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", bio->bi_iter.bi_size,
138 (unsigned long long)bio->bi_iter.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_iter.bi_sector = isect - be->be_f_offset +
161 bio->bi_bdev = be->be_mdev;
162 bio->bi_end_io = end_io;
163 bio->bi_private = par;
168 static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw,
169 sector_t isect, struct page *page,
170 struct pnfs_block_extent *be,
171 void (*end_io)(struct bio *, int err),
172 struct parallel_io *par,
173 unsigned int offset, int len)
175 isect = isect + (offset >> SECTOR_SHIFT);
176 dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
177 npg, rw, (unsigned long long)isect, offset, len);
180 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
182 return ERR_PTR(-ENOMEM);
184 if (bio_add_page(bio, page, len, offset) < len) {
185 bio = bl_submit_bio(rw, bio);
191 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
192 sector_t isect, struct page *page,
193 struct pnfs_block_extent *be,
194 void (*end_io)(struct bio *, int err),
195 struct parallel_io *par)
197 return do_add_page_to_bio(bio, npg, rw, isect, page, be,
198 end_io, par, 0, PAGE_CACHE_SIZE);
201 /* This is basically copied from mpage_end_io_read */
202 static void bl_end_io_read(struct bio *bio, int err)
204 struct parallel_io *par = bio->bi_private;
205 struct bio_vec *bvec;
209 bio_for_each_segment_all(bvec, bio, i)
210 SetPageUptodate(bvec->bv_page);
213 struct nfs_pgio_header *header = par->data;
215 if (!header->pnfs_error)
216 header->pnfs_error = -EIO;
217 pnfs_set_lo_fail(header->lseg);
223 static void bl_read_cleanup(struct work_struct *work)
225 struct rpc_task *task;
226 struct nfs_pgio_header *hdr;
227 dprintk("%s enter\n", __func__);
228 task = container_of(work, struct rpc_task, u.tk_work);
229 hdr = container_of(task, struct nfs_pgio_header, task);
230 pnfs_ld_read_done(hdr);
234 bl_end_par_io_read(void *data, int unused)
236 struct nfs_pgio_header *hdr = data;
238 hdr->task.tk_status = hdr->pnfs_error;
239 INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
240 schedule_work(&hdr->task.u.tk_work);
243 static enum pnfs_try_status
244 bl_read_pagelist(struct nfs_pgio_header *hdr)
246 struct nfs_pgio_header *header = hdr;
248 struct bio *bio = NULL;
249 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
250 sector_t isect, extent_length = 0;
251 struct parallel_io *par;
252 loff_t f_offset = hdr->args.offset;
253 size_t bytes_left = hdr->args.count;
254 unsigned int pg_offset, pg_len;
255 struct page **pages = hdr->args.pages;
256 int pg_index = hdr->args.pgbase >> PAGE_CACHE_SHIFT;
257 const bool is_dio = (header->dreq != NULL);
259 dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
260 hdr->page_array.npages, f_offset,
261 (unsigned int)hdr->args.count);
263 par = alloc_parallel(hdr);
266 par->pnfs_callback = bl_end_par_io_read;
267 /* At this point, we can no longer jump to use_mds */
269 isect = (sector_t) (f_offset >> SECTOR_SHIFT);
270 /* Code assumes extents are page-aligned */
271 for (i = pg_index; i < hdr->page_array.npages; i++) {
272 if (!extent_length) {
273 /* We've used up the previous extent */
275 bl_put_extent(cow_read);
276 bio = bl_submit_bio(READ, bio);
277 /* Get the next one */
278 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
281 header->pnfs_error = -EIO;
284 extent_length = be->be_length -
285 (isect - be->be_f_offset);
287 sector_t cow_length = cow_read->be_length -
288 (isect - cow_read->be_f_offset);
289 extent_length = min(extent_length, cow_length);
294 pg_offset = f_offset & ~PAGE_CACHE_MASK;
295 if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
296 pg_len = PAGE_CACHE_SIZE - pg_offset;
301 bytes_left -= pg_len;
302 isect += (pg_offset >> SECTOR_SHIFT);
305 pg_len = PAGE_CACHE_SIZE;
308 hole = is_hole(be, isect);
309 if (hole && !cow_read) {
310 bio = bl_submit_bio(READ, bio);
311 /* Fill hole w/ zeroes w/o accessing device */
312 dprintk("%s Zeroing page for hole\n", __func__);
313 zero_user_segment(pages[i], pg_offset, pg_len);
314 print_page(pages[i]);
315 SetPageUptodate(pages[i]);
317 struct pnfs_block_extent *be_read;
319 be_read = (hole && cow_read) ? cow_read : be;
320 bio = do_add_page_to_bio(bio,
321 hdr->page_array.npages - i,
323 isect, pages[i], be_read,
327 header->pnfs_error = PTR_ERR(bio);
332 isect += (pg_len >> SECTOR_SHIFT);
333 extent_length -= PAGE_CACHE_SECTORS;
335 if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
337 hdr->res.count = header->inode->i_size - hdr->args.offset;
339 hdr->res.count = (isect << SECTOR_SHIFT) - hdr->args.offset;
343 bl_put_extent(cow_read);
344 bl_submit_bio(READ, bio);
346 return PNFS_ATTEMPTED;
349 dprintk("Giving up and using normal NFS\n");
350 return PNFS_NOT_ATTEMPTED;
353 static void mark_extents_written(struct pnfs_block_layout *bl,
354 __u64 offset, __u32 count)
357 struct pnfs_block_extent *be;
358 struct pnfs_block_short_extent *se;
360 dprintk("%s(%llu, %u)\n", __func__, offset, count);
363 isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
364 end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
365 end >>= SECTOR_SHIFT;
366 while (isect < end) {
368 be = bl_find_get_extent(bl, isect, NULL);
369 BUG_ON(!be); /* FIXME */
370 len = min(end, be->be_f_offset + be->be_length) - isect;
371 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
372 se = bl_pop_one_short_extent(be->be_inval);
374 bl_mark_for_commit(be, isect, len, se);
381 static void bl_end_io_write_zero(struct bio *bio, int err)
383 struct parallel_io *par = bio->bi_private;
384 struct bio_vec *bvec;
387 bio_for_each_segment_all(bvec, bio, i) {
388 /* This is the zeroing page we added */
389 end_page_writeback(bvec->bv_page);
390 page_cache_release(bvec->bv_page);
394 struct nfs_pgio_header *header = par->data;
396 if (!header->pnfs_error)
397 header->pnfs_error = -EIO;
398 pnfs_set_lo_fail(header->lseg);
404 static void bl_end_io_write(struct bio *bio, int err)
406 struct parallel_io *par = bio->bi_private;
407 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
408 struct nfs_pgio_header *header = par->data;
411 if (!header->pnfs_error)
412 header->pnfs_error = -EIO;
413 pnfs_set_lo_fail(header->lseg);
419 /* Function scheduled for call during bl_end_par_io_write,
420 * it marks sectors as written and extends the commitlist.
422 static void bl_write_cleanup(struct work_struct *work)
424 struct rpc_task *task;
425 struct nfs_pgio_header *hdr;
426 dprintk("%s enter\n", __func__);
427 task = container_of(work, struct rpc_task, u.tk_work);
428 hdr = container_of(task, struct nfs_pgio_header, task);
429 if (likely(!hdr->pnfs_error)) {
430 /* Marks for LAYOUTCOMMIT */
431 mark_extents_written(BLK_LSEG2EXT(hdr->lseg),
432 hdr->args.offset, hdr->args.count);
434 pnfs_ld_write_done(hdr);
437 /* Called when last of bios associated with a bl_write_pagelist call finishes */
438 static void bl_end_par_io_write(void *data, int num_se)
440 struct nfs_pgio_header *hdr = data;
442 if (unlikely(hdr->pnfs_error)) {
443 bl_free_short_extents(&BLK_LSEG2EXT(hdr->lseg)->bl_inval,
447 hdr->task.tk_status = hdr->pnfs_error;
448 hdr->verf.committed = NFS_FILE_SYNC;
449 INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
450 schedule_work(&hdr->task.u.tk_work);
453 /* FIXME STUB - mark intersection of layout and page as bad, so is not
456 static void mark_bad_read(void)
462 * map_block: map a requested I/0 block (isect) into an offset in the LVM
466 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
468 dprintk("%s enter be=%p\n", __func__, be);
470 set_buffer_mapped(bh);
471 bh->b_bdev = be->be_mdev;
472 bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
473 (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
475 dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
476 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
482 bl_read_single_end_io(struct bio *bio, int error)
484 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
485 struct page *page = bvec->bv_page;
487 /* Only one page in bvec */
492 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
493 unsigned int offset, unsigned int len)
496 struct page *shadow_page;
498 char *kaddr, *kshadow_addr;
501 dprintk("%s: offset %u len %u\n", __func__, offset, len);
503 shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
504 if (shadow_page == NULL)
507 bio = bio_alloc(GFP_NOIO, 1);
511 isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
512 (offset / SECTOR_SIZE);
514 bio->bi_iter.bi_sector = isect - be->be_f_offset + be->be_v_offset;
515 bio->bi_bdev = be->be_mdev;
516 bio->bi_end_io = bl_read_single_end_io;
518 lock_page(shadow_page);
519 if (bio_add_page(bio, shadow_page,
520 SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
521 unlock_page(shadow_page);
526 submit_bio(READ, bio);
527 wait_on_page_locked(shadow_page);
528 if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
531 kaddr = kmap_atomic(page);
532 kshadow_addr = kmap_atomic(shadow_page);
533 memcpy(kaddr + offset, kshadow_addr + offset, len);
534 kunmap_atomic(kshadow_addr);
535 kunmap_atomic(kaddr);
537 __free_page(shadow_page);
544 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
545 unsigned int dirty_offset, unsigned int dirty_len,
549 unsigned int start, end;
553 end = PAGE_CACHE_SIZE;
555 start = round_down(dirty_offset, SECTOR_SIZE);
556 end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
559 dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
561 zero_user_segments(page, start, dirty_offset,
562 dirty_offset + dirty_len, end);
563 if (start == 0 && end == PAGE_CACHE_SIZE &&
564 trylock_page(page)) {
565 SetPageUptodate(page);
571 if (start != dirty_offset)
572 ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
574 if (!ret && (dirty_offset + dirty_len < end))
575 ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
576 end - dirty_offset - dirty_len);
581 /* Given an unmapped page, zero it or read in page for COW, page is locked
585 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
587 struct buffer_head *bh = NULL;
591 dprintk("%s enter, %p\n", __func__, page);
592 BUG_ON(PageUptodate(page));
594 zero_user_segment(page, 0, PAGE_SIZE);
595 SetPageUptodate(page);
599 bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
605 isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
606 map_block(bh, isect, cow_read);
607 if (!bh_uptodate_or_lock(bh))
608 ret = bh_submit_read(bh);
611 SetPageUptodate(page);
615 free_buffer_head(bh);
617 /* Need to mark layout with bad read...should now
618 * just use nfs4 for reads and writes.
625 /* Find or create a zeroing page marked being writeback.
626 * Return ERR_PTR on error, NULL to indicate skip this page and page itself
627 * to indicate write out.
630 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
631 struct pnfs_block_extent *cow_read)
635 page = find_get_page(inode->i_mapping, index);
639 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
640 if (unlikely(!page)) {
641 dprintk("%s oom\n", __func__);
642 return ERR_PTR(-ENOMEM);
647 /* PageDirty: Other will write this out
648 * PageWriteback: Other is writing this out
649 * PageUptodate: It was read before
651 if (PageDirty(page) || PageWriteback(page)) {
655 page_cache_release(page);
664 if (!PageUptodate(page)) {
665 /* New page, readin or zero it */
666 init_page_for_write(page, cow_read);
668 set_page_writeback(page);
674 static enum pnfs_try_status
675 bl_write_pagelist(struct nfs_pgio_header *header, int sync)
677 int i, ret, npg_zero, pg_index, last = 0;
678 struct bio *bio = NULL;
679 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
680 sector_t isect, last_isect = 0, extent_length = 0;
681 struct parallel_io *par = NULL;
682 loff_t offset = header->args.offset;
683 size_t count = header->args.count;
684 unsigned int pg_offset, pg_len, saved_len;
685 struct page **pages = header->args.pages;
690 NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
692 dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
694 if (header->dreq != NULL &&
695 (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
696 !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
697 dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
700 /* At this point, header->page_aray is a (sequential) list of nfs_pages.
701 * We want to write each, and if there is an error set pnfs_error
702 * to have it redone using nfs.
704 par = alloc_parallel(header);
707 par->pnfs_callback = bl_end_par_io_write;
708 /* At this point, have to be more careful with error handling */
710 isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
711 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
712 if (!be || !is_writable(be, isect)) {
713 dprintk("%s no matching extents!\n", __func__);
717 /* First page inside INVALID extent */
718 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
719 if (likely(!bl_push_one_short_extent(be->be_inval)))
723 temp = offset >> PAGE_CACHE_SHIFT;
724 npg_zero = do_div(temp, npg_per_block);
725 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
726 (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
727 extent_length = be->be_length - (isect - be->be_f_offset);
730 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
731 for (;npg_zero > 0; npg_zero--) {
732 if (bl_is_sector_init(be->be_inval, isect)) {
733 dprintk("isect %llu already init\n",
734 (unsigned long long)isect);
737 /* page ref released in bl_end_io_write_zero */
738 index = isect >> PAGE_CACHE_SECTOR_SHIFT;
739 dprintk("%s zero %dth page: index %lu isect %llu\n",
740 __func__, npg_zero, index,
741 (unsigned long long)isect);
742 page = bl_find_get_zeroing_page(header->inode, index,
744 if (unlikely(IS_ERR(page))) {
745 header->pnfs_error = PTR_ERR(page);
747 } else if (page == NULL)
750 ret = bl_mark_sectors_init(be->be_inval, isect,
753 dprintk("%s bl_mark_sectors_init fail %d\n",
755 end_page_writeback(page);
756 page_cache_release(page);
757 header->pnfs_error = ret;
760 if (likely(!bl_push_one_short_extent(be->be_inval)))
763 end_page_writeback(page);
764 page_cache_release(page);
765 header->pnfs_error = -ENOMEM;
768 /* FIXME: This should be done in bi_end_io */
769 mark_extents_written(BLK_LSEG2EXT(header->lseg),
770 page->index << PAGE_CACHE_SHIFT,
773 bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
775 bl_end_io_write_zero, par);
777 header->pnfs_error = PTR_ERR(bio);
782 isect += PAGE_CACHE_SECTORS;
783 extent_length -= PAGE_CACHE_SECTORS;
788 bio = bl_submit_bio(WRITE, bio);
791 pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT;
792 for (i = pg_index; i < header->page_array.npages; i++) {
793 if (!extent_length) {
794 /* We've used up the previous extent */
796 bl_put_extent(cow_read);
797 bio = bl_submit_bio(WRITE, bio);
798 /* Get the next one */
799 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
801 if (!be || !is_writable(be, isect)) {
802 header->pnfs_error = -EINVAL;
805 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
806 if (likely(!bl_push_one_short_extent(
810 header->pnfs_error = -ENOMEM;
814 extent_length = be->be_length -
815 (isect - be->be_f_offset);
818 dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
819 pg_offset = offset & ~PAGE_CACHE_MASK;
820 if (pg_offset + count > PAGE_CACHE_SIZE)
821 pg_len = PAGE_CACHE_SIZE - pg_offset;
826 if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
827 !bl_is_sector_init(be->be_inval, isect)) {
828 ret = bl_read_partial_page_sync(pages[i], cow_read,
829 pg_offset, pg_len, true);
831 dprintk("%s bl_read_partial_page_sync fail %d\n",
833 header->pnfs_error = ret;
837 ret = bl_mark_sectors_init(be->be_inval, isect,
840 dprintk("%s bl_mark_sectors_init fail %d\n",
842 header->pnfs_error = ret;
846 /* Expand to full page write */
848 pg_len = PAGE_CACHE_SIZE;
849 } else if ((pg_offset & (SECTOR_SIZE - 1)) ||
850 (pg_len & (SECTOR_SIZE - 1))){
851 /* ahh, nasty case. We have to do sync full sector
852 * read-modify-write cycles.
854 unsigned int saved_offset = pg_offset;
855 ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
857 pg_offset = round_down(pg_offset, SECTOR_SIZE);
858 pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
863 bio = do_add_page_to_bio(bio, header->page_array.npages - i,
866 bl_end_io_write, par,
869 header->pnfs_error = PTR_ERR(bio);
875 isect += PAGE_CACHE_SECTORS;
877 extent_length -= PAGE_CACHE_SECTORS;
880 /* Last page inside INVALID extent */
881 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
882 bio = bl_submit_bio(WRITE, bio);
883 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
884 npg_zero = npg_per_block - do_div(temp, npg_per_block);
885 if (npg_zero < npg_per_block) {
887 goto fill_invalid_ext;
892 header->res.count = header->args.count;
895 bl_put_extent(cow_read);
896 bl_submit_bio(WRITE, bio);
898 return PNFS_ATTEMPTED;
901 bl_put_extent(cow_read);
903 return PNFS_NOT_ATTEMPTED;
906 /* FIXME - range ignored */
908 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
911 struct pnfs_block_extent *be;
913 spin_lock(&bl->bl_ext_lock);
914 for (i = 0; i < EXTENT_LISTS; i++) {
915 while (!list_empty(&bl->bl_extents[i])) {
916 be = list_first_entry(&bl->bl_extents[i],
917 struct pnfs_block_extent,
919 list_del(&be->be_node);
923 spin_unlock(&bl->bl_ext_lock);
927 release_inval_marks(struct pnfs_inval_markings *marks)
929 struct pnfs_inval_tracking *pos, *temp;
930 struct pnfs_block_short_extent *se, *stemp;
932 list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
933 list_del(&pos->it_link);
937 list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
938 list_del(&se->bse_node);
944 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
946 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
948 dprintk("%s enter\n", __func__);
949 release_extents(bl, NULL);
950 release_inval_marks(&bl->bl_inval);
954 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
957 struct pnfs_block_layout *bl;
959 dprintk("%s enter\n", __func__);
960 bl = kzalloc(sizeof(*bl), gfp_flags);
963 spin_lock_init(&bl->bl_ext_lock);
964 INIT_LIST_HEAD(&bl->bl_extents[0]);
965 INIT_LIST_HEAD(&bl->bl_extents[1]);
966 INIT_LIST_HEAD(&bl->bl_commit);
967 INIT_LIST_HEAD(&bl->bl_committing);
969 bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
970 BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
971 return &bl->bl_layout;
974 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
976 dprintk("%s enter\n", __func__);
980 /* We pretty much ignore lseg, and store all data layout wide, so we
981 * can correctly merge.
983 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
984 struct nfs4_layoutget_res *lgr,
987 struct pnfs_layout_segment *lseg;
990 dprintk("%s enter\n", __func__);
991 lseg = kzalloc(sizeof(*lseg), gfp_flags);
993 return ERR_PTR(-ENOMEM);
994 status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
996 /* We don't want to call the full-blown bl_free_lseg,
997 * since on error extents were not touched.
1000 return ERR_PTR(status);
1006 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1007 const struct nfs4_layoutcommit_args *arg)
1009 dprintk("%s enter\n", __func__);
1010 encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1014 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1016 struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1018 dprintk("%s enter\n", __func__);
1019 clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1022 static void free_blk_mountid(struct block_mount_id *mid)
1025 struct pnfs_block_dev *dev, *tmp;
1027 /* No need to take bm_lock as we are last user freeing bm_devlist */
1028 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1029 list_del(&dev->bm_node);
1030 bl_free_block_dev(dev);
1036 /* This is mostly copied from the filelayout_get_device_info function.
1037 * It seems much of this should be at the generic pnfs level.
1039 static struct pnfs_block_dev *
1040 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1041 struct nfs4_deviceid *d_id)
1043 struct pnfs_device *dev;
1044 struct pnfs_block_dev *rv;
1047 struct page **pages = NULL;
1051 * Use the session max response size as the basis for setting
1052 * GETDEVICEINFO's maxcount
1054 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1055 max_pages = nfs_page_array_len(0, max_resp_sz);
1056 dprintk("%s max_resp_sz %u max_pages %d\n",
1057 __func__, max_resp_sz, max_pages);
1059 dev = kmalloc(sizeof(*dev), GFP_NOFS);
1061 dprintk("%s kmalloc failed\n", __func__);
1062 return ERR_PTR(-ENOMEM);
1065 pages = kcalloc(max_pages, sizeof(struct page *), GFP_NOFS);
1066 if (pages == NULL) {
1068 return ERR_PTR(-ENOMEM);
1070 for (i = 0; i < max_pages; i++) {
1071 pages[i] = alloc_page(GFP_NOFS);
1073 rv = ERR_PTR(-ENOMEM);
1078 memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1079 dev->layout_type = LAYOUT_BLOCK_VOLUME;
1082 dev->pglen = PAGE_SIZE * max_pages;
1084 dev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
1086 dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1087 rc = nfs4_proc_getdeviceinfo(server, dev, NULL);
1088 dprintk("%s getdevice info returns %d\n", __func__, rc);
1094 rv = nfs4_blk_decode_device(server, dev);
1096 for (i = 0; i < max_pages; i++)
1097 __free_page(pages[i]);
1104 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1106 struct block_mount_id *b_mt_id = NULL;
1107 struct pnfs_devicelist *dlist = NULL;
1108 struct pnfs_block_dev *bdev;
1109 LIST_HEAD(block_disklist);
1112 dprintk("%s enter\n", __func__);
1114 if (server->pnfs_blksize == 0) {
1115 dprintk("%s Server did not return blksize\n", __func__);
1118 b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1123 /* Initialize nfs4 block layout mount id */
1124 spin_lock_init(&b_mt_id->bm_lock);
1125 INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1127 dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1133 while (!dlist->eof) {
1134 status = nfs4_proc_getdevicelist(server, fh, dlist);
1137 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1138 __func__, dlist->num_devs, dlist->eof);
1139 for (i = 0; i < dlist->num_devs; i++) {
1140 bdev = nfs4_blk_get_deviceinfo(server, fh,
1143 status = PTR_ERR(bdev);
1146 spin_lock(&b_mt_id->bm_lock);
1147 list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1148 spin_unlock(&b_mt_id->bm_lock);
1151 dprintk("%s SUCCESS\n", __func__);
1152 server->pnfs_ld_data = b_mt_id;
1159 free_blk_mountid(b_mt_id);
1164 bl_clear_layoutdriver(struct nfs_server *server)
1166 struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1168 dprintk("%s enter\n", __func__);
1169 free_blk_mountid(b_mt_id);
1170 dprintk("%s RETURNS\n", __func__);
1175 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1177 return IS_ALIGNED(req->wb_offset, alignment) &&
1178 IS_ALIGNED(req->wb_bytes, alignment);
1182 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1184 if (pgio->pg_dreq != NULL &&
1185 !is_aligned_req(req, SECTOR_SIZE))
1186 nfs_pageio_reset_read_mds(pgio);
1188 pnfs_generic_pg_init_read(pgio, req);
1192 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1193 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1196 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1197 struct nfs_page *req)
1199 if (pgio->pg_dreq != NULL &&
1200 !is_aligned_req(req, SECTOR_SIZE))
1203 return pnfs_generic_pg_test(pgio, prev, req);
1207 * Return the number of contiguous bytes for a given inode
1208 * starting at page frame idx.
1210 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
1212 struct address_space *mapping = inode->i_mapping;
1215 /* Optimize common case that writes from 0 to end of file */
1216 end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE);
1217 if (end != NFS_I(inode)->npages) {
1219 end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
1224 return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT);
1226 return (end - idx) << PAGE_CACHE_SHIFT;
1230 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1232 if (pgio->pg_dreq != NULL &&
1233 !is_aligned_req(req, PAGE_CACHE_SIZE)) {
1234 nfs_pageio_reset_write_mds(pgio);
1237 if (pgio->pg_dreq == NULL)
1238 wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
1241 wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1243 pnfs_generic_pg_init_write(pgio, req, wb_size);
1248 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1249 * of bytes (maximum @req->wb_bytes) that can be coalesced.
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);