]>
Commit | Line | Data |
---|---|---|
155e7524 FI |
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 | */ | |
9549ec01 | 32 | |
155e7524 FI |
33 | #include <linux/module.h> |
34 | #include <linux/init.h> | |
fe0a9b74 JR |
35 | #include <linux/mount.h> |
36 | #include <linux/namei.h> | |
9549ec01 | 37 | #include <linux/bio.h> /* struct bio */ |
155e7524 FI |
38 | |
39 | #include "blocklayout.h" | |
40 | ||
41 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
42 | ||
43 | MODULE_LICENSE("GPL"); | |
44 | MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>"); | |
45 | MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver"); | |
46 | ||
fe0a9b74 JR |
47 | struct dentry *bl_device_pipe; |
48 | wait_queue_head_t bl_wq; | |
49 | ||
9549ec01 FI |
50 | static void print_page(struct page *page) |
51 | { | |
52 | dprintk("PRINTPAGE page %p\n", page); | |
53 | dprintk(" PagePrivate %d\n", PagePrivate(page)); | |
54 | dprintk(" PageUptodate %d\n", PageUptodate(page)); | |
55 | dprintk(" PageError %d\n", PageError(page)); | |
56 | dprintk(" PageDirty %d\n", PageDirty(page)); | |
57 | dprintk(" PageReferenced %d\n", PageReferenced(page)); | |
58 | dprintk(" PageLocked %d\n", PageLocked(page)); | |
59 | dprintk(" PageWriteback %d\n", PageWriteback(page)); | |
60 | dprintk(" PageMappedToDisk %d\n", PageMappedToDisk(page)); | |
61 | dprintk("\n"); | |
62 | } | |
63 | ||
64 | /* Given the be associated with isect, determine if page data needs to be | |
65 | * initialized. | |
66 | */ | |
67 | static int is_hole(struct pnfs_block_extent *be, sector_t isect) | |
68 | { | |
69 | if (be->be_state == PNFS_BLOCK_NONE_DATA) | |
70 | return 1; | |
71 | else if (be->be_state != PNFS_BLOCK_INVALID_DATA) | |
72 | return 0; | |
73 | else | |
74 | return !bl_is_sector_init(be->be_inval, isect); | |
75 | } | |
76 | ||
650e2d39 FI |
77 | /* Given the be associated with isect, determine if page data can be |
78 | * written to disk. | |
79 | */ | |
80 | static int is_writable(struct pnfs_block_extent *be, sector_t isect) | |
81 | { | |
82 | if (be->be_state == PNFS_BLOCK_READWRITE_DATA) | |
83 | return 1; | |
84 | else if (be->be_state != PNFS_BLOCK_INVALID_DATA) | |
85 | return 0; | |
86 | else | |
87 | return bl_is_sector_init(be->be_inval, isect); | |
88 | } | |
89 | ||
9549ec01 FI |
90 | /* The data we are handed might be spread across several bios. We need |
91 | * to track when the last one is finished. | |
92 | */ | |
93 | struct parallel_io { | |
94 | struct kref refcnt; | |
95 | struct rpc_call_ops call_ops; | |
96 | void (*pnfs_callback) (void *data); | |
97 | void *data; | |
98 | }; | |
99 | ||
100 | static inline struct parallel_io *alloc_parallel(void *data) | |
101 | { | |
102 | struct parallel_io *rv; | |
103 | ||
104 | rv = kmalloc(sizeof(*rv), GFP_NOFS); | |
105 | if (rv) { | |
106 | rv->data = data; | |
107 | kref_init(&rv->refcnt); | |
108 | } | |
109 | return rv; | |
110 | } | |
111 | ||
112 | static inline void get_parallel(struct parallel_io *p) | |
113 | { | |
114 | kref_get(&p->refcnt); | |
115 | } | |
116 | ||
117 | static void destroy_parallel(struct kref *kref) | |
118 | { | |
119 | struct parallel_io *p = container_of(kref, struct parallel_io, refcnt); | |
120 | ||
121 | dprintk("%s enter\n", __func__); | |
122 | p->pnfs_callback(p->data); | |
123 | kfree(p); | |
124 | } | |
125 | ||
126 | static inline void put_parallel(struct parallel_io *p) | |
127 | { | |
128 | kref_put(&p->refcnt, destroy_parallel); | |
129 | } | |
130 | ||
131 | static struct bio * | |
132 | bl_submit_bio(int rw, struct bio *bio) | |
133 | { | |
134 | if (bio) { | |
135 | get_parallel(bio->bi_private); | |
136 | dprintk("%s submitting %s bio %u@%llu\n", __func__, | |
137 | rw == READ ? "read" : "write", | |
138 | bio->bi_size, (unsigned long long)bio->bi_sector); | |
139 | submit_bio(rw, bio); | |
140 | } | |
141 | return NULL; | |
142 | } | |
143 | ||
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) | |
148 | { | |
149 | struct bio *bio; | |
150 | ||
151 | bio = bio_alloc(GFP_NOIO, npg); | |
152 | if (!bio) | |
153 | return NULL; | |
154 | ||
155 | bio->bi_sector = isect - be->be_f_offset + be->be_v_offset; | |
156 | bio->bi_bdev = be->be_mdev; | |
157 | bio->bi_end_io = end_io; | |
158 | bio->bi_private = par; | |
159 | return bio; | |
160 | } | |
161 | ||
162 | static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw, | |
163 | sector_t isect, struct page *page, | |
164 | struct pnfs_block_extent *be, | |
165 | void (*end_io)(struct bio *, int err), | |
166 | struct parallel_io *par) | |
167 | { | |
168 | retry: | |
169 | if (!bio) { | |
170 | bio = bl_alloc_init_bio(npg, isect, be, end_io, par); | |
171 | if (!bio) | |
172 | return ERR_PTR(-ENOMEM); | |
173 | } | |
174 | if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) { | |
175 | bio = bl_submit_bio(rw, bio); | |
176 | goto retry; | |
177 | } | |
178 | return bio; | |
179 | } | |
180 | ||
181 | static void bl_set_lo_fail(struct pnfs_layout_segment *lseg) | |
182 | { | |
183 | if (lseg->pls_range.iomode == IOMODE_RW) { | |
184 | dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__); | |
185 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
186 | } else { | |
187 | dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__); | |
188 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
189 | } | |
190 | } | |
191 | ||
192 | /* This is basically copied from mpage_end_io_read */ | |
193 | static void bl_end_io_read(struct bio *bio, int err) | |
194 | { | |
195 | struct parallel_io *par = bio->bi_private; | |
196 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
197 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
198 | struct nfs_read_data *rdata = (struct nfs_read_data *)par->data; | |
199 | ||
200 | do { | |
201 | struct page *page = bvec->bv_page; | |
202 | ||
203 | if (--bvec >= bio->bi_io_vec) | |
204 | prefetchw(&bvec->bv_page->flags); | |
205 | if (uptodate) | |
206 | SetPageUptodate(page); | |
207 | } while (bvec >= bio->bi_io_vec); | |
208 | if (!uptodate) { | |
209 | if (!rdata->pnfs_error) | |
210 | rdata->pnfs_error = -EIO; | |
211 | bl_set_lo_fail(rdata->lseg); | |
212 | } | |
213 | bio_put(bio); | |
214 | put_parallel(par); | |
215 | } | |
216 | ||
217 | static void bl_read_cleanup(struct work_struct *work) | |
218 | { | |
219 | struct rpc_task *task; | |
220 | struct nfs_read_data *rdata; | |
221 | dprintk("%s enter\n", __func__); | |
222 | task = container_of(work, struct rpc_task, u.tk_work); | |
223 | rdata = container_of(task, struct nfs_read_data, task); | |
224 | pnfs_ld_read_done(rdata); | |
225 | } | |
226 | ||
227 | static void | |
228 | bl_end_par_io_read(void *data) | |
229 | { | |
230 | struct nfs_read_data *rdata = data; | |
231 | ||
232 | INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup); | |
233 | schedule_work(&rdata->task.u.tk_work); | |
234 | } | |
235 | ||
236 | /* We don't want normal .rpc_call_done callback used, so we replace it | |
237 | * with this stub. | |
238 | */ | |
239 | static void bl_rpc_do_nothing(struct rpc_task *task, void *calldata) | |
240 | { | |
241 | return; | |
242 | } | |
243 | ||
155e7524 FI |
244 | static enum pnfs_try_status |
245 | bl_read_pagelist(struct nfs_read_data *rdata) | |
246 | { | |
9549ec01 FI |
247 | int i, hole; |
248 | struct bio *bio = NULL; | |
249 | struct pnfs_block_extent *be = NULL, *cow_read = NULL; | |
250 | sector_t isect, extent_length = 0; | |
251 | struct parallel_io *par; | |
252 | loff_t f_offset = rdata->args.offset; | |
253 | size_t count = rdata->args.count; | |
254 | struct page **pages = rdata->args.pages; | |
255 | int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT; | |
256 | ||
257 | dprintk("%s enter nr_pages %u offset %lld count %Zd\n", __func__, | |
258 | rdata->npages, f_offset, count); | |
259 | ||
260 | par = alloc_parallel(rdata); | |
261 | if (!par) | |
262 | goto use_mds; | |
263 | par->call_ops = *rdata->mds_ops; | |
264 | par->call_ops.rpc_call_done = bl_rpc_do_nothing; | |
265 | par->pnfs_callback = bl_end_par_io_read; | |
266 | /* At this point, we can no longer jump to use_mds */ | |
267 | ||
268 | isect = (sector_t) (f_offset >> SECTOR_SHIFT); | |
269 | /* Code assumes extents are page-aligned */ | |
270 | for (i = pg_index; i < rdata->npages; i++) { | |
271 | if (!extent_length) { | |
272 | /* We've used up the previous extent */ | |
273 | bl_put_extent(be); | |
274 | bl_put_extent(cow_read); | |
275 | bio = bl_submit_bio(READ, bio); | |
276 | /* Get the next one */ | |
277 | be = bl_find_get_extent(BLK_LSEG2EXT(rdata->lseg), | |
278 | isect, &cow_read); | |
279 | if (!be) { | |
280 | rdata->pnfs_error = -EIO; | |
281 | goto out; | |
282 | } | |
283 | extent_length = be->be_length - | |
284 | (isect - be->be_f_offset); | |
285 | if (cow_read) { | |
286 | sector_t cow_length = cow_read->be_length - | |
287 | (isect - cow_read->be_f_offset); | |
288 | extent_length = min(extent_length, cow_length); | |
289 | } | |
290 | } | |
291 | hole = is_hole(be, isect); | |
292 | if (hole && !cow_read) { | |
293 | bio = bl_submit_bio(READ, bio); | |
294 | /* Fill hole w/ zeroes w/o accessing device */ | |
295 | dprintk("%s Zeroing page for hole\n", __func__); | |
296 | zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE); | |
297 | print_page(pages[i]); | |
298 | SetPageUptodate(pages[i]); | |
299 | } else { | |
300 | struct pnfs_block_extent *be_read; | |
301 | ||
302 | be_read = (hole && cow_read) ? cow_read : be; | |
303 | bio = bl_add_page_to_bio(bio, rdata->npages - i, READ, | |
304 | isect, pages[i], be_read, | |
305 | bl_end_io_read, par); | |
306 | if (IS_ERR(bio)) { | |
307 | rdata->pnfs_error = PTR_ERR(bio); | |
308 | goto out; | |
309 | } | |
310 | } | |
311 | isect += PAGE_CACHE_SECTORS; | |
312 | extent_length -= PAGE_CACHE_SECTORS; | |
313 | } | |
314 | if ((isect << SECTOR_SHIFT) >= rdata->inode->i_size) { | |
315 | rdata->res.eof = 1; | |
316 | rdata->res.count = rdata->inode->i_size - f_offset; | |
317 | } else { | |
318 | rdata->res.count = (isect << SECTOR_SHIFT) - f_offset; | |
319 | } | |
320 | out: | |
321 | bl_put_extent(be); | |
322 | bl_put_extent(cow_read); | |
323 | bl_submit_bio(READ, bio); | |
324 | put_parallel(par); | |
325 | return PNFS_ATTEMPTED; | |
326 | ||
327 | use_mds: | |
328 | dprintk("Giving up and using normal NFS\n"); | |
155e7524 FI |
329 | return PNFS_NOT_ATTEMPTED; |
330 | } | |
331 | ||
650e2d39 FI |
332 | /* This is basically copied from mpage_end_io_read */ |
333 | static void bl_end_io_write(struct bio *bio, int err) | |
334 | { | |
335 | struct parallel_io *par = bio->bi_private; | |
336 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
337 | struct nfs_write_data *wdata = (struct nfs_write_data *)par->data; | |
338 | ||
339 | if (!uptodate) { | |
340 | if (!wdata->pnfs_error) | |
341 | wdata->pnfs_error = -EIO; | |
342 | bl_set_lo_fail(wdata->lseg); | |
343 | } | |
344 | bio_put(bio); | |
345 | put_parallel(par); | |
346 | } | |
347 | ||
348 | /* Function scheduled for call during bl_end_par_io_write, | |
349 | * it marks sectors as written and extends the commitlist. | |
350 | */ | |
351 | static void bl_write_cleanup(struct work_struct *work) | |
352 | { | |
353 | struct rpc_task *task; | |
354 | struct nfs_write_data *wdata; | |
355 | dprintk("%s enter\n", __func__); | |
356 | task = container_of(work, struct rpc_task, u.tk_work); | |
357 | wdata = container_of(task, struct nfs_write_data, task); | |
358 | pnfs_ld_write_done(wdata); | |
359 | } | |
360 | ||
361 | /* Called when last of bios associated with a bl_write_pagelist call finishes */ | |
362 | static void | |
363 | bl_end_par_io_write(void *data) | |
364 | { | |
365 | struct nfs_write_data *wdata = data; | |
366 | ||
367 | /* STUB - ignoring error handling */ | |
368 | wdata->task.tk_status = 0; | |
369 | wdata->verf.committed = NFS_FILE_SYNC; | |
370 | INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup); | |
371 | schedule_work(&wdata->task.u.tk_work); | |
372 | } | |
373 | ||
155e7524 | 374 | static enum pnfs_try_status |
650e2d39 | 375 | bl_write_pagelist(struct nfs_write_data *wdata, int sync) |
155e7524 | 376 | { |
650e2d39 FI |
377 | int i; |
378 | struct bio *bio = NULL; | |
379 | struct pnfs_block_extent *be = NULL; | |
380 | sector_t isect, extent_length = 0; | |
381 | struct parallel_io *par; | |
382 | loff_t offset = wdata->args.offset; | |
383 | size_t count = wdata->args.count; | |
384 | struct page **pages = wdata->args.pages; | |
385 | int pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT; | |
386 | ||
387 | dprintk("%s enter, %Zu@%lld\n", __func__, count, offset); | |
388 | /* At this point, wdata->pages is a (sequential) list of nfs_pages. | |
389 | * We want to write each, and if there is an error remove it from | |
390 | * list and call | |
391 | * nfs_retry_request(req) to have it redone using nfs. | |
392 | * QUEST? Do as block or per req? Think have to do per block | |
393 | * as part of end_bio | |
394 | */ | |
395 | par = alloc_parallel(wdata); | |
396 | if (!par) | |
397 | return PNFS_NOT_ATTEMPTED; | |
398 | par->call_ops = *wdata->mds_ops; | |
399 | par->call_ops.rpc_call_done = bl_rpc_do_nothing; | |
400 | par->pnfs_callback = bl_end_par_io_write; | |
401 | /* At this point, have to be more careful with error handling */ | |
402 | ||
403 | isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT); | |
404 | for (i = pg_index; i < wdata->npages ; i++) { | |
405 | if (!extent_length) { | |
406 | /* We've used up the previous extent */ | |
407 | bl_put_extent(be); | |
408 | bio = bl_submit_bio(WRITE, bio); | |
409 | /* Get the next one */ | |
410 | be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg), | |
411 | isect, NULL); | |
412 | if (!be || !is_writable(be, isect)) { | |
413 | wdata->pnfs_error = -ENOMEM; | |
414 | goto out; | |
415 | } | |
416 | extent_length = be->be_length - | |
417 | (isect - be->be_f_offset); | |
418 | } | |
419 | for (;;) { | |
420 | if (!bio) { | |
421 | bio = bio_alloc(GFP_NOIO, wdata->npages - i); | |
422 | if (!bio) { | |
423 | wdata->pnfs_error = -ENOMEM; | |
424 | goto out; | |
425 | } | |
426 | bio->bi_sector = isect - be->be_f_offset + | |
427 | be->be_v_offset; | |
428 | bio->bi_bdev = be->be_mdev; | |
429 | bio->bi_end_io = bl_end_io_write; | |
430 | bio->bi_private = par; | |
431 | } | |
432 | if (bio_add_page(bio, pages[i], PAGE_SIZE, 0)) | |
433 | break; | |
434 | bio = bl_submit_bio(WRITE, bio); | |
435 | } | |
436 | isect += PAGE_CACHE_SECTORS; | |
437 | extent_length -= PAGE_CACHE_SECTORS; | |
438 | } | |
439 | wdata->res.count = (isect << SECTOR_SHIFT) - (offset); | |
440 | if (count < wdata->res.count) | |
441 | wdata->res.count = count; | |
442 | out: | |
443 | bl_put_extent(be); | |
444 | bl_submit_bio(WRITE, bio); | |
445 | put_parallel(par); | |
446 | return PNFS_ATTEMPTED; | |
155e7524 FI |
447 | } |
448 | ||
9e692969 | 449 | /* FIXME - range ignored */ |
155e7524 | 450 | static void |
9e692969 | 451 | release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range) |
155e7524 | 452 | { |
9e692969 FI |
453 | int i; |
454 | struct pnfs_block_extent *be; | |
455 | ||
456 | spin_lock(&bl->bl_ext_lock); | |
457 | for (i = 0; i < EXTENT_LISTS; i++) { | |
458 | while (!list_empty(&bl->bl_extents[i])) { | |
459 | be = list_first_entry(&bl->bl_extents[i], | |
460 | struct pnfs_block_extent, | |
461 | be_node); | |
462 | list_del(&be->be_node); | |
463 | bl_put_extent(be); | |
464 | } | |
465 | } | |
466 | spin_unlock(&bl->bl_ext_lock); | |
155e7524 FI |
467 | } |
468 | ||
155e7524 FI |
469 | static void |
470 | release_inval_marks(struct pnfs_inval_markings *marks) | |
471 | { | |
c1c2a4cd FI |
472 | struct pnfs_inval_tracking *pos, *temp; |
473 | ||
474 | list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) { | |
475 | list_del(&pos->it_link); | |
476 | kfree(pos); | |
477 | } | |
155e7524 FI |
478 | return; |
479 | } | |
480 | ||
481 | static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
482 | { | |
483 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
484 | ||
485 | dprintk("%s enter\n", __func__); | |
486 | release_extents(bl, NULL); | |
487 | release_inval_marks(&bl->bl_inval); | |
488 | kfree(bl); | |
489 | } | |
490 | ||
491 | static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode, | |
492 | gfp_t gfp_flags) | |
493 | { | |
494 | struct pnfs_block_layout *bl; | |
495 | ||
496 | dprintk("%s enter\n", __func__); | |
497 | bl = kzalloc(sizeof(*bl), gfp_flags); | |
498 | if (!bl) | |
499 | return NULL; | |
500 | spin_lock_init(&bl->bl_ext_lock); | |
501 | INIT_LIST_HEAD(&bl->bl_extents[0]); | |
502 | INIT_LIST_HEAD(&bl->bl_extents[1]); | |
503 | INIT_LIST_HEAD(&bl->bl_commit); | |
504 | INIT_LIST_HEAD(&bl->bl_committing); | |
505 | bl->bl_count = 0; | |
506 | bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT; | |
507 | BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize); | |
508 | return &bl->bl_layout; | |
509 | } | |
510 | ||
a60d2ebd | 511 | static void bl_free_lseg(struct pnfs_layout_segment *lseg) |
155e7524 | 512 | { |
a60d2ebd FI |
513 | dprintk("%s enter\n", __func__); |
514 | kfree(lseg); | |
155e7524 FI |
515 | } |
516 | ||
a60d2ebd FI |
517 | /* We pretty much ignore lseg, and store all data layout wide, so we |
518 | * can correctly merge. | |
519 | */ | |
520 | static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo, | |
521 | struct nfs4_layoutget_res *lgr, | |
522 | gfp_t gfp_flags) | |
155e7524 | 523 | { |
a60d2ebd FI |
524 | struct pnfs_layout_segment *lseg; |
525 | int status; | |
526 | ||
527 | dprintk("%s enter\n", __func__); | |
528 | lseg = kzalloc(sizeof(*lseg), gfp_flags); | |
529 | if (!lseg) | |
530 | return ERR_PTR(-ENOMEM); | |
531 | status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags); | |
532 | if (status) { | |
533 | /* We don't want to call the full-blown bl_free_lseg, | |
534 | * since on error extents were not touched. | |
535 | */ | |
536 | kfree(lseg); | |
537 | return ERR_PTR(status); | |
538 | } | |
539 | return lseg; | |
155e7524 FI |
540 | } |
541 | ||
542 | static void | |
543 | bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr, | |
544 | const struct nfs4_layoutcommit_args *arg) | |
545 | { | |
90ace12a FI |
546 | dprintk("%s enter\n", __func__); |
547 | encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg); | |
155e7524 FI |
548 | } |
549 | ||
550 | static void | |
551 | bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata) | |
552 | { | |
b2be7811 FI |
553 | struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout; |
554 | ||
555 | dprintk("%s enter\n", __func__); | |
556 | clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status); | |
155e7524 FI |
557 | } |
558 | ||
2f9fd182 FI |
559 | static void free_blk_mountid(struct block_mount_id *mid) |
560 | { | |
561 | if (mid) { | |
562 | struct pnfs_block_dev *dev; | |
563 | spin_lock(&mid->bm_lock); | |
564 | while (!list_empty(&mid->bm_devlist)) { | |
565 | dev = list_first_entry(&mid->bm_devlist, | |
566 | struct pnfs_block_dev, | |
567 | bm_node); | |
568 | list_del(&dev->bm_node); | |
569 | bl_free_block_dev(dev); | |
570 | } | |
571 | spin_unlock(&mid->bm_lock); | |
572 | kfree(mid); | |
573 | } | |
574 | } | |
575 | ||
576 | /* This is mostly copied from the filelayout's get_device_info function. | |
577 | * It seems much of this should be at the generic pnfs level. | |
578 | */ | |
579 | static struct pnfs_block_dev * | |
580 | nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh, | |
581 | struct nfs4_deviceid *d_id) | |
582 | { | |
583 | struct pnfs_device *dev; | |
584 | struct pnfs_block_dev *rv = NULL; | |
585 | u32 max_resp_sz; | |
586 | int max_pages; | |
587 | struct page **pages = NULL; | |
588 | int i, rc; | |
589 | ||
590 | /* | |
591 | * Use the session max response size as the basis for setting | |
592 | * GETDEVICEINFO's maxcount | |
593 | */ | |
594 | max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz; | |
595 | max_pages = max_resp_sz >> PAGE_SHIFT; | |
596 | dprintk("%s max_resp_sz %u max_pages %d\n", | |
597 | __func__, max_resp_sz, max_pages); | |
598 | ||
599 | dev = kmalloc(sizeof(*dev), GFP_NOFS); | |
600 | if (!dev) { | |
601 | dprintk("%s kmalloc failed\n", __func__); | |
602 | return NULL; | |
603 | } | |
604 | ||
605 | pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS); | |
606 | if (pages == NULL) { | |
607 | kfree(dev); | |
608 | return NULL; | |
609 | } | |
610 | for (i = 0; i < max_pages; i++) { | |
611 | pages[i] = alloc_page(GFP_NOFS); | |
612 | if (!pages[i]) | |
613 | goto out_free; | |
614 | } | |
615 | ||
616 | memcpy(&dev->dev_id, d_id, sizeof(*d_id)); | |
617 | dev->layout_type = LAYOUT_BLOCK_VOLUME; | |
618 | dev->pages = pages; | |
619 | dev->pgbase = 0; | |
620 | dev->pglen = PAGE_SIZE * max_pages; | |
621 | dev->mincount = 0; | |
622 | ||
623 | dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data); | |
624 | rc = nfs4_proc_getdeviceinfo(server, dev); | |
625 | dprintk("%s getdevice info returns %d\n", __func__, rc); | |
626 | if (rc) | |
627 | goto out_free; | |
628 | ||
629 | rv = nfs4_blk_decode_device(server, dev); | |
630 | out_free: | |
631 | for (i = 0; i < max_pages; i++) | |
632 | __free_page(pages[i]); | |
633 | kfree(pages); | |
634 | kfree(dev); | |
635 | return rv; | |
636 | } | |
637 | ||
155e7524 FI |
638 | static int |
639 | bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh) | |
640 | { | |
2f9fd182 FI |
641 | struct block_mount_id *b_mt_id = NULL; |
642 | struct pnfs_devicelist *dlist = NULL; | |
643 | struct pnfs_block_dev *bdev; | |
644 | LIST_HEAD(block_disklist); | |
645 | int status = 0, i; | |
646 | ||
155e7524 | 647 | dprintk("%s enter\n", __func__); |
2f9fd182 FI |
648 | |
649 | if (server->pnfs_blksize == 0) { | |
650 | dprintk("%s Server did not return blksize\n", __func__); | |
651 | return -EINVAL; | |
652 | } | |
653 | b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS); | |
654 | if (!b_mt_id) { | |
655 | status = -ENOMEM; | |
656 | goto out_error; | |
657 | } | |
658 | /* Initialize nfs4 block layout mount id */ | |
659 | spin_lock_init(&b_mt_id->bm_lock); | |
660 | INIT_LIST_HEAD(&b_mt_id->bm_devlist); | |
661 | ||
662 | dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS); | |
663 | if (!dlist) { | |
664 | status = -ENOMEM; | |
665 | goto out_error; | |
666 | } | |
667 | dlist->eof = 0; | |
668 | while (!dlist->eof) { | |
669 | status = nfs4_proc_getdevicelist(server, fh, dlist); | |
670 | if (status) | |
671 | goto out_error; | |
672 | dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n", | |
673 | __func__, dlist->num_devs, dlist->eof); | |
674 | for (i = 0; i < dlist->num_devs; i++) { | |
675 | bdev = nfs4_blk_get_deviceinfo(server, fh, | |
676 | &dlist->dev_id[i]); | |
677 | if (!bdev) { | |
678 | status = -ENODEV; | |
679 | goto out_error; | |
680 | } | |
681 | spin_lock(&b_mt_id->bm_lock); | |
682 | list_add(&bdev->bm_node, &b_mt_id->bm_devlist); | |
683 | spin_unlock(&b_mt_id->bm_lock); | |
684 | } | |
685 | } | |
686 | dprintk("%s SUCCESS\n", __func__); | |
687 | server->pnfs_ld_data = b_mt_id; | |
688 | ||
689 | out_return: | |
690 | kfree(dlist); | |
691 | return status; | |
692 | ||
693 | out_error: | |
694 | free_blk_mountid(b_mt_id); | |
695 | goto out_return; | |
155e7524 FI |
696 | } |
697 | ||
698 | static int | |
699 | bl_clear_layoutdriver(struct nfs_server *server) | |
700 | { | |
2f9fd182 FI |
701 | struct block_mount_id *b_mt_id = server->pnfs_ld_data; |
702 | ||
155e7524 | 703 | dprintk("%s enter\n", __func__); |
2f9fd182 FI |
704 | free_blk_mountid(b_mt_id); |
705 | dprintk("%s RETURNS\n", __func__); | |
155e7524 FI |
706 | return 0; |
707 | } | |
708 | ||
e9643fe8 BH |
709 | static const struct nfs_pageio_ops bl_pg_read_ops = { |
710 | .pg_init = pnfs_generic_pg_init_read, | |
711 | .pg_test = pnfs_generic_pg_test, | |
712 | .pg_doio = pnfs_generic_pg_readpages, | |
713 | }; | |
714 | ||
715 | static const struct nfs_pageio_ops bl_pg_write_ops = { | |
716 | .pg_init = pnfs_generic_pg_init_write, | |
717 | .pg_test = pnfs_generic_pg_test, | |
718 | .pg_doio = pnfs_generic_pg_writepages, | |
719 | }; | |
720 | ||
155e7524 FI |
721 | static struct pnfs_layoutdriver_type blocklayout_type = { |
722 | .id = LAYOUT_BLOCK_VOLUME, | |
723 | .name = "LAYOUT_BLOCK_VOLUME", | |
724 | .read_pagelist = bl_read_pagelist, | |
725 | .write_pagelist = bl_write_pagelist, | |
726 | .alloc_layout_hdr = bl_alloc_layout_hdr, | |
727 | .free_layout_hdr = bl_free_layout_hdr, | |
728 | .alloc_lseg = bl_alloc_lseg, | |
729 | .free_lseg = bl_free_lseg, | |
730 | .encode_layoutcommit = bl_encode_layoutcommit, | |
731 | .cleanup_layoutcommit = bl_cleanup_layoutcommit, | |
732 | .set_layoutdriver = bl_set_layoutdriver, | |
733 | .clear_layoutdriver = bl_clear_layoutdriver, | |
e9643fe8 BH |
734 | .pg_read_ops = &bl_pg_read_ops, |
735 | .pg_write_ops = &bl_pg_write_ops, | |
155e7524 FI |
736 | }; |
737 | ||
fe0a9b74 JR |
738 | static const struct rpc_pipe_ops bl_upcall_ops = { |
739 | .upcall = bl_pipe_upcall, | |
740 | .downcall = bl_pipe_downcall, | |
741 | .destroy_msg = bl_pipe_destroy_msg, | |
742 | }; | |
743 | ||
155e7524 FI |
744 | static int __init nfs4blocklayout_init(void) |
745 | { | |
fe0a9b74 JR |
746 | struct vfsmount *mnt; |
747 | struct path path; | |
155e7524 FI |
748 | int ret; |
749 | ||
750 | dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__); | |
751 | ||
752 | ret = pnfs_register_layoutdriver(&blocklayout_type); | |
fe0a9b74 JR |
753 | if (ret) |
754 | goto out; | |
755 | ||
756 | init_waitqueue_head(&bl_wq); | |
757 | ||
758 | mnt = rpc_get_mount(); | |
759 | if (IS_ERR(mnt)) { | |
760 | ret = PTR_ERR(mnt); | |
761 | goto out_remove; | |
762 | } | |
763 | ||
764 | ret = vfs_path_lookup(mnt->mnt_root, | |
765 | mnt, | |
766 | NFS_PIPE_DIRNAME, 0, &path); | |
767 | if (ret) | |
768 | goto out_remove; | |
769 | ||
770 | bl_device_pipe = rpc_mkpipe(path.dentry, "blocklayout", NULL, | |
771 | &bl_upcall_ops, 0); | |
772 | if (IS_ERR(bl_device_pipe)) { | |
773 | ret = PTR_ERR(bl_device_pipe); | |
774 | goto out_remove; | |
775 | } | |
776 | out: | |
777 | return ret; | |
778 | ||
779 | out_remove: | |
780 | pnfs_unregister_layoutdriver(&blocklayout_type); | |
155e7524 FI |
781 | return ret; |
782 | } | |
783 | ||
784 | static void __exit nfs4blocklayout_exit(void) | |
785 | { | |
786 | dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n", | |
787 | __func__); | |
788 | ||
789 | pnfs_unregister_layoutdriver(&blocklayout_type); | |
fe0a9b74 | 790 | rpc_unlink(bl_device_pipe); |
155e7524 FI |
791 | } |
792 | ||
793 | MODULE_ALIAS("nfs-layouttype4-3"); | |
794 | ||
795 | module_init(nfs4blocklayout_init); | |
796 | module_exit(nfs4blocklayout_exit); |