Loading...
1/*
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
43static 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) {
49 printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__);
50 return -1;
51 }
52 *sp = s >> SECTOR_SHIFT;
53 return 0;
54}
55
56/* 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
82ssize_t bl_pipe_downcall(struct file *filp, const char __user *src,
83 size_t mlen)
84{
85 struct nfs_net *nn = net_generic(filp->f_dentry->d_sb->s_fs_info,
86 nfs_net_id);
87
88 if (mlen != sizeof (struct bl_dev_msg))
89 return -EINVAL;
90
91 if (copy_from_user(&nn->bl_mount_reply, src, mlen) != 0)
92 return -EFAULT;
93
94 wake_up(&nn->bl_wq);
95
96 return mlen;
97}
98
99void bl_pipe_destroy_msg(struct rpc_pipe_msg *msg)
100{
101 struct bl_pipe_msg *bl_pipe_msg = container_of(msg, struct bl_pipe_msg, msg);
102
103 if (msg->errno >= 0)
104 return;
105 wake_up(bl_pipe_msg->bl_wq);
106}
107
108/*
109 * Decodes pnfs_block_deviceaddr4 which is XDR encoded in dev->dev_addr_buf.
110 */
111struct pnfs_block_dev *
112nfs4_blk_decode_device(struct nfs_server *server,
113 struct pnfs_device *dev)
114{
115 struct pnfs_block_dev *rv;
116 struct block_device *bd = NULL;
117 struct bl_pipe_msg bl_pipe_msg;
118 struct rpc_pipe_msg *msg = &bl_pipe_msg.msg;
119 struct bl_msg_hdr bl_msg = {
120 .type = BL_DEVICE_MOUNT,
121 .totallen = dev->mincount,
122 };
123 uint8_t *dataptr;
124 DECLARE_WAITQUEUE(wq, current);
125 int offset, len, i, rc;
126 struct net *net = server->nfs_client->cl_net;
127 struct nfs_net *nn = net_generic(net, nfs_net_id);
128 struct bl_dev_msg *reply = &nn->bl_mount_reply;
129
130 dprintk("%s CREATING PIPEFS MESSAGE\n", __func__);
131 dprintk("%s: deviceid: %s, mincount: %d\n", __func__, dev->dev_id.data,
132 dev->mincount);
133
134 bl_pipe_msg.bl_wq = &nn->bl_wq;
135 memset(msg, 0, sizeof(*msg));
136 msg->data = kzalloc(sizeof(bl_msg) + dev->mincount, GFP_NOFS);
137 if (!msg->data) {
138 rv = ERR_PTR(-ENOMEM);
139 goto out;
140 }
141
142 memcpy(msg->data, &bl_msg, sizeof(bl_msg));
143 dataptr = (uint8_t *) msg->data;
144 len = dev->mincount;
145 offset = sizeof(bl_msg);
146 for (i = 0; len > 0; i++) {
147 memcpy(&dataptr[offset], page_address(dev->pages[i]),
148 len < PAGE_CACHE_SIZE ? len : PAGE_CACHE_SIZE);
149 len -= PAGE_CACHE_SIZE;
150 offset += PAGE_CACHE_SIZE;
151 }
152 msg->len = sizeof(bl_msg) + dev->mincount;
153
154 dprintk("%s CALLING USERSPACE DAEMON\n", __func__);
155 add_wait_queue(&nn->bl_wq, &wq);
156 rc = rpc_queue_upcall(nn->bl_device_pipe, msg);
157 if (rc < 0) {
158 remove_wait_queue(&nn->bl_wq, &wq);
159 rv = ERR_PTR(rc);
160 goto out;
161 }
162
163 set_current_state(TASK_UNINTERRUPTIBLE);
164 schedule();
165 __set_current_state(TASK_RUNNING);
166 remove_wait_queue(&nn->bl_wq, &wq);
167
168 if (reply->status != BL_DEVICE_REQUEST_PROC) {
169 dprintk("%s failed to open device: %d\n",
170 __func__, reply->status);
171 rv = ERR_PTR(-EINVAL);
172 goto out;
173 }
174
175 bd = nfs4_blkdev_get(MKDEV(reply->major, reply->minor));
176 if (IS_ERR(bd)) {
177 rc = PTR_ERR(bd);
178 dprintk("%s failed to open device : %d\n", __func__, rc);
179 rv = ERR_PTR(rc);
180 goto out;
181 }
182
183 rv = kzalloc(sizeof(*rv), GFP_NOFS);
184 if (!rv) {
185 rv = ERR_PTR(-ENOMEM);
186 goto out;
187 }
188
189 rv->bm_mdev = bd;
190 memcpy(&rv->bm_mdevid, &dev->dev_id, sizeof(struct nfs4_deviceid));
191 rv->net = net;
192 dprintk("%s Created device %s with bd_block_size %u\n",
193 __func__,
194 bd->bd_disk->disk_name,
195 bd->bd_block_size);
196
197out:
198 kfree(msg->data);
199 return rv;
200}
201
202/* Map deviceid returned by the server to constructed block_device */
203static struct block_device *translate_devid(struct pnfs_layout_hdr *lo,
204 struct nfs4_deviceid *id)
205{
206 struct block_device *rv = NULL;
207 struct block_mount_id *mid;
208 struct pnfs_block_dev *dev;
209
210 dprintk("%s enter, lo=%p, id=%p\n", __func__, lo, id);
211 mid = BLK_ID(lo);
212 spin_lock(&mid->bm_lock);
213 list_for_each_entry(dev, &mid->bm_devlist, bm_node) {
214 if (memcmp(id->data, dev->bm_mdevid.data,
215 NFS4_DEVICEID4_SIZE) == 0) {
216 rv = dev->bm_mdev;
217 goto out;
218 }
219 }
220 out:
221 spin_unlock(&mid->bm_lock);
222 dprintk("%s returning %p\n", __func__, rv);
223 return rv;
224}
225
226/* Tracks info needed to ensure extents in layout obey constraints of spec */
227struct layout_verification {
228 u32 mode; /* R or RW */
229 u64 start; /* Expected start of next non-COW extent */
230 u64 inval; /* Start of INVAL coverage */
231 u64 cowread; /* End of COW read coverage */
232};
233
234/* Verify the extent meets the layout requirements of the pnfs-block draft,
235 * section 2.3.1.
236 */
237static int verify_extent(struct pnfs_block_extent *be,
238 struct layout_verification *lv)
239{
240 if (lv->mode == IOMODE_READ) {
241 if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
242 be->be_state == PNFS_BLOCK_INVALID_DATA)
243 return -EIO;
244 if (be->be_f_offset != lv->start)
245 return -EIO;
246 lv->start += be->be_length;
247 return 0;
248 }
249 /* lv->mode == IOMODE_RW */
250 if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
251 if (be->be_f_offset != lv->start)
252 return -EIO;
253 if (lv->cowread > lv->start)
254 return -EIO;
255 lv->start += be->be_length;
256 lv->inval = lv->start;
257 return 0;
258 } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
259 if (be->be_f_offset != lv->start)
260 return -EIO;
261 lv->start += be->be_length;
262 return 0;
263 } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
264 if (be->be_f_offset > lv->start)
265 return -EIO;
266 if (be->be_f_offset < lv->inval)
267 return -EIO;
268 if (be->be_f_offset < lv->cowread)
269 return -EIO;
270 /* It looks like you might want to min this with lv->start,
271 * but you really don't.
272 */
273 lv->inval = lv->inval + be->be_length;
274 lv->cowread = be->be_f_offset + be->be_length;
275 return 0;
276 } else
277 return -EIO;
278}
279
280/* XDR decode pnfs_block_layout4 structure */
281int
282nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
283 struct nfs4_layoutget_res *lgr, gfp_t gfp_flags)
284{
285 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
286 int i, status = -EIO;
287 uint32_t count;
288 struct pnfs_block_extent *be = NULL, *save;
289 struct xdr_stream stream;
290 struct xdr_buf buf;
291 struct page *scratch;
292 __be32 *p;
293 struct layout_verification lv = {
294 .mode = lgr->range.iomode,
295 .start = lgr->range.offset >> SECTOR_SHIFT,
296 .inval = lgr->range.offset >> SECTOR_SHIFT,
297 .cowread = lgr->range.offset >> SECTOR_SHIFT,
298 };
299 LIST_HEAD(extents);
300
301 dprintk("---> %s\n", __func__);
302
303 scratch = alloc_page(gfp_flags);
304 if (!scratch)
305 return -ENOMEM;
306
307 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
308 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
309
310 p = xdr_inline_decode(&stream, 4);
311 if (unlikely(!p))
312 goto out_err;
313
314 count = be32_to_cpup(p++);
315
316 dprintk("%s enter, number of extents %i\n", __func__, count);
317 p = xdr_inline_decode(&stream, (28 + NFS4_DEVICEID4_SIZE) * count);
318 if (unlikely(!p))
319 goto out_err;
320
321 /* Decode individual extents, putting them in temporary
322 * staging area until whole layout is decoded to make error
323 * recovery easier.
324 */
325 for (i = 0; i < count; i++) {
326 be = bl_alloc_extent();
327 if (!be) {
328 status = -ENOMEM;
329 goto out_err;
330 }
331 memcpy(&be->be_devid, p, NFS4_DEVICEID4_SIZE);
332 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
333 be->be_mdev = translate_devid(lo, &be->be_devid);
334 if (!be->be_mdev)
335 goto out_err;
336
337 /* The next three values are read in as bytes,
338 * but stored as 512-byte sector lengths
339 */
340 if (decode_sector_number(&p, &be->be_f_offset) < 0)
341 goto out_err;
342 if (decode_sector_number(&p, &be->be_length) < 0)
343 goto out_err;
344 if (decode_sector_number(&p, &be->be_v_offset) < 0)
345 goto out_err;
346 be->be_state = be32_to_cpup(p++);
347 if (be->be_state == PNFS_BLOCK_INVALID_DATA)
348 be->be_inval = &bl->bl_inval;
349 if (verify_extent(be, &lv)) {
350 dprintk("%s verify failed\n", __func__);
351 goto out_err;
352 }
353 list_add_tail(&be->be_node, &extents);
354 }
355 if (lgr->range.offset + lgr->range.length !=
356 lv.start << SECTOR_SHIFT) {
357 dprintk("%s Final length mismatch\n", __func__);
358 be = NULL;
359 goto out_err;
360 }
361 if (lv.start < lv.cowread) {
362 dprintk("%s Final uncovered COW extent\n", __func__);
363 be = NULL;
364 goto out_err;
365 }
366 /* Extents decoded properly, now try to merge them in to
367 * existing layout extents.
368 */
369 spin_lock(&bl->bl_ext_lock);
370 list_for_each_entry_safe(be, save, &extents, be_node) {
371 list_del(&be->be_node);
372 status = bl_add_merge_extent(bl, be);
373 if (status) {
374 spin_unlock(&bl->bl_ext_lock);
375 /* This is a fairly catastrophic error, as the
376 * entire layout extent lists are now corrupted.
377 * We should have some way to distinguish this.
378 */
379 be = NULL;
380 goto out_err;
381 }
382 }
383 spin_unlock(&bl->bl_ext_lock);
384 status = 0;
385 out:
386 __free_page(scratch);
387 dprintk("%s returns %i\n", __func__, status);
388 return status;
389
390 out_err:
391 bl_put_extent(be);
392 while (!list_empty(&extents)) {
393 be = list_first_entry(&extents, struct pnfs_block_extent,
394 be_node);
395 list_del(&be->be_node);
396 bl_put_extent(be);
397 }
398 goto out;
399}