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>
41 #include "blocklayout.h"
43 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
45 MODULE_LICENSE("GPL");
46 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
47 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
49 wait_queue_head_t bl_wq;
51 static void print_page(struct page *page)
53 dprintk("PRINTPAGE page %p\n", page);
54 dprintk(" PagePrivate %d\n", PagePrivate(page));
55 dprintk(" PageUptodate %d\n", PageUptodate(page));
56 dprintk(" PageError %d\n", PageError(page));
57 dprintk(" PageDirty %d\n", PageDirty(page));
58 dprintk(" PageReferenced %d\n", PageReferenced(page));
59 dprintk(" PageLocked %d\n", PageLocked(page));
60 dprintk(" PageWriteback %d\n", PageWriteback(page));
61 dprintk(" PageMappedToDisk %d\n", PageMappedToDisk(page));
65 /* Given the be associated with isect, determine if page data needs to be
68 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
70 if (be->be_state == PNFS_BLOCK_NONE_DATA)
72 else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
75 return !bl_is_sector_init(be->be_inval, isect);
78 /* Given the be associated with isect, determine if page data can be
81 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
83 return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
84 be->be_state == PNFS_BLOCK_INVALID_DATA);
87 /* The data we are handed might be spread across several bios. We need
88 * to track when the last one is finished.
92 void (*pnfs_callback) (void *data, int num_se);
97 static inline struct parallel_io *alloc_parallel(void *data)
99 struct parallel_io *rv;
101 rv = kmalloc(sizeof(*rv), GFP_NOFS);
104 kref_init(&rv->refcnt);
110 static inline void get_parallel(struct parallel_io *p)
112 kref_get(&p->refcnt);
115 static void destroy_parallel(struct kref *kref)
117 struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
119 dprintk("%s enter\n", __func__);
120 p->pnfs_callback(p->data, p->bse_count);
124 static inline void put_parallel(struct parallel_io *p)
126 kref_put(&p->refcnt, destroy_parallel);
130 bl_submit_bio(int rw, struct bio *bio)
133 get_parallel(bio->bi_private);
134 dprintk("%s submitting %s bio %u@%llu\n", __func__,
135 rw == READ ? "read" : "write",
136 bio->bi_size, (unsigned long long)bio->bi_sector);
142 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
143 struct pnfs_block_extent *be,
144 void (*end_io)(struct bio *, int err),
145 struct parallel_io *par)
149 npg = min(npg, BIO_MAX_PAGES);
150 bio = bio_alloc(GFP_NOIO, npg);
151 if (!bio && (current->flags & PF_MEMALLOC)) {
152 while (!bio && (npg /= 2))
153 bio = bio_alloc(GFP_NOIO, npg);
157 bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
158 bio->bi_bdev = be->be_mdev;
159 bio->bi_end_io = end_io;
160 bio->bi_private = par;
165 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
166 sector_t isect, struct page *page,
167 struct pnfs_block_extent *be,
168 void (*end_io)(struct bio *, int err),
169 struct parallel_io *par)
173 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
175 return ERR_PTR(-ENOMEM);
177 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
178 bio = bl_submit_bio(rw, bio);
184 /* This is basically copied from mpage_end_io_read */
185 static void bl_end_io_read(struct bio *bio, int err)
187 struct parallel_io *par = bio->bi_private;
188 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
189 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
190 struct nfs_read_data *rdata = (struct nfs_read_data *)par->data;
193 struct page *page = bvec->bv_page;
195 if (--bvec >= bio->bi_io_vec)
196 prefetchw(&bvec->bv_page->flags);
198 SetPageUptodate(page);
199 } while (bvec >= bio->bi_io_vec);
201 if (!rdata->pnfs_error)
202 rdata->pnfs_error = -EIO;
203 pnfs_set_lo_fail(rdata->lseg);
209 static void bl_read_cleanup(struct work_struct *work)
211 struct rpc_task *task;
212 struct nfs_read_data *rdata;
213 dprintk("%s enter\n", __func__);
214 task = container_of(work, struct rpc_task, u.tk_work);
215 rdata = container_of(task, struct nfs_read_data, task);
216 pnfs_ld_read_done(rdata);
220 bl_end_par_io_read(void *data, int unused)
222 struct nfs_read_data *rdata = data;
224 rdata->task.tk_status = rdata->pnfs_error;
225 INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
226 schedule_work(&rdata->task.u.tk_work);
229 static enum pnfs_try_status
230 bl_read_pagelist(struct nfs_read_data *rdata)
233 struct bio *bio = NULL;
234 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
235 sector_t isect, extent_length = 0;
236 struct parallel_io *par;
237 loff_t f_offset = rdata->args.offset;
238 size_t count = rdata->args.count;
239 struct page **pages = rdata->args.pages;
240 int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
242 dprintk("%s enter nr_pages %u offset %lld count %Zd\n", __func__,
243 rdata->npages, f_offset, count);
245 par = alloc_parallel(rdata);
248 par->pnfs_callback = bl_end_par_io_read;
249 /* At this point, we can no longer jump to use_mds */
251 isect = (sector_t) (f_offset >> SECTOR_SHIFT);
252 /* Code assumes extents are page-aligned */
253 for (i = pg_index; i < rdata->npages; i++) {
254 if (!extent_length) {
255 /* We've used up the previous extent */
257 bl_put_extent(cow_read);
258 bio = bl_submit_bio(READ, bio);
259 /* Get the next one */
260 be = bl_find_get_extent(BLK_LSEG2EXT(rdata->lseg),
263 rdata->pnfs_error = -EIO;
266 extent_length = be->be_length -
267 (isect - be->be_f_offset);
269 sector_t cow_length = cow_read->be_length -
270 (isect - cow_read->be_f_offset);
271 extent_length = min(extent_length, cow_length);
274 hole = is_hole(be, isect);
275 if (hole && !cow_read) {
276 bio = bl_submit_bio(READ, bio);
277 /* Fill hole w/ zeroes w/o accessing device */
278 dprintk("%s Zeroing page for hole\n", __func__);
279 zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE);
280 print_page(pages[i]);
281 SetPageUptodate(pages[i]);
283 struct pnfs_block_extent *be_read;
285 be_read = (hole && cow_read) ? cow_read : be;
286 bio = bl_add_page_to_bio(bio, rdata->npages - i, READ,
287 isect, pages[i], be_read,
288 bl_end_io_read, par);
290 rdata->pnfs_error = PTR_ERR(bio);
295 isect += PAGE_CACHE_SECTORS;
296 extent_length -= PAGE_CACHE_SECTORS;
298 if ((isect << SECTOR_SHIFT) >= rdata->inode->i_size) {
300 rdata->res.count = rdata->inode->i_size - f_offset;
302 rdata->res.count = (isect << SECTOR_SHIFT) - f_offset;
306 bl_put_extent(cow_read);
307 bl_submit_bio(READ, bio);
309 return PNFS_ATTEMPTED;
312 dprintk("Giving up and using normal NFS\n");
313 return PNFS_NOT_ATTEMPTED;
316 static void mark_extents_written(struct pnfs_block_layout *bl,
317 __u64 offset, __u32 count)
320 struct pnfs_block_extent *be;
321 struct pnfs_block_short_extent *se;
323 dprintk("%s(%llu, %u)\n", __func__, offset, count);
326 isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
327 end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
328 end >>= SECTOR_SHIFT;
329 while (isect < end) {
331 be = bl_find_get_extent(bl, isect, NULL);
332 BUG_ON(!be); /* FIXME */
333 len = min(end, be->be_f_offset + be->be_length) - isect;
334 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
335 se = bl_pop_one_short_extent(be->be_inval);
337 bl_mark_for_commit(be, isect, len, se);
344 static void bl_end_io_write_zero(struct bio *bio, int err)
346 struct parallel_io *par = bio->bi_private;
347 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
348 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
349 struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
352 struct page *page = bvec->bv_page;
354 if (--bvec >= bio->bi_io_vec)
355 prefetchw(&bvec->bv_page->flags);
356 /* This is the zeroing page we added */
357 end_page_writeback(page);
358 page_cache_release(page);
359 } while (bvec >= bio->bi_io_vec);
361 if (unlikely(!uptodate)) {
362 if (!wdata->pnfs_error)
363 wdata->pnfs_error = -EIO;
364 pnfs_set_lo_fail(wdata->lseg);
370 static void bl_end_io_write(struct bio *bio, int err)
372 struct parallel_io *par = bio->bi_private;
373 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
374 struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
377 if (!wdata->pnfs_error)
378 wdata->pnfs_error = -EIO;
379 pnfs_set_lo_fail(wdata->lseg);
385 /* Function scheduled for call during bl_end_par_io_write,
386 * it marks sectors as written and extends the commitlist.
388 static void bl_write_cleanup(struct work_struct *work)
390 struct rpc_task *task;
391 struct nfs_write_data *wdata;
392 dprintk("%s enter\n", __func__);
393 task = container_of(work, struct rpc_task, u.tk_work);
394 wdata = container_of(task, struct nfs_write_data, task);
395 if (likely(!wdata->pnfs_error)) {
396 /* Marks for LAYOUTCOMMIT */
397 mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
398 wdata->args.offset, wdata->args.count);
400 pnfs_ld_write_done(wdata);
403 /* Called when last of bios associated with a bl_write_pagelist call finishes */
404 static void bl_end_par_io_write(void *data, int num_se)
406 struct nfs_write_data *wdata = data;
408 if (unlikely(wdata->pnfs_error)) {
409 bl_free_short_extents(&BLK_LSEG2EXT(wdata->lseg)->bl_inval,
413 wdata->task.tk_status = wdata->pnfs_error;
414 wdata->verf.committed = NFS_FILE_SYNC;
415 INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
416 schedule_work(&wdata->task.u.tk_work);
419 /* FIXME STUB - mark intersection of layout and page as bad, so is not
422 static void mark_bad_read(void)
428 * map_block: map a requested I/0 block (isect) into an offset in the LVM
432 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
434 dprintk("%s enter be=%p\n", __func__, be);
436 set_buffer_mapped(bh);
437 bh->b_bdev = be->be_mdev;
438 bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
439 (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
441 dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
442 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
447 /* Given an unmapped page, zero it or read in page for COW, page is locked
451 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
453 struct buffer_head *bh = NULL;
457 dprintk("%s enter, %p\n", __func__, page);
458 BUG_ON(PageUptodate(page));
460 zero_user_segment(page, 0, PAGE_SIZE);
461 SetPageUptodate(page);
465 bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
471 isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
472 map_block(bh, isect, cow_read);
473 if (!bh_uptodate_or_lock(bh))
474 ret = bh_submit_read(bh);
477 SetPageUptodate(page);
480 bl_put_extent(cow_read);
482 free_buffer_head(bh);
484 /* Need to mark layout with bad read...should now
485 * just use nfs4 for reads and writes.
492 /* Find or create a zeroing page marked being writeback.
493 * Return ERR_PTR on error, NULL to indicate skip this page and page itself
494 * to indicate write out.
497 bl_find_get_zeroing_page(struct inode *inode, pgoff_t index,
498 struct pnfs_block_extent *cow_read)
502 page = find_get_page(inode->i_mapping, index);
506 page = find_or_create_page(inode->i_mapping, index, GFP_NOFS);
507 if (unlikely(!page)) {
508 dprintk("%s oom\n", __func__);
509 return ERR_PTR(-ENOMEM);
514 /* PageDirty: Other will write this out
515 * PageWriteback: Other is writing this out
516 * PageUptodate: It was read before
518 if (PageDirty(page) || PageWriteback(page)) {
522 page_cache_release(page);
531 if (!PageUptodate(page)) {
532 /* New page, readin or zero it */
533 init_page_for_write(page, cow_read);
535 set_page_writeback(page);
541 static enum pnfs_try_status
542 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
544 int i, ret, npg_zero, pg_index, last = 0;
545 struct bio *bio = NULL;
546 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
547 sector_t isect, last_isect = 0, extent_length = 0;
548 struct parallel_io *par;
549 loff_t offset = wdata->args.offset;
550 size_t count = wdata->args.count;
551 struct page **pages = wdata->args.pages;
556 NFS_SERVER(wdata->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
558 dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
559 /* At this point, wdata->pages is a (sequential) list of nfs_pages.
560 * We want to write each, and if there is an error set pnfs_error
561 * to have it redone using nfs.
563 par = alloc_parallel(wdata);
566 par->pnfs_callback = bl_end_par_io_write;
567 /* At this point, have to be more careful with error handling */
569 isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
570 be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg), isect, &cow_read);
571 if (!be || !is_writable(be, isect)) {
572 dprintk("%s no matching extents!\n", __func__);
576 /* First page inside INVALID extent */
577 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
578 if (likely(!bl_push_one_short_extent(be->be_inval)))
582 temp = offset >> PAGE_CACHE_SHIFT;
583 npg_zero = do_div(temp, npg_per_block);
584 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
585 (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
586 extent_length = be->be_length - (isect - be->be_f_offset);
589 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
590 for (;npg_zero > 0; npg_zero--) {
591 if (bl_is_sector_init(be->be_inval, isect)) {
592 dprintk("isect %llu already init\n",
593 (unsigned long long)isect);
596 /* page ref released in bl_end_io_write_zero */
597 index = isect >> PAGE_CACHE_SECTOR_SHIFT;
598 dprintk("%s zero %dth page: index %lu isect %llu\n",
599 __func__, npg_zero, index,
600 (unsigned long long)isect);
601 page = bl_find_get_zeroing_page(wdata->inode, index,
603 if (unlikely(IS_ERR(page))) {
604 wdata->pnfs_error = PTR_ERR(page);
606 } else if (page == NULL)
609 ret = bl_mark_sectors_init(be->be_inval, isect,
612 dprintk("%s bl_mark_sectors_init fail %d\n",
614 end_page_writeback(page);
615 page_cache_release(page);
616 wdata->pnfs_error = ret;
619 if (likely(!bl_push_one_short_extent(be->be_inval)))
622 end_page_writeback(page);
623 page_cache_release(page);
624 wdata->pnfs_error = -ENOMEM;
627 /* FIXME: This should be done in bi_end_io */
628 mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
629 page->index << PAGE_CACHE_SHIFT,
632 bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
634 bl_end_io_write_zero, par);
636 wdata->pnfs_error = PTR_ERR(bio);
641 isect += PAGE_CACHE_SECTORS;
642 extent_length -= PAGE_CACHE_SECTORS;
647 bio = bl_submit_bio(WRITE, bio);
650 pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
651 for (i = pg_index; i < wdata->npages; i++) {
652 if (!extent_length) {
653 /* We've used up the previous extent */
655 bio = bl_submit_bio(WRITE, bio);
656 /* Get the next one */
657 be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg),
659 if (!be || !is_writable(be, isect)) {
660 wdata->pnfs_error = -EINVAL;
663 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
664 if (likely(!bl_push_one_short_extent(
668 wdata->pnfs_error = -ENOMEM;
672 extent_length = be->be_length -
673 (isect - be->be_f_offset);
675 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
676 ret = bl_mark_sectors_init(be->be_inval, isect,
679 dprintk("%s bl_mark_sectors_init fail %d\n",
681 wdata->pnfs_error = ret;
685 bio = bl_add_page_to_bio(bio, wdata->npages - i, WRITE,
687 bl_end_io_write, par);
689 wdata->pnfs_error = PTR_ERR(bio);
693 isect += PAGE_CACHE_SECTORS;
695 extent_length -= PAGE_CACHE_SECTORS;
698 /* Last page inside INVALID extent */
699 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
700 bio = bl_submit_bio(WRITE, bio);
701 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
702 npg_zero = npg_per_block - do_div(temp, npg_per_block);
703 if (npg_zero < npg_per_block) {
705 goto fill_invalid_ext;
710 wdata->res.count = (last_isect << SECTOR_SHIFT) - (offset);
711 if (count < wdata->res.count) {
712 wdata->res.count = count;
716 bl_submit_bio(WRITE, bio);
718 return PNFS_ATTEMPTED;
722 return PNFS_NOT_ATTEMPTED;
725 /* FIXME - range ignored */
727 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
730 struct pnfs_block_extent *be;
732 spin_lock(&bl->bl_ext_lock);
733 for (i = 0; i < EXTENT_LISTS; i++) {
734 while (!list_empty(&bl->bl_extents[i])) {
735 be = list_first_entry(&bl->bl_extents[i],
736 struct pnfs_block_extent,
738 list_del(&be->be_node);
742 spin_unlock(&bl->bl_ext_lock);
746 release_inval_marks(struct pnfs_inval_markings *marks)
748 struct pnfs_inval_tracking *pos, *temp;
749 struct pnfs_block_short_extent *se, *stemp;
751 list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
752 list_del(&pos->it_link);
756 list_for_each_entry_safe(se, stemp, &marks->im_extents, bse_node) {
757 list_del(&se->bse_node);
763 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
765 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
767 dprintk("%s enter\n", __func__);
768 release_extents(bl, NULL);
769 release_inval_marks(&bl->bl_inval);
773 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
776 struct pnfs_block_layout *bl;
778 dprintk("%s enter\n", __func__);
779 bl = kzalloc(sizeof(*bl), gfp_flags);
782 spin_lock_init(&bl->bl_ext_lock);
783 INIT_LIST_HEAD(&bl->bl_extents[0]);
784 INIT_LIST_HEAD(&bl->bl_extents[1]);
785 INIT_LIST_HEAD(&bl->bl_commit);
786 INIT_LIST_HEAD(&bl->bl_committing);
788 bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
789 BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
790 return &bl->bl_layout;
793 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
795 dprintk("%s enter\n", __func__);
799 /* We pretty much ignore lseg, and store all data layout wide, so we
800 * can correctly merge.
802 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
803 struct nfs4_layoutget_res *lgr,
806 struct pnfs_layout_segment *lseg;
809 dprintk("%s enter\n", __func__);
810 lseg = kzalloc(sizeof(*lseg), gfp_flags);
812 return ERR_PTR(-ENOMEM);
813 status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
815 /* We don't want to call the full-blown bl_free_lseg,
816 * since on error extents were not touched.
819 return ERR_PTR(status);
825 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
826 const struct nfs4_layoutcommit_args *arg)
828 dprintk("%s enter\n", __func__);
829 encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
833 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
835 struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
837 dprintk("%s enter\n", __func__);
838 clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
841 static void free_blk_mountid(struct block_mount_id *mid)
844 struct pnfs_block_dev *dev, *tmp;
846 /* No need to take bm_lock as we are last user freeing bm_devlist */
847 list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) {
848 list_del(&dev->bm_node);
849 bl_free_block_dev(dev);
855 /* This is mostly copied from the filelayout's get_device_info function.
856 * It seems much of this should be at the generic pnfs level.
858 static struct pnfs_block_dev *
859 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
860 struct nfs4_deviceid *d_id)
862 struct pnfs_device *dev;
863 struct pnfs_block_dev *rv;
866 struct page **pages = NULL;
870 * Use the session max response size as the basis for setting
871 * GETDEVICEINFO's maxcount
873 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
874 max_pages = max_resp_sz >> PAGE_SHIFT;
875 dprintk("%s max_resp_sz %u max_pages %d\n",
876 __func__, max_resp_sz, max_pages);
878 dev = kmalloc(sizeof(*dev), GFP_NOFS);
880 dprintk("%s kmalloc failed\n", __func__);
881 return ERR_PTR(-ENOMEM);
884 pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
887 return ERR_PTR(-ENOMEM);
889 for (i = 0; i < max_pages; i++) {
890 pages[i] = alloc_page(GFP_NOFS);
892 rv = ERR_PTR(-ENOMEM);
897 memcpy(&dev->dev_id, d_id, sizeof(*d_id));
898 dev->layout_type = LAYOUT_BLOCK_VOLUME;
901 dev->pglen = PAGE_SIZE * max_pages;
904 dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
905 rc = nfs4_proc_getdeviceinfo(server, dev);
906 dprintk("%s getdevice info returns %d\n", __func__, rc);
912 rv = nfs4_blk_decode_device(server, dev);
914 for (i = 0; i < max_pages; i++)
915 __free_page(pages[i]);
922 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
924 struct block_mount_id *b_mt_id = NULL;
925 struct pnfs_devicelist *dlist = NULL;
926 struct pnfs_block_dev *bdev;
927 LIST_HEAD(block_disklist);
930 dprintk("%s enter\n", __func__);
932 if (server->pnfs_blksize == 0) {
933 dprintk("%s Server did not return blksize\n", __func__);
936 b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
941 /* Initialize nfs4 block layout mount id */
942 spin_lock_init(&b_mt_id->bm_lock);
943 INIT_LIST_HEAD(&b_mt_id->bm_devlist);
945 dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
951 while (!dlist->eof) {
952 status = nfs4_proc_getdevicelist(server, fh, dlist);
955 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
956 __func__, dlist->num_devs, dlist->eof);
957 for (i = 0; i < dlist->num_devs; i++) {
958 bdev = nfs4_blk_get_deviceinfo(server, fh,
961 status = PTR_ERR(bdev);
964 spin_lock(&b_mt_id->bm_lock);
965 list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
966 spin_unlock(&b_mt_id->bm_lock);
969 dprintk("%s SUCCESS\n", __func__);
970 server->pnfs_ld_data = b_mt_id;
977 free_blk_mountid(b_mt_id);
982 bl_clear_layoutdriver(struct nfs_server *server)
984 struct block_mount_id *b_mt_id = server->pnfs_ld_data;
986 dprintk("%s enter\n", __func__);
987 free_blk_mountid(b_mt_id);
988 dprintk("%s RETURNS\n", __func__);
992 static const struct nfs_pageio_ops bl_pg_read_ops = {
993 .pg_init = pnfs_generic_pg_init_read,
994 .pg_test = pnfs_generic_pg_test,
995 .pg_doio = pnfs_generic_pg_readpages,
998 static const struct nfs_pageio_ops bl_pg_write_ops = {
999 .pg_init = pnfs_generic_pg_init_write,
1000 .pg_test = pnfs_generic_pg_test,
1001 .pg_doio = pnfs_generic_pg_writepages,
1004 static struct pnfs_layoutdriver_type blocklayout_type = {
1005 .id = LAYOUT_BLOCK_VOLUME,
1006 .name = "LAYOUT_BLOCK_VOLUME",
1007 .read_pagelist = bl_read_pagelist,
1008 .write_pagelist = bl_write_pagelist,
1009 .alloc_layout_hdr = bl_alloc_layout_hdr,
1010 .free_layout_hdr = bl_free_layout_hdr,
1011 .alloc_lseg = bl_alloc_lseg,
1012 .free_lseg = bl_free_lseg,
1013 .encode_layoutcommit = bl_encode_layoutcommit,
1014 .cleanup_layoutcommit = bl_cleanup_layoutcommit,
1015 .set_layoutdriver = bl_set_layoutdriver,
1016 .clear_layoutdriver = bl_clear_layoutdriver,
1017 .pg_read_ops = &bl_pg_read_ops,
1018 .pg_write_ops = &bl_pg_write_ops,
1021 static const struct rpc_pipe_ops bl_upcall_ops = {
1022 .upcall = rpc_pipe_generic_upcall,
1023 .downcall = bl_pipe_downcall,
1024 .destroy_msg = bl_pipe_destroy_msg,
1027 static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb,
1028 struct rpc_pipe *pipe)
1030 struct dentry *dir, *dentry;
1032 dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME);
1034 return ERR_PTR(-ENOENT);
1035 dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe);
1040 static void nfs4blocklayout_unregister_sb(struct super_block *sb,
1041 struct rpc_pipe *pipe)
1044 rpc_unlink(pipe->dentry);
1047 static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event,
1050 struct super_block *sb = ptr;
1051 struct net *net = sb->s_fs_info;
1052 struct nfs_net *nn = net_generic(net, nfs_net_id);
1053 struct dentry *dentry;
1056 if (!try_module_get(THIS_MODULE))
1059 if (nn->bl_device_pipe == NULL) {
1060 module_put(THIS_MODULE);
1065 case RPC_PIPEFS_MOUNT:
1066 dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe);
1067 if (IS_ERR(dentry)) {
1068 ret = PTR_ERR(dentry);
1071 nn->bl_device_pipe->dentry = dentry;
1073 case RPC_PIPEFS_UMOUNT:
1074 if (nn->bl_device_pipe->dentry)
1075 nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe);
1081 module_put(THIS_MODULE);
1085 static struct notifier_block nfs4blocklayout_block = {
1086 .notifier_call = rpc_pipefs_event,
1089 static struct dentry *nfs4blocklayout_register_net(struct net *net,
1090 struct rpc_pipe *pipe)
1092 struct super_block *pipefs_sb;
1093 struct dentry *dentry;
1095 pipefs_sb = rpc_get_sb_net(net);
1097 return ERR_PTR(-ENOENT);
1098 dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe);
1099 rpc_put_sb_net(net);
1103 static void nfs4blocklayout_unregister_net(struct net *net,
1104 struct rpc_pipe *pipe)
1106 struct super_block *pipefs_sb;
1108 pipefs_sb = rpc_get_sb_net(net);
1110 nfs4blocklayout_unregister_sb(pipefs_sb, pipe);
1111 rpc_put_sb_net(net);
1115 static int nfs4blocklayout_net_init(struct net *net)
1117 struct nfs_net *nn = net_generic(net, nfs_net_id);
1118 struct dentry *dentry;
1120 nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0);
1121 if (IS_ERR(nn->bl_device_pipe))
1122 return PTR_ERR(nn->bl_device_pipe);
1123 dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe);
1124 if (IS_ERR(dentry)) {
1125 rpc_destroy_pipe_data(nn->bl_device_pipe);
1126 return PTR_ERR(dentry);
1128 nn->bl_device_pipe->dentry = dentry;
1132 static void nfs4blocklayout_net_exit(struct net *net)
1134 struct nfs_net *nn = net_generic(net, nfs_net_id);
1136 nfs4blocklayout_unregister_net(net, nn->bl_device_pipe);
1137 rpc_destroy_pipe_data(nn->bl_device_pipe);
1138 nn->bl_device_pipe = NULL;
1141 static struct pernet_operations nfs4blocklayout_net_ops = {
1142 .init = nfs4blocklayout_net_init,
1143 .exit = nfs4blocklayout_net_exit,
1146 static int __init nfs4blocklayout_init(void)
1148 struct vfsmount *mnt;
1151 dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
1153 ret = pnfs_register_layoutdriver(&blocklayout_type);
1157 init_waitqueue_head(&bl_wq);
1159 mnt = rpc_get_mount();
1164 ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block);
1167 ret = register_pernet_subsys(&nfs4blocklayout_net_ops);
1174 rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1176 pnfs_unregister_layoutdriver(&blocklayout_type);
1180 static void __exit nfs4blocklayout_exit(void)
1182 dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1185 rpc_pipefs_notifier_unregister(&nfs4blocklayout_block);
1186 unregister_pernet_subsys(&nfs4blocklayout_net_ops);
1187 pnfs_unregister_layoutdriver(&blocklayout_type);
1190 MODULE_ALIAS("nfs-layouttype4-3");
1192 module_init(nfs4blocklayout_init);
1193 module_exit(nfs4blocklayout_exit);