]>
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 */ |
88c9e421 | 38 | #include <linux/prefetch.h> |
6296556f | 39 | #include <linux/pagevec.h> |
155e7524 | 40 | |
10bd295a | 41 | #include "../pnfs.h" |
76e697ba | 42 | #include "../nfs4session.h" |
10bd295a | 43 | #include "../internal.h" |
155e7524 FI |
44 | #include "blocklayout.h" |
45 | ||
46 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
47 | ||
48 | MODULE_LICENSE("GPL"); | |
49 | MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>"); | |
50 | MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver"); | |
51 | ||
8067253c | 52 | static bool is_hole(struct pnfs_block_extent *be) |
9549ec01 | 53 | { |
8067253c CH |
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 | } | |
650e2d39 FI |
62 | } |
63 | ||
9549ec01 FI |
64 | /* The data we are handed might be spread across several bios. We need |
65 | * to track when the last one is finished. | |
66 | */ | |
67 | struct parallel_io { | |
68 | struct kref refcnt; | |
8067253c | 69 | void (*pnfs_callback) (void *data); |
9549ec01 FI |
70 | void *data; |
71 | }; | |
72 | ||
73 | static 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 | ||
85 | static inline void get_parallel(struct parallel_io *p) | |
86 | { | |
87 | kref_get(&p->refcnt); | |
88 | } | |
89 | ||
90 | static 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__); | |
8067253c | 95 | p->pnfs_callback(p->data); |
9549ec01 FI |
96 | kfree(p); |
97 | } | |
98 | ||
99 | static inline void put_parallel(struct parallel_io *p) | |
100 | { | |
101 | kref_put(&p->refcnt, destroy_parallel); | |
102 | } | |
103 | ||
104 | static struct bio * | |
105 | bl_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__, | |
4f024f37 KO |
110 | rw == READ ? "read" : "write", bio->bi_iter.bi_size, |
111 | (unsigned long long)bio->bi_iter.bi_sector); | |
9549ec01 FI |
112 | submit_bio(rw, bio); |
113 | } | |
114 | return NULL; | |
115 | } | |
116 | ||
117 | static struct bio *bl_alloc_init_bio(int npg, sector_t isect, | |
118 | struct pnfs_block_extent *be, | |
119 | void (*end_io)(struct bio *, int err), | |
120 | struct parallel_io *par) | |
121 | { | |
122 | struct bio *bio; | |
123 | ||
74a6eeb4 | 124 | npg = min(npg, BIO_MAX_PAGES); |
9549ec01 | 125 | bio = bio_alloc(GFP_NOIO, npg); |
74a6eeb4 PT |
126 | if (!bio && (current->flags & PF_MEMALLOC)) { |
127 | while (!bio && (npg /= 2)) | |
128 | bio = bio_alloc(GFP_NOIO, npg); | |
129 | } | |
9549ec01 | 130 | |
74a6eeb4 | 131 | if (bio) { |
4f024f37 KO |
132 | bio->bi_iter.bi_sector = isect - be->be_f_offset + |
133 | be->be_v_offset; | |
74a6eeb4 PT |
134 | bio->bi_bdev = be->be_mdev; |
135 | bio->bi_end_io = end_io; | |
136 | bio->bi_private = par; | |
137 | } | |
9549ec01 FI |
138 | return bio; |
139 | } | |
140 | ||
fe6e1e8d | 141 | static struct bio *do_add_page_to_bio(struct bio *bio, int npg, int rw, |
9549ec01 FI |
142 | sector_t isect, struct page *page, |
143 | struct pnfs_block_extent *be, | |
144 | void (*end_io)(struct bio *, int err), | |
fe6e1e8d PT |
145 | struct parallel_io *par, |
146 | unsigned int offset, int len) | |
9549ec01 | 147 | { |
fe6e1e8d PT |
148 | isect = isect + (offset >> SECTOR_SHIFT); |
149 | dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__, | |
150 | npg, rw, (unsigned long long)isect, offset, len); | |
9549ec01 FI |
151 | retry: |
152 | if (!bio) { | |
153 | bio = bl_alloc_init_bio(npg, isect, be, end_io, par); | |
154 | if (!bio) | |
155 | return ERR_PTR(-ENOMEM); | |
156 | } | |
fe6e1e8d | 157 | if (bio_add_page(bio, page, len, offset) < len) { |
9549ec01 FI |
158 | bio = bl_submit_bio(rw, bio); |
159 | goto retry; | |
160 | } | |
161 | return bio; | |
162 | } | |
163 | ||
9549ec01 FI |
164 | static void bl_end_io_read(struct bio *bio, int err) |
165 | { | |
166 | struct parallel_io *par = bio->bi_private; | |
9549ec01 | 167 | |
2c30c71b | 168 | if (err) { |
d45f60c6 | 169 | struct nfs_pgio_header *header = par->data; |
cd841605 FI |
170 | |
171 | if (!header->pnfs_error) | |
172 | header->pnfs_error = -EIO; | |
173 | pnfs_set_lo_fail(header->lseg); | |
9549ec01 | 174 | } |
8c792ea9 | 175 | |
9549ec01 FI |
176 | bio_put(bio); |
177 | put_parallel(par); | |
178 | } | |
179 | ||
180 | static void bl_read_cleanup(struct work_struct *work) | |
181 | { | |
182 | struct rpc_task *task; | |
d45f60c6 | 183 | struct nfs_pgio_header *hdr; |
9549ec01 FI |
184 | dprintk("%s enter\n", __func__); |
185 | task = container_of(work, struct rpc_task, u.tk_work); | |
d45f60c6 WAA |
186 | hdr = container_of(task, struct nfs_pgio_header, task); |
187 | pnfs_ld_read_done(hdr); | |
9549ec01 FI |
188 | } |
189 | ||
190 | static void | |
8067253c | 191 | bl_end_par_io_read(void *data) |
9549ec01 | 192 | { |
d45f60c6 | 193 | struct nfs_pgio_header *hdr = data; |
9549ec01 | 194 | |
d45f60c6 WAA |
195 | hdr->task.tk_status = hdr->pnfs_error; |
196 | INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup); | |
197 | schedule_work(&hdr->task.u.tk_work); | |
9549ec01 FI |
198 | } |
199 | ||
155e7524 | 200 | static enum pnfs_try_status |
8067253c | 201 | bl_read_pagelist(struct nfs_pgio_header *header) |
155e7524 | 202 | { |
8067253c | 203 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
9549ec01 | 204 | struct bio *bio = NULL; |
8067253c | 205 | struct pnfs_block_extent be; |
9549ec01 FI |
206 | sector_t isect, extent_length = 0; |
207 | struct parallel_io *par; | |
8067253c CH |
208 | loff_t f_offset = header->args.offset; |
209 | size_t bytes_left = header->args.count; | |
f742dc4a | 210 | unsigned int pg_offset, pg_len; |
8067253c CH |
211 | struct page **pages = header->args.pages; |
212 | int pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT; | |
f742dc4a | 213 | const bool is_dio = (header->dreq != NULL); |
be98fd0a | 214 | struct blk_plug plug; |
8067253c | 215 | int i; |
9549ec01 | 216 | |
6f00866d | 217 | dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__, |
8067253c CH |
218 | header->page_array.npages, f_offset, |
219 | (unsigned int)header->args.count); | |
9549ec01 | 220 | |
8067253c | 221 | par = alloc_parallel(header); |
9549ec01 | 222 | if (!par) |
8067253c | 223 | return PNFS_NOT_ATTEMPTED; |
9549ec01 | 224 | par->pnfs_callback = bl_end_par_io_read; |
9549ec01 | 225 | |
be98fd0a CH |
226 | blk_start_plug(&plug); |
227 | ||
9549ec01 FI |
228 | isect = (sector_t) (f_offset >> SECTOR_SHIFT); |
229 | /* Code assumes extents are page-aligned */ | |
8067253c | 230 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 231 | if (extent_length <= 0) { |
9549ec01 | 232 | /* We've used up the previous extent */ |
9549ec01 | 233 | bio = bl_submit_bio(READ, bio); |
8067253c | 234 | |
9549ec01 | 235 | /* Get the next one */ |
8067253c | 236 | if (!ext_tree_lookup(bl, isect, &be, false)) { |
cd841605 | 237 | header->pnfs_error = -EIO; |
9549ec01 FI |
238 | goto out; |
239 | } | |
8067253c | 240 | extent_length = be.be_length - (isect - be.be_f_offset); |
9549ec01 | 241 | } |
f742dc4a | 242 | |
3a6fd1f0 | 243 | pg_offset = f_offset & ~PAGE_CACHE_MASK; |
f742dc4a | 244 | if (is_dio) { |
f742dc4a PT |
245 | if (pg_offset + bytes_left > PAGE_CACHE_SIZE) |
246 | pg_len = PAGE_CACHE_SIZE - pg_offset; | |
247 | else | |
248 | pg_len = bytes_left; | |
249 | ||
250 | f_offset += pg_len; | |
251 | bytes_left -= pg_len; | |
252 | isect += (pg_offset >> SECTOR_SHIFT); | |
921b81a8 | 253 | extent_length -= (pg_offset >> SECTOR_SHIFT); |
f742dc4a | 254 | } else { |
3a6fd1f0 | 255 | BUG_ON(pg_offset != 0); |
f742dc4a PT |
256 | pg_len = PAGE_CACHE_SIZE; |
257 | } | |
258 | ||
8067253c | 259 | if (is_hole(&be)) { |
9549ec01 FI |
260 | bio = bl_submit_bio(READ, bio); |
261 | /* Fill hole w/ zeroes w/o accessing device */ | |
262 | dprintk("%s Zeroing page for hole\n", __func__); | |
f742dc4a | 263 | zero_user_segment(pages[i], pg_offset, pg_len); |
9549ec01 | 264 | } else { |
823b0c9d | 265 | bio = do_add_page_to_bio(bio, |
8067253c | 266 | header->page_array.npages - i, |
30dd374f | 267 | READ, |
8067253c | 268 | isect, pages[i], &be, |
f742dc4a PT |
269 | bl_end_io_read, par, |
270 | pg_offset, pg_len); | |
9549ec01 | 271 | if (IS_ERR(bio)) { |
cd841605 | 272 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 273 | bio = NULL; |
9549ec01 FI |
274 | goto out; |
275 | } | |
276 | } | |
f742dc4a | 277 | isect += (pg_len >> SECTOR_SHIFT); |
921b81a8 | 278 | extent_length -= (pg_len >> SECTOR_SHIFT); |
9549ec01 | 279 | } |
cd841605 | 280 | if ((isect << SECTOR_SHIFT) >= header->inode->i_size) { |
8067253c CH |
281 | header->res.eof = 1; |
282 | header->res.count = header->inode->i_size - header->args.offset; | |
9549ec01 | 283 | } else { |
8067253c | 284 | header->res.count = (isect << SECTOR_SHIFT) - header->args.offset; |
9549ec01 FI |
285 | } |
286 | out: | |
9549ec01 | 287 | bl_submit_bio(READ, bio); |
be98fd0a | 288 | blk_finish_plug(&plug); |
9549ec01 FI |
289 | put_parallel(par); |
290 | return PNFS_ATTEMPTED; | |
31e6306a FI |
291 | } |
292 | ||
650e2d39 FI |
293 | static void bl_end_io_write(struct bio *bio, int err) |
294 | { | |
295 | struct parallel_io *par = bio->bi_private; | |
296 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
d45f60c6 | 297 | struct nfs_pgio_header *header = par->data; |
650e2d39 FI |
298 | |
299 | if (!uptodate) { | |
cd841605 FI |
300 | if (!header->pnfs_error) |
301 | header->pnfs_error = -EIO; | |
302 | pnfs_set_lo_fail(header->lseg); | |
650e2d39 FI |
303 | } |
304 | bio_put(bio); | |
305 | put_parallel(par); | |
306 | } | |
307 | ||
308 | /* Function scheduled for call during bl_end_par_io_write, | |
309 | * it marks sectors as written and extends the commitlist. | |
310 | */ | |
311 | static void bl_write_cleanup(struct work_struct *work) | |
312 | { | |
8067253c CH |
313 | struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work); |
314 | struct nfs_pgio_header *hdr = | |
315 | container_of(task, struct nfs_pgio_header, task); | |
316 | ||
650e2d39 | 317 | dprintk("%s enter\n", __func__); |
8067253c | 318 | |
d45f60c6 | 319 | if (likely(!hdr->pnfs_error)) { |
8067253c CH |
320 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg); |
321 | u64 start = hdr->args.offset & (loff_t)PAGE_CACHE_MASK; | |
322 | u64 end = (hdr->args.offset + hdr->args.count + | |
323 | PAGE_CACHE_SIZE - 1) & (loff_t)PAGE_CACHE_MASK; | |
324 | ||
325 | ext_tree_mark_written(bl, start >> SECTOR_SHIFT, | |
326 | (end - start) >> SECTOR_SHIFT); | |
31e6306a | 327 | } |
8067253c | 328 | |
d45f60c6 | 329 | pnfs_ld_write_done(hdr); |
650e2d39 FI |
330 | } |
331 | ||
332 | /* Called when last of bios associated with a bl_write_pagelist call finishes */ | |
8067253c | 333 | static void bl_end_par_io_write(void *data) |
650e2d39 | 334 | { |
d45f60c6 | 335 | struct nfs_pgio_header *hdr = data; |
650e2d39 | 336 | |
d45f60c6 | 337 | hdr->task.tk_status = hdr->pnfs_error; |
c65e6254 | 338 | hdr->verf.committed = NFS_FILE_SYNC; |
d45f60c6 WAA |
339 | INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup); |
340 | schedule_work(&hdr->task.u.tk_work); | |
650e2d39 FI |
341 | } |
342 | ||
155e7524 | 343 | static enum pnfs_try_status |
d45f60c6 | 344 | bl_write_pagelist(struct nfs_pgio_header *header, int sync) |
155e7524 | 345 | { |
8067253c | 346 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
650e2d39 | 347 | struct bio *bio = NULL; |
8067253c | 348 | struct pnfs_block_extent be; |
3a6fd1f0 | 349 | sector_t isect, extent_length = 0; |
96c9eae6 | 350 | struct parallel_io *par = NULL; |
d45f60c6 WAA |
351 | loff_t offset = header->args.offset; |
352 | size_t count = header->args.count; | |
d45f60c6 | 353 | struct page **pages = header->args.pages; |
3a6fd1f0 | 354 | int pg_index = pg_index = header->args.pgbase >> PAGE_CACHE_SHIFT; |
be98fd0a | 355 | struct blk_plug plug; |
8067253c | 356 | int i; |
650e2d39 FI |
357 | |
358 | dprintk("%s enter, %Zu@%lld\n", __func__, count, offset); | |
96c9eae6 | 359 | |
d45f60c6 | 360 | /* At this point, header->page_aray is a (sequential) list of nfs_pages. |
71cdd40f PT |
361 | * We want to write each, and if there is an error set pnfs_error |
362 | * to have it redone using nfs. | |
650e2d39 | 363 | */ |
d45f60c6 | 364 | par = alloc_parallel(header); |
650e2d39 | 365 | if (!par) |
8067253c | 366 | return PNFS_NOT_ATTEMPTED; |
650e2d39 | 367 | par->pnfs_callback = bl_end_par_io_write; |
650e2d39 | 368 | |
3a6fd1f0 | 369 | blk_start_plug(&plug); |
71cdd40f | 370 | |
3a6fd1f0 CH |
371 | /* we always write out the whole page */ |
372 | offset = offset & (loff_t)PAGE_CACHE_MASK; | |
373 | isect = offset >> SECTOR_SHIFT; | |
71cdd40f | 374 | |
d45f60c6 | 375 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 376 | if (extent_length <= 0) { |
650e2d39 | 377 | /* We've used up the previous extent */ |
650e2d39 FI |
378 | bio = bl_submit_bio(WRITE, bio); |
379 | /* Get the next one */ | |
8067253c | 380 | if (!ext_tree_lookup(bl, isect, &be, true)) { |
cd841605 | 381 | header->pnfs_error = -EINVAL; |
650e2d39 FI |
382 | goto out; |
383 | } | |
fe6e1e8d | 384 | |
8067253c | 385 | extent_length = be.be_length - (isect - be.be_f_offset); |
71cdd40f | 386 | } |
fe6e1e8d | 387 | |
d45f60c6 | 388 | bio = do_add_page_to_bio(bio, header->page_array.npages - i, |
8067253c | 389 | WRITE, isect, pages[i], &be, |
fe6e1e8d | 390 | bl_end_io_write, par, |
3a6fd1f0 | 391 | 0, PAGE_CACHE_SIZE); |
71cdd40f | 392 | if (IS_ERR(bio)) { |
cd841605 | 393 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 394 | bio = NULL; |
71cdd40f | 395 | goto out; |
650e2d39 | 396 | } |
3a6fd1f0 CH |
397 | offset += PAGE_CACHE_SIZE; |
398 | count -= PAGE_CACHE_SIZE; | |
650e2d39 FI |
399 | isect += PAGE_CACHE_SECTORS; |
400 | extent_length -= PAGE_CACHE_SECTORS; | |
401 | } | |
71cdd40f | 402 | |
d45f60c6 | 403 | header->res.count = header->args.count; |
650e2d39 | 404 | out: |
650e2d39 | 405 | bl_submit_bio(WRITE, bio); |
be98fd0a | 406 | blk_finish_plug(&plug); |
650e2d39 FI |
407 | put_parallel(par); |
408 | return PNFS_ATTEMPTED; | |
155e7524 FI |
409 | } |
410 | ||
411 | static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
412 | { | |
413 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
8067253c | 414 | int err; |
155e7524 FI |
415 | |
416 | dprintk("%s enter\n", __func__); | |
8067253c CH |
417 | |
418 | err = ext_tree_remove(bl, true, 0, LLONG_MAX); | |
419 | WARN_ON(err); | |
420 | ||
155e7524 FI |
421 | kfree(bl); |
422 | } | |
423 | ||
424 | static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode, | |
425 | gfp_t gfp_flags) | |
426 | { | |
427 | struct pnfs_block_layout *bl; | |
428 | ||
429 | dprintk("%s enter\n", __func__); | |
430 | bl = kzalloc(sizeof(*bl), gfp_flags); | |
431 | if (!bl) | |
432 | return NULL; | |
8067253c CH |
433 | |
434 | bl->bl_ext_rw = RB_ROOT; | |
435 | bl->bl_ext_ro = RB_ROOT; | |
155e7524 | 436 | spin_lock_init(&bl->bl_ext_lock); |
8067253c | 437 | |
155e7524 FI |
438 | return &bl->bl_layout; |
439 | } | |
440 | ||
a60d2ebd | 441 | static void bl_free_lseg(struct pnfs_layout_segment *lseg) |
155e7524 | 442 | { |
a60d2ebd FI |
443 | dprintk("%s enter\n", __func__); |
444 | kfree(lseg); | |
155e7524 FI |
445 | } |
446 | ||
a60d2ebd FI |
447 | /* We pretty much ignore lseg, and store all data layout wide, so we |
448 | * can correctly merge. | |
449 | */ | |
450 | static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo, | |
451 | struct nfs4_layoutget_res *lgr, | |
452 | gfp_t gfp_flags) | |
155e7524 | 453 | { |
a60d2ebd FI |
454 | struct pnfs_layout_segment *lseg; |
455 | int status; | |
456 | ||
457 | dprintk("%s enter\n", __func__); | |
458 | lseg = kzalloc(sizeof(*lseg), gfp_flags); | |
459 | if (!lseg) | |
460 | return ERR_PTR(-ENOMEM); | |
461 | status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags); | |
462 | if (status) { | |
463 | /* We don't want to call the full-blown bl_free_lseg, | |
464 | * since on error extents were not touched. | |
465 | */ | |
466 | kfree(lseg); | |
467 | return ERR_PTR(status); | |
468 | } | |
469 | return lseg; | |
155e7524 FI |
470 | } |
471 | ||
71d5b763 CH |
472 | static void |
473 | bl_return_range(struct pnfs_layout_hdr *lo, | |
474 | struct pnfs_layout_range *range) | |
475 | { | |
476 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
477 | sector_t offset = range->offset >> SECTOR_SHIFT, end; | |
478 | int err; | |
479 | ||
480 | if (range->offset % 8) { | |
481 | dprintk("%s: offset %lld not block size aligned\n", | |
482 | __func__, range->offset); | |
483 | return; | |
484 | } | |
485 | ||
486 | if (range->length != NFS4_MAX_UINT64) { | |
487 | if (range->length % 8) { | |
488 | dprintk("%s: length %lld not block size aligned\n", | |
489 | __func__, range->length); | |
490 | return; | |
491 | } | |
492 | ||
493 | end = offset + (range->length >> SECTOR_SHIFT); | |
494 | } else { | |
495 | end = round_down(NFS4_MAX_UINT64, PAGE_SIZE); | |
496 | } | |
497 | ||
498 | err = ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end); | |
499 | } | |
500 | ||
155e7524 FI |
501 | static void |
502 | bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr, | |
503 | const struct nfs4_layoutcommit_args *arg) | |
504 | { | |
90ace12a | 505 | dprintk("%s enter\n", __func__); |
8067253c | 506 | ext_tree_encode_commit(BLK_LO2EXT(lo), xdr); |
155e7524 FI |
507 | } |
508 | ||
509 | static void | |
510 | bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata) | |
511 | { | |
b2be7811 FI |
512 | struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout; |
513 | ||
514 | dprintk("%s enter\n", __func__); | |
8067253c | 515 | ext_tree_mark_committed(BLK_LO2EXT(lo), lcdata->res.status); |
155e7524 FI |
516 | } |
517 | ||
2f9fd182 FI |
518 | static void free_blk_mountid(struct block_mount_id *mid) |
519 | { | |
520 | if (mid) { | |
93a3844e PT |
521 | struct pnfs_block_dev *dev, *tmp; |
522 | ||
523 | /* No need to take bm_lock as we are last user freeing bm_devlist */ | |
524 | list_for_each_entry_safe(dev, tmp, &mid->bm_devlist, bm_node) { | |
2f9fd182 FI |
525 | list_del(&dev->bm_node); |
526 | bl_free_block_dev(dev); | |
527 | } | |
2f9fd182 FI |
528 | kfree(mid); |
529 | } | |
530 | } | |
531 | ||
78e4e05c | 532 | /* This is mostly copied from the filelayout_get_device_info function. |
2f9fd182 FI |
533 | * It seems much of this should be at the generic pnfs level. |
534 | */ | |
535 | static struct pnfs_block_dev * | |
536 | nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh, | |
537 | struct nfs4_deviceid *d_id) | |
538 | { | |
539 | struct pnfs_device *dev; | |
516f2e24 | 540 | struct pnfs_block_dev *rv; |
2f9fd182 FI |
541 | u32 max_resp_sz; |
542 | int max_pages; | |
543 | struct page **pages = NULL; | |
544 | int i, rc; | |
545 | ||
546 | /* | |
547 | * Use the session max response size as the basis for setting | |
548 | * GETDEVICEINFO's maxcount | |
549 | */ | |
550 | max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz; | |
10bd295a | 551 | max_pages = nfs_page_array_len(0, max_resp_sz); |
2f9fd182 FI |
552 | dprintk("%s max_resp_sz %u max_pages %d\n", |
553 | __func__, max_resp_sz, max_pages); | |
554 | ||
555 | dev = kmalloc(sizeof(*dev), GFP_NOFS); | |
556 | if (!dev) { | |
557 | dprintk("%s kmalloc failed\n", __func__); | |
516f2e24 | 558 | return ERR_PTR(-ENOMEM); |
2f9fd182 FI |
559 | } |
560 | ||
f15b5041 | 561 | pages = kcalloc(max_pages, sizeof(struct page *), GFP_NOFS); |
2f9fd182 FI |
562 | if (pages == NULL) { |
563 | kfree(dev); | |
516f2e24 | 564 | return ERR_PTR(-ENOMEM); |
2f9fd182 FI |
565 | } |
566 | for (i = 0; i < max_pages; i++) { | |
567 | pages[i] = alloc_page(GFP_NOFS); | |
516f2e24 JR |
568 | if (!pages[i]) { |
569 | rv = ERR_PTR(-ENOMEM); | |
2f9fd182 | 570 | goto out_free; |
516f2e24 | 571 | } |
2f9fd182 FI |
572 | } |
573 | ||
574 | memcpy(&dev->dev_id, d_id, sizeof(*d_id)); | |
575 | dev->layout_type = LAYOUT_BLOCK_VOLUME; | |
576 | dev->pages = pages; | |
577 | dev->pgbase = 0; | |
578 | dev->pglen = PAGE_SIZE * max_pages; | |
579 | dev->mincount = 0; | |
968fe252 | 580 | dev->maxcount = max_resp_sz - nfs41_maxgetdevinfo_overhead; |
2f9fd182 FI |
581 | |
582 | dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data); | |
cd5875fe | 583 | rc = nfs4_proc_getdeviceinfo(server, dev, NULL); |
2f9fd182 | 584 | dprintk("%s getdevice info returns %d\n", __func__, rc); |
516f2e24 JR |
585 | if (rc) { |
586 | rv = ERR_PTR(rc); | |
2f9fd182 | 587 | goto out_free; |
516f2e24 | 588 | } |
2f9fd182 FI |
589 | |
590 | rv = nfs4_blk_decode_device(server, dev); | |
591 | out_free: | |
592 | for (i = 0; i < max_pages; i++) | |
593 | __free_page(pages[i]); | |
594 | kfree(pages); | |
595 | kfree(dev); | |
596 | return rv; | |
597 | } | |
598 | ||
155e7524 FI |
599 | static int |
600 | bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh) | |
601 | { | |
2f9fd182 FI |
602 | struct block_mount_id *b_mt_id = NULL; |
603 | struct pnfs_devicelist *dlist = NULL; | |
604 | struct pnfs_block_dev *bdev; | |
605 | LIST_HEAD(block_disklist); | |
516f2e24 | 606 | int status, i; |
2f9fd182 | 607 | |
155e7524 | 608 | dprintk("%s enter\n", __func__); |
2f9fd182 FI |
609 | |
610 | if (server->pnfs_blksize == 0) { | |
611 | dprintk("%s Server did not return blksize\n", __func__); | |
612 | return -EINVAL; | |
613 | } | |
e3aaf7f2 CH |
614 | if (server->pnfs_blksize > PAGE_SIZE) { |
615 | printk(KERN_ERR "%s: pNFS blksize %d not supported.\n", | |
616 | __func__, server->pnfs_blksize); | |
617 | return -EINVAL; | |
618 | } | |
619 | ||
2f9fd182 FI |
620 | b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS); |
621 | if (!b_mt_id) { | |
622 | status = -ENOMEM; | |
623 | goto out_error; | |
624 | } | |
625 | /* Initialize nfs4 block layout mount id */ | |
626 | spin_lock_init(&b_mt_id->bm_lock); | |
627 | INIT_LIST_HEAD(&b_mt_id->bm_devlist); | |
628 | ||
629 | dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS); | |
630 | if (!dlist) { | |
631 | status = -ENOMEM; | |
632 | goto out_error; | |
633 | } | |
634 | dlist->eof = 0; | |
635 | while (!dlist->eof) { | |
636 | status = nfs4_proc_getdevicelist(server, fh, dlist); | |
637 | if (status) | |
638 | goto out_error; | |
639 | dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n", | |
640 | __func__, dlist->num_devs, dlist->eof); | |
641 | for (i = 0; i < dlist->num_devs; i++) { | |
642 | bdev = nfs4_blk_get_deviceinfo(server, fh, | |
643 | &dlist->dev_id[i]); | |
516f2e24 JR |
644 | if (IS_ERR(bdev)) { |
645 | status = PTR_ERR(bdev); | |
2f9fd182 FI |
646 | goto out_error; |
647 | } | |
648 | spin_lock(&b_mt_id->bm_lock); | |
649 | list_add(&bdev->bm_node, &b_mt_id->bm_devlist); | |
650 | spin_unlock(&b_mt_id->bm_lock); | |
651 | } | |
652 | } | |
653 | dprintk("%s SUCCESS\n", __func__); | |
654 | server->pnfs_ld_data = b_mt_id; | |
655 | ||
656 | out_return: | |
657 | kfree(dlist); | |
658 | return status; | |
659 | ||
660 | out_error: | |
661 | free_blk_mountid(b_mt_id); | |
662 | goto out_return; | |
155e7524 FI |
663 | } |
664 | ||
665 | static int | |
666 | bl_clear_layoutdriver(struct nfs_server *server) | |
667 | { | |
2f9fd182 FI |
668 | struct block_mount_id *b_mt_id = server->pnfs_ld_data; |
669 | ||
155e7524 | 670 | dprintk("%s enter\n", __func__); |
2f9fd182 FI |
671 | free_blk_mountid(b_mt_id); |
672 | dprintk("%s RETURNS\n", __func__); | |
155e7524 FI |
673 | return 0; |
674 | } | |
675 | ||
f742dc4a | 676 | static bool |
3a6fd1f0 CH |
677 | is_aligned_req(struct nfs_pageio_descriptor *pgio, |
678 | struct nfs_page *req, unsigned int alignment) | |
f742dc4a | 679 | { |
3a6fd1f0 CH |
680 | /* |
681 | * Always accept buffered writes, higher layers take care of the | |
682 | * right alignment. | |
683 | */ | |
684 | if (pgio->pg_dreq == NULL) | |
685 | return true; | |
686 | ||
687 | if (!IS_ALIGNED(req->wb_offset, alignment)) | |
688 | return false; | |
689 | ||
690 | if (IS_ALIGNED(req->wb_bytes, alignment)) | |
691 | return true; | |
692 | ||
693 | if (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode)) { | |
694 | /* | |
695 | * If the write goes up to the inode size, just write | |
696 | * the full page. Data past the inode size is | |
697 | * guaranteed to be zeroed by the higher level client | |
698 | * code, and this behaviour is mandated by RFC 5663 | |
699 | * section 2.3.2. | |
700 | */ | |
701 | return true; | |
702 | } | |
703 | ||
704 | return false; | |
f742dc4a PT |
705 | } |
706 | ||
707 | static void | |
708 | bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
709 | { | |
3a6fd1f0 | 710 | if (!is_aligned_req(pgio, req, SECTOR_SIZE)) { |
f742dc4a | 711 | nfs_pageio_reset_read_mds(pgio); |
3a6fd1f0 CH |
712 | return; |
713 | } | |
714 | ||
715 | pnfs_generic_pg_init_read(pgio, req); | |
f742dc4a PT |
716 | } |
717 | ||
b4fdac1a WAA |
718 | /* |
719 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
720 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
721 | */ | |
722 | static size_t | |
f742dc4a PT |
723 | bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
724 | struct nfs_page *req) | |
725 | { | |
3a6fd1f0 | 726 | if (!is_aligned_req(pgio, req, SECTOR_SIZE)) |
b4fdac1a | 727 | return 0; |
f742dc4a PT |
728 | return pnfs_generic_pg_test(pgio, prev, req); |
729 | } | |
730 | ||
6296556f PT |
731 | /* |
732 | * Return the number of contiguous bytes for a given inode | |
733 | * starting at page frame idx. | |
734 | */ | |
735 | static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx) | |
736 | { | |
737 | struct address_space *mapping = inode->i_mapping; | |
738 | pgoff_t end; | |
739 | ||
740 | /* Optimize common case that writes from 0 to end of file */ | |
741 | end = DIV_ROUND_UP(i_size_read(inode), PAGE_CACHE_SIZE); | |
742 | if (end != NFS_I(inode)->npages) { | |
743 | rcu_read_lock(); | |
e7b563bb | 744 | end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX); |
6296556f PT |
745 | rcu_read_unlock(); |
746 | } | |
747 | ||
748 | if (!end) | |
749 | return i_size_read(inode) - (idx << PAGE_CACHE_SHIFT); | |
750 | else | |
751 | return (end - idx) << PAGE_CACHE_SHIFT; | |
752 | } | |
753 | ||
6f018efa | 754 | static void |
96c9eae6 PT |
755 | bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) |
756 | { | |
3a6fd1f0 CH |
757 | u64 wb_size; |
758 | ||
759 | if (!is_aligned_req(pgio, req, PAGE_SIZE)) { | |
96c9eae6 | 760 | nfs_pageio_reset_write_mds(pgio); |
3a6fd1f0 | 761 | return; |
6296556f | 762 | } |
3a6fd1f0 CH |
763 | |
764 | if (pgio->pg_dreq == NULL) | |
765 | wb_size = pnfs_num_cont_bytes(pgio->pg_inode, | |
766 | req->wb_index); | |
767 | else | |
768 | wb_size = nfs_dreq_bytes_left(pgio->pg_dreq); | |
769 | ||
770 | pnfs_generic_pg_init_write(pgio, req, wb_size); | |
96c9eae6 PT |
771 | } |
772 | ||
b4fdac1a WAA |
773 | /* |
774 | * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number | |
775 | * of bytes (maximum @req->wb_bytes) that can be coalesced. | |
776 | */ | |
777 | static size_t | |
96c9eae6 PT |
778 | bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
779 | struct nfs_page *req) | |
780 | { | |
3a6fd1f0 | 781 | if (!is_aligned_req(pgio, req, PAGE_SIZE)) |
b4fdac1a | 782 | return 0; |
96c9eae6 PT |
783 | return pnfs_generic_pg_test(pgio, prev, req); |
784 | } | |
785 | ||
e9643fe8 | 786 | static const struct nfs_pageio_ops bl_pg_read_ops = { |
f742dc4a PT |
787 | .pg_init = bl_pg_init_read, |
788 | .pg_test = bl_pg_test_read, | |
e9643fe8 BH |
789 | .pg_doio = pnfs_generic_pg_readpages, |
790 | }; | |
791 | ||
792 | static const struct nfs_pageio_ops bl_pg_write_ops = { | |
96c9eae6 PT |
793 | .pg_init = bl_pg_init_write, |
794 | .pg_test = bl_pg_test_write, | |
e9643fe8 BH |
795 | .pg_doio = pnfs_generic_pg_writepages, |
796 | }; | |
797 | ||
155e7524 FI |
798 | static struct pnfs_layoutdriver_type blocklayout_type = { |
799 | .id = LAYOUT_BLOCK_VOLUME, | |
800 | .name = "LAYOUT_BLOCK_VOLUME", | |
5a12cca6 | 801 | .owner = THIS_MODULE, |
3a6fd1f0 | 802 | .flags = PNFS_READ_WHOLE_PAGE, |
155e7524 FI |
803 | .read_pagelist = bl_read_pagelist, |
804 | .write_pagelist = bl_write_pagelist, | |
805 | .alloc_layout_hdr = bl_alloc_layout_hdr, | |
806 | .free_layout_hdr = bl_free_layout_hdr, | |
807 | .alloc_lseg = bl_alloc_lseg, | |
808 | .free_lseg = bl_free_lseg, | |
71d5b763 | 809 | .return_range = bl_return_range, |
155e7524 FI |
810 | .encode_layoutcommit = bl_encode_layoutcommit, |
811 | .cleanup_layoutcommit = bl_cleanup_layoutcommit, | |
812 | .set_layoutdriver = bl_set_layoutdriver, | |
813 | .clear_layoutdriver = bl_clear_layoutdriver, | |
e9643fe8 BH |
814 | .pg_read_ops = &bl_pg_read_ops, |
815 | .pg_write_ops = &bl_pg_write_ops, | |
155e7524 FI |
816 | }; |
817 | ||
fe0a9b74 | 818 | static const struct rpc_pipe_ops bl_upcall_ops = { |
c1225158 | 819 | .upcall = rpc_pipe_generic_upcall, |
fe0a9b74 JR |
820 | .downcall = bl_pipe_downcall, |
821 | .destroy_msg = bl_pipe_destroy_msg, | |
822 | }; | |
823 | ||
332dfab6 SK |
824 | static struct dentry *nfs4blocklayout_register_sb(struct super_block *sb, |
825 | struct rpc_pipe *pipe) | |
826 | { | |
827 | struct dentry *dir, *dentry; | |
828 | ||
829 | dir = rpc_d_lookup_sb(sb, NFS_PIPE_DIRNAME); | |
830 | if (dir == NULL) | |
831 | return ERR_PTR(-ENOENT); | |
832 | dentry = rpc_mkpipe_dentry(dir, "blocklayout", NULL, pipe); | |
833 | dput(dir); | |
834 | return dentry; | |
835 | } | |
836 | ||
837 | static void nfs4blocklayout_unregister_sb(struct super_block *sb, | |
838 | struct rpc_pipe *pipe) | |
839 | { | |
840 | if (pipe->dentry) | |
841 | rpc_unlink(pipe->dentry); | |
842 | } | |
843 | ||
627f3066 SK |
844 | static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event, |
845 | void *ptr) | |
846 | { | |
847 | struct super_block *sb = ptr; | |
848 | struct net *net = sb->s_fs_info; | |
849 | struct nfs_net *nn = net_generic(net, nfs_net_id); | |
850 | struct dentry *dentry; | |
851 | int ret = 0; | |
852 | ||
853 | if (!try_module_get(THIS_MODULE)) | |
854 | return 0; | |
855 | ||
856 | if (nn->bl_device_pipe == NULL) { | |
857 | module_put(THIS_MODULE); | |
858 | return 0; | |
859 | } | |
860 | ||
861 | switch (event) { | |
862 | case RPC_PIPEFS_MOUNT: | |
863 | dentry = nfs4blocklayout_register_sb(sb, nn->bl_device_pipe); | |
864 | if (IS_ERR(dentry)) { | |
865 | ret = PTR_ERR(dentry); | |
866 | break; | |
867 | } | |
868 | nn->bl_device_pipe->dentry = dentry; | |
869 | break; | |
870 | case RPC_PIPEFS_UMOUNT: | |
871 | if (nn->bl_device_pipe->dentry) | |
872 | nfs4blocklayout_unregister_sb(sb, nn->bl_device_pipe); | |
873 | break; | |
874 | default: | |
875 | ret = -ENOTSUPP; | |
876 | break; | |
877 | } | |
878 | module_put(THIS_MODULE); | |
879 | return ret; | |
880 | } | |
881 | ||
882 | static struct notifier_block nfs4blocklayout_block = { | |
883 | .notifier_call = rpc_pipefs_event, | |
884 | }; | |
885 | ||
332dfab6 SK |
886 | static struct dentry *nfs4blocklayout_register_net(struct net *net, |
887 | struct rpc_pipe *pipe) | |
888 | { | |
889 | struct super_block *pipefs_sb; | |
890 | struct dentry *dentry; | |
891 | ||
892 | pipefs_sb = rpc_get_sb_net(net); | |
893 | if (!pipefs_sb) | |
2561d618 | 894 | return NULL; |
332dfab6 SK |
895 | dentry = nfs4blocklayout_register_sb(pipefs_sb, pipe); |
896 | rpc_put_sb_net(net); | |
897 | return dentry; | |
898 | } | |
899 | ||
900 | static void nfs4blocklayout_unregister_net(struct net *net, | |
901 | struct rpc_pipe *pipe) | |
902 | { | |
903 | struct super_block *pipefs_sb; | |
904 | ||
905 | pipefs_sb = rpc_get_sb_net(net); | |
906 | if (pipefs_sb) { | |
907 | nfs4blocklayout_unregister_sb(pipefs_sb, pipe); | |
908 | rpc_put_sb_net(net); | |
909 | } | |
910 | } | |
911 | ||
9e2e74db SK |
912 | static int nfs4blocklayout_net_init(struct net *net) |
913 | { | |
914 | struct nfs_net *nn = net_generic(net, nfs_net_id); | |
915 | struct dentry *dentry; | |
916 | ||
5ffaf855 | 917 | init_waitqueue_head(&nn->bl_wq); |
9e2e74db SK |
918 | nn->bl_device_pipe = rpc_mkpipe_data(&bl_upcall_ops, 0); |
919 | if (IS_ERR(nn->bl_device_pipe)) | |
920 | return PTR_ERR(nn->bl_device_pipe); | |
921 | dentry = nfs4blocklayout_register_net(net, nn->bl_device_pipe); | |
922 | if (IS_ERR(dentry)) { | |
923 | rpc_destroy_pipe_data(nn->bl_device_pipe); | |
924 | return PTR_ERR(dentry); | |
925 | } | |
926 | nn->bl_device_pipe->dentry = dentry; | |
927 | return 0; | |
928 | } | |
929 | ||
930 | static void nfs4blocklayout_net_exit(struct net *net) | |
931 | { | |
932 | struct nfs_net *nn = net_generic(net, nfs_net_id); | |
933 | ||
934 | nfs4blocklayout_unregister_net(net, nn->bl_device_pipe); | |
935 | rpc_destroy_pipe_data(nn->bl_device_pipe); | |
936 | nn->bl_device_pipe = NULL; | |
937 | } | |
938 | ||
939 | static struct pernet_operations nfs4blocklayout_net_ops = { | |
940 | .init = nfs4blocklayout_net_init, | |
941 | .exit = nfs4blocklayout_net_exit, | |
942 | }; | |
943 | ||
155e7524 FI |
944 | static int __init nfs4blocklayout_init(void) |
945 | { | |
946 | int ret; | |
947 | ||
948 | dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__); | |
949 | ||
950 | ret = pnfs_register_layoutdriver(&blocklayout_type); | |
fe0a9b74 JR |
951 | if (ret) |
952 | goto out; | |
953 | ||
627f3066 | 954 | ret = rpc_pipefs_notifier_register(&nfs4blocklayout_block); |
9e2e74db SK |
955 | if (ret) |
956 | goto out_remove; | |
627f3066 SK |
957 | ret = register_pernet_subsys(&nfs4blocklayout_net_ops); |
958 | if (ret) | |
959 | goto out_notifier; | |
fe0a9b74 JR |
960 | out: |
961 | return ret; | |
962 | ||
627f3066 SK |
963 | out_notifier: |
964 | rpc_pipefs_notifier_unregister(&nfs4blocklayout_block); | |
fe0a9b74 JR |
965 | out_remove: |
966 | pnfs_unregister_layoutdriver(&blocklayout_type); | |
155e7524 FI |
967 | return ret; |
968 | } | |
969 | ||
970 | static void __exit nfs4blocklayout_exit(void) | |
971 | { | |
972 | dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n", | |
973 | __func__); | |
974 | ||
627f3066 | 975 | rpc_pipefs_notifier_unregister(&nfs4blocklayout_block); |
9e2e74db | 976 | unregister_pernet_subsys(&nfs4blocklayout_net_ops); |
155e7524 FI |
977 | pnfs_unregister_layoutdriver(&blocklayout_type); |
978 | } | |
979 | ||
980 | MODULE_ALIAS("nfs-layouttype4-3"); | |
981 | ||
982 | module_init(nfs4blocklayout_init); | |
983 | module_exit(nfs4blocklayout_exit); |