2 * Device operations for the pnfs nfs4 file layout driver.
4 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
6 * Tao Peng <bergwolf@primarydata.com>
9 #include <linux/nfs_fs.h>
10 #include <linux/vmalloc.h>
11 #include <linux/module.h>
12 #include <linux/sunrpc/addr.h>
14 #include "../internal.h"
15 #include "../nfs4session.h"
16 #include "flexfilelayout.h"
18 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
20 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
21 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
23 void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
26 nfs4_put_deviceid_node(&mirror_ds->id_node);
29 void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
31 nfs4_print_deviceid(&mirror_ds->id_node.deviceid);
32 nfs4_pnfs_ds_put(mirror_ds->ds);
33 kfree_rcu(mirror_ds, id_node.rcu);
36 /* Decode opaque device data and construct new_ds using it */
37 struct nfs4_ff_layout_ds *
38 nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
41 struct xdr_stream stream;
44 struct list_head dsaddrs;
45 struct nfs4_pnfs_ds_addr *da;
46 struct nfs4_ff_layout_ds *new_ds = NULL;
47 struct nfs4_ff_ds_version *ds_versions = NULL;
53 /* set up xdr stream */
54 scratch = alloc_page(gfp_flags);
58 new_ds = kzalloc(sizeof(struct nfs4_ff_layout_ds), gfp_flags);
62 nfs4_init_deviceid_node(&new_ds->id_node,
65 INIT_LIST_HEAD(&dsaddrs);
67 xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
68 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
71 p = xdr_inline_decode(&stream, 4);
73 goto out_err_drain_dsaddrs;
74 mp_count = be32_to_cpup(p);
75 dprintk("%s: multipath ds count %d\n", __func__, mp_count);
77 for (i = 0; i < mp_count; i++) {
79 da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
82 list_add_tail(&da->da_node, &dsaddrs);
84 if (list_empty(&dsaddrs)) {
85 dprintk("%s: no suitable DS addresses found\n",
88 goto out_err_drain_dsaddrs;
92 p = xdr_inline_decode(&stream, 4);
94 goto out_err_drain_dsaddrs;
95 version_count = be32_to_cpup(p);
96 dprintk("%s: version count %d\n", __func__, version_count);
98 ds_versions = kzalloc(version_count * sizeof(struct nfs4_ff_ds_version),
103 for (i = 0; i < version_count; i++) {
104 /* 20 = version(4) + minor_version(4) + rsize(4) + wsize(4) +
105 * tightly_coupled(4) */
106 p = xdr_inline_decode(&stream, 20);
108 goto out_err_drain_dsaddrs;
109 ds_versions[i].version = be32_to_cpup(p++);
110 ds_versions[i].minor_version = be32_to_cpup(p++);
111 ds_versions[i].rsize = nfs_block_size(be32_to_cpup(p++), NULL);
112 ds_versions[i].wsize = nfs_block_size(be32_to_cpup(p++), NULL);
113 ds_versions[i].tightly_coupled = be32_to_cpup(p);
115 if (ds_versions[i].rsize > NFS_MAX_FILE_IO_SIZE)
116 ds_versions[i].rsize = NFS_MAX_FILE_IO_SIZE;
117 if (ds_versions[i].wsize > NFS_MAX_FILE_IO_SIZE)
118 ds_versions[i].wsize = NFS_MAX_FILE_IO_SIZE;
120 if (ds_versions[i].version != 3 || ds_versions[i].minor_version != 0) {
121 dprintk("%s: [%d] unsupported ds version %d-%d\n", __func__,
122 i, ds_versions[i].version,
123 ds_versions[i].minor_version);
124 ret = -EPROTONOSUPPORT;
125 goto out_err_drain_dsaddrs;
128 dprintk("%s: [%d] vers %u minor_ver %u rsize %u wsize %u coupled %d\n",
129 __func__, i, ds_versions[i].version,
130 ds_versions[i].minor_version,
131 ds_versions[i].rsize,
132 ds_versions[i].wsize,
133 ds_versions[i].tightly_coupled);
136 new_ds->ds_versions = ds_versions;
137 new_ds->ds_versions_cnt = version_count;
139 new_ds->ds = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
141 goto out_err_drain_dsaddrs;
143 /* If DS was already in cache, free ds addrs */
144 while (!list_empty(&dsaddrs)) {
145 da = list_first_entry(&dsaddrs,
146 struct nfs4_pnfs_ds_addr,
148 list_del_init(&da->da_node);
149 kfree(da->da_remotestr);
153 __free_page(scratch);
156 out_err_drain_dsaddrs:
157 while (!list_empty(&dsaddrs)) {
158 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
160 list_del_init(&da->da_node);
161 kfree(da->da_remotestr);
167 __free_page(scratch);
171 dprintk("%s ERROR: returning %d\n", __func__, ret);
175 static void ff_layout_mark_devid_invalid(struct pnfs_layout_segment *lseg,
176 struct nfs4_deviceid_node *devid)
178 nfs4_mark_deviceid_unavailable(devid);
179 if (!ff_layout_has_available_ds(lseg))
180 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
184 static bool ff_layout_mirror_valid(struct pnfs_layout_segment *lseg,
185 struct nfs4_ff_layout_mirror *mirror)
187 if (mirror == NULL || mirror->mirror_ds == NULL) {
188 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
192 if (mirror->mirror_ds->ds == NULL) {
193 struct nfs4_deviceid_node *devid;
194 devid = &mirror->mirror_ds->id_node;
195 ff_layout_mark_devid_invalid(lseg, devid);
202 end_offset(u64 start, u64 len)
207 return end >= start ? end : NFS4_MAX_UINT64;
210 static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
211 u64 offset, u64 length)
215 end = max_t(u64, end_offset(err->offset, err->length),
216 end_offset(offset, length));
217 err->offset = min_t(u64, err->offset, offset);
218 err->length = end - err->offset;
221 static bool ds_error_can_merge(struct nfs4_ff_layout_ds_err *err, u64 offset,
222 u64 length, int status, enum nfs_opnum4 opnum,
223 nfs4_stateid *stateid,
224 struct nfs4_deviceid *deviceid)
226 return err->status == status && err->opnum == opnum &&
227 nfs4_stateid_match(&err->stateid, stateid) &&
228 !memcmp(&err->deviceid, deviceid, sizeof(*deviceid)) &&
229 end_offset(err->offset, err->length) >= offset &&
230 err->offset <= end_offset(offset, length);
233 static bool merge_ds_error(struct nfs4_ff_layout_ds_err *old,
234 struct nfs4_ff_layout_ds_err *new)
236 if (!ds_error_can_merge(old, new->offset, new->length, new->status,
237 new->opnum, &new->stateid, &new->deviceid))
240 extend_ds_error(old, new->offset, new->length);
245 ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo,
246 struct nfs4_ff_layout_ds_err *dserr)
248 struct nfs4_ff_layout_ds_err *err;
250 list_for_each_entry(err, &flo->error_list, list) {
251 if (merge_ds_error(err, dserr)) {
256 list_add(&dserr->list, &flo->error_list);
261 ff_layout_update_ds_error(struct nfs4_flexfile_layout *flo, u64 offset,
262 u64 length, int status, enum nfs_opnum4 opnum,
263 nfs4_stateid *stateid, struct nfs4_deviceid *deviceid)
266 struct nfs4_ff_layout_ds_err *err;
268 list_for_each_entry(err, &flo->error_list, list) {
269 if (ds_error_can_merge(err, offset, length, status, opnum,
270 stateid, deviceid)) {
272 extend_ds_error(err, offset, length);
280 int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
281 struct nfs4_ff_layout_mirror *mirror, u64 offset,
282 u64 length, int status, enum nfs_opnum4 opnum,
285 struct nfs4_ff_layout_ds_err *dserr;
291 if (mirror->mirror_ds == NULL)
294 spin_lock(&flo->generic_hdr.plh_inode->i_lock);
295 if (ff_layout_update_ds_error(flo, offset, length, status, opnum,
297 &mirror->mirror_ds->id_node.deviceid)) {
298 spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
301 spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
302 dserr = kmalloc(sizeof(*dserr), gfp_flags);
306 INIT_LIST_HEAD(&dserr->list);
307 dserr->offset = offset;
308 dserr->length = length;
309 dserr->status = status;
310 dserr->opnum = opnum;
311 nfs4_stateid_copy(&dserr->stateid, &mirror->stateid);
312 memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid,
313 NFS4_DEVICEID4_SIZE);
315 spin_lock(&flo->generic_hdr.plh_inode->i_lock);
316 needfree = ff_layout_add_ds_error_locked(flo, dserr);
317 spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
324 /* currently we only support AUTH_NONE and AUTH_SYS */
325 static rpc_authflavor_t
326 nfs4_ff_layout_choose_authflavor(struct nfs4_ff_layout_mirror *mirror)
328 if (mirror->uid == (u32)-1)
329 return RPC_AUTH_NULL;
330 return RPC_AUTH_UNIX;
333 /* fetch cred for NFSv3 DS */
334 static int ff_layout_update_mirror_cred(struct nfs4_ff_layout_mirror *mirror,
335 struct nfs4_pnfs_ds *ds)
337 if (ds->ds_clp && !mirror->cred &&
338 mirror->mirror_ds->ds_versions[0].version == 3) {
339 struct rpc_auth *auth = ds->ds_clp->cl_rpcclient->cl_auth;
340 struct rpc_cred *cred;
341 struct auth_cred acred = {
342 .uid = make_kuid(&init_user_ns, mirror->uid),
343 .gid = make_kgid(&init_user_ns, mirror->gid),
346 /* AUTH_NULL ignores acred */
347 cred = auth->au_ops->lookup_cred(auth, &acred, 0);
349 dprintk("%s: lookup_cred failed with %ld\n",
350 __func__, PTR_ERR(cred));
351 return PTR_ERR(cred);
353 if (cmpxchg(&mirror->cred, NULL, cred))
361 nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
363 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
364 struct nfs_fh *fh = NULL;
366 if (!ff_layout_mirror_valid(lseg, mirror)) {
367 pr_err_ratelimited("NFS: %s: No data server for mirror offset index %d\n",
368 __func__, mirror_idx);
372 /* FIXME: For now assume there is only 1 version available for the DS */
373 fh = &mirror->fh_versions[0];
378 /* Upon return, either ds is connected, or ds is NULL */
379 struct nfs4_pnfs_ds *
380 nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
383 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
384 struct nfs4_pnfs_ds *ds = NULL;
385 struct nfs4_deviceid_node *devid;
386 struct inode *ino = lseg->pls_layout->plh_inode;
387 struct nfs_server *s = NFS_SERVER(ino);
388 unsigned int max_payload;
389 rpc_authflavor_t flavor;
391 if (!ff_layout_mirror_valid(lseg, mirror)) {
392 pr_err_ratelimited("NFS: %s: No data server for offset index %d\n",
397 devid = &mirror->mirror_ds->id_node;
398 if (ff_layout_test_devid_unavailable(devid))
401 ds = mirror->mirror_ds->ds;
402 /* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
405 goto out_update_creds;
407 flavor = nfs4_ff_layout_choose_authflavor(mirror);
409 /* FIXME: For now we assume the server sent only one version of NFS
412 nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
414 mirror->mirror_ds->ds_versions[0].version,
415 mirror->mirror_ds->ds_versions[0].minor_version,
418 /* connect success, check rsize/wsize limit */
421 nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
423 if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
424 mirror->mirror_ds->ds_versions[0].rsize = max_payload;
425 if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
426 mirror->mirror_ds->ds_versions[0].wsize = max_payload;
428 ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
429 mirror, lseg->pls_range.offset,
430 lseg->pls_range.length, NFS4ERR_NXIO,
431 OP_ILLEGAL, GFP_NOIO);
433 if (ff_layout_has_available_ds(lseg))
434 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
435 &lseg->pls_layout->plh_flags);
437 pnfs_error_mark_layout_for_return(ino, lseg);
439 pnfs_error_mark_layout_for_return(ino, lseg);
442 if (ff_layout_update_mirror_cred(mirror, ds))
449 ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
450 struct rpc_cred *mdscred)
452 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
453 struct rpc_cred *cred = ERR_PTR(-EINVAL);
455 if (!nfs4_ff_layout_prepare_ds(lseg, ds_idx, true))
458 if (mirror && mirror->cred)
467 * Find or create a DS rpc client with th MDS server rpc client auth flavor
468 * in the nfs_client cl_ds_clients list.
471 nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
472 struct nfs_client *ds_clp, struct inode *inode)
474 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
476 switch (mirror->mirror_ds->ds_versions[0].version) {
478 /* For NFSv3 DS, flavor is set when creating DS connections */
479 return ds_clp->cl_rpcclient;
481 return nfs4_find_or_create_ds_client(ds_clp, inode);
487 static bool is_range_intersecting(u64 offset1, u64 length1,
488 u64 offset2, u64 length2)
490 u64 end1 = end_offset(offset1, length1);
491 u64 end2 = end_offset(offset2, length2);
493 return (end1 == NFS4_MAX_UINT64 || end1 > offset2) &&
494 (end2 == NFS4_MAX_UINT64 || end2 > offset1);
497 /* called with inode i_lock held */
498 int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
499 struct xdr_stream *xdr, int *count,
500 const struct pnfs_layout_range *range)
502 struct nfs4_ff_layout_ds_err *err, *n;
505 list_for_each_entry_safe(err, n, &flo->error_list, list) {
506 if (!is_range_intersecting(err->offset, err->length,
507 range->offset, range->length))
509 /* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
510 * + array length + deviceid(NFS4_DEVICEID4_SIZE)
511 * + status(4) + opnum(4)
513 p = xdr_reserve_space(xdr,
514 28 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
517 p = xdr_encode_hyper(p, err->offset);
518 p = xdr_encode_hyper(p, err->length);
519 p = xdr_encode_opaque_fixed(p, &err->stateid,
522 *p++ = cpu_to_be32(1);
523 p = xdr_encode_opaque_fixed(p, &err->deviceid,
524 NFS4_DEVICEID4_SIZE);
525 *p++ = cpu_to_be32(err->status);
526 *p++ = cpu_to_be32(err->opnum);
528 list_del(&err->list);
529 dprintk("%s: offset %llu length %llu status %d op %d count %d\n",
530 __func__, err->offset, err->length, err->status,
538 static bool ff_read_layout_has_available_ds(struct pnfs_layout_segment *lseg)
540 struct nfs4_ff_layout_mirror *mirror;
541 struct nfs4_deviceid_node *devid;
544 for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
545 mirror = FF_LAYOUT_COMP(lseg, idx);
546 if (mirror && mirror->mirror_ds) {
547 devid = &mirror->mirror_ds->id_node;
548 if (!ff_layout_test_devid_unavailable(devid))
556 static bool ff_rw_layout_has_available_ds(struct pnfs_layout_segment *lseg)
558 struct nfs4_ff_layout_mirror *mirror;
559 struct nfs4_deviceid_node *devid;
562 for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
563 mirror = FF_LAYOUT_COMP(lseg, idx);
564 if (!mirror || !mirror->mirror_ds)
566 devid = &mirror->mirror_ds->id_node;
567 if (ff_layout_test_devid_unavailable(devid))
571 return FF_LAYOUT_MIRROR_COUNT(lseg) != 0;
574 bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
576 if (lseg->pls_range.iomode == IOMODE_READ)
577 return ff_read_layout_has_available_ds(lseg);
578 /* Note: RW layout needs all mirrors available */
579 return ff_rw_layout_has_available_ds(lseg);
582 module_param(dataserver_retrans, uint, 0644);
583 MODULE_PARM_DESC(dataserver_retrans, "The number of times the NFSv4.1 client "
584 "retries a request before it attempts further "
585 " recovery action.");
586 module_param(dataserver_timeo, uint, 0644);
587 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
588 "NFSv4.1 client waits for a response from a "
589 " data server before it retries an NFS request.");