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_data *rdata = par->data;
214 struct nfs_pgio_header *header = rdata->header;
216 if (!header->pnfs_error)
217 header->pnfs_error = -EIO;
218 pnfs_set_lo_fail(header->lseg);
224 static void bl_read_cleanup(struct work_struct *work)
226 struct rpc_task *task;
227 struct nfs_pgio_data *rdata;
228 dprintk("%s enter\n", __func__);
229 task = container_of(work, struct rpc_task, u.tk_work);
230 rdata = container_of(task, struct nfs_pgio_data, task);
231 pnfs_ld_read_done(rdata);
235 bl_end_par_io_read(void *data, int unused)
237 struct nfs_pgio_data *rdata = data;
239 rdata->task.tk_status = rdata->header->pnfs_error;
240 INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
241 schedule_work(&rdata->task.u.tk_work);
244 static enum pnfs_try_status
245 bl_read_pagelist(struct nfs_pgio_data *rdata)
247 struct nfs_pgio_header *header = rdata->header;
249 struct bio *bio = NULL;
250 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
251 sector_t isect, extent_length = 0;
252 struct parallel_io *par;
253 loff_t f_offset = rdata->args.offset;
254 size_t bytes_left = rdata->args.count;
255 unsigned int pg_offset, pg_len;
256 struct page **pages = rdata->args.pages;
257 int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
258 const bool is_dio = (header->dreq != NULL);
260 dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
261 rdata->page_array.npages, f_offset,
262 (unsigned int)rdata->args.count);
264 par = alloc_parallel(rdata);
267 par->pnfs_callback = bl_end_par_io_read;
268 /* At this point, we can no longer jump to use_mds */
270 isect = (sector_t) (f_offset >> SECTOR_SHIFT);
271 /* Code assumes extents are page-aligned */
272 for (i = pg_index; i < rdata->page_array.npages; i++) {
273 if (!extent_length) {
274 /* We've used up the previous extent */
276 bl_put_extent(cow_read);
277 bio = bl_submit_bio(READ, bio);
278 /* Get the next one */
279 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
282 header->pnfs_error = -EIO;
285 extent_length = be->be_length -
286 (isect - be->be_f_offset);
288 sector_t cow_length = cow_read->be_length -
289 (isect - cow_read->be_f_offset);
290 extent_length = min(extent_length, cow_length);
295 pg_offset = f_offset & ~PAGE_CACHE_MASK;
296 if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
297 pg_len = PAGE_CACHE_SIZE - pg_offset;
302 bytes_left -= pg_len;
303 isect += (pg_offset >> SECTOR_SHIFT);
306 pg_len = PAGE_CACHE_SIZE;
309 hole = is_hole(be, isect);
310 if (hole && !cow_read) {
311 bio = bl_submit_bio(READ, bio);
312 /* Fill hole w/ zeroes w/o accessing device */
313 dprintk("%s Zeroing page for hole\n", __func__);
314 zero_user_segment(pages[i], pg_offset, pg_len);
315 print_page(pages[i]);
316 SetPageUptodate(pages[i]);
318 struct pnfs_block_extent *be_read;
320 be_read = (hole && cow_read) ? cow_read : be;
321 bio = do_add_page_to_bio(bio,
322 rdata->page_array.npages - i,
324 isect, pages[i], be_read,
328 header->pnfs_error = PTR_ERR(bio);
333 isect += (pg_len >> SECTOR_SHIFT);
334 extent_length -= PAGE_CACHE_SECTORS;
336 if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
338 rdata->res.count = header->inode->i_size - rdata->args.offset;
340 rdata->res.count = (isect << SECTOR_SHIFT) - rdata->args.offset;
344 bl_put_extent(cow_read);
345 bl_submit_bio(READ, bio);
347 return PNFS_ATTEMPTED;
350 dprintk("Giving up and using normal NFS\n");
351 return PNFS_NOT_ATTEMPTED;
354 static void mark_extents_written(struct pnfs_block_layout *bl,
355 __u64 offset, __u32 count)
358 struct pnfs_block_extent *be;
359 struct pnfs_block_short_extent *se;
361 dprintk("%s(%llu, %u)\n", __func__, offset, count);
364 isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
365 end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
366 end >>= SECTOR_SHIFT;
367 while (isect < end) {
369 be = bl_find_get_extent(bl, isect, NULL);
370 BUG_ON(!be); /* FIXME */
371 len = min(end, be->be_f_offset + be->be_length) - isect;
372 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
373 se = bl_pop_one_short_extent(be->be_inval);
375 bl_mark_for_commit(be, isect, len, se);
382 static void bl_end_io_write_zero(struct bio *bio, int err)
384 struct parallel_io *par = bio->bi_private;
385 struct bio_vec *bvec;
388 bio_for_each_segment_all(bvec, bio, i) {
389 /* This is the zeroing page we added */
390 end_page_writeback(bvec->bv_page);
391 page_cache_release(bvec->bv_page);
395 struct nfs_pgio_data *data = par->data;
396 struct nfs_pgio_header *header = data->header;
398 if (!header->pnfs_error)
399 header->pnfs_error = -EIO;
400 pnfs_set_lo_fail(header->lseg);
406 static void bl_end_io_write(struct bio *bio, int err)
408 struct parallel_io *par = bio->bi_private;
409 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
410 struct nfs_pgio_data *data = par->data;
411 struct nfs_pgio_header *header = data->header;
414 if (!header->pnfs_error)
415 header->pnfs_error = -EIO;
416 pnfs_set_lo_fail(header->lseg);
422 /* Function scheduled for call during bl_end_par_io_write,
423 * it marks sectors as written and extends the commitlist.
425 static void bl_write_cleanup(struct work_struct *work)
427 struct rpc_task *task;
428 struct nfs_pgio_data *wdata;
429 dprintk("%s enter\n", __func__);
430 task = container_of(work, struct rpc_task, u.tk_work);
431 wdata = container_of(task, struct nfs_pgio_data, task);
432 if (likely(!wdata->header->pnfs_error)) {
433 /* Marks for LAYOUTCOMMIT */
434 mark_extents_written(BLK_LSEG2EXT(wdata->header->lseg),
435 wdata->args.offset, wdata->args.count);
437 pnfs_ld_write_done(wdata);
440 /* Called when last of bios associated with a bl_write_pagelist call finishes */
441 static void bl_end_par_io_write(void *data, int num_se)
443 struct nfs_pgio_data *wdata = data;
445 if (unlikely(wdata->header->pnfs_error)) {
446 bl_free_short_extents(&BLK_LSEG2EXT(wdata->header->lseg)->bl_inval,
450 wdata->task.tk_status = wdata->header->pnfs_error;
451 wdata->writeverf.committed = NFS_FILE_SYNC;
452 INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
453 schedule_work(&wdata->task.u.tk_work);
456 /* FIXME STUB - mark intersection of layout and page as bad, so is not
459 static void mark_bad_read(void)
465 * map_block: map a requested I/0 block (isect) into an offset in the LVM
469 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
471 dprintk("%s enter be=%p\n", __func__, be);
473 set_buffer_mapped(bh);
474 bh->b_bdev = be->be_mdev;
475 bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
476 (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
478 dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
479 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
485 bl_read_single_end_io(struct bio *bio, int error)
487 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
488 struct page *page = bvec->bv_page;
490 /* Only one page in bvec */
495 bl_do_readpage_sync(struct page *page, struct pnfs_block_extent *be,
496 unsigned int offset, unsigned int len)
499 struct page *shadow_page;
501 char *kaddr, *kshadow_addr;
504 dprintk("%s: offset %u len %u\n", __func__, offset, len);
506 shadow_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
507 if (shadow_page == NULL)
510 bio = bio_alloc(GFP_NOIO, 1);
514 isect = (page->index << PAGE_CACHE_SECTOR_SHIFT) +
515 (offset / SECTOR_SIZE);
517 bio->bi_iter.bi_sector = isect - be->be_f_offset + be->be_v_offset;
518 bio->bi_bdev = be->be_mdev;
519 bio->bi_end_io = bl_read_single_end_io;
521 lock_page(shadow_page);
522 if (bio_add_page(bio, shadow_page,
523 SECTOR_SIZE, round_down(offset, SECTOR_SIZE)) == 0) {
524 unlock_page(shadow_page);
529 submit_bio(READ, bio);
530 wait_on_page_locked(shadow_page);
531 if (unlikely(!test_bit(BIO_UPTODATE, &bio->bi_flags))) {
534 kaddr = kmap_atomic(page);
535 kshadow_addr = kmap_atomic(shadow_page);
536 memcpy(kaddr + offset, kshadow_addr + offset, len);
537 kunmap_atomic(kshadow_addr);
538 kunmap_atomic(kaddr);
540 __free_page(shadow_page);
547 bl_read_partial_page_sync(struct page *page, struct pnfs_block_extent *be,
548 unsigned int dirty_offset, unsigned int dirty_len,
552 unsigned int start, end;
556 end = PAGE_CACHE_SIZE;
558 start = round_down(dirty_offset, SECTOR_SIZE);
559 end = round_up(dirty_offset + dirty_len, SECTOR_SIZE);
562 dprintk("%s: offset %u len %d\n", __func__, dirty_offset, dirty_len);
564 zero_user_segments(page, start, dirty_offset,
565 dirty_offset + dirty_len, end);
566 if (start == 0 && end == PAGE_CACHE_SIZE &&
567 trylock_page(page)) {
568 SetPageUptodate(page);
574 if (start != dirty_offset)
575 ret = bl_do_readpage_sync(page, be, start, dirty_offset - start);
577 if (!ret && (dirty_offset + dirty_len < end))
578 ret = bl_do_readpage_sync(page, be, dirty_offset + dirty_len,
579 end - dirty_offset - dirty_len);
584 /* Given an unmapped page, zero it or read in page for COW, page is locked
588 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
590 struct buffer_head *bh = NULL;
594 dprintk("%s enter, %p\n", __func__, page);
595 BUG_ON(PageUptodate(page));
597 zero_user_segment(page, 0, PAGE_SIZE);
598 SetPageUptodate(page);
602 bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
608 isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
609 map_block(bh, isect, cow_read);
610 if (!bh_uptodate_or_lock(bh))
611 ret = bh_submit_read(bh);
614 SetPageUptodate(page);
618 free_buffer_head(bh);
620 /* Need to mark layout with bad read...should now
621 * just use nfs4 for reads and writes.
628 /* Find or create a zeroing page marked being writeback.
629 * Return ERR_PTR on error, NULL to indicate skip this page and page itself
630 * to indicate write out.
633 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
634 struct pnfs_block_extent *cow_read)
638 page = find_get_page(inode->i_mapping, index);
642 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
643 if (unlikely(!page)) {
644 dprintk("%s oom\n", __func__);
645 return ERR_PTR(-ENOMEM);
650 /* PageDirty: Other will write this out
651 * PageWriteback: Other is writing this out
652 * PageUptodate: It was read before
654 if (PageDirty(page) || PageWriteback(page)) {
658 page_cache_release(page);
667 if (!PageUptodate(page)) {
668 /* New page, readin or zero it */
669 init_page_for_write(page, cow_read);
671 set_page_writeback(page);
677 static enum pnfs_try_status
678 bl_write_pagelist(struct nfs_pgio_data *wdata, int sync)
680 struct nfs_pgio_header *header = wdata->header;
681 int i, ret, npg_zero, pg_index, last = 0;
682 struct bio *bio = NULL;
683 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
684 sector_t isect, last_isect = 0, extent_length = 0;
685 struct parallel_io *par = NULL;
686 loff_t offset = wdata->args.offset;
687 size_t count = wdata->args.count;
688 unsigned int pg_offset, pg_len, saved_len;
689 struct page **pages = wdata->args.pages;
694 NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
696 dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
698 if (header->dreq != NULL &&
699 (!IS_ALIGNED(offset, NFS_SERVER(header->inode)->pnfs_blksize) ||
700 !IS_ALIGNED(count, NFS_SERVER(header->inode)->pnfs_blksize))) {
701 dprintk("pnfsblock nonblock aligned DIO writes. Resend MDS\n");
704 /* At this point, wdata->page_aray is a (sequential) list of nfs_pages.
705 * We want to write each, and if there is an error set pnfs_error
706 * to have it redone using nfs.
708 par = alloc_parallel(wdata);
711 par->pnfs_callback = bl_end_par_io_write;
712 /* At this point, have to be more careful with error handling */
714 isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
715 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg), isect, &cow_read);
716 if (!be || !is_writable(be, isect)) {
717 dprintk("%s no matching extents!\n", __func__);
721 /* First page inside INVALID extent */
722 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
723 if (likely(!bl_push_one_short_extent(be->be_inval)))
727 temp = offset >> PAGE_CACHE_SHIFT;
728 npg_zero = do_div(temp, npg_per_block);
729 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
730 (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
731 extent_length = be->be_length - (isect - be->be_f_offset);
734 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
735 for (;npg_zero > 0; npg_zero--) {
736 if (bl_is_sector_init(be->be_inval, isect)) {
737 dprintk("isect %llu already init\n",
738 (unsigned long long)isect);
741 /* page ref released in bl_end_io_write_zero */
742 index = isect >> PAGE_CACHE_SECTOR_SHIFT;
743 dprintk("%s zero %dth page: index %lu isect %llu\n",
744 __func__, npg_zero, index,
745 (unsigned long long)isect);
746 page = bl_find_get_zeroing_page(header->inode, index,
748 if (unlikely(IS_ERR(page))) {
749 header->pnfs_error = PTR_ERR(page);
751 } else if (page == NULL)
754 ret = bl_mark_sectors_init(be->be_inval, isect,
757 dprintk("%s bl_mark_sectors_init fail %d\n",
759 end_page_writeback(page);
760 page_cache_release(page);
761 header->pnfs_error = ret;
764 if (likely(!bl_push_one_short_extent(be->be_inval)))
767 end_page_writeback(page);
768 page_cache_release(page);
769 header->pnfs_error = -ENOMEM;
772 /* FIXME: This should be done in bi_end_io */
773 mark_extents_written(BLK_LSEG2EXT(header->lseg),
774 page->index << PAGE_CACHE_SHIFT,
777 bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
779 bl_end_io_write_zero, par);
781 header->pnfs_error = PTR_ERR(bio);
786 isect += PAGE_CACHE_SECTORS;
787 extent_length -= PAGE_CACHE_SECTORS;
792 bio = bl_submit_bio(WRITE, bio);
795 pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
796 for (i = pg_index; i < wdata->page_array.npages; i++) {
797 if (!extent_length) {
798 /* We've used up the previous extent */
800 bl_put_extent(cow_read);
801 bio = bl_submit_bio(WRITE, bio);
802 /* Get the next one */
803 be = bl_find_get_extent(BLK_LSEG2EXT(header->lseg),
805 if (!be || !is_writable(be, isect)) {
806 header->pnfs_error = -EINVAL;
809 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
810 if (likely(!bl_push_one_short_extent(
814 header->pnfs_error = -ENOMEM;
818 extent_length = be->be_length -
819 (isect - be->be_f_offset);
822 dprintk("%s offset %lld count %Zu\n", __func__, offset, count);
823 pg_offset = offset & ~PAGE_CACHE_MASK;
824 if (pg_offset + count > PAGE_CACHE_SIZE)
825 pg_len = PAGE_CACHE_SIZE - pg_offset;
830 if (be->be_state == PNFS_BLOCK_INVALID_DATA &&
831 !bl_is_sector_init(be->be_inval, isect)) {
832 ret = bl_read_partial_page_sync(pages[i], cow_read,
833 pg_offset, pg_len, true);
835 dprintk("%s bl_read_partial_page_sync fail %d\n",
837 header->pnfs_error = ret;
841 ret = bl_mark_sectors_init(be->be_inval, isect,
844 dprintk("%s bl_mark_sectors_init fail %d\n",
846 header->pnfs_error = ret;
850 /* Expand to full page write */
852 pg_len = PAGE_CACHE_SIZE;
853 } else if ((pg_offset & (SECTOR_SIZE - 1)) ||
854 (pg_len & (SECTOR_SIZE - 1))){
855 /* ahh, nasty case. We have to do sync full sector
856 * read-modify-write cycles.
858 unsigned int saved_offset = pg_offset;
859 ret = bl_read_partial_page_sync(pages[i], be, pg_offset,
861 pg_offset = round_down(pg_offset, SECTOR_SIZE);
862 pg_len = round_up(saved_offset + pg_len, SECTOR_SIZE)
867 bio = do_add_page_to_bio(bio, wdata->page_array.npages - i,
870 bl_end_io_write, par,
873 header->pnfs_error = PTR_ERR(bio);
879 isect += PAGE_CACHE_SECTORS;
881 extent_length -= PAGE_CACHE_SECTORS;
884 /* Last page inside INVALID extent */
885 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
886 bio = bl_submit_bio(WRITE, bio);
887 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
888 npg_zero = npg_per_block - do_div(temp, npg_per_block);
889 if (npg_zero < npg_per_block) {
891 goto fill_invalid_ext;
896 wdata->res.count = wdata->args.count;
899 bl_put_extent(cow_read);
900 bl_submit_bio(WRITE, bio);
902 return PNFS_ATTEMPTED;
905 bl_put_extent(cow_read);
907 return PNFS_NOT_ATTEMPTED;
910 /* FIXME - range ignored */
912 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
915 struct pnfs_block_extent *be;
917 spin_lock(&bl->bl_ext_lock);
918 for (i = 0; i < EXTENT_LISTS; i++) {
919 while (!list_empty(&bl->bl_extents[i])) {
920 be = list_first_entry(&bl->bl_extents[i],
921 struct pnfs_block_extent,
923 list_del(&be->be_node);
927 spin_unlock(&bl->bl_ext_lock);
931 release_inval_marks(struct pnfs_inval_markings *marks)
933 struct pnfs_inval_tracking *pos, *temp;
934 struct pnfs_block_short_extent *se, *stemp;
936 list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
937 list_del(&pos->it_link);
941 list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
942 list_del(&se->bse_node);
948 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
950 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
952 dprintk("%s enter\n", __func__);
953 release_extents(bl, NULL);
954 release_inval_marks(&bl->bl_inval);
958 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
961 struct pnfs_block_layout *bl;
963 dprintk("%s enter\n", __func__);
964 bl = kzalloc(sizeof(*bl), gfp_flags);
967 spin_lock_init(&bl->bl_ext_lock);
968 INIT_LIST_HEAD(&bl->bl_extents[0]);
969 INIT_LIST_HEAD(&bl->bl_extents[1]);
970 INIT_LIST_HEAD(&bl->bl_commit);
971 INIT_LIST_HEAD(&bl->bl_committing);
973 bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
974 BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
975 return &bl->bl_layout;
978 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
980 dprintk("%s enter\n", __func__);
984 /* We pretty much ignore lseg, and store all data layout wide, so we
985 * can correctly merge.
987 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
988 struct nfs4_layoutget_res *lgr,
991 struct pnfs_layout_segment *lseg;
994 dprintk("%s enter\n", __func__);
995 lseg = kzalloc(sizeof(*lseg), gfp_flags);
997 return ERR_PTR(-ENOMEM);
998 status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
1000 /* We don't want to call the full-blown bl_free_lseg,
1001 * since on error extents were not touched.
1004 return ERR_PTR(status);
1010 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
1011 const struct nfs4_layoutcommit_args *arg)
1013 dprintk("%s enter\n", __func__);
1014 encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
1018 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
1020 struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
1022 dprintk("%s enter\n", __func__);
1023 clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
1026 static void free_blk_mountid(struct block_mount_id *mid)
1029 struct pnfs_block_dev *dev, *tmp;
1031 /* No need to take bm_lock as we are last user freeing bm_devlist */
1032 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
1033 list_del(&dev->bm_node);
1034 bl_free_block_dev(dev);
1040 /* This is mostly copied from the filelayout_get_device_info function.
1041 * It seems much of this should be at the generic pnfs level.
1043 static struct pnfs_block_dev *
1044 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
1045 struct nfs4_deviceid *d_id)
1047 struct pnfs_device *dev;
1048 struct pnfs_block_dev *rv;
1051 struct page **pages = NULL;
1055 * Use the session max response size as the basis for setting
1056 * GETDEVICEINFO's maxcount
1058 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
1059 max_pages = nfs_page_array_len(0, max_resp_sz);
1060 dprintk("%s max_resp_sz %u max_pages %d\n",
1061 __func__, max_resp_sz, max_pages);
1063 dev = kmalloc(sizeof(*dev), GFP_NOFS);
1065 dprintk("%s kmalloc failed\n", __func__);
1066 return ERR_PTR(-ENOMEM);
1069 pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
1070 if (pages == NULL) {
1072 return ERR_PTR(-ENOMEM);
1074 for (i = 0; i < max_pages; i++) {
1075 pages[i] = alloc_page(GFP_NOFS);
1077 rv = ERR_PTR(-ENOMEM);
1082 memcpy(&dev->dev_id, d_id, sizeof(*d_id));
1083 dev->layout_type = LAYOUT_BLOCK_VOLUME;
1086 dev->pglen = PAGE_SIZE * max_pages;
1088 dev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead;
1090 dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
1091 rc = nfs4_proc_getdeviceinfo(server, dev, NULL);
1092 dprintk("%s getdevice info returns %d\n", __func__, rc);
1098 rv = nfs4_blk_decode_device(server, dev);
1100 for (i = 0; i < max_pages; i++)
1101 __free_page(pages[i]);
1108 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
1110 struct block_mount_id *b_mt_id = NULL;
1111 struct pnfs_devicelist *dlist = NULL;
1112 struct pnfs_block_dev *bdev;
1113 LIST_HEAD(block_disklist);
1116 dprintk("%s enter\n", __func__);
1118 if (server->pnfs_blksize == 0) {
1119 dprintk("%s Server did not return blksize\n", __func__);
1122 b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
1127 /* Initialize nfs4 block layout mount id */
1128 spin_lock_init(&b_mt_id->bm_lock);
1129 INIT_LIST_HEAD(&b_mt_id->bm_devlist);
1131 dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
1137 while (!dlist->eof) {
1138 status = nfs4_proc_getdevicelist(server, fh, dlist);
1141 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
1142 __func__, dlist->num_devs, dlist->eof);
1143 for (i = 0; i < dlist->num_devs; i++) {
1144 bdev = nfs4_blk_get_deviceinfo(server, fh,
1147 status = PTR_ERR(bdev);
1150 spin_lock(&b_mt_id->bm_lock);
1151 list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
1152 spin_unlock(&b_mt_id->bm_lock);
1155 dprintk("%s SUCCESS\n", __func__);
1156 server->pnfs_ld_data = b_mt_id;
1163 free_blk_mountid(b_mt_id);
1168 bl_clear_layoutdriver(struct nfs_server *server)
1170 struct block_mount_id *b_mt_id = server->pnfs_ld_data;
1172 dprintk("%s enter\n", __func__);
1173 free_blk_mountid(b_mt_id);
1174 dprintk("%s RETURNS\n", __func__);
1179 is_aligned_req(struct nfs_page *req, unsigned int alignment)
1181 return IS_ALIGNED(req->wb_offset, alignment) &&
1182 IS_ALIGNED(req->wb_bytes, alignment);
1186 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1188 if (pgio->pg_dreq != NULL &&
1189 !is_aligned_req(req, SECTOR_SIZE))
1190 nfs_pageio_reset_read_mds(pgio);
1192 pnfs_generic_pg_init_read(pgio, req);
1196 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1197 * of bytes (maximum @req->wb_bytes) that can be coalesced.
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 = page_cache_next_hole(mapping, 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 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1253 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1256 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
1257 struct nfs_page *req)
1259 if (pgio->pg_dreq != NULL &&
1260 !is_aligned_req(req, PAGE_CACHE_SIZE))
1263 return pnfs_generic_pg_test(pgio, prev, req);
1266 static const struct nfs_pageio_ops bl_pg_read_ops = {
1267 .pg_init = bl_pg_init_read,
1268 .pg_test = bl_pg_test_read,
1269 .pg_doio = pnfs_generic_pg_readpages,
1272 static const struct nfs_pageio_ops bl_pg_write_ops = {
1273 .pg_init = bl_pg_init_write,
1274 .pg_test = bl_pg_test_write,
1275 .pg_doio = pnfs_generic_pg_writepages,
1278 static struct pnfs_layoutdriver_type blocklayout_type = {
1279 .id = LAYOUT_BLOCK_VOLUME,
1280 .name = "LAYOUT_BLOCK_VOLUME",
1281 .owner = THIS_MODULE,
1282 .read_pagelist = bl_read_pagelist,
1283 .write_pagelist = bl_write_pagelist,
1284 .alloc_layout_hdr = bl_alloc_layout_hdr,
1285 .free_layout_hdr = bl_free_layout_hdr,
1286 .alloc_lseg = bl_alloc_lseg,
1287 .free_lseg = bl_free_lseg,
1288 .encode_layoutcommit = bl_encode_layoutcommit,
1289 .cleanup_layoutcommit = bl_cleanup_layoutcommit,
1290 .set_layoutdriver = bl_set_layoutdriver,
1291 .clear_layoutdriver = bl_clear_layoutdriver,
1292 .pg_read_ops = &bl_pg_read_ops,
1293 .pg_write_ops = &bl_pg_write_ops,
1296 static const struct rpc_pipe_ops bl_upcall_ops = {
1297 .upcall = rpc_pipe_generic_upcall,
1298 .downcall = bl_pipe_downcall,
1299 .destroy_msg = bl_pipe_destroy_msg,
1302 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1303 struct rpc_pipe *pipe)
1305 struct dentry *dir, *dentry;
1307 dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1309 return ERR_PTR(-ENOENT);
1310 dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1315 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1316 struct rpc_pipe *pipe)
1319 rpc_unlink(pipe->dentry);
1322 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1325 struct super_block *sb = ptr;
1326 struct net *net = sb->s_fs_info;
1327 struct nfs_net *nn = net_generic(net, nfs_net_id);
1328 struct dentry *dentry;
1331 if (!try_module_get(THIS_MODULE))
1334 if (nn->bl_device_pipe == NULL) {
1335 module_put(THIS_MODULE);
1340 case RPC_PIPEFS_MOUNT:
1341 dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1342 if (IS_ERR(dentry)) {
1343 ret = PTR_ERR(dentry);
1346 nn->bl_device_pipe->dentry = dentry;
1348 case RPC_PIPEFS_UMOUNT:
1349 if (nn->bl_device_pipe->dentry)
1350 nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1356 module_put(THIS_MODULE);
1360 static struct notifier_block nfs4blocklayout_block = {
1361 .notifier_call = rpc_pipefs_event,
1364 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1365 struct rpc_pipe *pipe)
1367 struct super_block *pipefs_sb;
1368 struct dentry *dentry;
1370 pipefs_sb = rpc_get_sb_net(net);
1373 dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1374 rpc_put_sb_net(net);
1378 static void nfs4blocklayout_unregister_net(struct net *net,
1379 struct rpc_pipe *pipe)
1381 struct super_block *pipefs_sb;
1383 pipefs_sb = rpc_get_sb_net(net);
1385 nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1386 rpc_put_sb_net(net);
1390 static int nfs4blocklayout_net_init(struct net *net)
1392 struct nfs_net *nn = net_generic(net, nfs_net_id);
1393 struct dentry *dentry;
1395 init_waitqueue_head(&nn->bl_wq);
1396 nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1397 if (IS_ERR(nn->bl_device_pipe))
1398 return PTR_ERR(nn->bl_device_pipe);
1399 dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1400 if (IS_ERR(dentry)) {
1401 rpc_destroy_pipe_data(nn->bl_device_pipe);
1402 return PTR_ERR(dentry);
1404 nn->bl_device_pipe->dentry = dentry;
1408 static void nfs4blocklayout_net_exit(struct net *net)
1410 struct nfs_net *nn = net_generic(net, nfs_net_id);
1412 nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1413 rpc_destroy_pipe_data(nn->bl_device_pipe);
1414 nn->bl_device_pipe = NULL;
1417 static struct pernet_operations nfs4blocklayout_net_ops = {
1418 .init = nfs4blocklayout_net_init,
1419 .exit = nfs4blocklayout_net_exit,
1422 static int __init nfs4blocklayout_init(void)
1426 dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1428 ret = pnfs_register_layoutdriver(&blocklayout_type);
1432 ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1435 ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1442 rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1444 pnfs_unregister_layoutdriver(&blocklayout_type);
1448 static void __exit nfs4blocklayout_exit(void)
1450 dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1453 rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1454 unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1455 pnfs_unregister_layoutdriver(&blocklayout_type);
1458 MODULE_ALIAS("nfs-layouttype4-3");
1460 module_init(nfs4blocklayout_init);
1461 module_exit(nfs4blocklayout_exit);