent_length -= PAGE_CACHE_SECTORS;
337 if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
339 rdata->res.count = header->inode->i_size - rdata->args.offset;
341 rdata->res.count = (isect << SECTOR_SHIFT) - rdata->args.offset;
345 bl_put_extent(cow_read);
346 bl_submit_bio(READ, bio);
348 return PNFS_ATTEMPTED;
351 dprintk("Giving up and using normal NFS\n");
352 return PNFS_NOT_ATTEMPTED;
355 static void mark_extents_written(struct pnfs_block_layout *bl,
356 __u64 offset, __u32 count)
359 struct pnfs_block_extent *be;
360 struct pnfs_block_short_extent *se;
362 dprintk("%s(%llu, %u)\n", __func__, offset, count);
365 isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
366 end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
367 end >>= SECTOR_SHIFT;
368 while (isect < end) {
370 be = bl_find_get_extent(bl, isect, NULL);
371 BUG_ON(!be); /* FIXME */
372 len = min(end, be->be_f_offset + be->be_length) - isect;
373 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
374 se = bl_pop_one_short_extent(be->be_inval);
376 bl_mark_for_commit(be, isect, len, se);
383 static void bl_end_io_write_zero(struct bio *bio, int err)
385 struct parallel_io *par = bio->bi_private;
386 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
387 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
390 struct page *page = bvec->bv_page;
392 if (--bvec >= bio->bi_io_vec)
393 prefetchw(&bvec->bv_page->flags);
394 /* This is the zeroing page we added */
395 end_page_writeback(page);
396 page_cache_release(page);
397 } while (bvec >= bio->bi_io_vec);
399 if (unlikely(!uptodate)) {
400 struct nfs_write_data *data = par->data;
401 struct nfs_pgio_header *header = data->header;
403 if (!header->pnfs_error)
404 header->pnfs_error = -EIO;
405 pnfs_set_lo_fail(header->lseg);
411 static void bl_end_io_write(struct bio *bio, int err)
413 struct parallel_io *par = bio->bi_private;
414 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
415 struct nfs_write_data *data = par->data;
416 struct nfs_pgio_header *header = data->header;
419 if (!header->pnfs_error)
420 header->pnfs_error = -EIO;
421 pnfs_set_lo_fail(header->lseg);
427 /* Function scheduled for call during bl_end_par_io_write,
428 * it marks sectors as written and extends the commitlist.
430 static void bl_write_cleanup(struct work_struct *work)
432 struct rpc_task *task;
433 struct nfs_write_data *wdata;
434 dprintk("%s enter\n", __func__);
435 task = container_of(work, struct rpc_task, u.tk_work);
436 wdata = container_of(task, struct nfs_write_data, task);
437 if (likely(!wdata->header->pnfs_error)) {
438 /* Marks for LAYOUTCOMMIT */
439 mark_extents_written(BLK_LSEG2EXT(wdata->header->lseg),
440 wdata->args.offset, wdata->args.count);
442 pnfs_ld_write_done(wdata);
445 /* Called when last of bios associated with a bl_write_pagelist call finishes */
446 static void bl_end_par_io_write(void *data, int num_se)
448 struct nfs_write_data *wdata = data;
450 if (unlikely(wdata->header->pnfs_error)) {
451 bl_free_short_extents(&BLK_LSEG2EXT(wdata->header->lseg)->bl_inval,
455 wdata->task.tk_status = wdata->header->pnfs_error;
456 wdata->verf.committed = NFS_FILE_SYNC;
457 INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
458 schedule_work(&wdata->task.u.tk_work);
461 /* FIXME STUB - mark intersection of layout and page as bad, so is not
464 static void mark_bad_read(void)
470 * map_block: map a requested I/0 block (isect) into an offset in the LVM
474 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
476 dprintk("%s enter be=%p\n", __func__, be);
478 set_buffer_mapped(bh);
479 bh->b_bdev = be->be_mdev;
480 bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
481 (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
483 dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
484 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
490 bl_read_single_end_io(struct bio *bio, int error)
492 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
493 struct page *page = bvec->bv_page;
495 /* Only one page in bvec */
500 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
501 unsigned int offset, unsigned int len)
504 struct page *shadow_page;
506 char *kaddr, *kshadow_addr;
509 dprintk("%s: offset %u len %u\n", __func__, offset, len);
511 shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
512 if (shadow_page == NULL)
515 bio = bio_alloc(GFP_NOIO, 1);
519 isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
520 (offset / SECTOR_SIZE);
522 bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
523 bio->bi_bdev = be->be_mdev;
524 bio->bi_end_io = bl_read_single_end_io;
526 lock_page(shadow_page);
527 if (bio_add_page(bio, shadow_page,
528 SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
529 unlock_page(shadow_page);
534 submit_bio(READ, bio);
535 wait_on_page_locked(shadow_page);
536 if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
539 kaddr = kmap_atomic(page);
540 kshadow_addr = kmap_atomic(shadow_page);
541 memcpy(kaddr + offset, kshadow_addr + offset, len);
542 kunmap_atomic(kshadow_addr);
543 kunmap_atomic(kaddr);
545 __free_page(shadow_page);
552 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
553 unsigned int dirty_offset, unsigned int dirty_len,
557 unsigned int start, end;
561 end = PAGE_CACHE_SIZE;
563 start = round_down(dirty_offset, SECTOR_SIZE);
564 end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
567 dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
569 zero_user_segments(page, start, dirty_offset,
570 dirty_offset + dirty_len, end);
571 if (start == 0 && end == PAGE_CACHE_SIZE &&
572 trylock_page(page)) {
573 SetPageUptodate(page);
579 if (start != dirty_offset)
580 ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
582 if (!ret && (dirty_offset + dirty_len < end))
583 ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
584 end - dirty_offset - dirty_len);
589 /* Given an unmapped page, zero it or read in page for COW, page is locked
593 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
595 struct buffer_head *bh = NULL;
599 dprintk("%s enter, %p\n", __func__, page);
600 BUG_ON(PageUptodate(page));
602 zero_user_segment(page, 0, PAGE_SIZE);
603 SetPageUptodate(page);
607 bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
613 isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
614 map_block(bh, isect, cow_read);
615 if (!bh_uptodate_or_lock(bh))
616 ret = bh_submit_read(bh);
619 SetPageUptodate(page);
623 free_buffer_head(bh);
625 /* Need to mark layout with bad read...should now
626 * just use nfs4 for reads and writes.
633 /* Find or create a zeroing page marked being writeback.
634 * Return ERR_PTR on error, NULL to indicate skip this page and page itself
635 * to indicate write out.
638 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
639 struct pnfs_block_extent *cow_read)
643 page = find_get_page(inode->i_mapping, index);
647 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
648 if (unlikely(!page)) {
649 dprintk("%s oom\n", __func__);
650 return ERR_PTR(-ENOMEM);
655 /* PageDirty: Other will write this out
656 * PageWriteback: Other is writing this out
657 * PageUptodate: It was read before
659 if (PageDirty(page) || PageWriteback(page)) {
663 page_cache_release(page);
672 if (!PageUptodate(page)) {
673 /* New page, readin or zero it */
674 init_page_for_write(page, cow_read);
676 set_page_writeback(page);
682 static enum pnfs_try_status
683 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
685 struct nfs_pgio_header *header = wdata->header;
686 int i, ret, npg_zero, pg_index, last = 0;
687 struct bio *bio = NULL;
688 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
689 sector_t isect, last_isect = 0, extent_length = 0;
690 struct parallel_io *par = NULL;
691 loff_t offset = wdata->args.offset;
692 size_t count = wdata->args.count;
693 unsigned int pg_offset, pg_len, saved_len;
694 struct page **pages = wdata->args.pages;
699 NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
701 dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
703 if (header->dreq != NULL &&
704 (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
705 !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
706 dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
709 /* At this point, wdata->pages is a (sequential) list of nfs_pages.
710 * We want to write each, and if there is an error set pnfs_error
711 * to have it redone using nfs.
713 par = alloc_parallel(wdata);
716 par->pnfs_callback = bl_end_par_io_write;
717 /* At this point, have to be more careful with error handling */
719 isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
720 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
721 if (!be || !is_writable(be, isect)) {
722 dprintk("%s no matching extents!\n", __func__);
726 /* First page inside INVALID extent */
727 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
728 if (likely(!bl_push_one_short_extent(be->be_inval)))
732 temp = offset >> PAGE_CACHE_SHIFT;
733 npg_zero = do_div(temp, npg_per_block);
734 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
735 (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
736 extent_length = be->be_length - (isect - be->be_f_offset);
739 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
740 for (;npg_zero > 0; npg_zero--) {
741 if (bl_is_sector_init(be->be_inval, isect)) {
742 dprintk("isect %llu already init\n",
743 (unsigned long long)isect);
746 /* page ref released in bl_end_io_write_zero */
747 index = isect >> PAGE_CACHE_SECTOR_SHIFT;
748 dprintk("%s zero %dth page: index %lu isect %llu\n",
749 __func__, npg_zero, index,
750 (unsigned long long)isect);
751 page = bl_find_get_zeroing_page(header->inode, index,
753 if (unlikely(IS_ERR(page))) {
754 header->pnfs_error = PTR_ERR(page);
756 } else if (page == NULL)
759 ret = bl_mark_sectors_init(be->be_inval, isect,
762 dprintk("%s bl_mark_sectors_init fail %d\n",
764 end_page_writeback(page);
765 page_cache_release(page);
766 header->pnfs_error = ret;
769 if (likely(!bl_push_one_short_extent(be->be_inval)))
772 end_page_writeback(page);
773 page_cache_release(page);
774 header->pnfs_error = -ENOMEM;
777 /* FIXME: This should be done in bi_end_io */
778 mark_extents_written(BLK_LSEG2EXT(header->lseg),
779 page->index << PAGE_CACHE_SHIFT,
782 bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
784 bl_end_io_write_zero, par);
786 header->pnfs_error = PTR_ERR(bio);
791 isect += PAGE_CACHE_SECTORS;
792 extent_length -= PAGE_CACHE_SECTORS;
797 bio = bl_submit_bio(WRITE, bio);
800 pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
801 for (i = pg_index; i < wdata->pages.npages; i++) {
802 if (!extent_length) {
803 /* We've used up the previous extent */
805 bl_put_extent(cow_read);
806 bio = bl_submit_bio(WRITE, bio);
807 /* Get the next one */
808 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
810 if (!be || !is_writable(be, isect)) {
811 header->pnfs_error = -EINVAL;
814 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
815 if (likely(!bl_push_one_short_extent(
819 header->pnfs_error = -ENOMEM;
823 extent_length = be->be_length -
824 (isect - be->be_f_offset);
827 dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
828 pg_offset = offset & ~PAGE_CACHE_MASK;
829 if (pg_offset + count > PAGE_CACHE_SIZE)
830 pg_len = PAGE_CACHE_SIZE - pg_offset;
835 if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
836 !bl_is_sector_init(be->be_inval, isect)) {
837 ret = bl_read_partial_page_sync(pages[i], cow_read,
838 pg_offset, pg_len, true);
840 dprintk("%s bl_read_partial_page_sync fail %d\n",
842 header->pnfs_error = ret;
846 ret = bl_mark_sectors_init(be->be_inval, isect,
849 dprintk("%s bl_mark_sectors_init fail %d\n",
851 header->pnfs_error = ret;
855 /* Expand to full page write */
857 pg_len = PAGE_CACHE_SIZE;
858 } else if ((pg_offset & (SECTOR_SIZE - 1)) ||
859 (pg_len & (SECTOR_SIZE - 1))){
860 /* ahh, nasty case. We have to do sync full sector
861 * read-modify-write cycles.
863 unsigned int saved_offset = pg_offset;
864 ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
866 pg_offset = round_down(pg_offset, SECTOR_SIZE);
867 pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
872 bio = do_add_page_to_bio(bio, wdata->pages.npages - i, WRITE,
874 bl_end_io_write, par,
877 header->pnfs_error = PTR_ERR(bio);
883 isect += PAGE_CACHE_SECTORS;
885 extent_length -= PAGE_CACHE_SECTORS;
888 /* Last page inside INVALID extent */
889 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
890 bio = bl_submit_bio(WRITE, bio);
891 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
892 npg_zero = npg_per_block - do_div(temp, npg_per_block);
893 if (npg_zero < npg_per_block) {
895 goto fill_invalid_ext;
900 wdata->res.count = wdata->args.count;
903 bl_put_extent(cow_read);
904 bl_submit_bio(WRITE, bio);
906 return PNFS_ATTEMPTED;
909 bl_put_extent(cow_read);
911 return PNFS_NOT_ATTEMPTED;
914 /* FIXME - range ignored */
916 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
919 struct pnfs_block_extent *be;
921 spin_lock(&bl->bl_ext_lock);
922 for (i = 0; i < EXTENT_LISTS; i++) {
923 while (!list_empty(&bl->bl_extents[i])) {
924 be = list_first_entry(&bl->bl_extents[i],
925 struct pnfs_block_extent,
927 list_del(&be->be_node);
931 spin_unlock(&bl->bl_ext_lock);
935 release_inval_marks(struct pnfs_inval_markings *marks)
937 struct pnfs_inval_tracking *pos, *temp;
938 struct pnfs_block_short_extent *se, *stemp;
940 list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
941 list_del(&pos->it_link);
945 list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
946 list_del(&se->bse_node);
952 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
954 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
956 dprintk("%s enter\n", __func__);
957 release_extents(bl, NULL);
958 release_inval_marks(&bl->bl_inval);
962 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
965 struct pnfs_block_layout *bl;
967 dprintk("%s enter\n", __func__);
968 bl = kzalloc(sizeof(*bl), gfp_flags);
971 spin_lock_init(&bl->bl_ext_lock);
972 INIT_LIST_HEAD(&bl->bl_extents[0]);
973 INIT_LIST_HEAD(&bl->bl_extents[1]);
974 INIT_LIST_HEAD(&bl->bl_commit);
975 INIT_LIST_HEAD(&bl->bl_committing);
977 bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
978 BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
979 return &bl->bl_layout;
982 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
984 dprintk("%s enter\n", __func__);
988 /* We pretty much ignore lseg, and store all data layout wide, so we
989 * can correctly merge.
991 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
992 struct nfs4_layoutget_res *lgr,
995 struct pnfs_layout_segment *lseg;
998 dprintk("%s enter\n", __func__);
999 lseg = kzalloc(sizeof(*lseg), gfp_flags);
1001 return ERR_PTR(-ENOMEM);
1002 status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
1004 /* We don't want to call the full-blown bl_free_lseg,
1005 * since on error extents were not touched.
1008 return ERR_PTR(status);
1014 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1015 const struct nfs4_layoutcommit_args *arg)
1017 dprintk("%s enter\n", __func__);
1018 encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1022 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1024 struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1026 dprintk("%s enter\n", __func__);
1027 clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1030 static void free_blk_mountid(struct block_mount_id *mid)
1033 struct pnfs_block_dev *dev, *tmp;
1035 /* No need to take bm_lock as we are last user freeing bm_devlist */
1036 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1037 list_del(&dev->bm_node);
1038 bl_free_block_dev(dev);
1044 /* This is mostly copied from the filelayout_get_device_info function.
1045 * It seems much of this should be at the generic pnfs level.
1047 static struct pnfs_block_dev *
1048 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1049 struct nfs4_deviceid *d_id)
1051 struct pnfs_device *dev;
1052 struct pnfs_block_dev *rv;
1055 struct page **pages = NULL;
1059 * Use the session max response size as the basis for setting
1060 * GETDEVICEINFO's maxcount
1062 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1063 max_pages = nfs_page_array_len(0, max_resp_sz);
1064 dprintk("%s max_resp_sz %u max_pages %d\n",
1065 __func__, max_resp_sz, max_pages);
1067 dev = kmalloc(sizeof(*dev), GFP_NOFS);
1069 dprintk("%s kmalloc failed\n", __func__);
1070 return ERR_PTR(-ENOMEM);
1073 pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
1074 if (pages == NULL) {
1076 return ERR_PTR(-ENOMEM);
1078 for (i = 0; i < max_pages; i++) {
1079 pages[i] = alloc_page(GFP_NOFS);
1081 rv = ERR_PTR(-ENOMEM);
1086 memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1087 dev->layout_type = LAYOUT_BLOCK_VOLUME;
1090 dev->pglen = PAGE_SIZE * max_pages;
1092 dev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
1094 dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1095 rc = nfs4_proc_getdeviceinfo(server, dev, NULL);
1096 dprintk("%s getdevice info returns %d\n", __func__, rc);
1102 rv = nfs4_blk_decode_device(server, dev);
1104 for (i = 0; i < max_pages; i++)
1105 __free_page(pages[i]);
1112 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1114 struct block_mount_id *b_mt_id = NULL;
1115 struct pnfs_devicelist *dlist = NULL;
1116 struct pnfs_block_dev *bdev;
1117 LIST_HEAD(block_disklist);
1120 dprintk("%s enter\n", __func__);
1122 if (server->pnfs_blksize == 0) {
1123 dprintk("%s Server did not return blksize\n", __func__);
1126 b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1131 /* Initialize nfs4 block layout mount id */
1132 spin_lock_init(&b_mt_id->bm_lock);
1133 INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1135 dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1141 while (!dlist->eof) {
1142 status = nfs4_proc_getdevicelist(server, fh, dlist);
1145 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1146 __func__, dlist->num_devs, dlist->eof);
1147 for (i = 0; i < dlist->num_devs; i++) {
1148 bdev = nfs4_blk_get_deviceinfo(server, fh,
1151 status = PTR_ERR(bdev);
1154 spin_lock(&b_mt_id->bm_lock);
1155 list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1156 spin_unlock(&b_mt_id->bm_lock);
1159 dprintk("%s SUCCESS\n", __func__);
1160 server->pnfs_ld_data = b_mt_id;
1167 free_blk_mountid(b_mt_id);
1172 bl_clear_layoutdriver(struct nfs_server *server)
1174 struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1176 dprintk("%s enter\n", __func__);
1177 free_blk_mountid(b_mt_id);
1178 dprintk("%s RETURNS\n", __func__);
1183 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1185 return IS_ALIGNED(req->wb_offset, alignment) &&
1186 IS_ALIGNED(req->wb_bytes, alignment);
1190 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1192 if (pgio->pg_dreq != NULL &&
1193 !is_aligned_req(req, SECTOR_SIZE))
1194 nfs_pageio_reset_read_mds(pgio);
1196 pnfs_generic_pg_init_read(pgio, req);
1200 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1201 struct nfs_page *req)
1203 if (pgio->pg_dreq != NULL &&
1204 !is_aligned_req(req, SECTOR_SIZE))
1207 return pnfs_generic_pg_test(pgio, prev, req);
1211 * Return the number of contiguous bytes for a given inode
1212 * starting at page frame idx.
1214 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
1216 struct address_space *mapping = inode->i_mapping;
1219 /* Optimize common case that writes from 0 to end of file */
1220 end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
1221 if (end != NFS_I(inode)->npages) {
1223 end = radix_tree_next_hole(&mapping->page_tree, idx + 1, ULONG_MAX);
1228 return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
1230 return (end - idx) << PAGE_CACHE_SHIFT;
1234 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1236 if (pgio->pg_dreq != NULL &&
1237 !is_aligned_req(req, PAGE_CACHE_SIZE)) {
1238 nfs_pageio_reset_write_mds(pgio);
1241 if (pgio->pg_dreq == NULL)
1242 wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
1245 wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1247 pnfs_generic_pg_init_write(pgio, req, wb_size);
1252 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1253 struct nfs_page *req)
1255 if (pgio->pg_dreq != NULL &&
1256 !is_aligned_req(req, PAGE_CACHE_SIZE))
1259 return pnfs_generic_pg_test(pgio, prev, req);
1262 static const struct nfs_pageio_ops bl_pg_read_ops = {
1263 .pg_init = bl_pg_init_read,
1264 .pg_test = bl_pg_test_read,
1265 .pg_doio = pnfs_generic_pg_readpages,
1268 static const struct nfs_pageio_ops bl_pg_write_ops = {
1269 .pg_init = bl_pg_init_write,
1270 .pg_test = bl_pg_test_write,
1271 .pg_doio = pnfs_generic_pg_writepages,
1274 static struct pnfs_layoutdriver_type blocklayout_type = {
1275 .id = LAYOUT_BLOCK_VOLUME,
1276 .name = "LAYOUT_BLOCK_VOLUME",
1277 .owner = THIS_MODULE,
1278 .read_pagelist = bl_read_pagelist,
1279 .write_pagelist = bl_write_pagelist,
1280 .alloc_layout_hdr = bl_alloc_layout_hdr,
1281 .free_layout_hdr = bl_free_layout_hdr,
1282 .alloc_lseg = bl_alloc_lseg,
1283 .free_lseg = bl_free_lseg,
1284 .encode_layoutcommit = bl_encode_layoutcommit,
1285 .cleanup_layoutcommit = bl_cleanup_layoutcommit,
1286 .set_layoutdriver = bl_set_layoutdriver,
1287 .clear_layoutdriver = bl_clear_layoutdriver,
1288 .pg_read_ops = &bl_pg_read_ops,
1289 .pg_write_ops = &bl_pg_write_ops,
1292 static const struct rpc_pipe_ops bl_upcall_ops = {
1293 .upcall = rpc_pipe_generic_upcall,
1294 .downcall = bl_pipe_downcall,
1295 .destroy_msg = bl_pipe_destroy_msg,
1298 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1299 struct rpc_pipe *pipe)
1301 struct dentry *dir, *dentry;
1303 dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1305 return ERR_PTR(-ENOENT);
1306 dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1311 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1312 struct rpc_pipe *pipe)
1315 rpc_unlink(pipe->dentry);
1318 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1321 struct super_block *sb = ptr;
1322 struct net *net = sb->s_fs_info;
1323 struct nfs_net *nn = net_generic(net, nfs_net_id);
1324 struct dentry *dentry;
1327 if (!try_module_get(THIS_MODULE))
1330 if (nn->bl_device_pipe == NULL) {
1331 module_put(THIS_MODULE);
1336 case RPC_PIPEFS_MOUNT:
1337 dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1338 if (IS_ERR(dentry)) {
1339 ret = PTR_ERR(dentry);
1342 nn->bl_device_pipe->dentry = dentry;
1344 case RPC_PIPEFS_UMOUNT:
1345 if (nn->bl_device_pipe->dentry)
1346 nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1352 module_put(THIS_MODULE);
1356 static struct notifier_block nfs4blocklayout_block = {
1357 .notifier_call = rpc_pipefs_event,
1360 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1361 struct rpc_pipe *pipe)
1363 struct super_block *pipefs_sb;
1364 struct dentry *dentry;
1366 pipefs_sb = rpc_get_sb_net(net);
1369 dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1370 rpc_put_sb_net(net);
1374 static void nfs4blocklayout_unregister_net(struct net *net,
1375 struct rpc_pipe *pipe)
1377 struct super_block *pipefs_sb;
1379 pipefs_sb = rpc_get_sb_net(net);
1381 nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1382 rpc_put_sb_net(net);
1386 static int nfs4blocklayout_net_init(struct net *net)
1388 struct nfs_net *nn = net_generic(net, nfs_net_id);
1389 struct dentry *dentry;
1391 init_waitqueue_head(&nn->bl_wq);
1392 nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1393 if (IS_ERR(nn->bl_device_pipe))
1394 return PTR_ERR(nn->bl_device_pipe);
1395 dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1396 if (IS_ERR(dentry)) {
1397 rpc_destroy_pipe_data(nn->bl_device_pipe);
1398 return PTR_ERR(dentry);
1400 nn->bl_device_pipe->dentry = dentry;
1404 static void nfs4blocklayout_net_exit(struct net *net)
1406 struct nfs_net *nn = net_generic(net, nfs_net_id);
1408 nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1409 rpc_destroy_pipe_data(nn->bl_device_pipe);
1410 nn->bl_device_pipe = NULL;
1413 static struct pernet_operations nfs4blocklayout_net_ops = {
1414 .init = nfs4blocklayout_net_init,
1415 .exit = nfs4blocklayout_net_exit,
1418 static int __init nfs4blocklayout_init(void)
1422 dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1424 ret = pnfs_register_layoutdriver(&blocklayout_type);
1428 ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1431 ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1438 rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1440 pnfs_unregister_layoutdriver(&blocklayout_type);
1444 static void __exit nfs4blocklayout_exit(void)
1446 dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1449 rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1450 unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1451 pnfs_unregister_layoutdriver(&blocklayout_type);
1454 MODULE_ALIAS("nfs-layouttype4-3");
1456 module_init(nfs4blocklayout_init);
1457 module_exit(nfs4blocklayout_exit);