blob: b48f782a94aded0a39a08f3533b8e843a30587af [file] [log] [blame]
Jim Reesfe0a9b72011-07-30 20:52:42 -04001/*
2 * linux/fs/nfs/blocklayout/blocklayoutdev.c
3 *
4 * Device operations for the pnfs nfs4 file layout driver.
5 *
6 * Copyright (c) 2006 The Regents of the University of Michigan.
7 * All rights reserved.
8 *
9 * Andy Adamson <andros@citi.umich.edu>
10 * Fred Isaman <iisaman@umich.edu>
11 *
12 * permission is granted to use, copy, create derivative works and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the university of michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization. if
17 * the above copyright notice or any other identification of the
18 * university of michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
20 *
21 * this software is provided as is, without representation from the
22 * university of michigan as to its fitness for any purpose, and without
23 * warranty by the university of michigan of any kind, either express
24 * or implied, including without limitation the implied warranties of
25 * merchantability and fitness for a particular purpose. the regents
26 * of the university of michigan shall not be liable for any damages,
27 * including special, indirect, incidental, or consequential damages,
28 * with respect to any claim arising out or in connection with the use
29 * of the software, even if it has been or is hereafter advised of the
30 * possibility of such damages.
31 */
32#include <linux/module.h>
33#include <linux/buffer_head.h> /* __bread */
34
35#include <linux/genhd.h>
36#include <linux/blkdev.h>
37#include <linux/hash.h>
38
39#include "blocklayout.h"
40
41#define NFSDBG_FACILITY NFSDBG_PNFS_LD
42
Fred Isamane9437cc2011-07-30 20:52:47 -040043static int decode_sector_number(__be32 **rp, sector_t *sp)
44{
45 uint64_t s;
46
47 *rp = xdr_decode_hyper(*rp, &s);
48 if (s & 0x1ff) {
Weston Andros Adamsona0308892012-01-26 13:32:23 -050049 printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
Fred Isamane9437cc2011-07-30 20:52:47 -040050 return -1;
51 }
52 *sp = s >> SECTOR_SHIFT;
53 return 0;
54}
55
Jim Reesfe0a9b72011-07-30 20:52:42 -040056/* Open a block_device by device number. */
57struct block_device *nfs4_blkdev_get(dev_t dev)
58{
59 struct block_device *bd;
60
61 dprintk("%s enter\n", __func__);
62 bd = blkdev_get_by_dev(dev, FMODE_READ, NULL);
63 if (IS_ERR(bd))
64 goto fail;
65 return bd;
66fail:
67 dprintk("%s failed to open device : %ld\n",
68 __func__, PTR_ERR(bd));
69 return NULL;
70}
71
72/*
73 * Release the block device
74 */
75int nfs4_blkdev_put(struct block_device *bdev)
76{
77 dprintk("%s for device %d:%d\n", __func__, MAJOR(bdev->bd_dev),
78 MINOR(bdev->bd_dev));
79 return blkdev_put(bdev, FMODE_READ);
80}
81
Jim Reesfe0a9b72011-07-30 20:52:42 -040082static struct bl_dev_msg bl_mount_reply;
83
84ssize_t bl_pipe_downcall(struct file *filp, const char __user *src,
85 size_t mlen)
86{
87 if (mlen != sizeof (struct bl_dev_msg))
88 return -EINVAL;
89
90 if (copy_from_user(&bl_mount_reply, src, mlen) != 0)
91 return -EFAULT;
92
93 wake_up(&bl_wq);
94
95 return mlen;
96}
97
98void bl_pipe_destroy_msg(struct rpc_pipe_msg *msg)
99{
100 if (msg->errno >= 0)
101 return;
102 wake_up(&bl_wq);
103}
104
105/*
106 * Decodes pnfs_block_deviceaddr4 which is XDR encoded in dev->dev_addr_buf.
107 */
108struct pnfs_block_dev *
109nfs4_blk_decode_device(struct nfs_server *server,
Fred Isaman2f9fd182011-07-30 20:52:46 -0400110 struct pnfs_device *dev)
Jim Reesfe0a9b72011-07-30 20:52:42 -0400111{
Jim Rees516f2e22011-09-22 21:50:08 -0400112 struct pnfs_block_dev *rv;
Jim Reesfe0a9b72011-07-30 20:52:42 -0400113 struct block_device *bd = NULL;
114 struct rpc_pipe_msg msg;
115 struct bl_msg_hdr bl_msg = {
116 .type = BL_DEVICE_MOUNT,
117 .totallen = dev->mincount,
118 };
119 uint8_t *dataptr;
120 DECLARE_WAITQUEUE(wq, current);
121 struct bl_dev_msg *reply = &bl_mount_reply;
Jim Rees516f2e22011-09-22 21:50:08 -0400122 int offset, len, i, rc;
Stanislav Kinsbursky9e2e74d2012-01-10 17:04:24 +0400123 struct net *net = server->nfs_client->net;
124 struct nfs_net *nn = net_generic(net, nfs_net_id);
Jim Reesfe0a9b72011-07-30 20:52:42 -0400125
126 dprintk("%s CREATING PIPEFS MESSAGE\n", __func__);
127 dprintk("%s: deviceid: %s, mincount: %d\n", __func__, dev->dev_id.data,
128 dev->mincount);
129
130 memset(&msg, 0, sizeof(msg));
131 msg.data = kzalloc(sizeof(bl_msg) + dev->mincount, GFP_NOFS);
132 if (!msg.data) {
133 rv = ERR_PTR(-ENOMEM);
134 goto out;
135 }
136
137 memcpy(msg.data, &bl_msg, sizeof(bl_msg));
138 dataptr = (uint8_t *) msg.data;
Fred Isaman2f9fd182011-07-30 20:52:46 -0400139 len = dev->mincount;
140 offset = sizeof(bl_msg);
141 for (i = 0; len > 0; i++) {
142 memcpy(&dataptr[offset], page_address(dev->pages[i]),
143 len < PAGE_CACHE_SIZE ? len : PAGE_CACHE_SIZE);
144 len -= PAGE_CACHE_SIZE;
145 offset += PAGE_CACHE_SIZE;
146 }
Jim Reesfe0a9b72011-07-30 20:52:42 -0400147 msg.len = sizeof(bl_msg) + dev->mincount;
148
149 dprintk("%s CALLING USERSPACE DAEMON\n", __func__);
150 add_wait_queue(&bl_wq, &wq);
Stanislav Kinsbursky9e2e74d2012-01-10 17:04:24 +0400151 rc = rpc_queue_upcall(nn->bl_device_pipe, &msg);
Jim Rees516f2e22011-09-22 21:50:08 -0400152 if (rc < 0) {
Jim Reesfe0a9b72011-07-30 20:52:42 -0400153 remove_wait_queue(&bl_wq, &wq);
Jim Rees516f2e22011-09-22 21:50:08 -0400154 rv = ERR_PTR(rc);
Jim Reesfe0a9b72011-07-30 20:52:42 -0400155 goto out;
156 }
157
158 set_current_state(TASK_UNINTERRUPTIBLE);
159 schedule();
160 __set_current_state(TASK_RUNNING);
161 remove_wait_queue(&bl_wq, &wq);
162
163 if (reply->status != BL_DEVICE_REQUEST_PROC) {
164 dprintk("%s failed to open device: %d\n",
165 __func__, reply->status);
166 rv = ERR_PTR(-EINVAL);
167 goto out;
168 }
169
170 bd = nfs4_blkdev_get(MKDEV(reply->major, reply->minor));
171 if (IS_ERR(bd)) {
Jim Rees516f2e22011-09-22 21:50:08 -0400172 rc = PTR_ERR(bd);
173 dprintk("%s failed to open device : %d\n", __func__, rc);
174 rv = ERR_PTR(rc);
Jim Reesfe0a9b72011-07-30 20:52:42 -0400175 goto out;
176 }
177
178 rv = kzalloc(sizeof(*rv), GFP_NOFS);
179 if (!rv) {
180 rv = ERR_PTR(-ENOMEM);
181 goto out;
182 }
183
184 rv->bm_mdev = bd;
185 memcpy(&rv->bm_mdevid, &dev->dev_id, sizeof(struct nfs4_deviceid));
Stanislav Kinsbursky9e2e74d2012-01-10 17:04:24 +0400186 rv->net = net;
Jim Reesfe0a9b72011-07-30 20:52:42 -0400187 dprintk("%s Created device %s with bd_block_size %u\n",
188 __func__,
189 bd->bd_disk->disk_name,
190 bd->bd_block_size);
191
192out:
193 kfree(msg.data);
194 return rv;
195}
Fred Isamana60d2eb2011-07-30 20:52:44 -0400196
Fred Isamane9437cc2011-07-30 20:52:47 -0400197/* Map deviceid returned by the server to constructed block_device */
198static struct block_device *translate_devid(struct pnfs_layout_hdr *lo,
199 struct nfs4_deviceid *id)
200{
201 struct block_device *rv = NULL;
202 struct block_mount_id *mid;
203 struct pnfs_block_dev *dev;
204
205 dprintk("%s enter, lo=%p, id=%p\n", __func__, lo, id);
206 mid = BLK_ID(lo);
207 spin_lock(&mid->bm_lock);
208 list_for_each_entry(dev, &mid->bm_devlist, bm_node) {
209 if (memcmp(id->data, dev->bm_mdevid.data,
210 NFS4_DEVICEID4_SIZE) == 0) {
211 rv = dev->bm_mdev;
212 goto out;
213 }
214 }
215 out:
216 spin_unlock(&mid->bm_lock);
217 dprintk("%s returning %p\n", __func__, rv);
218 return rv;
219}
220
221/* Tracks info needed to ensure extents in layout obey constraints of spec */
222struct layout_verification {
223 u32 mode; /* R or RW */
224 u64 start; /* Expected start of next non-COW extent */
225 u64 inval; /* Start of INVAL coverage */
226 u64 cowread; /* End of COW read coverage */
227};
228
229/* Verify the extent meets the layout requirements of the pnfs-block draft,
230 * section 2.3.1.
231 */
232static int verify_extent(struct pnfs_block_extent *be,
233 struct layout_verification *lv)
234{
235 if (lv->mode == IOMODE_READ) {
236 if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
237 be->be_state == PNFS_BLOCK_INVALID_DATA)
238 return -EIO;
239 if (be->be_f_offset != lv->start)
240 return -EIO;
241 lv->start += be->be_length;
242 return 0;
243 }
244 /* lv->mode == IOMODE_RW */
245 if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
246 if (be->be_f_offset != lv->start)
247 return -EIO;
248 if (lv->cowread > lv->start)
249 return -EIO;
250 lv->start += be->be_length;
251 lv->inval = lv->start;
252 return 0;
253 } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
254 if (be->be_f_offset != lv->start)
255 return -EIO;
256 lv->start += be->be_length;
257 return 0;
258 } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
259 if (be->be_f_offset > lv->start)
260 return -EIO;
261 if (be->be_f_offset < lv->inval)
262 return -EIO;
263 if (be->be_f_offset < lv->cowread)
264 return -EIO;
265 /* It looks like you might want to min this with lv->start,
266 * but you really don't.
267 */
268 lv->inval = lv->inval + be->be_length;
269 lv->cowread = be->be_f_offset + be->be_length;
270 return 0;
271 } else
272 return -EIO;
273}
274
275/* XDR decode pnfs_block_layout4 structure */
Fred Isamana60d2eb2011-07-30 20:52:44 -0400276int
277nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
278 struct nfs4_layoutget_res *lgr, gfp_t gfp_flags)
279{
Fred Isamane9437cc2011-07-30 20:52:47 -0400280 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
281 int i, status = -EIO;
282 uint32_t count;
283 struct pnfs_block_extent *be = NULL, *save;
284 struct xdr_stream stream;
285 struct xdr_buf buf;
286 struct page *scratch;
287 __be32 *p;
288 struct layout_verification lv = {
289 .mode = lgr->range.iomode,
290 .start = lgr->range.offset >> SECTOR_SHIFT,
291 .inval = lgr->range.offset >> SECTOR_SHIFT,
292 .cowread = lgr->range.offset >> SECTOR_SHIFT,
293 };
294 LIST_HEAD(extents);
295
296 dprintk("---> %s\n", __func__);
297
298 scratch = alloc_page(gfp_flags);
299 if (!scratch)
300 return -ENOMEM;
301
302 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
303 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
304
305 p = xdr_inline_decode(&stream, 4);
306 if (unlikely(!p))
307 goto out_err;
308
309 count = be32_to_cpup(p++);
310
311 dprintk("%s enter, number of extents %i\n", __func__, count);
312 p = xdr_inline_decode(&stream, (28 + NFS4_DEVICEID4_SIZE) * count);
313 if (unlikely(!p))
314 goto out_err;
315
316 /* Decode individual extents, putting them in temporary
317 * staging area until whole layout is decoded to make error
318 * recovery easier.
319 */
320 for (i = 0; i < count; i++) {
321 be = bl_alloc_extent();
322 if (!be) {
323 status = -ENOMEM;
324 goto out_err;
325 }
326 memcpy(&be->be_devid, p, NFS4_DEVICEID4_SIZE);
327 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
328 be->be_mdev = translate_devid(lo, &be->be_devid);
329 if (!be->be_mdev)
330 goto out_err;
331
332 /* The next three values are read in as bytes,
333 * but stored as 512-byte sector lengths
334 */
335 if (decode_sector_number(&p, &be->be_f_offset) < 0)
336 goto out_err;
337 if (decode_sector_number(&p, &be->be_length) < 0)
338 goto out_err;
339 if (decode_sector_number(&p, &be->be_v_offset) < 0)
340 goto out_err;
341 be->be_state = be32_to_cpup(p++);
342 if (be->be_state == PNFS_BLOCK_INVALID_DATA)
343 be->be_inval = &bl->bl_inval;
344 if (verify_extent(be, &lv)) {
345 dprintk("%s verify failed\n", __func__);
346 goto out_err;
347 }
348 list_add_tail(&be->be_node, &extents);
349 }
350 if (lgr->range.offset + lgr->range.length !=
351 lv.start << SECTOR_SHIFT) {
352 dprintk("%s Final length mismatch\n", __func__);
353 be = NULL;
354 goto out_err;
355 }
356 if (lv.start < lv.cowread) {
357 dprintk("%s Final uncovered COW extent\n", __func__);
358 be = NULL;
359 goto out_err;
360 }
361 /* Extents decoded properly, now try to merge them in to
362 * existing layout extents.
363 */
364 spin_lock(&bl->bl_ext_lock);
365 list_for_each_entry_safe(be, save, &extents, be_node) {
366 list_del(&be->be_node);
367 status = bl_add_merge_extent(bl, be);
368 if (status) {
369 spin_unlock(&bl->bl_ext_lock);
370 /* This is a fairly catastrophic error, as the
371 * entire layout extent lists are now corrupted.
372 * We should have some way to distinguish this.
373 */
374 be = NULL;
375 goto out_err;
376 }
377 }
378 spin_unlock(&bl->bl_ext_lock);
379 status = 0;
380 out:
381 __free_page(scratch);
382 dprintk("%s returns %i\n", __func__, status);
383 return status;
384
385 out_err:
386 bl_put_extent(be);
387 while (!list_empty(&extents)) {
388 be = list_first_entry(&extents, struct pnfs_block_extent,
389 be_node);
390 list_del(&be->be_node);
391 bl_put_extent(be);
392 }
393 goto out;
Fred Isamana60d2eb2011-07-30 20:52:44 -0400394}