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 | ||
5c83746a CH |
117 | static struct bio * |
118 | bl_alloc_init_bio(int npg, struct block_device *bdev, sector_t disk_sector, | |
4246a0b6 | 119 | bio_end_io_t end_io, struct parallel_io *par) |
9549ec01 FI |
120 | { |
121 | struct bio *bio; | |
122 | ||
74a6eeb4 | 123 | npg = min(npg, BIO_MAX_PAGES); |
9549ec01 | 124 | bio = bio_alloc(GFP_NOIO, npg); |
74a6eeb4 PT |
125 | if (!bio && (current->flags & PF_MEMALLOC)) { |
126 | while (!bio && (npg /= 2)) | |
127 | bio = bio_alloc(GFP_NOIO, npg); | |
128 | } | |
9549ec01 | 129 | |
74a6eeb4 | 130 | if (bio) { |
5c83746a CH |
131 | bio->bi_iter.bi_sector = disk_sector; |
132 | bio->bi_bdev = bdev; | |
74a6eeb4 PT |
133 | bio->bi_end_io = end_io; |
134 | bio->bi_private = par; | |
135 | } | |
9549ec01 FI |
136 | return bio; |
137 | } | |
138 | ||
5c83746a CH |
139 | static struct bio * |
140 | do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect, | |
141 | struct page *page, struct pnfs_block_dev_map *map, | |
4246a0b6 | 142 | struct pnfs_block_extent *be, bio_end_io_t end_io, |
5c83746a | 143 | struct parallel_io *par, unsigned int offset, int *len) |
9549ec01 | 144 | { |
5c83746a CH |
145 | struct pnfs_block_dev *dev = |
146 | container_of(be->be_device, struct pnfs_block_dev, node); | |
147 | u64 disk_addr, end; | |
148 | ||
fe6e1e8d | 149 | dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__, |
5c83746a CH |
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 | ||
9549ec01 FI |
171 | retry: |
172 | if (!bio) { | |
5c83746a CH |
173 | bio = bl_alloc_init_bio(npg, map->bdev, |
174 | disk_addr >> SECTOR_SHIFT, end_io, par); | |
9549ec01 FI |
175 | if (!bio) |
176 | return ERR_PTR(-ENOMEM); | |
177 | } | |
5c83746a | 178 | if (bio_add_page(bio, page, *len, offset) < *len) { |
9549ec01 FI |
179 | bio = bl_submit_bio(rw, bio); |
180 | goto retry; | |
181 | } | |
182 | return bio; | |
183 | } | |
184 | ||
4246a0b6 | 185 | static void bl_end_io_read(struct bio *bio) |
9549ec01 FI |
186 | { |
187 | struct parallel_io *par = bio->bi_private; | |
9549ec01 | 188 | |
4246a0b6 | 189 | if (bio->bi_error) { |
d45f60c6 | 190 | struct nfs_pgio_header *header = par->data; |
cd841605 FI |
191 | |
192 | if (!header->pnfs_error) | |
193 | header->pnfs_error = -EIO; | |
194 | pnfs_set_lo_fail(header->lseg); | |
9549ec01 | 195 | } |
8c792ea9 | 196 | |
9549ec01 FI |
197 | bio_put(bio); |
198 | put_parallel(par); | |
199 | } | |
200 | ||
201 | static void bl_read_cleanup(struct work_struct *work) | |
202 | { | |
203 | struct rpc_task *task; | |
d45f60c6 | 204 | struct nfs_pgio_header *hdr; |
9549ec01 FI |
205 | dprintk("%s enter\n", __func__); |
206 | task = container_of(work, struct rpc_task, u.tk_work); | |
d45f60c6 WAA |
207 | hdr = container_of(task, struct nfs_pgio_header, task); |
208 | pnfs_ld_read_done(hdr); | |
9549ec01 FI |
209 | } |
210 | ||
211 | static void | |
8067253c | 212 | bl_end_par_io_read(void *data) |
9549ec01 | 213 | { |
d45f60c6 | 214 | struct nfs_pgio_header *hdr = data; |
9549ec01 | 215 | |
d45f60c6 WAA |
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); | |
9549ec01 FI |
219 | } |
220 | ||
155e7524 | 221 | static enum pnfs_try_status |
8067253c | 222 | bl_read_pagelist(struct nfs_pgio_header *header) |
155e7524 | 223 | { |
8067253c | 224 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
5c83746a | 225 | struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 }; |
9549ec01 | 226 | struct bio *bio = NULL; |
8067253c | 227 | struct pnfs_block_extent be; |
9549ec01 FI |
228 | sector_t isect, extent_length = 0; |
229 | struct parallel_io *par; | |
8067253c CH |
230 | loff_t f_offset = header->args.offset; |
231 | size_t bytes_left = header->args.count; | |
15ae2c7b | 232 | unsigned int pg_offset = header->args.pgbase, pg_len; |
8067253c | 233 | struct page **pages = header->args.pages; |
09cbfeaf | 234 | int pg_index = header->args.pgbase >> PAGE_SHIFT; |
f742dc4a | 235 | const bool is_dio = (header->dreq != NULL); |
be98fd0a | 236 | struct blk_plug plug; |
8067253c | 237 | int i; |
9549ec01 | 238 | |
6f00866d | 239 | dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__, |
8067253c CH |
240 | header->page_array.npages, f_offset, |
241 | (unsigned int)header->args.count); | |
9549ec01 | 242 | |
8067253c | 243 | par = alloc_parallel(header); |
9549ec01 | 244 | if (!par) |
8067253c | 245 | return PNFS_NOT_ATTEMPTED; |
9549ec01 | 246 | par->pnfs_callback = bl_end_par_io_read; |
9549ec01 | 247 | |
be98fd0a CH |
248 | blk_start_plug(&plug); |
249 | ||
9549ec01 FI |
250 | isect = (sector_t) (f_offset >> SECTOR_SHIFT); |
251 | /* Code assumes extents are page-aligned */ | |
8067253c | 252 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 253 | if (extent_length <= 0) { |
9549ec01 | 254 | /* We've used up the previous extent */ |
9549ec01 | 255 | bio = bl_submit_bio(READ, bio); |
8067253c | 256 | |
9549ec01 | 257 | /* Get the next one */ |
8067253c | 258 | if (!ext_tree_lookup(bl, isect, &be, false)) { |
cd841605 | 259 | header->pnfs_error = -EIO; |
9549ec01 FI |
260 | goto out; |
261 | } | |
8067253c | 262 | extent_length = be.be_length - (isect - be.be_f_offset); |
9549ec01 | 263 | } |
f742dc4a PT |
264 | |
265 | if (is_dio) { | |
09cbfeaf KS |
266 | if (pg_offset + bytes_left > PAGE_SIZE) |
267 | pg_len = PAGE_SIZE - pg_offset; | |
f742dc4a PT |
268 | else |
269 | pg_len = bytes_left; | |
f742dc4a | 270 | } else { |
3a6fd1f0 | 271 | BUG_ON(pg_offset != 0); |
09cbfeaf | 272 | pg_len = PAGE_SIZE; |
f742dc4a PT |
273 | } |
274 | ||
8067253c | 275 | if (is_hole(&be)) { |
9549ec01 FI |
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__); | |
f742dc4a | 279 | zero_user_segment(pages[i], pg_offset, pg_len); |
5c83746a CH |
280 | |
281 | /* invalidate map */ | |
282 | map.start = NFS4_MAX_UINT64; | |
9549ec01 | 283 | } else { |
823b0c9d | 284 | bio = do_add_page_to_bio(bio, |
8067253c | 285 | header->page_array.npages - i, |
30dd374f | 286 | READ, |
5c83746a | 287 | isect, pages[i], &map, &be, |
f742dc4a | 288 | bl_end_io_read, par, |
5c83746a | 289 | pg_offset, &pg_len); |
9549ec01 | 290 | if (IS_ERR(bio)) { |
cd841605 | 291 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 292 | bio = NULL; |
9549ec01 FI |
293 | goto out; |
294 | } | |
295 | } | |
f742dc4a | 296 | isect += (pg_len >> SECTOR_SHIFT); |
921b81a8 | 297 | extent_length -= (pg_len >> SECTOR_SHIFT); |
5c83746a CH |
298 | f_offset += pg_len; |
299 | bytes_left -= pg_len; | |
15ae2c7b | 300 | pg_offset = 0; |
9549ec01 | 301 | } |
cd841605 | 302 | if ((isect << SECTOR_SHIFT) >= header->inode->i_size) { |
8067253c CH |
303 | header->res.eof = 1; |
304 | header->res.count = header->inode->i_size - header->args.offset; | |
9549ec01 | 305 | } else { |
8067253c | 306 | header->res.count = (isect << SECTOR_SHIFT) - header->args.offset; |
9549ec01 FI |
307 | } |
308 | out: | |
9549ec01 | 309 | bl_submit_bio(READ, bio); |
be98fd0a | 310 | blk_finish_plug(&plug); |
9549ec01 FI |
311 | put_parallel(par); |
312 | return PNFS_ATTEMPTED; | |
31e6306a FI |
313 | } |
314 | ||
4246a0b6 | 315 | static void bl_end_io_write(struct bio *bio) |
650e2d39 FI |
316 | { |
317 | struct parallel_io *par = bio->bi_private; | |
d45f60c6 | 318 | struct nfs_pgio_header *header = par->data; |
650e2d39 | 319 | |
4246a0b6 | 320 | if (bio->bi_error) { |
cd841605 FI |
321 | if (!header->pnfs_error) |
322 | header->pnfs_error = -EIO; | |
323 | pnfs_set_lo_fail(header->lseg); | |
650e2d39 FI |
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 | */ | |
332 | static void bl_write_cleanup(struct work_struct *work) | |
333 | { | |
8067253c CH |
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 | ||
650e2d39 | 338 | dprintk("%s enter\n", __func__); |
8067253c | 339 | |
d45f60c6 | 340 | if (likely(!hdr->pnfs_error)) { |
8067253c | 341 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg); |
09cbfeaf | 342 | u64 start = hdr->args.offset & (loff_t)PAGE_MASK; |
8067253c | 343 | u64 end = (hdr->args.offset + hdr->args.count + |
09cbfeaf | 344 | PAGE_SIZE - 1) & (loff_t)PAGE_MASK; |
8067253c CH |
345 | |
346 | ext_tree_mark_written(bl, start >> SECTOR_SHIFT, | |
347 | (end - start) >> SECTOR_SHIFT); | |
31e6306a | 348 | } |
8067253c | 349 | |
d45f60c6 | 350 | pnfs_ld_write_done(hdr); |
650e2d39 FI |
351 | } |
352 | ||
353 | /* Called when last of bios associated with a bl_write_pagelist call finishes */ | |
8067253c | 354 | static void bl_end_par_io_write(void *data) |
650e2d39 | 355 | { |
d45f60c6 | 356 | struct nfs_pgio_header *hdr = data; |
650e2d39 | 357 | |
d45f60c6 | 358 | hdr->task.tk_status = hdr->pnfs_error; |
c65e6254 | 359 | hdr->verf.committed = NFS_FILE_SYNC; |
d45f60c6 WAA |
360 | INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup); |
361 | schedule_work(&hdr->task.u.tk_work); | |
650e2d39 FI |
362 | } |
363 | ||
155e7524 | 364 | static enum pnfs_try_status |
d45f60c6 | 365 | bl_write_pagelist(struct nfs_pgio_header *header, int sync) |
155e7524 | 366 | { |
8067253c | 367 | struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); |
5c83746a | 368 | struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 }; |
650e2d39 | 369 | struct bio *bio = NULL; |
8067253c | 370 | struct pnfs_block_extent be; |
3a6fd1f0 | 371 | sector_t isect, extent_length = 0; |
96c9eae6 | 372 | struct parallel_io *par = NULL; |
d45f60c6 WAA |
373 | loff_t offset = header->args.offset; |
374 | size_t count = header->args.count; | |
d45f60c6 | 375 | struct page **pages = header->args.pages; |
09cbfeaf | 376 | int pg_index = header->args.pgbase >> PAGE_SHIFT; |
5c83746a | 377 | unsigned int pg_len; |
be98fd0a | 378 | struct blk_plug plug; |
8067253c | 379 | int i; |
650e2d39 FI |
380 | |
381 | dprintk("%s enter, %Zu@%lld\n", __func__, count, offset); | |
96c9eae6 | 382 | |
d45f60c6 | 383 | /* At this point, header->page_aray is a (sequential) list of nfs_pages. |
71cdd40f PT |
384 | * We want to write each, and if there is an error set pnfs_error |
385 | * to have it redone using nfs. | |
650e2d39 | 386 | */ |
d45f60c6 | 387 | par = alloc_parallel(header); |
650e2d39 | 388 | if (!par) |
8067253c | 389 | return PNFS_NOT_ATTEMPTED; |
650e2d39 | 390 | par->pnfs_callback = bl_end_par_io_write; |
650e2d39 | 391 | |
3a6fd1f0 | 392 | blk_start_plug(&plug); |
71cdd40f | 393 | |
3a6fd1f0 | 394 | /* we always write out the whole page */ |
09cbfeaf | 395 | offset = offset & (loff_t)PAGE_MASK; |
3a6fd1f0 | 396 | isect = offset >> SECTOR_SHIFT; |
71cdd40f | 397 | |
d45f60c6 | 398 | for (i = pg_index; i < header->page_array.npages; i++) { |
921b81a8 | 399 | if (extent_length <= 0) { |
650e2d39 | 400 | /* We've used up the previous extent */ |
650e2d39 FI |
401 | bio = bl_submit_bio(WRITE, bio); |
402 | /* Get the next one */ | |
8067253c | 403 | if (!ext_tree_lookup(bl, isect, &be, true)) { |
cd841605 | 404 | header->pnfs_error = -EINVAL; |
650e2d39 FI |
405 | goto out; |
406 | } | |
fe6e1e8d | 407 | |
8067253c | 408 | extent_length = be.be_length - (isect - be.be_f_offset); |
71cdd40f | 409 | } |
fe6e1e8d | 410 | |
09cbfeaf | 411 | pg_len = PAGE_SIZE; |
d45f60c6 | 412 | bio = do_add_page_to_bio(bio, header->page_array.npages - i, |
5c83746a | 413 | WRITE, isect, pages[i], &map, &be, |
fe6e1e8d | 414 | bl_end_io_write, par, |
5c83746a | 415 | 0, &pg_len); |
71cdd40f | 416 | if (IS_ERR(bio)) { |
cd841605 | 417 | header->pnfs_error = PTR_ERR(bio); |
e6d05a75 | 418 | bio = NULL; |
71cdd40f | 419 | goto out; |
650e2d39 | 420 | } |
5c83746a CH |
421 | |
422 | offset += pg_len; | |
423 | count -= pg_len; | |
424 | isect += (pg_len >> SECTOR_SHIFT); | |
425 | extent_length -= (pg_len >> SECTOR_SHIFT); | |
650e2d39 | 426 | } |
71cdd40f | 427 | |
d45f60c6 | 428 | header->res.count = header->args.count; |
650e2d39 | 429 | out: |
650e2d39 | 430 | bl_submit_bio(WRITE, bio); |
be98fd0a | 431 | blk_finish_plug(&plug); |
650e2d39 FI |
432 | put_parallel(par); |
433 | return PNFS_ATTEMPTED; | |
155e7524 FI |
434 | } |
435 | ||
436 | static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo) | |
437 | { | |
438 | struct pnfs_block_layout *bl = BLK_LO2EXT(lo); | |
8067253c | 439 | int err; |
155e7524 FI |
440 | |
441 | dprintk("%s enter\n", __func__); | |
8067253c CH |
442 | |
443 | err = ext_tree_remove(bl, true, 0, LLONG_MAX); | |
444 | WARN_ON(err); | |
445 | ||
155e7524 FI |
446 | kfree(bl); |
447 | } | |
448 | ||
d9186c03 CH |
449 | static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode, |
450 | gfp_t gfp_flags, bool is_scsi_layout) | |
155e7524 FI |
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; | |
8067253c CH |
458 | |
459 | bl->bl_ext_rw = RB_ROOT; | |
460 | bl->bl_ext_ro = RB_ROOT; | |
155e7524 | 461 | spin_lock_init(&bl->bl_ext_lock); |
8067253c | 462 | |
d9186c03 | 463 | bl->bl_scsi_layout = is_scsi_layout; |
155e7524 FI |
464 | return &bl->bl_layout; |
465 | } | |
466 | ||
d9186c03 CH |
467 | static 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 | ||
473 | static 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 | ||
a60d2ebd | 479 | static void bl_free_lseg(struct pnfs_layout_segment *lseg) |
155e7524 | 480 | { |
a60d2ebd FI |
481 | dprintk("%s enter\n", __func__); |
482 | kfree(lseg); | |
155e7524 FI |
483 | } |
484 | ||
9cc47541 CH |
485 | /* Tracks info needed to ensure extents in layout obey constraints of spec */ |
486 | struct 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 | */ | |
496 | static 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 | ||
539 | static 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 | ||
9cc47541 | 552 | static int |
ca0fe1df CH |
553 | bl_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) | |
9cc47541 | 556 | { |
ca0fe1df CH |
557 | struct pnfs_block_extent *be; |
558 | struct nfs4_deviceid id; | |
559 | int error; | |
9cc47541 | 560 | __be32 *p; |
ca0fe1df CH |
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 | ||
600 | out_put_deviceid: | |
601 | nfs4_put_deviceid_node(be->be_device); | |
602 | out_free_be: | |
603 | kfree(be); | |
604 | return error; | |
605 | } | |
606 | ||
607 | static struct pnfs_layout_segment * | |
608 | bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr, | |
609 | gfp_t gfp_mask) | |
610 | { | |
9cc47541 CH |
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 | }; | |
ca0fe1df CH |
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; | |
9cc47541 CH |
625 | LIST_HEAD(extents); |
626 | ||
627 | dprintk("---> %s\n", __func__); | |
628 | ||
ca0fe1df CH |
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); | |
9cc47541 | 635 | if (!scratch) |
ca0fe1df | 636 | goto out; |
9cc47541 | 637 | |
ca0fe1df CH |
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); | |
9cc47541 | 641 | |
ca0fe1df CH |
642 | status = -EIO; |
643 | p = xdr_inline_decode(&xdr, 4); | |
9cc47541 | 644 | if (unlikely(!p)) |
ca0fe1df | 645 | goto out_free_scratch; |
9cc47541 CH |
646 | |
647 | count = be32_to_cpup(p++); | |
ca0fe1df | 648 | dprintk("%s: number of extents %d\n", __func__, count); |
9cc47541 | 649 | |
ca0fe1df CH |
650 | /* |
651 | * Decode individual extents, putting them in temporary staging area | |
652 | * until whole layout is decoded to make error recovery easier. | |
9cc47541 CH |
653 | */ |
654 | for (i = 0; i < count; i++) { | |
ca0fe1df CH |
655 | status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask); |
656 | if (status) | |
657 | goto process_extents; | |
9cc47541 | 658 | } |
ca0fe1df | 659 | |
9cc47541 CH |
660 | if (lgr->range.offset + lgr->range.length != |
661 | lv.start << SECTOR_SHIFT) { | |
662 | dprintk("%s Final length mismatch\n", __func__); | |
ca0fe1df CH |
663 | status = -EIO; |
664 | goto process_extents; | |
9cc47541 | 665 | } |
ca0fe1df | 666 | |
9cc47541 CH |
667 | if (lv.start < lv.cowread) { |
668 | dprintk("%s Final uncovered COW extent\n", __func__); | |
ca0fe1df | 669 | status = -EIO; |
9cc47541 | 670 | } |
9cc47541 | 671 | |
ca0fe1df | 672 | process_extents: |
9cc47541 | 673 | while (!list_empty(&extents)) { |
ca0fe1df CH |
674 | struct pnfs_block_extent *be = |
675 | list_first_entry(&extents, struct pnfs_block_extent, | |
676 | be_list); | |
9cc47541 | 677 | list_del(&be->be_list); |
9cc47541 | 678 | |
ca0fe1df CH |
679 | if (!status) |
680 | status = ext_tree_insert(bl, be); | |
a60d2ebd | 681 | |
ca0fe1df CH |
682 | if (status) { |
683 | nfs4_put_deviceid_node(be->be_device); | |
684 | kfree(be); | |
685 | } | |
686 | } | |
687 | ||
688 | out_free_scratch: | |
689 | __free_page(scratch); | |
690 | out: | |
691 | dprintk("%s returns %d\n", __func__, status); | |
a60d2ebd | 692 | if (status) { |
a60d2ebd FI |
693 | kfree(lseg); |
694 | return ERR_PTR(status); | |
695 | } | |
696 | return lseg; | |
155e7524 FI |
697 | } |
698 | ||
71d5b763 CH |
699 | static void |
700 | bl_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; | |
71d5b763 CH |
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 | ||
164ae58c | 724 | ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end); |
71d5b763 CH |
725 | } |
726 | ||
34dc93c2 CH |
727 | static int |
728 | bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg) | |
155e7524 | 729 | { |
34dc93c2 | 730 | return ext_tree_prepare_commit(arg); |
155e7524 FI |
731 | } |
732 | ||
733 | static void | |
734 | bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata) | |
735 | { | |
34dc93c2 | 736 | ext_tree_mark_committed(&lcdata->args, lcdata->res.status); |
155e7524 FI |
737 | } |
738 | ||
739 | static int | |
740 | bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh) | |
741 | { | |
742 | dprintk("%s enter\n", __func__); | |
2f9fd182 FI |
743 | |
744 | if (server->pnfs_blksize == 0) { | |
745 | dprintk("%s Server did not return blksize\n", __func__); | |
746 | return -EINVAL; | |
747 | } | |
e3aaf7f2 CH |
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 | ||
d4b18c3e | 754 | return 0; |
155e7524 FI |
755 | } |
756 | ||
f742dc4a | 757 | static bool |
3a6fd1f0 | 758 | is_aligned_req(struct nfs_pageio_descriptor *pgio, |
f35592a9 | 759 | struct nfs_page *req, unsigned int alignment, bool is_write) |
f742dc4a | 760 | { |
3a6fd1f0 CH |
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 | ||
f35592a9 KM |
774 | if (is_write && |
775 | (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) { | |
3a6fd1f0 CH |
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; | |
f742dc4a PT |
787 | } |
788 | ||
789 | static void | |
790 | bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) | |
791 | { | |
f35592a9 | 792 | if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) { |
f742dc4a | 793 | nfs_pageio_reset_read_mds(pgio); |
3a6fd1f0 CH |
794 | return; |
795 | } | |
796 | ||
797 | pnfs_generic_pg_init_read(pgio, req); | |
f742dc4a PT |
798 | } |
799 | ||
b4fdac1a WAA |
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 | */ | |
804 | static size_t | |
f742dc4a PT |
805 | bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
806 | struct nfs_page *req) | |
807 | { | |
f35592a9 | 808 | if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) |
b4fdac1a | 809 | return 0; |
f742dc4a PT |
810 | return pnfs_generic_pg_test(pgio, prev, req); |
811 | } | |
812 | ||
6296556f PT |
813 | /* |
814 | * Return the number of contiguous bytes for a given inode | |
815 | * starting at page frame idx. | |
816 | */ | |
817 | static 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 */ | |
09cbfeaf | 823 | end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
6a74c0c9 | 824 | if (end != inode->i_mapping->nrpages) { |
6296556f | 825 | rcu_read_lock(); |
e7b563bb | 826 | end = page_cache_next_hole(mapping, idx + 1, ULONG_MAX); |
6296556f PT |
827 | rcu_read_unlock(); |
828 | } | |
829 | ||
830 | if (!end) | |
09cbfeaf | 831 | return i_size_read(inode) - (idx << PAGE_SHIFT); |
6296556f | 832 | else |
09cbfeaf | 833 | return (end - idx) << PAGE_SHIFT; |
6296556f PT |
834 | } |
835 | ||
6f018efa | 836 | static void |
96c9eae6 PT |
837 | bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) |
838 | { | |
3a6fd1f0 CH |
839 | u64 wb_size; |
840 | ||
f35592a9 | 841 | if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) { |
96c9eae6 | 842 | nfs_pageio_reset_write_mds(pgio); |
3a6fd1f0 | 843 | return; |
6296556f | 844 | } |
3a6fd1f0 CH |
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); | |
96c9eae6 PT |
853 | } |
854 | ||
b4fdac1a WAA |
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 | */ | |
859 | static size_t | |
96c9eae6 PT |
860 | bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, |
861 | struct nfs_page *req) | |
862 | { | |
f35592a9 | 863 | if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) |
b4fdac1a | 864 | return 0; |
96c9eae6 PT |
865 | return pnfs_generic_pg_test(pgio, prev, req); |
866 | } | |
867 | ||
e9643fe8 | 868 | static const struct nfs_pageio_ops bl_pg_read_ops = { |
f742dc4a PT |
869 | .pg_init = bl_pg_init_read, |
870 | .pg_test = bl_pg_test_read, | |
e9643fe8 | 871 | .pg_doio = pnfs_generic_pg_readpages, |
180bb5ec | 872 | .pg_cleanup = pnfs_generic_pg_cleanup, |
e9643fe8 BH |
873 | }; |
874 | ||
875 | static const struct nfs_pageio_ops bl_pg_write_ops = { | |
96c9eae6 PT |
876 | .pg_init = bl_pg_init_write, |
877 | .pg_test = bl_pg_test_write, | |
e9643fe8 | 878 | .pg_doio = pnfs_generic_pg_writepages, |
180bb5ec | 879 | .pg_cleanup = pnfs_generic_pg_cleanup, |
e9643fe8 BH |
880 | }; |
881 | ||
155e7524 FI |
882 | static struct pnfs_layoutdriver_type blocklayout_type = { |
883 | .id = LAYOUT_BLOCK_VOLUME, | |
884 | .name = "LAYOUT_BLOCK_VOLUME", | |
5a12cca6 | 885 | .owner = THIS_MODULE, |
848746bd CH |
886 | .flags = PNFS_LAYOUTRET_ON_SETATTR | |
887 | PNFS_READ_WHOLE_PAGE, | |
155e7524 FI |
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, | |
71d5b763 | 894 | .return_range = bl_return_range, |
34dc93c2 | 895 | .prepare_layoutcommit = bl_prepare_layoutcommit, |
155e7524 FI |
896 | .cleanup_layoutcommit = bl_cleanup_layoutcommit, |
897 | .set_layoutdriver = bl_set_layoutdriver, | |
20d655d6 CH |
898 | .alloc_deviceid_node = bl_alloc_deviceid_node, |
899 | .free_deviceid_node = bl_free_deviceid_node, | |
e9643fe8 BH |
900 | .pg_read_ops = &bl_pg_read_ops, |
901 | .pg_write_ops = &bl_pg_write_ops, | |
5bb89b47 | 902 | .sync = pnfs_generic_sync, |
155e7524 FI |
903 | }; |
904 | ||
d9186c03 CH |
905 | static 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 | ||
155e7524 FI |
929 | static int __init nfs4blocklayout_init(void) |
930 | { | |
931 | int ret; | |
932 | ||
933 | dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__); | |
934 | ||
d9186c03 | 935 | ret = bl_init_pipefs(); |
fe0a9b74 JR |
936 | if (ret) |
937 | goto out; | |
d9186c03 CH |
938 | |
939 | ret = pnfs_register_layoutdriver(&blocklayout_type); | |
627f3066 | 940 | if (ret) |
d9186c03 CH |
941 | goto out_cleanup_pipe; |
942 | ||
943 | ret = pnfs_register_layoutdriver(&scsilayout_type); | |
944 | if (ret) | |
945 | goto out_unregister_block; | |
871760ce | 946 | return 0; |
fe0a9b74 | 947 | |
d9186c03 | 948 | out_unregister_block: |
fe0a9b74 | 949 | pnfs_unregister_layoutdriver(&blocklayout_type); |
d9186c03 CH |
950 | out_cleanup_pipe: |
951 | bl_cleanup_pipefs(); | |
871760ce | 952 | out: |
155e7524 FI |
953 | return ret; |
954 | } | |
955 | ||
956 | static void __exit nfs4blocklayout_exit(void) | |
957 | { | |
958 | dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n", | |
959 | __func__); | |
960 | ||
d9186c03 | 961 | pnfs_unregister_layoutdriver(&scsilayout_type); |
155e7524 | 962 | pnfs_unregister_layoutdriver(&blocklayout_type); |
d9186c03 | 963 | bl_cleanup_pipefs(); |
155e7524 FI |
964 | } |
965 | ||
966 | MODULE_ALIAS("nfs-layouttype4-3"); | |
967 | ||
968 | module_init(nfs4blocklayout_init); | |
969 | module_exit(nfs4blocklayout_exit); |