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 "%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. */
nfs4_blkdev_get(dev_t dev)57 struct 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;
66 fail:
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 */
nfs4_blkdev_put(struct block_device * bdev)75 int 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
82 static struct bl_dev_msg bl_mount_reply;
83
bl_pipe_downcall(struct file * filp,const char __user * src,size_t mlen)84 ssize_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
bl_pipe_destroy_msg(struct rpc_pipe_msg * msg)98 void 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 */
108 struct pnfs_block_dev *
nfs4_blk_decode_device(struct nfs_server * server,struct pnfs_device * dev)109 nfs4_blk_decode_device(struct nfs_server *server,
110 struct pnfs_device *dev)
111 {
112 struct pnfs_block_dev *rv;
113 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;
122 int offset, len, i, rc;
123
124 dprintk("%s CREATING PIPEFS MESSAGE\n", __func__);
125 dprintk("%s: deviceid: %s, mincount: %d\n", __func__, dev->dev_id.data,
126 dev->mincount);
127
128 memset(&msg, 0, sizeof(msg));
129 msg.data = kzalloc(sizeof(bl_msg) + dev->mincount, GFP_NOFS);
130 if (!msg.data) {
131 rv = ERR_PTR(-ENOMEM);
132 goto out;
133 }
134
135 memcpy(msg.data, &bl_msg, sizeof(bl_msg));
136 dataptr = (uint8_t *) msg.data;
137 len = dev->mincount;
138 offset = sizeof(bl_msg);
139 for (i = 0; len > 0; i++) {
140 memcpy(&dataptr[offset], page_address(dev->pages[i]),
141 len < PAGE_CACHE_SIZE ? len : PAGE_CACHE_SIZE);
142 len -= PAGE_CACHE_SIZE;
143 offset += PAGE_CACHE_SIZE;
144 }
145 msg.len = sizeof(bl_msg) + dev->mincount;
146
147 dprintk("%s CALLING USERSPACE DAEMON\n", __func__);
148 add_wait_queue(&bl_wq, &wq);
149 rc = rpc_queue_upcall(bl_device_pipe->d_inode, &msg);
150 if (rc < 0) {
151 remove_wait_queue(&bl_wq, &wq);
152 rv = ERR_PTR(rc);
153 goto out;
154 }
155
156 set_current_state(TASK_UNINTERRUPTIBLE);
157 schedule();
158 __set_current_state(TASK_RUNNING);
159 remove_wait_queue(&bl_wq, &wq);
160
161 if (reply->status != BL_DEVICE_REQUEST_PROC) {
162 dprintk("%s failed to open device: %d\n",
163 __func__, reply->status);
164 rv = ERR_PTR(-EINVAL);
165 goto out;
166 }
167
168 bd = nfs4_blkdev_get(MKDEV(reply->major, reply->minor));
169 if (IS_ERR(bd)) {
170 rc = PTR_ERR(bd);
171 dprintk("%s failed to open device : %d\n", __func__, rc);
172 rv = ERR_PTR(rc);
173 goto out;
174 }
175
176 rv = kzalloc(sizeof(*rv), GFP_NOFS);
177 if (!rv) {
178 rv = ERR_PTR(-ENOMEM);
179 goto out;
180 }
181
182 rv->bm_mdev = bd;
183 memcpy(&rv->bm_mdevid, &dev->dev_id, sizeof(struct nfs4_deviceid));
184 dprintk("%s Created device %s with bd_block_size %u\n",
185 __func__,
186 bd->bd_disk->disk_name,
187 bd->bd_block_size);
188
189 out:
190 kfree(msg.data);
191 return rv;
192 }
193
194 /* Map deviceid returned by the server to constructed block_device */
translate_devid(struct pnfs_layout_hdr * lo,struct nfs4_deviceid * id)195 static struct block_device *translate_devid(struct pnfs_layout_hdr *lo,
196 struct nfs4_deviceid *id)
197 {
198 struct block_device *rv = NULL;
199 struct block_mount_id *mid;
200 struct pnfs_block_dev *dev;
201
202 dprintk("%s enter, lo=%p, id=%p\n", __func__, lo, id);
203 mid = BLK_ID(lo);
204 spin_lock(&mid->bm_lock);
205 list_for_each_entry(dev, &mid->bm_devlist, bm_node) {
206 if (memcmp(id->data, dev->bm_mdevid.data,
207 NFS4_DEVICEID4_SIZE) == 0) {
208 rv = dev->bm_mdev;
209 goto out;
210 }
211 }
212 out:
213 spin_unlock(&mid->bm_lock);
214 dprintk("%s returning %p\n", __func__, rv);
215 return rv;
216 }
217
218 /* Tracks info needed to ensure extents in layout obey constraints of spec */
219 struct layout_verification {
220 u32 mode; /* R or RW */
221 u64 start; /* Expected start of next non-COW extent */
222 u64 inval; /* Start of INVAL coverage */
223 u64 cowread; /* End of COW read coverage */
224 };
225
226 /* Verify the extent meets the layout requirements of the pnfs-block draft,
227 * section 2.3.1.
228 */
verify_extent(struct pnfs_block_extent * be,struct layout_verification * lv)229 static int verify_extent(struct pnfs_block_extent *be,
230 struct layout_verification *lv)
231 {
232 if (lv->mode == IOMODE_READ) {
233 if (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
234 be->be_state == PNFS_BLOCK_INVALID_DATA)
235 return -EIO;
236 if (be->be_f_offset != lv->start)
237 return -EIO;
238 lv->start += be->be_length;
239 return 0;
240 }
241 /* lv->mode == IOMODE_RW */
242 if (be->be_state == PNFS_BLOCK_READWRITE_DATA) {
243 if (be->be_f_offset != lv->start)
244 return -EIO;
245 if (lv->cowread > lv->start)
246 return -EIO;
247 lv->start += be->be_length;
248 lv->inval = lv->start;
249 return 0;
250 } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
251 if (be->be_f_offset != lv->start)
252 return -EIO;
253 lv->start += be->be_length;
254 return 0;
255 } else if (be->be_state == PNFS_BLOCK_READ_DATA) {
256 if (be->be_f_offset > lv->start)
257 return -EIO;
258 if (be->be_f_offset < lv->inval)
259 return -EIO;
260 if (be->be_f_offset < lv->cowread)
261 return -EIO;
262 /* It looks like you might want to min this with lv->start,
263 * but you really don't.
264 */
265 lv->inval = lv->inval + be->be_length;
266 lv->cowread = be->be_f_offset + be->be_length;
267 return 0;
268 } else
269 return -EIO;
270 }
271
272 /* XDR decode pnfs_block_layout4 structure */
273 int
nfs4_blk_process_layoutget(struct pnfs_layout_hdr * lo,struct nfs4_layoutget_res * lgr,gfp_t gfp_flags)274 nfs4_blk_process_layoutget(struct pnfs_layout_hdr *lo,
275 struct nfs4_layoutget_res *lgr, gfp_t gfp_flags)
276 {
277 struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
278 int i, status = -EIO;
279 uint32_t count;
280 struct pnfs_block_extent *be = NULL, *save;
281 struct xdr_stream stream;
282 struct xdr_buf buf;
283 struct page *scratch;
284 __be32 *p;
285 struct layout_verification lv = {
286 .mode = lgr->range.iomode,
287 .start = lgr->range.offset >> SECTOR_SHIFT,
288 .inval = lgr->range.offset >> SECTOR_SHIFT,
289 .cowread = lgr->range.offset >> SECTOR_SHIFT,
290 };
291 LIST_HEAD(extents);
292
293 dprintk("---> %s\n", __func__);
294
295 scratch = alloc_page(gfp_flags);
296 if (!scratch)
297 return -ENOMEM;
298
299 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, lgr->layoutp->len);
300 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
301
302 p = xdr_inline_decode(&stream, 4);
303 if (unlikely(!p))
304 goto out_err;
305
306 count = be32_to_cpup(p++);
307
308 dprintk("%s enter, number of extents %i\n", __func__, count);
309 p = xdr_inline_decode(&stream, (28 + NFS4_DEVICEID4_SIZE) * count);
310 if (unlikely(!p))
311 goto out_err;
312
313 /* Decode individual extents, putting them in temporary
314 * staging area until whole layout is decoded to make error
315 * recovery easier.
316 */
317 for (i = 0; i < count; i++) {
318 be = bl_alloc_extent();
319 if (!be) {
320 status = -ENOMEM;
321 goto out_err;
322 }
323 memcpy(&be->be_devid, p, NFS4_DEVICEID4_SIZE);
324 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
325 be->be_mdev = translate_devid(lo, &be->be_devid);
326 if (!be->be_mdev)
327 goto out_err;
328
329 /* The next three values are read in as bytes,
330 * but stored as 512-byte sector lengths
331 */
332 if (decode_sector_number(&p, &be->be_f_offset) < 0)
333 goto out_err;
334 if (decode_sector_number(&p, &be->be_length) < 0)
335 goto out_err;
336 if (decode_sector_number(&p, &be->be_v_offset) < 0)
337 goto out_err;
338 be->be_state = be32_to_cpup(p++);
339 if (be->be_state == PNFS_BLOCK_INVALID_DATA)
340 be->be_inval = &bl->bl_inval;
341 if (verify_extent(be, &lv)) {
342 dprintk("%s verify failed\n", __func__);
343 goto out_err;
344 }
345 list_add_tail(&be->be_node, &extents);
346 }
347 if (lgr->range.offset + lgr->range.length !=
348 lv.start << SECTOR_SHIFT) {
349 dprintk("%s Final length mismatch\n", __func__);
350 be = NULL;
351 goto out_err;
352 }
353 if (lv.start < lv.cowread) {
354 dprintk("%s Final uncovered COW extent\n", __func__);
355 be = NULL;
356 goto out_err;
357 }
358 /* Extents decoded properly, now try to merge them in to
359 * existing layout extents.
360 */
361 spin_lock(&bl->bl_ext_lock);
362 list_for_each_entry_safe(be, save, &extents, be_node) {
363 list_del(&be->be_node);
364 status = bl_add_merge_extent(bl, be);
365 if (status) {
366 spin_unlock(&bl->bl_ext_lock);
367 /* This is a fairly catastrophic error, as the
368 * entire layout extent lists are now corrupted.
369 * We should have some way to distinguish this.
370 */
371 be = NULL;
372 goto out_err;
373 }
374 }
375 spin_unlock(&bl->bl_ext_lock);
376 status = 0;
377 out:
378 __free_page(scratch);
379 dprintk("%s returns %i\n", __func__, status);
380 return status;
381
382 out_err:
383 bl_put_extent(be);
384 while (!list_empty(&extents)) {
385 be = list_first_entry(&extents, struct pnfs_block_extent,
386 be_node);
387 list_del(&be->be_node);
388 bl_put_extent(be);
389 }
390 goto out;
391 }
392