Loading...
1/*
2 * linux/fs/nfs/blocklayout/blocklayout.c
3 *
4 * Module for the NFSv4.1 pNFS block layout driver.
5 *
6 * Copyright (c) 2006 The Regents of the University of Michigan.
7 * All rights reserved.
8 *
9 * Andy Adamson <andros@citi.umich.edu>
10 * Fred Isaman <iisaman@umich.edu>
11 *
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.
20 *
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.
31 */
32
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/prefetch.h>
39#include <linux/pagevec.h>
40
41#include "../pnfs.h"
42#include "../nfs4session.h"
43#include "../internal.h"
44#include "blocklayout.h"
45
46#define NFSDBG_FACILITY NFSDBG_PNFS_LD
47
48MODULE_LICENSE("GPL");
49MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
50MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
51
52static bool is_hole(struct pnfs_block_extent *be)
53{
54 switch (be->be_state) {
55 case PNFS_BLOCK_NONE_DATA:
56 return true;
57 case PNFS_BLOCK_INVALID_DATA:
58 return be->be_tag ? false : true;
59 default:
60 return false;
61 }
62}
63
64/* The data we are handed might be spread across several bios. We need
65 * to track when the last one is finished.
66 */
67struct parallel_io {
68 struct kref refcnt;
69 void (*pnfs_callback) (void *data);
70 void *data;
71};
72
73static inline struct parallel_io *alloc_parallel(void *data)
74{
75 struct parallel_io *rv;
76
77 rv = kmalloc(sizeof(*rv), GFP_NOFS);
78 if (rv) {
79 rv->data = data;
80 kref_init(&rv->refcnt);
81 }
82 return rv;
83}
84
85static inline void get_parallel(struct parallel_io *p)
86{
87 kref_get(&p->refcnt);
88}
89
90static void destroy_parallel(struct kref *kref)
91{
92 struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
93
94 dprintk("%s enter\n", __func__);
95 p->pnfs_callback(p->data);
96 kfree(p);
97}
98
99static inline void put_parallel(struct parallel_io *p)
100{
101 kref_put(&p->refcnt, destroy_parallel);
102}
103
104static struct bio *
105bl_submit_bio(int rw, struct bio *bio)
106{
107 if (bio) {
108 get_parallel(bio->bi_private);
109 dprintk("%s submitting %s bio %u@%llu\n", __func__,
110 rw == READ ? "read" : "write", bio->bi_iter.bi_size,
111 (unsigned long long)bio->bi_iter.bi_sector);
112 submit_bio(rw, bio);
113 }
114 return NULL;
115}
116
117static struct bio *
118bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector,
119 bio_end_io_t end_io, struct parallel_io *par)
120{
121 struct bio *bio;
122
123 npg = min(npg, BIO_MAX_PAGES);
124 bio = bio_alloc(GFP_NOIO, npg);
125 if (!bio && (current->flags & PF_MEMALLOC)) {
126 while (!bio && (npg /= 2))
127 bio = bio_alloc(GFP_NOIO, npg);
128 }
129
130 if (bio) {
131 bio->bi_iter.bi_sector = disk_sector;
132 bio->bi_bdev = bdev;
133 bio->bi_end_io = end_io;
134 bio->bi_private = par;
135 }
136 return bio;
137}
138
139static struct bio *
140do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect,
141 struct page *page, struct pnfs_block_dev_map *map,
142 struct pnfs_block_extent *be, bio_end_io_t end_io,
143 struct parallel_io *par, unsigned int offset, int *len)
144{
145 struct pnfs_block_dev *dev =
146 container_of(be->be_device, struct pnfs_block_dev, node);
147 u64 disk_addr, end;
148
149 dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__,
150 npg, rw, (unsigned long long)isect, offset, *len);
151
152 /* translate to device offset */
153 isect += be->be_v_offset;
154 isect -= be->be_f_offset;
155
156 /* translate to physical disk offset */
157 disk_addr = (u64)isect << SECTOR_SHIFT;
158 if (disk_addr < map->start || disk_addr >= map->start + map->len) {
159 if (!dev->map(dev, disk_addr, map))
160 return ERR_PTR(-EIO);
161 bio = bl_submit_bio(rw, bio);
162 }
163 disk_addr += map->disk_offset;
164 disk_addr -= map->start;
165
166 /* limit length to what the device mapping allows */
167 end = disk_addr + *len;
168 if (end >= map->start + map->len)
169 *len = map->start + map->len - disk_addr;
170
171retry:
172 if (!bio) {
173 bio = bl_alloc_init_bio(npg, map->bdev,
174 disk_addr >> SECTOR_SHIFT, end_io, par);
175 if (!bio)
176 return ERR_PTR(-ENOMEM);
177 }
178 if (bio_add_page(bio, page, *len, offset) < *len) {
179 bio = bl_submit_bio(rw, bio);
180 goto retry;
181 }
182 return bio;
183}
184
185static void bl_end_io_read(struct bio *bio)
186{
187 struct parallel_io *par = bio->bi_private;
188
189 if (bio->bi_error) {
190 struct nfs_pgio_header *header = par->data;
191
192 if (!header->pnfs_error)
193 header->pnfs_error = -EIO;
194 pnfs_set_lo_fail(header->lseg);
195 }
196
197 bio_put(bio);
198 put_parallel(par);
199}
200
201static void bl_read_cleanup(struct work_struct *work)
202{
203 struct rpc_task *task;
204 struct nfs_pgio_header *hdr;
205 dprintk("%s enter\n", __func__);
206 task = container_of(work, struct rpc_task, u.tk_work);
207 hdr = container_of(task, struct nfs_pgio_header, task);
208 pnfs_ld_read_done(hdr);
209}
210
211static void
212bl_end_par_io_read(void *data)
213{
214 struct nfs_pgio_header *hdr = data;
215
216 hdr->task.tk_status = hdr->pnfs_error;
217 INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup);
218 schedule_work(&hdr->task.u.tk_work);
219}
220
221static enum pnfs_try_status
222bl_read_pagelist(struct nfs_pgio_header *header)
223{
224 struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
225 struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
226 struct bio *bio = NULL;
227 struct pnfs_block_extent be;
228 sector_t isect, extent_length = 0;
229 struct parallel_io *par;
230 loff_t f_offset = header->args.offset;
231 size_t bytes_left = header->args.count;
232 unsigned int pg_offset = header->args.pgbase, pg_len;
233 struct page **pages = header->args.pages;
234 int pg_index = header->args.pgbase >> PAGE_SHIFT;
235 const bool is_dio = (header->dreq != NULL);
236 struct blk_plug plug;
237 int i;
238
239 dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
240 header->page_array.npages, f_offset,
241 (unsigned int)header->args.count);
242
243 par = alloc_parallel(header);
244 if (!par)
245 return PNFS_NOT_ATTEMPTED;
246 par->pnfs_callback = bl_end_par_io_read;
247
248 blk_start_plug(&plug);
249
250 isect = (sector_t) (f_offset >> SECTOR_SHIFT);
251 /* Code assumes extents are page-aligned */
252 for (i = pg_index; i < header->page_array.npages; i++) {
253 if (extent_length <= 0) {
254 /* We've used up the previous extent */
255 bio = bl_submit_bio(READ, bio);
256
257 /* Get the next one */
258 if (!ext_tree_lookup(bl, isect, &be, false)) {
259 header->pnfs_error = -EIO;
260 goto out;
261 }
262 extent_length = be.be_length - (isect - be.be_f_offset);
263 }
264
265 if (is_dio) {
266 if (pg_offset + bytes_left > PAGE_SIZE)
267 pg_len = PAGE_SIZE - pg_offset;
268 else
269 pg_len = bytes_left;
270 } else {
271 BUG_ON(pg_offset != 0);
272 pg_len = PAGE_SIZE;
273 }
274
275 if (is_hole(&be)) {
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], pg_offset, pg_len);
280
281 /* invalidate map */
282 map.start = NFS4_MAX_UINT64;
283 } else {
284 bio = do_add_page_to_bio(bio,
285 header->page_array.npages - i,
286 READ,
287 isect, pages[i], &map, &be,
288 bl_end_io_read, par,
289 pg_offset, &pg_len);
290 if (IS_ERR(bio)) {
291 header->pnfs_error = PTR_ERR(bio);
292 bio = NULL;
293 goto out;
294 }
295 }
296 isect += (pg_len >> SECTOR_SHIFT);
297 extent_length -= (pg_len >> SECTOR_SHIFT);
298 f_offset += pg_len;
299 bytes_left -= pg_len;
300 pg_offset = 0;
301 }
302 if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
303 header->res.eof = 1;
304 header->res.count = header->inode->i_size - header->args.offset;
305 } else {
306 header->res.count = (isect << SECTOR_SHIFT) - header->args.offset;
307 }
308out:
309 bl_submit_bio(READ, bio);
310 blk_finish_plug(&plug);
311 put_parallel(par);
312 return PNFS_ATTEMPTED;
313}
314
315static void bl_end_io_write(struct bio *bio)
316{
317 struct parallel_io *par = bio->bi_private;
318 struct nfs_pgio_header *header = par->data;
319
320 if (bio->bi_error) {
321 if (!header->pnfs_error)
322 header->pnfs_error = -EIO;
323 pnfs_set_lo_fail(header->lseg);
324 }
325 bio_put(bio);
326 put_parallel(par);
327}
328
329/* Function scheduled for call during bl_end_par_io_write,
330 * it marks sectors as written and extends the commitlist.
331 */
332static void bl_write_cleanup(struct work_struct *work)
333{
334 struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work);
335 struct nfs_pgio_header *hdr =
336 container_of(task, struct nfs_pgio_header, task);
337
338 dprintk("%s enter\n", __func__);
339
340 if (likely(!hdr->pnfs_error)) {
341 struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg);
342 u64 start = hdr->args.offset & (loff_t)PAGE_MASK;
343 u64 end = (hdr->args.offset + hdr->args.count +
344 PAGE_SIZE - 1) & (loff_t)PAGE_MASK;
345
346 ext_tree_mark_written(bl, start >> SECTOR_SHIFT,
347 (end - start) >> SECTOR_SHIFT);
348 }
349
350 pnfs_ld_write_done(hdr);
351}
352
353/* Called when last of bios associated with a bl_write_pagelist call finishes */
354static void bl_end_par_io_write(void *data)
355{
356 struct nfs_pgio_header *hdr = data;
357
358 hdr->task.tk_status = hdr->pnfs_error;
359 hdr->verf.committed = NFS_FILE_SYNC;
360 INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup);
361 schedule_work(&hdr->task.u.tk_work);
362}
363
364static enum pnfs_try_status
365bl_write_pagelist(struct nfs_pgio_header *header, int sync)
366{
367 struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg);
368 struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 };
369 struct bio *bio = NULL;
370 struct pnfs_block_extent be;
371 sector_t isect, extent_length = 0;
372 struct parallel_io *par = NULL;
373 loff_t offset = header->args.offset;
374 size_t count = header->args.count;
375 struct page **pages = header->args.pages;
376 int pg_index = header->args.pgbase >> PAGE_SHIFT;
377 unsigned int pg_len;
378 struct blk_plug plug;
379 int i;
380
381 dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
382
383 /* At this point, header->page_aray is a (sequential) list of nfs_pages.
384 * We want to write each, and if there is an error set pnfs_error
385 * to have it redone using nfs.
386 */
387 par = alloc_parallel(header);
388 if (!par)
389 return PNFS_NOT_ATTEMPTED;
390 par->pnfs_callback = bl_end_par_io_write;
391
392 blk_start_plug(&plug);
393
394 /* we always write out the whole page */
395 offset = offset & (loff_t)PAGE_MASK;
396 isect = offset >> SECTOR_SHIFT;
397
398 for (i = pg_index; i < header->page_array.npages; i++) {
399 if (extent_length <= 0) {
400 /* We've used up the previous extent */
401 bio = bl_submit_bio(WRITE, bio);
402 /* Get the next one */
403 if (!ext_tree_lookup(bl, isect, &be, true)) {
404 header->pnfs_error = -EINVAL;
405 goto out;
406 }
407
408 extent_length = be.be_length - (isect - be.be_f_offset);
409 }
410
411 pg_len = PAGE_SIZE;
412 bio = do_add_page_to_bio(bio, header->page_array.npages - i,
413 WRITE, isect, pages[i], &map, &be,
414 bl_end_io_write, par,
415 0, &pg_len);
416 if (IS_ERR(bio)) {
417 header->pnfs_error = PTR_ERR(bio);
418 bio = NULL;
419 goto out;
420 }
421
422 offset += pg_len;
423 count -= pg_len;
424 isect += (pg_len >> SECTOR_SHIFT);
425 extent_length -= (pg_len >> SECTOR_SHIFT);
426 }
427
428 header->res.count = header->args.count;
429out:
430 bl_submit_bio(WRITE, bio);
431 blk_finish_plug(&plug);
432 put_parallel(par);
433 return PNFS_ATTEMPTED;
434}
435
436static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
437{
438 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
439 int err;
440
441 dprintk("%s enter\n", __func__);
442
443 err = ext_tree_remove(bl, true, 0, LLONG_MAX);
444 WARN_ON(err);
445
446 kfree(bl);
447}
448
449static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode,
450 gfp_t gfp_flags, bool is_scsi_layout)
451{
452 struct pnfs_block_layout *bl;
453
454 dprintk("%s enter\n", __func__);
455 bl = kzalloc(sizeof(*bl), gfp_flags);
456 if (!bl)
457 return NULL;
458
459 bl->bl_ext_rw = RB_ROOT;
460 bl->bl_ext_ro = RB_ROOT;
461 spin_lock_init(&bl->bl_ext_lock);
462
463 bl->bl_scsi_layout = is_scsi_layout;
464 return &bl->bl_layout;
465}
466
467static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
468 gfp_t gfp_flags)
469{
470 return __bl_alloc_layout_hdr(inode, gfp_flags, false);
471}
472
473static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode,
474 gfp_t gfp_flags)
475{
476 return __bl_alloc_layout_hdr(inode, gfp_flags, true);
477}
478
479static void bl_free_lseg(struct pnfs_layout_segment *lseg)
480{
481 dprintk("%s enter\n", __func__);
482 kfree(lseg);
483}
484
485/* Tracks info needed to ensure extents in layout obey constraints of spec */
486struct layout_verification {
487 u32 mode; /* R or RW */
488 u64 start; /* Expected start of next non-COW extent */
489 u64 inval; /* Start of INVAL coverage */
490 u64 cowread; /* End of COW read coverage */
491};
492
493/* Verify the extent meets the layout requirements of the pnfs-block draft,
494 * section 2.3.1.
495 */
496static int verify_extent(struct pnfs_block_extent *be,
497 struct layout_verification *lv)
498{
499 if (lv->mode == IOMODE_READ) {
500 if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
501 be->be_state == PNFS_BLOCK_INVALID_DATA)
502 return -EIO;
503 if (be->be_f_offset != lv->start)
504 return -EIO;
505 lv->start += be->be_length;
506 return 0;
507 }
508 /* lv->mode == IOMODE_RW */
509 if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
510 if (be->be_f_offset != lv->start)
511 return -EIO;
512 if (lv->cowread > lv->start)
513 return -EIO;
514 lv->start += be->be_length;
515 lv->inval = lv->start;
516 return 0;
517 } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
518 if (be->be_f_offset != lv->start)
519 return -EIO;
520 lv->start += be->be_length;
521 return 0;
522 } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
523 if (be->be_f_offset > lv->start)
524 return -EIO;
525 if (be->be_f_offset < lv->inval)
526 return -EIO;
527 if (be->be_f_offset < lv->cowread)
528 return -EIO;
529 /* It looks like you might want to min this with lv->start,
530 * but you really don't.
531 */
532 lv->inval = lv->inval + be->be_length;
533 lv->cowread = be->be_f_offset + be->be_length;
534 return 0;
535 } else
536 return -EIO;
537}
538
539static int decode_sector_number(__be32 **rp, sector_t *sp)
540{
541 uint64_t s;
542
543 *rp = xdr_decode_hyper(*rp, &s);
544 if (s & 0x1ff) {
545 printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
546 return -1;
547 }
548 *sp = s >> SECTOR_SHIFT;
549 return 0;
550}
551
552static int
553bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo,
554 struct layout_verification *lv, struct list_head *extents,
555 gfp_t gfp_mask)
556{
557 struct pnfs_block_extent *be;
558 struct nfs4_deviceid id;
559 int error;
560 __be32 *p;
561
562 p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE);
563 if (!p)
564 return -EIO;
565
566 be = kzalloc(sizeof(*be), GFP_NOFS);
567 if (!be)
568 return -ENOMEM;
569
570 memcpy(&id, p, NFS4_DEVICEID4_SIZE);
571 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
572
573 error = -EIO;
574 be->be_device = nfs4_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id,
575 lo->plh_lc_cred, gfp_mask);
576 if (!be->be_device)
577 goto out_free_be;
578
579 /*
580 * The next three values are read in as bytes, but stored in the
581 * extent structure in 512-byte granularity.
582 */
583 if (decode_sector_number(&p, &be->be_f_offset) < 0)
584 goto out_put_deviceid;
585 if (decode_sector_number(&p, &be->be_length) < 0)
586 goto out_put_deviceid;
587 if (decode_sector_number(&p, &be->be_v_offset) < 0)
588 goto out_put_deviceid;
589 be->be_state = be32_to_cpup(p++);
590
591 error = verify_extent(be, lv);
592 if (error) {
593 dprintk("%s: extent verification failed\n", __func__);
594 goto out_put_deviceid;
595 }
596
597 list_add_tail(&be->be_list, extents);
598 return 0;
599
600out_put_deviceid:
601 nfs4_put_deviceid_node(be->be_device);
602out_free_be:
603 kfree(be);
604 return error;
605}
606
607static struct pnfs_layout_segment *
608bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr,
609 gfp_t gfp_mask)
610{
611 struct layout_verification lv = {
612 .mode = lgr->range.iomode,
613 .start = lgr->range.offset >> SECTOR_SHIFT,
614 .inval = lgr->range.offset >> SECTOR_SHIFT,
615 .cowread = lgr->range.offset >> SECTOR_SHIFT,
616 };
617 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
618 struct pnfs_layout_segment *lseg;
619 struct xdr_buf buf;
620 struct xdr_stream xdr;
621 struct page *scratch;
622 int status, i;
623 uint32_t count;
624 __be32 *p;
625 LIST_HEAD(extents);
626
627 dprintk("---> %s\n", __func__);
628
629 lseg = kzalloc(sizeof(*lseg), gfp_mask);
630 if (!lseg)
631 return ERR_PTR(-ENOMEM);
632
633 status = -ENOMEM;
634 scratch = alloc_page(gfp_mask);
635 if (!scratch)
636 goto out;
637
638 xdr_init_decode_pages(&xdr, &buf,
639 lgr->layoutp->pages, lgr->layoutp->len);
640 xdr_set_scratch_buffer(&xdr, page_address(scratch), PAGE_SIZE);
641
642 status = -EIO;
643 p = xdr_inline_decode(&xdr, 4);
644 if (unlikely(!p))
645 goto out_free_scratch;
646
647 count = be32_to_cpup(p++);
648 dprintk("%s: number of extents %d\n", __func__, count);
649
650 /*
651 * Decode individual extents, putting them in temporary staging area
652 * until whole layout is decoded to make error recovery easier.
653 */
654 for (i = 0; i < count; i++) {
655 status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask);
656 if (status)
657 goto process_extents;
658 }
659
660 if (lgr->range.offset + lgr->range.length !=
661 lv.start << SECTOR_SHIFT) {
662 dprintk("%s Final length mismatch\n", __func__);
663 status = -EIO;
664 goto process_extents;
665 }
666
667 if (lv.start < lv.cowread) {
668 dprintk("%s Final uncovered COW extent\n", __func__);
669 status = -EIO;
670 }
671
672process_extents:
673 while (!list_empty(&extents)) {
674 struct pnfs_block_extent *be =
675 list_first_entry(&extents, struct pnfs_block_extent,
676 be_list);
677 list_del(&be->be_list);
678
679 if (!status)
680 status = ext_tree_insert(bl, be);
681
682 if (status) {
683 nfs4_put_deviceid_node(be->be_device);
684 kfree(be);
685 }
686 }
687
688out_free_scratch:
689 __free_page(scratch);
690out:
691 dprintk("%s returns %d\n", __func__, status);
692 if (status) {
693 kfree(lseg);
694 return ERR_PTR(status);
695 }
696 return lseg;
697}
698
699static void
700bl_return_range(struct pnfs_layout_hdr *lo,
701 struct pnfs_layout_range *range)
702{
703 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
704 sector_t offset = range->offset >> SECTOR_SHIFT, end;
705
706 if (range->offset % 8) {
707 dprintk("%s: offset %lld not block size aligned\n",
708 __func__, range->offset);
709 return;
710 }
711
712 if (range->length != NFS4_MAX_UINT64) {
713 if (range->length % 8) {
714 dprintk("%s: length %lld not block size aligned\n",
715 __func__, range->length);
716 return;
717 }
718
719 end = offset + (range->length >> SECTOR_SHIFT);
720 } else {
721 end = round_down(NFS4_MAX_UINT64, PAGE_SIZE);
722 }
723
724 ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end);
725}
726
727static int
728bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg)
729{
730 return ext_tree_prepare_commit(arg);
731}
732
733static void
734bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
735{
736 ext_tree_mark_committed(&lcdata->args, lcdata->res.status);
737}
738
739static int
740bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
741{
742 dprintk("%s enter\n", __func__);
743
744 if (server->pnfs_blksize == 0) {
745 dprintk("%s Server did not return blksize\n", __func__);
746 return -EINVAL;
747 }
748 if (server->pnfs_blksize > PAGE_SIZE) {
749 printk(KERN_ERR "%s: pNFS blksize %d not supported.\n",
750 __func__, server->pnfs_blksize);
751 return -EINVAL;
752 }
753
754 return 0;
755}
756
757static bool
758is_aligned_req(struct nfs_pageio_descriptor *pgio,
759 struct nfs_page *req, unsigned int alignment, bool is_write)
760{
761 /*
762 * Always accept buffered writes, higher layers take care of the
763 * right alignment.
764 */
765 if (pgio->pg_dreq == NULL)
766 return true;
767
768 if (!IS_ALIGNED(req->wb_offset, alignment))
769 return false;
770
771 if (IS_ALIGNED(req->wb_bytes, alignment))
772 return true;
773
774 if (is_write &&
775 (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) {
776 /*
777 * If the write goes up to the inode size, just write
778 * the full page. Data past the inode size is
779 * guaranteed to be zeroed by the higher level client
780 * code, and this behaviour is mandated by RFC 5663
781 * section 2.3.2.
782 */
783 return true;
784 }
785
786 return false;
787}
788
789static void
790bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
791{
792 if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) {
793 nfs_pageio_reset_read_mds(pgio);
794 return;
795 }
796
797 pnfs_generic_pg_init_read(pgio, req);
798}
799
800/*
801 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
802 * of bytes (maximum @req->wb_bytes) that can be coalesced.
803 */
804static size_t
805bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
806 struct nfs_page *req)
807{
808 if (!is_aligned_req(pgio, req, SECTOR_SIZE, false))
809 return 0;
810 return pnfs_generic_pg_test(pgio, prev, req);
811}
812
813/*
814 * Return the number of contiguous bytes for a given inode
815 * starting at page frame idx.
816 */
817static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx)
818{
819 struct address_space *mapping = inode->i_mapping;
820 pgoff_t end;
821
822 /* Optimize common case that writes from 0 to end of file */
823 end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
824 if (end != inode->i_mapping->nrpages) {
825 rcu_read_lock();
826 end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX);
827 rcu_read_unlock();
828 }
829
830 if (!end)
831 return i_size_read(inode) - (idx << PAGE_SHIFT);
832 else
833 return (end - idx) << PAGE_SHIFT;
834}
835
836static void
837bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
838{
839 u64 wb_size;
840
841 if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) {
842 nfs_pageio_reset_write_mds(pgio);
843 return;
844 }
845
846 if (pgio->pg_dreq == NULL)
847 wb_size = pnfs_num_cont_bytes(pgio->pg_inode,
848 req->wb_index);
849 else
850 wb_size = nfs_dreq_bytes_left(pgio->pg_dreq);
851
852 pnfs_generic_pg_init_write(pgio, req, wb_size);
853}
854
855/*
856 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
857 * of bytes (maximum @req->wb_bytes) that can be coalesced.
858 */
859static size_t
860bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
861 struct nfs_page *req)
862{
863 if (!is_aligned_req(pgio, req, PAGE_SIZE, true))
864 return 0;
865 return pnfs_generic_pg_test(pgio, prev, req);
866}
867
868static const struct nfs_pageio_ops bl_pg_read_ops = {
869 .pg_init = bl_pg_init_read,
870 .pg_test = bl_pg_test_read,
871 .pg_doio = pnfs_generic_pg_readpages,
872 .pg_cleanup = pnfs_generic_pg_cleanup,
873};
874
875static const struct nfs_pageio_ops bl_pg_write_ops = {
876 .pg_init = bl_pg_init_write,
877 .pg_test = bl_pg_test_write,
878 .pg_doio = pnfs_generic_pg_writepages,
879 .pg_cleanup = pnfs_generic_pg_cleanup,
880};
881
882static struct pnfs_layoutdriver_type blocklayout_type = {
883 .id = LAYOUT_BLOCK_VOLUME,
884 .name = "LAYOUT_BLOCK_VOLUME",
885 .owner = THIS_MODULE,
886 .flags = PNFS_LAYOUTRET_ON_SETATTR |
887 PNFS_READ_WHOLE_PAGE,
888 .read_pagelist = bl_read_pagelist,
889 .write_pagelist = bl_write_pagelist,
890 .alloc_layout_hdr = bl_alloc_layout_hdr,
891 .free_layout_hdr = bl_free_layout_hdr,
892 .alloc_lseg = bl_alloc_lseg,
893 .free_lseg = bl_free_lseg,
894 .return_range = bl_return_range,
895 .prepare_layoutcommit = bl_prepare_layoutcommit,
896 .cleanup_layoutcommit = bl_cleanup_layoutcommit,
897 .set_layoutdriver = bl_set_layoutdriver,
898 .alloc_deviceid_node = bl_alloc_deviceid_node,
899 .free_deviceid_node = bl_free_deviceid_node,
900 .pg_read_ops = &bl_pg_read_ops,
901 .pg_write_ops = &bl_pg_write_ops,
902 .sync = pnfs_generic_sync,
903};
904
905static struct pnfs_layoutdriver_type scsilayout_type = {
906 .id = LAYOUT_SCSI,
907 .name = "LAYOUT_SCSI",
908 .owner = THIS_MODULE,
909 .flags = PNFS_LAYOUTRET_ON_SETATTR |
910 PNFS_READ_WHOLE_PAGE,
911 .read_pagelist = bl_read_pagelist,
912 .write_pagelist = bl_write_pagelist,
913 .alloc_layout_hdr = sl_alloc_layout_hdr,
914 .free_layout_hdr = bl_free_layout_hdr,
915 .alloc_lseg = bl_alloc_lseg,
916 .free_lseg = bl_free_lseg,
917 .return_range = bl_return_range,
918 .prepare_layoutcommit = bl_prepare_layoutcommit,
919 .cleanup_layoutcommit = bl_cleanup_layoutcommit,
920 .set_layoutdriver = bl_set_layoutdriver,
921 .alloc_deviceid_node = bl_alloc_deviceid_node,
922 .free_deviceid_node = bl_free_deviceid_node,
923 .pg_read_ops = &bl_pg_read_ops,
924 .pg_write_ops = &bl_pg_write_ops,
925 .sync = pnfs_generic_sync,
926};
927
928
929static int __init nfs4blocklayout_init(void)
930{
931 int ret;
932
933 dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
934
935 ret = bl_init_pipefs();
936 if (ret)
937 goto out;
938
939 ret = pnfs_register_layoutdriver(&blocklayout_type);
940 if (ret)
941 goto out_cleanup_pipe;
942
943 ret = pnfs_register_layoutdriver(&scsilayout_type);
944 if (ret)
945 goto out_unregister_block;
946 return 0;
947
948out_unregister_block:
949 pnfs_unregister_layoutdriver(&blocklayout_type);
950out_cleanup_pipe:
951 bl_cleanup_pipefs();
952out:
953 return ret;
954}
955
956static void __exit nfs4blocklayout_exit(void)
957{
958 dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
959 __func__);
960
961 pnfs_unregister_layoutdriver(&scsilayout_type);
962 pnfs_unregister_layoutdriver(&blocklayout_type);
963 bl_cleanup_pipefs();
964}
965
966MODULE_ALIAS("nfs-layouttype4-3");
967
968module_init(nfs4blocklayout_init);
969module_exit(nfs4blocklayout_exit);
1/*
2 * linux/fs/nfs/blocklayout/blocklayout.c
3 *
4 * Module for the NFSv4.1 pNFS block layout driver.
5 *
6 * Copyright (c) 2006 The Regents of the University of Michigan.
7 * All rights reserved.
8 *
9 * Andy Adamson <andros@citi.umich.edu>
10 * Fred Isaman <iisaman@umich.edu>
11 *
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.
20 *
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.
31 */
32
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
41#include "blocklayout.h"
42
43#define NFSDBG_FACILITY NFSDBG_PNFS_LD
44
45MODULE_LICENSE("GPL");
46MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
47MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
48
49struct dentry *bl_device_pipe;
50wait_queue_head_t bl_wq;
51
52static void print_page(struct page *page)
53{
54 dprintk("PRINTPAGE page %p\n", page);
55 dprintk(" PagePrivate %d\n", PagePrivate(page));
56 dprintk(" PageUptodate %d\n", PageUptodate(page));
57 dprintk(" PageError %d\n", PageError(page));
58 dprintk(" PageDirty %d\n", PageDirty(page));
59 dprintk(" PageReferenced %d\n", PageReferenced(page));
60 dprintk(" PageLocked %d\n", PageLocked(page));
61 dprintk(" PageWriteback %d\n", PageWriteback(page));
62 dprintk(" PageMappedToDisk %d\n", PageMappedToDisk(page));
63 dprintk("\n");
64}
65
66/* Given the be associated with isect, determine if page data needs to be
67 * initialized.
68 */
69static int is_hole(struct pnfs_block_extent *be, sector_t isect)
70{
71 if (be->be_state == PNFS_BLOCK_NONE_DATA)
72 return 1;
73 else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
74 return 0;
75 else
76 return !bl_is_sector_init(be->be_inval, isect);
77}
78
79/* Given the be associated with isect, determine if page data can be
80 * written to disk.
81 */
82static int is_writable(struct pnfs_block_extent *be, sector_t isect)
83{
84 return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
85 be->be_state == PNFS_BLOCK_INVALID_DATA);
86}
87
88/* The data we are handed might be spread across several bios. We need
89 * to track when the last one is finished.
90 */
91struct parallel_io {
92 struct kref refcnt;
93 struct rpc_call_ops call_ops;
94 void (*pnfs_callback) (void *data);
95 void *data;
96};
97
98static inline struct parallel_io *alloc_parallel(void *data)
99{
100 struct parallel_io *rv;
101
102 rv = kmalloc(sizeof(*rv), GFP_NOFS);
103 if (rv) {
104 rv->data = data;
105 kref_init(&rv->refcnt);
106 }
107 return rv;
108}
109
110static inline void get_parallel(struct parallel_io *p)
111{
112 kref_get(&p->refcnt);
113}
114
115static void destroy_parallel(struct kref *kref)
116{
117 struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
118
119 dprintk("%s enter\n", __func__);
120 p->pnfs_callback(p->data);
121 kfree(p);
122}
123
124static inline void put_parallel(struct parallel_io *p)
125{
126 kref_put(&p->refcnt, destroy_parallel);
127}
128
129static struct bio *
130bl_submit_bio(int rw, struct bio *bio)
131{
132 if (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);
137 submit_bio(rw, bio);
138 }
139 return NULL;
140}
141
142static 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)
146{
147 struct bio *bio;
148
149 bio = bio_alloc(GFP_NOIO, npg);
150 if (!bio)
151 return NULL;
152
153 bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
154 bio->bi_bdev = be->be_mdev;
155 bio->bi_end_io = end_io;
156 bio->bi_private = par;
157 return bio;
158}
159
160static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
161 sector_t isect, struct page *page,
162 struct pnfs_block_extent *be,
163 void (*end_io)(struct bio *, int err),
164 struct parallel_io *par)
165{
166retry:
167 if (!bio) {
168 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
169 if (!bio)
170 return ERR_PTR(-ENOMEM);
171 }
172 if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
173 bio = bl_submit_bio(rw, bio);
174 goto retry;
175 }
176 return bio;
177}
178
179static void bl_set_lo_fail(struct pnfs_layout_segment *lseg)
180{
181 if (lseg->pls_range.iomode == IOMODE_RW) {
182 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
183 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
184 } else {
185 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
186 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
187 }
188}
189
190/* This is basically copied from mpage_end_io_read */
191static void bl_end_io_read(struct bio *bio, int err)
192{
193 struct parallel_io *par = bio->bi_private;
194 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
195 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
196 struct nfs_read_data *rdata = (struct nfs_read_data *)par->data;
197
198 do {
199 struct page *page = bvec->bv_page;
200
201 if (--bvec >= bio->bi_io_vec)
202 prefetchw(&bvec->bv_page->flags);
203 if (uptodate)
204 SetPageUptodate(page);
205 } while (bvec >= bio->bi_io_vec);
206 if (!uptodate) {
207 if (!rdata->pnfs_error)
208 rdata->pnfs_error = -EIO;
209 bl_set_lo_fail(rdata->lseg);
210 }
211 bio_put(bio);
212 put_parallel(par);
213}
214
215static void bl_read_cleanup(struct work_struct *work)
216{
217 struct rpc_task *task;
218 struct nfs_read_data *rdata;
219 dprintk("%s enter\n", __func__);
220 task = container_of(work, struct rpc_task, u.tk_work);
221 rdata = container_of(task, struct nfs_read_data, task);
222 pnfs_ld_read_done(rdata);
223}
224
225static void
226bl_end_par_io_read(void *data)
227{
228 struct nfs_read_data *rdata = data;
229
230 INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
231 schedule_work(&rdata->task.u.tk_work);
232}
233
234/* We don't want normal .rpc_call_done callback used, so we replace it
235 * with this stub.
236 */
237static void bl_rpc_do_nothing(struct rpc_task *task, void *calldata)
238{
239 return;
240}
241
242static enum pnfs_try_status
243bl_read_pagelist(struct nfs_read_data *rdata)
244{
245 int i, hole;
246 struct bio *bio = NULL;
247 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
248 sector_t isect, extent_length = 0;
249 struct parallel_io *par;
250 loff_t f_offset = rdata->args.offset;
251 size_t count = rdata->args.count;
252 struct page **pages = rdata->args.pages;
253 int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
254
255 dprintk("%s enter nr_pages %u offset %lld count %Zd\n", __func__,
256 rdata->npages, f_offset, count);
257
258 par = alloc_parallel(rdata);
259 if (!par)
260 goto use_mds;
261 par->call_ops = *rdata->mds_ops;
262 par->call_ops.rpc_call_done = bl_rpc_do_nothing;
263 par->pnfs_callback = bl_end_par_io_read;
264 /* At this point, we can no longer jump to use_mds */
265
266 isect = (sector_t) (f_offset >> SECTOR_SHIFT);
267 /* Code assumes extents are page-aligned */
268 for (i = pg_index; i < rdata->npages; i++) {
269 if (!extent_length) {
270 /* We've used up the previous extent */
271 bl_put_extent(be);
272 bl_put_extent(cow_read);
273 bio = bl_submit_bio(READ, bio);
274 /* Get the next one */
275 be = bl_find_get_extent(BLK_LSEG2EXT(rdata->lseg),
276 isect, &cow_read);
277 if (!be) {
278 rdata->pnfs_error = -EIO;
279 goto out;
280 }
281 extent_length = be->be_length -
282 (isect - be->be_f_offset);
283 if (cow_read) {
284 sector_t cow_length = cow_read->be_length -
285 (isect - cow_read->be_f_offset);
286 extent_length = min(extent_length, cow_length);
287 }
288 }
289 hole = is_hole(be, isect);
290 if (hole && !cow_read) {
291 bio = bl_submit_bio(READ, bio);
292 /* Fill hole w/ zeroes w/o accessing device */
293 dprintk("%s Zeroing page for hole\n", __func__);
294 zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE);
295 print_page(pages[i]);
296 SetPageUptodate(pages[i]);
297 } else {
298 struct pnfs_block_extent *be_read;
299
300 be_read = (hole && cow_read) ? cow_read : be;
301 bio = bl_add_page_to_bio(bio, rdata->npages - i, READ,
302 isect, pages[i], be_read,
303 bl_end_io_read, par);
304 if (IS_ERR(bio)) {
305 rdata->pnfs_error = PTR_ERR(bio);
306 goto out;
307 }
308 }
309 isect += PAGE_CACHE_SECTORS;
310 extent_length -= PAGE_CACHE_SECTORS;
311 }
312 if ((isect << SECTOR_SHIFT) >= rdata->inode->i_size) {
313 rdata->res.eof = 1;
314 rdata->res.count = rdata->inode->i_size - f_offset;
315 } else {
316 rdata->res.count = (isect << SECTOR_SHIFT) - f_offset;
317 }
318out:
319 bl_put_extent(be);
320 bl_put_extent(cow_read);
321 bl_submit_bio(READ, bio);
322 put_parallel(par);
323 return PNFS_ATTEMPTED;
324
325 use_mds:
326 dprintk("Giving up and using normal NFS\n");
327 return PNFS_NOT_ATTEMPTED;
328}
329
330static void mark_extents_written(struct pnfs_block_layout *bl,
331 __u64 offset, __u32 count)
332{
333 sector_t isect, end;
334 struct pnfs_block_extent *be;
335
336 dprintk("%s(%llu, %u)\n", __func__, offset, count);
337 if (count == 0)
338 return;
339 isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
340 end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
341 end >>= SECTOR_SHIFT;
342 while (isect < end) {
343 sector_t len;
344 be = bl_find_get_extent(bl, isect, NULL);
345 BUG_ON(!be); /* FIXME */
346 len = min(end, be->be_f_offset + be->be_length) - isect;
347 if (be->be_state == PNFS_BLOCK_INVALID_DATA)
348 bl_mark_for_commit(be, isect, len); /* What if fails? */
349 isect += len;
350 bl_put_extent(be);
351 }
352}
353
354static void bl_end_io_write_zero(struct bio *bio, int err)
355{
356 struct parallel_io *par = bio->bi_private;
357 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
358 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
359 struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
360
361 do {
362 struct page *page = bvec->bv_page;
363
364 if (--bvec >= bio->bi_io_vec)
365 prefetchw(&bvec->bv_page->flags);
366 /* This is the zeroing page we added */
367 end_page_writeback(page);
368 page_cache_release(page);
369 } while (bvec >= bio->bi_io_vec);
370 if (!uptodate) {
371 if (!wdata->pnfs_error)
372 wdata->pnfs_error = -EIO;
373 bl_set_lo_fail(wdata->lseg);
374 }
375 bio_put(bio);
376 put_parallel(par);
377}
378
379/* This is basically copied from mpage_end_io_read */
380static void bl_end_io_write(struct bio *bio, int err)
381{
382 struct parallel_io *par = bio->bi_private;
383 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
384 struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
385
386 if (!uptodate) {
387 if (!wdata->pnfs_error)
388 wdata->pnfs_error = -EIO;
389 bl_set_lo_fail(wdata->lseg);
390 }
391 bio_put(bio);
392 put_parallel(par);
393}
394
395/* Function scheduled for call during bl_end_par_io_write,
396 * it marks sectors as written and extends the commitlist.
397 */
398static void bl_write_cleanup(struct work_struct *work)
399{
400 struct rpc_task *task;
401 struct nfs_write_data *wdata;
402 dprintk("%s enter\n", __func__);
403 task = container_of(work, struct rpc_task, u.tk_work);
404 wdata = container_of(task, struct nfs_write_data, task);
405 if (!wdata->pnfs_error) {
406 /* Marks for LAYOUTCOMMIT */
407 mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
408 wdata->args.offset, wdata->args.count);
409 }
410 pnfs_ld_write_done(wdata);
411}
412
413/* Called when last of bios associated with a bl_write_pagelist call finishes */
414static void bl_end_par_io_write(void *data)
415{
416 struct nfs_write_data *wdata = data;
417
418 wdata->task.tk_status = 0;
419 wdata->verf.committed = NFS_FILE_SYNC;
420 INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
421 schedule_work(&wdata->task.u.tk_work);
422}
423
424/* FIXME STUB - mark intersection of layout and page as bad, so is not
425 * used again.
426 */
427static void mark_bad_read(void)
428{
429 return;
430}
431
432/*
433 * map_block: map a requested I/0 block (isect) into an offset in the LVM
434 * block_device
435 */
436static void
437map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
438{
439 dprintk("%s enter be=%p\n", __func__, be);
440
441 set_buffer_mapped(bh);
442 bh->b_bdev = be->be_mdev;
443 bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
444 (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
445
446 dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
447 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
448 bh->b_size);
449 return;
450}
451
452/* Given an unmapped page, zero it or read in page for COW, page is locked
453 * by caller.
454 */
455static int
456init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
457{
458 struct buffer_head *bh = NULL;
459 int ret = 0;
460 sector_t isect;
461
462 dprintk("%s enter, %p\n", __func__, page);
463 BUG_ON(PageUptodate(page));
464 if (!cow_read) {
465 zero_user_segment(page, 0, PAGE_SIZE);
466 SetPageUptodate(page);
467 goto cleanup;
468 }
469
470 bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
471 if (!bh) {
472 ret = -ENOMEM;
473 goto cleanup;
474 }
475
476 isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
477 map_block(bh, isect, cow_read);
478 if (!bh_uptodate_or_lock(bh))
479 ret = bh_submit_read(bh);
480 if (ret)
481 goto cleanup;
482 SetPageUptodate(page);
483
484cleanup:
485 bl_put_extent(cow_read);
486 if (bh)
487 free_buffer_head(bh);
488 if (ret) {
489 /* Need to mark layout with bad read...should now
490 * just use nfs4 for reads and writes.
491 */
492 mark_bad_read();
493 }
494 return ret;
495}
496
497static enum pnfs_try_status
498bl_write_pagelist(struct nfs_write_data *wdata, int sync)
499{
500 int i, ret, npg_zero, pg_index, last = 0;
501 struct bio *bio = NULL;
502 struct pnfs_block_extent *be = NULL, *cow_read = NULL;
503 sector_t isect, last_isect = 0, extent_length = 0;
504 struct parallel_io *par;
505 loff_t offset = wdata->args.offset;
506 size_t count = wdata->args.count;
507 struct page **pages = wdata->args.pages;
508 struct page *page;
509 pgoff_t index;
510 u64 temp;
511 int npg_per_block =
512 NFS_SERVER(wdata->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
513
514 dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
515 /* At this point, wdata->pages is a (sequential) list of nfs_pages.
516 * We want to write each, and if there is an error set pnfs_error
517 * to have it redone using nfs.
518 */
519 par = alloc_parallel(wdata);
520 if (!par)
521 return PNFS_NOT_ATTEMPTED;
522 par->call_ops = *wdata->mds_ops;
523 par->call_ops.rpc_call_done = bl_rpc_do_nothing;
524 par->pnfs_callback = bl_end_par_io_write;
525 /* At this point, have to be more careful with error handling */
526
527 isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
528 be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg), isect, &cow_read);
529 if (!be || !is_writable(be, isect)) {
530 dprintk("%s no matching extents!\n", __func__);
531 wdata->pnfs_error = -EINVAL;
532 goto out;
533 }
534
535 /* First page inside INVALID extent */
536 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
537 temp = offset >> PAGE_CACHE_SHIFT;
538 npg_zero = do_div(temp, npg_per_block);
539 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
540 (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
541 extent_length = be->be_length - (isect - be->be_f_offset);
542
543fill_invalid_ext:
544 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
545 for (;npg_zero > 0; npg_zero--) {
546 /* page ref released in bl_end_io_write_zero */
547 index = isect >> PAGE_CACHE_SECTOR_SHIFT;
548 dprintk("%s zero %dth page: index %lu isect %llu\n",
549 __func__, npg_zero, index,
550 (unsigned long long)isect);
551 page =
552 find_or_create_page(wdata->inode->i_mapping, index,
553 GFP_NOFS);
554 if (!page) {
555 dprintk("%s oom\n", __func__);
556 wdata->pnfs_error = -ENOMEM;
557 goto out;
558 }
559
560 /* PageDirty: Other will write this out
561 * PageWriteback: Other is writing this out
562 * PageUptodate: It was read before
563 * sector_initialized: already written out
564 */
565 if (PageDirty(page) || PageWriteback(page) ||
566 bl_is_sector_init(be->be_inval, isect)) {
567 print_page(page);
568 unlock_page(page);
569 page_cache_release(page);
570 goto next_page;
571 }
572 if (!PageUptodate(page)) {
573 /* New page, readin or zero it */
574 init_page_for_write(page, cow_read);
575 }
576 set_page_writeback(page);
577 unlock_page(page);
578
579 ret = bl_mark_sectors_init(be->be_inval, isect,
580 PAGE_CACHE_SECTORS,
581 NULL);
582 if (unlikely(ret)) {
583 dprintk("%s bl_mark_sectors_init fail %d\n",
584 __func__, ret);
585 end_page_writeback(page);
586 page_cache_release(page);
587 wdata->pnfs_error = ret;
588 goto out;
589 }
590 bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
591 isect, page, be,
592 bl_end_io_write_zero, par);
593 if (IS_ERR(bio)) {
594 wdata->pnfs_error = PTR_ERR(bio);
595 goto out;
596 }
597 /* FIXME: This should be done in bi_end_io */
598 mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
599 page->index << PAGE_CACHE_SHIFT,
600 PAGE_CACHE_SIZE);
601next_page:
602 isect += PAGE_CACHE_SECTORS;
603 extent_length -= PAGE_CACHE_SECTORS;
604 }
605 if (last)
606 goto write_done;
607 }
608 bio = bl_submit_bio(WRITE, bio);
609
610 /* Middle pages */
611 pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
612 for (i = pg_index; i < wdata->npages; i++) {
613 if (!extent_length) {
614 /* We've used up the previous extent */
615 bl_put_extent(be);
616 bio = bl_submit_bio(WRITE, bio);
617 /* Get the next one */
618 be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg),
619 isect, NULL);
620 if (!be || !is_writable(be, isect)) {
621 wdata->pnfs_error = -EINVAL;
622 goto out;
623 }
624 extent_length = be->be_length -
625 (isect - be->be_f_offset);
626 }
627 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
628 ret = bl_mark_sectors_init(be->be_inval, isect,
629 PAGE_CACHE_SECTORS,
630 NULL);
631 if (unlikely(ret)) {
632 dprintk("%s bl_mark_sectors_init fail %d\n",
633 __func__, ret);
634 wdata->pnfs_error = ret;
635 goto out;
636 }
637 }
638 bio = bl_add_page_to_bio(bio, wdata->npages - i, WRITE,
639 isect, pages[i], be,
640 bl_end_io_write, par);
641 if (IS_ERR(bio)) {
642 wdata->pnfs_error = PTR_ERR(bio);
643 goto out;
644 }
645 isect += PAGE_CACHE_SECTORS;
646 last_isect = isect;
647 extent_length -= PAGE_CACHE_SECTORS;
648 }
649
650 /* Last page inside INVALID extent */
651 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
652 bio = bl_submit_bio(WRITE, bio);
653 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
654 npg_zero = npg_per_block - do_div(temp, npg_per_block);
655 if (npg_zero < npg_per_block) {
656 last = 1;
657 goto fill_invalid_ext;
658 }
659 }
660
661write_done:
662 wdata->res.count = (last_isect << SECTOR_SHIFT) - (offset);
663 if (count < wdata->res.count) {
664 wdata->res.count = count;
665 }
666out:
667 bl_put_extent(be);
668 bl_submit_bio(WRITE, bio);
669 put_parallel(par);
670 return PNFS_ATTEMPTED;
671}
672
673/* FIXME - range ignored */
674static void
675release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
676{
677 int i;
678 struct pnfs_block_extent *be;
679
680 spin_lock(&bl->bl_ext_lock);
681 for (i = 0; i < EXTENT_LISTS; i++) {
682 while (!list_empty(&bl->bl_extents[i])) {
683 be = list_first_entry(&bl->bl_extents[i],
684 struct pnfs_block_extent,
685 be_node);
686 list_del(&be->be_node);
687 bl_put_extent(be);
688 }
689 }
690 spin_unlock(&bl->bl_ext_lock);
691}
692
693static void
694release_inval_marks(struct pnfs_inval_markings *marks)
695{
696 struct pnfs_inval_tracking *pos, *temp;
697
698 list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
699 list_del(&pos->it_link);
700 kfree(pos);
701 }
702 return;
703}
704
705static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
706{
707 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
708
709 dprintk("%s enter\n", __func__);
710 release_extents(bl, NULL);
711 release_inval_marks(&bl->bl_inval);
712 kfree(bl);
713}
714
715static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
716 gfp_t gfp_flags)
717{
718 struct pnfs_block_layout *bl;
719
720 dprintk("%s enter\n", __func__);
721 bl = kzalloc(sizeof(*bl), gfp_flags);
722 if (!bl)
723 return NULL;
724 spin_lock_init(&bl->bl_ext_lock);
725 INIT_LIST_HEAD(&bl->bl_extents[0]);
726 INIT_LIST_HEAD(&bl->bl_extents[1]);
727 INIT_LIST_HEAD(&bl->bl_commit);
728 INIT_LIST_HEAD(&bl->bl_committing);
729 bl->bl_count = 0;
730 bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
731 BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
732 return &bl->bl_layout;
733}
734
735static void bl_free_lseg(struct pnfs_layout_segment *lseg)
736{
737 dprintk("%s enter\n", __func__);
738 kfree(lseg);
739}
740
741/* We pretty much ignore lseg, and store all data layout wide, so we
742 * can correctly merge.
743 */
744static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
745 struct nfs4_layoutget_res *lgr,
746 gfp_t gfp_flags)
747{
748 struct pnfs_layout_segment *lseg;
749 int status;
750
751 dprintk("%s enter\n", __func__);
752 lseg = kzalloc(sizeof(*lseg), gfp_flags);
753 if (!lseg)
754 return ERR_PTR(-ENOMEM);
755 status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
756 if (status) {
757 /* We don't want to call the full-blown bl_free_lseg,
758 * since on error extents were not touched.
759 */
760 kfree(lseg);
761 return ERR_PTR(status);
762 }
763 return lseg;
764}
765
766static void
767bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
768 const struct nfs4_layoutcommit_args *arg)
769{
770 dprintk("%s enter\n", __func__);
771 encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
772}
773
774static void
775bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
776{
777 struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
778
779 dprintk("%s enter\n", __func__);
780 clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
781}
782
783static void free_blk_mountid(struct block_mount_id *mid)
784{
785 if (mid) {
786 struct pnfs_block_dev *dev;
787 spin_lock(&mid->bm_lock);
788 while (!list_empty(&mid->bm_devlist)) {
789 dev = list_first_entry(&mid->bm_devlist,
790 struct pnfs_block_dev,
791 bm_node);
792 list_del(&dev->bm_node);
793 bl_free_block_dev(dev);
794 }
795 spin_unlock(&mid->bm_lock);
796 kfree(mid);
797 }
798}
799
800/* This is mostly copied from the filelayout's get_device_info function.
801 * It seems much of this should be at the generic pnfs level.
802 */
803static struct pnfs_block_dev *
804nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
805 struct nfs4_deviceid *d_id)
806{
807 struct pnfs_device *dev;
808 struct pnfs_block_dev *rv = NULL;
809 u32 max_resp_sz;
810 int max_pages;
811 struct page **pages = NULL;
812 int i, rc;
813
814 /*
815 * Use the session max response size as the basis for setting
816 * GETDEVICEINFO's maxcount
817 */
818 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
819 max_pages = max_resp_sz >> PAGE_SHIFT;
820 dprintk("%s max_resp_sz %u max_pages %d\n",
821 __func__, max_resp_sz, max_pages);
822
823 dev = kmalloc(sizeof(*dev), GFP_NOFS);
824 if (!dev) {
825 dprintk("%s kmalloc failed\n", __func__);
826 return NULL;
827 }
828
829 pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
830 if (pages == NULL) {
831 kfree(dev);
832 return NULL;
833 }
834 for (i = 0; i < max_pages; i++) {
835 pages[i] = alloc_page(GFP_NOFS);
836 if (!pages[i])
837 goto out_free;
838 }
839
840 memcpy(&dev->dev_id, d_id, sizeof(*d_id));
841 dev->layout_type = LAYOUT_BLOCK_VOLUME;
842 dev->pages = pages;
843 dev->pgbase = 0;
844 dev->pglen = PAGE_SIZE * max_pages;
845 dev->mincount = 0;
846
847 dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
848 rc = nfs4_proc_getdeviceinfo(server, dev);
849 dprintk("%s getdevice info returns %d\n", __func__, rc);
850 if (rc)
851 goto out_free;
852
853 rv = nfs4_blk_decode_device(server, dev);
854 out_free:
855 for (i = 0; i < max_pages; i++)
856 __free_page(pages[i]);
857 kfree(pages);
858 kfree(dev);
859 return rv;
860}
861
862static int
863bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
864{
865 struct block_mount_id *b_mt_id = NULL;
866 struct pnfs_devicelist *dlist = NULL;
867 struct pnfs_block_dev *bdev;
868 LIST_HEAD(block_disklist);
869 int status = 0, i;
870
871 dprintk("%s enter\n", __func__);
872
873 if (server->pnfs_blksize == 0) {
874 dprintk("%s Server did not return blksize\n", __func__);
875 return -EINVAL;
876 }
877 b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
878 if (!b_mt_id) {
879 status = -ENOMEM;
880 goto out_error;
881 }
882 /* Initialize nfs4 block layout mount id */
883 spin_lock_init(&b_mt_id->bm_lock);
884 INIT_LIST_HEAD(&b_mt_id->bm_devlist);
885
886 dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
887 if (!dlist) {
888 status = -ENOMEM;
889 goto out_error;
890 }
891 dlist->eof = 0;
892 while (!dlist->eof) {
893 status = nfs4_proc_getdevicelist(server, fh, dlist);
894 if (status)
895 goto out_error;
896 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
897 __func__, dlist->num_devs, dlist->eof);
898 for (i = 0; i < dlist->num_devs; i++) {
899 bdev = nfs4_blk_get_deviceinfo(server, fh,
900 &dlist->dev_id[i]);
901 if (!bdev) {
902 status = -ENODEV;
903 goto out_error;
904 }
905 spin_lock(&b_mt_id->bm_lock);
906 list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
907 spin_unlock(&b_mt_id->bm_lock);
908 }
909 }
910 dprintk("%s SUCCESS\n", __func__);
911 server->pnfs_ld_data = b_mt_id;
912
913 out_return:
914 kfree(dlist);
915 return status;
916
917 out_error:
918 free_blk_mountid(b_mt_id);
919 goto out_return;
920}
921
922static int
923bl_clear_layoutdriver(struct nfs_server *server)
924{
925 struct block_mount_id *b_mt_id = server->pnfs_ld_data;
926
927 dprintk("%s enter\n", __func__);
928 free_blk_mountid(b_mt_id);
929 dprintk("%s RETURNS\n", __func__);
930 return 0;
931}
932
933static const struct nfs_pageio_ops bl_pg_read_ops = {
934 .pg_init = pnfs_generic_pg_init_read,
935 .pg_test = pnfs_generic_pg_test,
936 .pg_doio = pnfs_generic_pg_readpages,
937};
938
939static const struct nfs_pageio_ops bl_pg_write_ops = {
940 .pg_init = pnfs_generic_pg_init_write,
941 .pg_test = pnfs_generic_pg_test,
942 .pg_doio = pnfs_generic_pg_writepages,
943};
944
945static struct pnfs_layoutdriver_type blocklayout_type = {
946 .id = LAYOUT_BLOCK_VOLUME,
947 .name = "LAYOUT_BLOCK_VOLUME",
948 .read_pagelist = bl_read_pagelist,
949 .write_pagelist = bl_write_pagelist,
950 .alloc_layout_hdr = bl_alloc_layout_hdr,
951 .free_layout_hdr = bl_free_layout_hdr,
952 .alloc_lseg = bl_alloc_lseg,
953 .free_lseg = bl_free_lseg,
954 .encode_layoutcommit = bl_encode_layoutcommit,
955 .cleanup_layoutcommit = bl_cleanup_layoutcommit,
956 .set_layoutdriver = bl_set_layoutdriver,
957 .clear_layoutdriver = bl_clear_layoutdriver,
958 .pg_read_ops = &bl_pg_read_ops,
959 .pg_write_ops = &bl_pg_write_ops,
960};
961
962static const struct rpc_pipe_ops bl_upcall_ops = {
963 .upcall = bl_pipe_upcall,
964 .downcall = bl_pipe_downcall,
965 .destroy_msg = bl_pipe_destroy_msg,
966};
967
968static int __init nfs4blocklayout_init(void)
969{
970 struct vfsmount *mnt;
971 struct path path;
972 int ret;
973
974 dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
975
976 ret = pnfs_register_layoutdriver(&blocklayout_type);
977 if (ret)
978 goto out;
979
980 init_waitqueue_head(&bl_wq);
981
982 mnt = rpc_get_mount();
983 if (IS_ERR(mnt)) {
984 ret = PTR_ERR(mnt);
985 goto out_remove;
986 }
987
988 ret = vfs_path_lookup(mnt->mnt_root,
989 mnt,
990 NFS_PIPE_DIRNAME, 0, &path);
991 if (ret)
992 goto out_remove;
993
994 bl_device_pipe = rpc_mkpipe(path.dentry, "blocklayout", NULL,
995 &bl_upcall_ops, 0);
996 if (IS_ERR(bl_device_pipe)) {
997 ret = PTR_ERR(bl_device_pipe);
998 goto out_remove;
999 }
1000out:
1001 return ret;
1002
1003out_remove:
1004 pnfs_unregister_layoutdriver(&blocklayout_type);
1005 return ret;
1006}
1007
1008static void __exit nfs4blocklayout_exit(void)
1009{
1010 dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1011 __func__);
1012
1013 pnfs_unregister_layoutdriver(&blocklayout_type);
1014 rpc_unlink(bl_device_pipe);
1015}
1016
1017MODULE_ALIAS("nfs-layouttype4-3");
1018
1019module_init(nfs4blocklayout_init);
1020module_exit(nfs4blocklayout_exit);