Commit | Line | Data |
---|---|---|
7ab672ce DH |
1 | /* |
2 | * Module for the pnfs nfs4 file layout driver. | |
3 | * Defines all I/O and Policy interface operations, plus code | |
4 | * to register itself with the pNFS client. | |
5 | * | |
6 | * Copyright (c) 2002 | |
7 | * The Regents of the University of Michigan | |
8 | * All Rights Reserved | |
9 | * | |
10 | * Dean Hildebrand <dhildebz@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 or warranty | |
22 | * of any kind either express or implied, including without limitation | |
23 | * the implied warranties of merchantability, fitness for a particular | |
24 | * purpose, or noninfringement. The Regents of the University of | |
25 | * Michigan shall not be liable for any damages, including special, | |
26 | * indirect, incidental, or consequential damages, with respect to any | |
27 | * claim arising out of or in connection with the use of the software, | |
28 | * even if it has been or is hereafter advised of the possibility of | |
29 | * such damages. | |
30 | */ | |
31 | ||
32 | #include <linux/nfs_fs.h> | |
16b374ca AA |
33 | |
34 | #include "internal.h" | |
35 | #include "nfs4filelayout.h" | |
7ab672ce DH |
36 | |
37 | #define NFSDBG_FACILITY NFSDBG_PNFS_LD | |
38 | ||
39 | MODULE_LICENSE("GPL"); | |
40 | MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>"); | |
41 | MODULE_DESCRIPTION("The NFSv4 file layout driver"); | |
42 | ||
cbdabc7f AA |
43 | #define FILELAYOUT_POLL_RETRY_MAX (15*HZ) |
44 | ||
cfe7f412 FI |
45 | static loff_t |
46 | filelayout_get_dense_offset(struct nfs4_filelayout_segment *flseg, | |
47 | loff_t offset) | |
48 | { | |
49 | u32 stripe_width = flseg->stripe_unit * flseg->dsaddr->stripe_count; | |
50 | u64 tmp; | |
51 | ||
52 | offset -= flseg->pattern_offset; | |
53 | tmp = offset; | |
54 | do_div(tmp, stripe_width); | |
55 | ||
56 | return tmp * flseg->stripe_unit + do_div(offset, flseg->stripe_unit); | |
57 | } | |
58 | ||
59 | /* This function is used by the layout driver to calculate the | |
60 | * offset of the file on the dserver based on whether the | |
61 | * layout type is STRIPE_DENSE or STRIPE_SPARSE | |
62 | */ | |
63 | static loff_t | |
64 | filelayout_get_dserver_offset(struct pnfs_layout_segment *lseg, loff_t offset) | |
65 | { | |
66 | struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg); | |
67 | ||
68 | switch (flseg->stripe_type) { | |
69 | case STRIPE_SPARSE: | |
70 | return offset; | |
71 | ||
72 | case STRIPE_DENSE: | |
73 | return filelayout_get_dense_offset(flseg, offset); | |
74 | } | |
75 | ||
76 | BUG(); | |
77 | } | |
78 | ||
cbdabc7f AA |
79 | /* For data server errors we don't recover from */ |
80 | static void | |
81 | filelayout_set_lo_fail(struct pnfs_layout_segment *lseg) | |
82 | { | |
83 | if (lseg->pls_range.iomode == IOMODE_RW) { | |
84 | dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__); | |
85 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
86 | } else { | |
87 | dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__); | |
88 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
89 | } | |
90 | } | |
91 | ||
92 | static int filelayout_async_handle_error(struct rpc_task *task, | |
93 | struct nfs4_state *state, | |
94 | struct nfs_client *clp, | |
95 | int *reset) | |
96 | { | |
97 | if (task->tk_status >= 0) | |
98 | return 0; | |
99 | ||
100 | *reset = 0; | |
101 | ||
102 | switch (task->tk_status) { | |
103 | case -NFS4ERR_BADSESSION: | |
104 | case -NFS4ERR_BADSLOT: | |
105 | case -NFS4ERR_BAD_HIGH_SLOT: | |
106 | case -NFS4ERR_DEADSESSION: | |
107 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: | |
108 | case -NFS4ERR_SEQ_FALSE_RETRY: | |
109 | case -NFS4ERR_SEQ_MISORDERED: | |
110 | dprintk("%s ERROR %d, Reset session. Exchangeid " | |
111 | "flags 0x%x\n", __func__, task->tk_status, | |
112 | clp->cl_exchange_flags); | |
113 | nfs4_schedule_session_recovery(clp->cl_session); | |
114 | break; | |
115 | case -NFS4ERR_DELAY: | |
116 | case -NFS4ERR_GRACE: | |
117 | case -EKEYEXPIRED: | |
118 | rpc_delay(task, FILELAYOUT_POLL_RETRY_MAX); | |
119 | break; | |
120 | default: | |
121 | dprintk("%s DS error. Retry through MDS %d\n", __func__, | |
122 | task->tk_status); | |
123 | *reset = 1; | |
124 | break; | |
125 | } | |
126 | task->tk_status = 0; | |
127 | return -EAGAIN; | |
128 | } | |
129 | ||
130 | /* NFS_PROTO call done callback routines */ | |
131 | ||
132 | static int filelayout_read_done_cb(struct rpc_task *task, | |
133 | struct nfs_read_data *data) | |
134 | { | |
135 | struct nfs_client *clp = data->ds_clp; | |
136 | int reset = 0; | |
137 | ||
138 | dprintk("%s DS read\n", __func__); | |
139 | ||
140 | if (filelayout_async_handle_error(task, data->args.context->state, | |
141 | data->ds_clp, &reset) == -EAGAIN) { | |
142 | dprintk("%s calling restart ds_clp %p ds_clp->cl_session %p\n", | |
143 | __func__, data->ds_clp, data->ds_clp->cl_session); | |
144 | if (reset) { | |
145 | filelayout_set_lo_fail(data->lseg); | |
146 | nfs4_reset_read(task, data); | |
147 | clp = NFS_SERVER(data->inode)->nfs_client; | |
148 | } | |
149 | nfs_restart_rpc(task, clp); | |
150 | return -EAGAIN; | |
151 | } | |
152 | ||
153 | return 0; | |
154 | } | |
155 | ||
dc70d7b3 AA |
156 | /* |
157 | * Call ops for the async read/write cases | |
158 | * In the case of dense layouts, the offset needs to be reset to its | |
159 | * original value. | |
160 | */ | |
161 | static void filelayout_read_prepare(struct rpc_task *task, void *data) | |
162 | { | |
163 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
164 | ||
cbdabc7f AA |
165 | rdata->read_done_cb = filelayout_read_done_cb; |
166 | ||
dc70d7b3 AA |
167 | if (nfs41_setup_sequence(rdata->ds_clp->cl_session, |
168 | &rdata->args.seq_args, &rdata->res.seq_res, | |
169 | 0, task)) | |
170 | return; | |
171 | ||
172 | rpc_call_start(task); | |
173 | } | |
174 | ||
175 | static void filelayout_read_call_done(struct rpc_task *task, void *data) | |
176 | { | |
177 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
178 | ||
179 | dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status); | |
180 | ||
181 | /* Note this may cause RPC to be resent */ | |
182 | rdata->mds_ops->rpc_call_done(task, data); | |
183 | } | |
184 | ||
185 | static void filelayout_read_release(void *data) | |
186 | { | |
187 | struct nfs_read_data *rdata = (struct nfs_read_data *)data; | |
188 | ||
189 | rdata->mds_ops->rpc_release(data); | |
190 | } | |
191 | ||
192 | struct rpc_call_ops filelayout_read_call_ops = { | |
193 | .rpc_call_prepare = filelayout_read_prepare, | |
194 | .rpc_call_done = filelayout_read_call_done, | |
195 | .rpc_release = filelayout_read_release, | |
196 | }; | |
197 | ||
198 | static enum pnfs_try_status | |
199 | filelayout_read_pagelist(struct nfs_read_data *data) | |
200 | { | |
201 | struct pnfs_layout_segment *lseg = data->lseg; | |
202 | struct nfs4_pnfs_ds *ds; | |
203 | loff_t offset = data->args.offset; | |
204 | u32 j, idx; | |
205 | struct nfs_fh *fh; | |
206 | int status; | |
207 | ||
208 | dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n", | |
209 | __func__, data->inode->i_ino, | |
210 | data->args.pgbase, (size_t)data->args.count, offset); | |
211 | ||
212 | /* Retrieve the correct rpc_client for the byte range */ | |
213 | j = nfs4_fl_calc_j_index(lseg, offset); | |
214 | idx = nfs4_fl_calc_ds_index(lseg, j); | |
215 | ds = nfs4_fl_prepare_ds(lseg, idx); | |
216 | if (!ds) { | |
568e8c49 AA |
217 | /* Either layout fh index faulty, or ds connect failed */ |
218 | set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags); | |
219 | set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags); | |
dc70d7b3 AA |
220 | return PNFS_NOT_ATTEMPTED; |
221 | } | |
222 | dprintk("%s USE DS:ip %x %hu\n", __func__, | |
223 | ntohl(ds->ds_ip_addr), ntohs(ds->ds_port)); | |
224 | ||
225 | /* No multipath support. Use first DS */ | |
226 | data->ds_clp = ds->ds_clp; | |
227 | fh = nfs4_fl_select_ds_fh(lseg, j); | |
228 | if (fh) | |
229 | data->args.fh = fh; | |
230 | ||
231 | data->args.offset = filelayout_get_dserver_offset(lseg, offset); | |
232 | data->mds_offset = offset; | |
233 | ||
234 | /* Perform an asynchronous read to ds */ | |
235 | status = nfs_initiate_read(data, ds->ds_clp->cl_rpcclient, | |
236 | &filelayout_read_call_ops); | |
237 | BUG_ON(status != 0); | |
238 | return PNFS_ATTEMPTED; | |
239 | } | |
240 | ||
16b374ca AA |
241 | /* |
242 | * filelayout_check_layout() | |
243 | * | |
244 | * Make sure layout segment parameters are sane WRT the device. | |
245 | * At this point no generic layer initialization of the lseg has occurred, | |
246 | * and nothing has been added to the layout_hdr cache. | |
247 | * | |
248 | */ | |
249 | static int | |
250 | filelayout_check_layout(struct pnfs_layout_hdr *lo, | |
251 | struct nfs4_filelayout_segment *fl, | |
252 | struct nfs4_layoutget_res *lgr, | |
253 | struct nfs4_deviceid *id) | |
254 | { | |
255 | struct nfs4_file_layout_dsaddr *dsaddr; | |
256 | int status = -EINVAL; | |
b7edfaa1 | 257 | struct nfs_server *nfss = NFS_SERVER(lo->plh_inode); |
16b374ca AA |
258 | |
259 | dprintk("--> %s\n", __func__); | |
260 | ||
261 | if (fl->pattern_offset > lgr->range.offset) { | |
262 | dprintk("%s pattern_offset %lld to large\n", | |
263 | __func__, fl->pattern_offset); | |
264 | goto out; | |
265 | } | |
266 | ||
267 | if (fl->stripe_unit % PAGE_SIZE) { | |
268 | dprintk("%s Stripe unit (%u) not page aligned\n", | |
269 | __func__, fl->stripe_unit); | |
270 | goto out; | |
271 | } | |
272 | ||
273 | /* find and reference the deviceid */ | |
ea8eecdd | 274 | dsaddr = nfs4_fl_find_get_deviceid(id); |
16b374ca | 275 | if (dsaddr == NULL) { |
b7edfaa1 | 276 | dsaddr = get_device_info(lo->plh_inode, id); |
16b374ca AA |
277 | if (dsaddr == NULL) |
278 | goto out; | |
279 | } | |
280 | fl->dsaddr = dsaddr; | |
281 | ||
282 | if (fl->first_stripe_index < 0 || | |
283 | fl->first_stripe_index >= dsaddr->stripe_count) { | |
284 | dprintk("%s Bad first_stripe_index %d\n", | |
285 | __func__, fl->first_stripe_index); | |
286 | goto out_put; | |
287 | } | |
288 | ||
289 | if ((fl->stripe_type == STRIPE_SPARSE && | |
290 | fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) || | |
291 | (fl->stripe_type == STRIPE_DENSE && | |
292 | fl->num_fh != dsaddr->stripe_count)) { | |
293 | dprintk("%s num_fh %u not valid for given packing\n", | |
294 | __func__, fl->num_fh); | |
295 | goto out_put; | |
296 | } | |
297 | ||
298 | if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) { | |
299 | dprintk("%s Stripe unit (%u) not aligned with rsize %u " | |
300 | "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize, | |
301 | nfss->wsize); | |
302 | } | |
303 | ||
304 | status = 0; | |
305 | out: | |
306 | dprintk("--> %s returns %d\n", __func__, status); | |
307 | return status; | |
308 | out_put: | |
ea8eecdd | 309 | nfs4_fl_put_deviceid(dsaddr); |
16b374ca AA |
310 | goto out; |
311 | } | |
312 | ||
313 | static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl) | |
314 | { | |
315 | int i; | |
316 | ||
317 | for (i = 0; i < fl->num_fh; i++) { | |
318 | if (!fl->fh_array[i]) | |
319 | break; | |
320 | kfree(fl->fh_array[i]); | |
321 | } | |
322 | kfree(fl->fh_array); | |
323 | fl->fh_array = NULL; | |
324 | } | |
325 | ||
326 | static void | |
327 | _filelayout_free_lseg(struct nfs4_filelayout_segment *fl) | |
328 | { | |
329 | filelayout_free_fh_array(fl); | |
330 | kfree(fl); | |
331 | } | |
332 | ||
333 | static int | |
334 | filelayout_decode_layout(struct pnfs_layout_hdr *flo, | |
335 | struct nfs4_filelayout_segment *fl, | |
336 | struct nfs4_layoutget_res *lgr, | |
337 | struct nfs4_deviceid *id) | |
338 | { | |
339 | uint32_t *p = (uint32_t *)lgr->layout.buf; | |
340 | uint32_t nfl_util; | |
341 | int i; | |
342 | ||
343 | dprintk("%s: set_layout_map Begin\n", __func__); | |
344 | ||
345 | memcpy(id, p, sizeof(*id)); | |
346 | p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); | |
347 | print_deviceid(id); | |
348 | ||
349 | nfl_util = be32_to_cpup(p++); | |
350 | if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS) | |
351 | fl->commit_through_mds = 1; | |
352 | if (nfl_util & NFL4_UFLG_DENSE) | |
353 | fl->stripe_type = STRIPE_DENSE; | |
354 | else | |
355 | fl->stripe_type = STRIPE_SPARSE; | |
356 | fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK; | |
357 | ||
358 | fl->first_stripe_index = be32_to_cpup(p++); | |
359 | p = xdr_decode_hyper(p, &fl->pattern_offset); | |
360 | fl->num_fh = be32_to_cpup(p++); | |
361 | ||
362 | dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n", | |
363 | __func__, nfl_util, fl->num_fh, fl->first_stripe_index, | |
364 | fl->pattern_offset); | |
365 | ||
366 | fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *), | |
367 | GFP_KERNEL); | |
368 | if (!fl->fh_array) | |
369 | return -ENOMEM; | |
370 | ||
371 | for (i = 0; i < fl->num_fh; i++) { | |
372 | /* Do we want to use a mempool here? */ | |
373 | fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL); | |
374 | if (!fl->fh_array[i]) { | |
375 | filelayout_free_fh_array(fl); | |
376 | return -ENOMEM; | |
377 | } | |
378 | fl->fh_array[i]->size = be32_to_cpup(p++); | |
379 | if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) { | |
380 | printk(KERN_ERR "Too big fh %d received %d\n", | |
381 | i, fl->fh_array[i]->size); | |
382 | filelayout_free_fh_array(fl); | |
383 | return -EIO; | |
384 | } | |
385 | memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size); | |
386 | p += XDR_QUADLEN(fl->fh_array[i]->size); | |
387 | dprintk("DEBUG: %s: fh len %d\n", __func__, | |
388 | fl->fh_array[i]->size); | |
389 | } | |
390 | ||
391 | return 0; | |
392 | } | |
393 | ||
394 | static struct pnfs_layout_segment * | |
395 | filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid, | |
396 | struct nfs4_layoutget_res *lgr) | |
397 | { | |
398 | struct nfs4_filelayout_segment *fl; | |
399 | int rc; | |
400 | struct nfs4_deviceid id; | |
401 | ||
402 | dprintk("--> %s\n", __func__); | |
403 | fl = kzalloc(sizeof(*fl), GFP_KERNEL); | |
404 | if (!fl) | |
405 | return NULL; | |
406 | ||
407 | rc = filelayout_decode_layout(layoutid, fl, lgr, &id); | |
408 | if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id)) { | |
409 | _filelayout_free_lseg(fl); | |
410 | return NULL; | |
411 | } | |
412 | return &fl->generic_hdr; | |
413 | } | |
414 | ||
415 | static void | |
416 | filelayout_free_lseg(struct pnfs_layout_segment *lseg) | |
417 | { | |
16b374ca AA |
418 | struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg); |
419 | ||
420 | dprintk("--> %s\n", __func__); | |
ea8eecdd | 421 | nfs4_fl_put_deviceid(fl->dsaddr); |
16b374ca AA |
422 | _filelayout_free_lseg(fl); |
423 | } | |
424 | ||
94ad1c80 FI |
425 | /* |
426 | * filelayout_pg_test(). Called by nfs_can_coalesce_requests() | |
427 | * | |
428 | * return 1 : coalesce page | |
429 | * return 0 : don't coalesce page | |
430 | */ | |
431 | int | |
432 | filelayout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, | |
433 | struct nfs_page *req) | |
434 | { | |
435 | u64 p_stripe, r_stripe; | |
436 | u32 stripe_unit; | |
437 | ||
438 | if (!pgio->pg_lseg) | |
439 | return 1; | |
440 | p_stripe = (u64)prev->wb_index << PAGE_CACHE_SHIFT; | |
441 | r_stripe = (u64)req->wb_index << PAGE_CACHE_SHIFT; | |
442 | stripe_unit = FILELAYOUT_LSEG(pgio->pg_lseg)->stripe_unit; | |
443 | ||
444 | do_div(p_stripe, stripe_unit); | |
445 | do_div(r_stripe, stripe_unit); | |
446 | ||
447 | return (p_stripe == r_stripe); | |
448 | } | |
449 | ||
7ab672ce | 450 | static struct pnfs_layoutdriver_type filelayout_type = { |
ea8eecdd CH |
451 | .id = LAYOUT_NFSV4_1_FILES, |
452 | .name = "LAYOUT_NFSV4_1_FILES", | |
453 | .owner = THIS_MODULE, | |
454 | .alloc_lseg = filelayout_alloc_lseg, | |
455 | .free_lseg = filelayout_free_lseg, | |
94ad1c80 | 456 | .pg_test = filelayout_pg_test, |
dc70d7b3 | 457 | .read_pagelist = filelayout_read_pagelist, |
7ab672ce DH |
458 | }; |
459 | ||
460 | static int __init nfs4filelayout_init(void) | |
461 | { | |
462 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n", | |
463 | __func__); | |
464 | return pnfs_register_layoutdriver(&filelayout_type); | |
465 | } | |
466 | ||
467 | static void __exit nfs4filelayout_exit(void) | |
468 | { | |
469 | printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n", | |
470 | __func__); | |
471 | pnfs_unregister_layoutdriver(&filelayout_type); | |
472 | } | |
473 | ||
474 | module_init(nfs4filelayout_init); | |
475 | module_exit(nfs4filelayout_exit); |