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
decode_sector_number(__be32 ** rp,sector_t * sp)43 static 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 /*
57 * Release the block device
58 */
nfs4_blkdev_put(struct block_device * bdev)59 void nfs4_blkdev_put(struct block_device *bdev)
60 {
61 dprintk("%s for device %d:%d\n", __func__, MAJOR(bdev->bd_dev),
62 MINOR(bdev->bd_dev));
63 blkdev_put(bdev, FMODE_READ);
64 }
65
bl_pipe_downcall(struct file * filp,const char __user * src,size_t mlen)66 ssize_t bl_pipe_downcall(struct file *filp, const char __user *src,
67 size_t mlen)
68 {
69 struct nfs_net *nn = net_generic(filp->f_dentry->d_sb->s_fs_info,
70 nfs_net_id);
71
72 if (mlen != sizeof (struct bl_dev_msg))
73 return -EINVAL;
74
75 if (copy_from_user(&nn->bl_mount_reply, src, mlen) != 0)
76 return -EFAULT;
77
78 wake_up(&nn->bl_wq);
79
80 return mlen;
81 }
82
bl_pipe_destroy_msg(struct rpc_pipe_msg * msg)83 void bl_pipe_destroy_msg(struct rpc_pipe_msg *msg)
84 {
85 struct bl_pipe_msg *bl_pipe_msg = container_of(msg, struct bl_pipe_msg, msg);
86
87 if (msg->errno >= 0)
88 return;
89 wake_up(bl_pipe_msg->bl_wq);
90 }
91
92 /*
93 * Decodes pnfs_block_deviceaddr4 which is XDR encoded in dev->dev_addr_buf.
94 */
95 struct pnfs_block_dev *
nfs4_blk_decode_device(struct nfs_server * server,struct pnfs_device * dev)96 nfs4_blk_decode_device(struct nfs_server *server,
97 struct pnfs_device *dev)
98 {
99 struct pnfs_block_dev *rv;
100 struct block_device *bd = NULL;
101 struct bl_pipe_msg bl_pipe_msg;
102 struct rpc_pipe_msg *msg = &bl_pipe_msg.msg;
103 struct bl_msg_hdr bl_msg = {
104 .type = BL_DEVICE_MOUNT,
105 .totallen = dev->mincount,
106 };
107 uint8_t *dataptr;
108 DECLARE_WAITQUEUE(wq, current);
109 int offset, len, i, rc;
110 struct net *net = server->nfs_client->cl_net;
111 struct nfs_net *nn = net_generic(net, nfs_net_id);
112 struct bl_dev_msg *reply = &nn->bl_mount_reply;
113
114 dprintk("%s CREATING PIPEFS MESSAGE\n", __func__);
115 dprintk("%s: deviceid: %s, mincount: %d\n", __func__, dev->dev_id.data,
116 dev->mincount);
117
118 bl_pipe_msg.bl_wq = &nn->bl_wq;
119 memset(msg, 0, sizeof(*msg));
120 msg->data = kzalloc(sizeof(bl_msg) + dev->mincount, GFP_NOFS);
121 if (!msg->data) {
122 rv = ERR_PTR(-ENOMEM);
123 goto out;
124 }
125
126 memcpy(msg->data, &bl_msg, sizeof(bl_msg));
127 dataptr = (uint8_t *) msg->data;
128 len = dev->mincount;
129 offset = sizeof(bl_msg);
130 for (i = 0; len > 0; i++) {
131 memcpy(&dataptr[offset], page_address(dev->pages[i]),
132 len < PAGE_CACHE_SIZE ? len : PAGE_CACHE_SIZE);
133 len -= PAGE_CACHE_SIZE;
134 offset += PAGE_CACHE_SIZE;
135 }
136 msg->len = sizeof(bl_msg) + dev->mincount;
137
138 dprintk("%s CALLING USERSPACE DAEMON\n", __func__);
139 add_wait_queue(&nn->bl_wq, &wq);
140 rc = rpc_queue_upcall(nn->bl_device_pipe, msg);
141 if (rc < 0) {
142 remove_wait_queue(&nn->bl_wq, &wq);
143 rv = ERR_PTR(rc);
144 goto out;
145 }
146
147 set_current_state(TASK_UNINTERRUPTIBLE);
148 schedule();
149 __set_current_state(TASK_RUNNING);
150 remove_wait_queue(&nn->bl_wq, &wq);
151
152 if (reply->status != BL_DEVICE_REQUEST_PROC) {
153 dprintk("%s failed to open device: %d\n",
154 __func__, reply->status);
155 rv = ERR_PTR(-EINVAL);
156 goto out;
157 }
158
159 bd = blkdev_get_by_dev(MKDEV(reply->major, reply->minor),
160 FMODE_READ, NULL);
161 if (IS_ERR(bd)) {
162 dprintk("%s failed to open device : %ld\n", __func__,
163 PTR_ERR(bd));
164 rv = ERR_CAST(bd);
165 goto out;
166 }
167
168 rv = kzalloc(sizeof(*rv), GFP_NOFS);
169 if (!rv) {
170 rv = ERR_PTR(-ENOMEM);
171 goto out;
172 }
173
174 rv->bm_mdev = bd;
175 memcpy(&rv->bm_mdevid, &dev->dev_id, sizeof(struct nfs4_deviceid));
176 rv->net = net;
177 dprintk("%s Created device %s with bd_block_size %u\n",
178 __func__,
179 bd->bd_disk->disk_name,
180 bd->bd_block_size);
181
182 out:
183 kfree(msg->data);
184 return rv;
185 }
186
187 /* Map deviceid returned by the server to constructed block_device */
translate_devid(struct pnfs_layout_hdr * lo,struct nfs4_deviceid * id)188 static struct block_device *translate_devid(struct pnfs_layout_hdr *lo,
189 struct nfs4_deviceid *id)
190 {
191 struct block_device *rv = NULL;
192 struct block_mount_id *mid;
193 struct pnfs_block_dev *dev;
194
195 dprintk("%s enter, lo=%p, id=%p\n", __func__, lo, id);
196 mid = BLK_ID(lo);
197 spin_lock(&mid->bm_lock);
198 list_for_each_entry(dev, &mid->bm_devlist, bm_node) {
199 if (memcmp(id->data, dev->bm_mdevid.data,
200 NFS4_DEVICEID4_SIZE) == 0) {
201 rv = dev->bm_mdev;
202 goto out;
203 }
204 }
205 out:
206 spin_unlock(&mid->bm_lock);
207 dprintk("%s returning %p\n", __func__, rv);
208 return rv;
209 }
210
211 /* Tracks info needed to ensure extents in layout obey constraints of spec */
212 struct layout_verification {
213 u32 mode; /* R or RW */
214 u64 start; /* Expected start of next non-COW extent */
215 u64 inval; /* Start of INVAL coverage */
216 u64 cowread; /* End of COW read coverage */
217 };
218
219 /* Verify the extent meets the layout requirements of the pnfs-block draft,
220 * section 2.3.1.
221 */
verify_extent(struct pnfs_block_extent * be,struct layout_verification * lv)222 static int verify_extent(struct pnfs_block_extent *be,
223 struct layout_verification *lv)
224 {
225 if (lv->mode == IOMODE_READ) {
226 if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
227 be->be_state == PNFS_BLOCK_INVALID_DATA)
228 return -EIO;
229 if (be->be_f_offset != lv->start)
230 return -EIO;
231 lv->start += be->be_length;
232 return 0;
233 }
234 /* lv->mode == IOMODE_RW */
235 if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
236 if (be->be_f_offset != lv->start)
237 return -EIO;
238 if (lv->cowread > lv->start)
239 return -EIO;
240 lv->start += be->be_length;
241 lv->inval = lv->start;
242 return 0;
243 } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
244 if (be->be_f_offset != lv->start)
245 return -EIO;
246 lv->start += be->be_length;
247 return 0;
248 } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
249 if (be->be_f_offset > lv->start)
250 return -EIO;
251 if (be->be_f_offset < lv->inval)
252 return -EIO;
253 if (be->be_f_offset < lv->cowread)
254 return -EIO;
255 /* It looks like you might want to min this with lv->start,
256 * but you really don't.
257 */
258 lv->inval = lv->inval + be->be_length;
259 lv->cowread = be->be_f_offset + be->be_length;
260 return 0;
261 } else
262 return -EIO;
263 }
264
265 /* XDR decode pnfs_block_layout4 structure */
266 int
nfs4_blk_process_layoutget(struct pnfs_layout_hdr * lo,struct nfs4_layoutget_res * lgr,gfp_t gfp_flags)267 nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
268 struct nfs4_layoutget_res *lgr, gfp_t gfp_flags)
269 {
270 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
271 int i, status = -EIO;
272 uint32_t count;
273 struct pnfs_block_extent *be = NULL, *save;
274 struct xdr_stream stream;
275 struct xdr_buf buf;
276 struct page *scratch;
277 __be32 *p;
278 struct layout_verification lv = {
279 .mode = lgr->range.iomode,
280 .start = lgr->range.offset >> SECTOR_SHIFT,
281 .inval = lgr->range.offset >> SECTOR_SHIFT,
282 .cowread = lgr->range.offset >> SECTOR_SHIFT,
283 };
284 LIST_HEAD(extents);
285
286 dprintk("---> %s\n", __func__);
287
288 scratch = alloc_page(gfp_flags);
289 if (!scratch)
290 return -ENOMEM;
291
292 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
293 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
294
295 p = xdr_inline_decode(&stream, 4);
296 if (unlikely(!p))
297 goto out_err;
298
299 count = be32_to_cpup(p++);
300
301 dprintk("%s enter, number of extents %i\n", __func__, count);
302 p = xdr_inline_decode(&stream, (28 + NFS4_DEVICEID4_SIZE) * count);
303 if (unlikely(!p))
304 goto out_err;
305
306 /* Decode individual extents, putting them in temporary
307 * staging area until whole layout is decoded to make error
308 * recovery easier.
309 */
310 for (i = 0; i < count; i++) {
311 be = bl_alloc_extent();
312 if (!be) {
313 status = -ENOMEM;
314 goto out_err;
315 }
316 memcpy(&be->be_devid, p, NFS4_DEVICEID4_SIZE);
317 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
318 be->be_mdev = translate_devid(lo, &be->be_devid);
319 if (!be->be_mdev)
320 goto out_err;
321
322 /* The next three values are read in as bytes,
323 * but stored as 512-byte sector lengths
324 */
325 if (decode_sector_number(&p, &be->be_f_offset) < 0)
326 goto out_err;
327 if (decode_sector_number(&p, &be->be_length) < 0)
328 goto out_err;
329 if (decode_sector_number(&p, &be->be_v_offset) < 0)
330 goto out_err;
331 be->be_state = be32_to_cpup(p++);
332 if (be->be_state == PNFS_BLOCK_INVALID_DATA)
333 be->be_inval = &bl->bl_inval;
334 if (verify_extent(be, &lv)) {
335 dprintk("%s verify failed\n", __func__);
336 goto out_err;
337 }
338 list_add_tail(&be->be_node, &extents);
339 }
340 if (lgr->range.offset + lgr->range.length !=
341 lv.start << SECTOR_SHIFT) {
342 dprintk("%s Final length mismatch\n", __func__);
343 be = NULL;
344 goto out_err;
345 }
346 if (lv.start < lv.cowread) {
347 dprintk("%s Final uncovered COW extent\n", __func__);
348 be = NULL;
349 goto out_err;
350 }
351 /* Extents decoded properly, now try to merge them in to
352 * existing layout extents.
353 */
354 spin_lock(&bl->bl_ext_lock);
355 list_for_each_entry_safe(be, save, &extents, be_node) {
356 list_del(&be->be_node);
357 status = bl_add_merge_extent(bl, be);
358 if (status) {
359 spin_unlock(&bl->bl_ext_lock);
360 /* This is a fairly catastrophic error, as the
361 * entire layout extent lists are now corrupted.
362 * We should have some way to distinguish this.
363 */
364 be = NULL;
365 goto out_err;
366 }
367 }
368 spin_unlock(&bl->bl_ext_lock);
369 status = 0;
370 out:
371 __free_page(scratch);
372 dprintk("%s returns %i\n", __func__, status);
373 return status;
374
375 out_err:
376 bl_put_extent(be);
377 while (!list_empty(&extents)) {
378 be = list_first_entry(&extents, struct pnfs_block_extent,
379 be_node);
380 list_del(&be->be_node);
381 bl_put_extent(be);
382 }
383 goto out;
384 }
385