2 * pNFS Objects layout driver high level definitions
4 * Copyright (C) 2007 Panasas Inc. [year of first publication]
7 * Benny Halevy <bhalevy@panasas.com>
8 * Boaz Harrosh <bharrosh@panasas.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2
12 * See the file COPYING included with this distribution for more details.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
18 * 1. Redistributions of source code must retain the above copyright
19 * notice, this list of conditions and the following disclaimer.
20 * 2. Redistributions in binary form must reproduce the above copyright
21 * notice, this list of conditions and the following disclaimer in the
22 * documentation and/or other materials provided with the distribution.
23 * 3. Neither the name of the Panasas company nor the names of its
24 * contributors may be used to endorse or promote products derived
25 * from this software without specific prior written permission.
27 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
28 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
29 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
30 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
34 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
35 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
36 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
37 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 #include <linux/kmod.h>
41 #include <linux/moduleparam.h>
42 #include <linux/ratelimit.h>
43 #include <scsi/osd_initiator.h>
44 #include "objlayout.h"
46 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
48 * Create a objlayout layout structure for the given inode and return it.
50 struct pnfs_layout_hdr *
51 objlayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
53 struct objlayout *objlay;
55 objlay = kzalloc(sizeof(struct objlayout), gfp_flags);
57 spin_lock_init(&objlay->lock);
58 INIT_LIST_HEAD(&objlay->err_list);
60 dprintk("%s: Return %p\n", __func__, objlay);
61 return &objlay->pnfs_layout;
65 * Free an objlayout layout structure
68 objlayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
70 struct objlayout *objlay = OBJLAYOUT(lo);
72 dprintk("%s: objlay %p\n", __func__, objlay);
74 WARN_ON(!list_empty(&objlay->err_list));
79 * Unmarshall layout and store it in pnfslay.
81 struct pnfs_layout_segment *
82 objlayout_alloc_lseg(struct pnfs_layout_hdr *pnfslay,
83 struct nfs4_layoutget_res *lgr,
87 struct xdr_stream stream;
88 struct xdr_buf buf = {
89 .pages = lgr->layoutp->pages,
90 .page_len = lgr->layoutp->len,
91 .buflen = lgr->layoutp->len,
92 .len = lgr->layoutp->len,
95 struct pnfs_layout_segment *lseg;
97 dprintk("%s: Begin pnfslay %p\n", __func__, pnfslay);
99 scratch = alloc_page(gfp_flags);
103 xdr_init_decode(&stream, &buf, NULL);
104 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
106 status = objio_alloc_lseg(&lseg, pnfslay, &lgr->range, &stream, gfp_flags);
107 if (unlikely(status)) {
108 dprintk("%s: objio_alloc_lseg Return err %d\n", __func__,
113 __free_page(scratch);
115 dprintk("%s: Return %p\n", __func__, lseg);
119 __free_page(scratch);
121 dprintk("%s: Err Return=>%d\n", __func__, status);
122 return ERR_PTR(status);
126 * Free a layout segement
129 objlayout_free_lseg(struct pnfs_layout_segment *lseg)
131 dprintk("%s: freeing layout segment %p\n", __func__, lseg);
136 objio_free_lseg(lseg);
143 end_offset(u64 start, u64 len)
148 return end >= start ? end : NFS4_MAX_UINT64;
151 /* last octet in a range */
153 last_byte_offset(u64 start, u64 len)
159 return end > start ? end - 1 : NFS4_MAX_UINT64;
162 static void _fix_verify_io_params(struct pnfs_layout_segment *lseg,
163 struct page ***p_pages, unsigned *p_pgbase,
164 u64 offset, unsigned long count)
168 BUG_ON(offset < lseg->pls_range.offset);
169 lseg_end_offset = end_offset(lseg->pls_range.offset,
170 lseg->pls_range.length);
171 BUG_ON(offset >= lseg_end_offset);
172 WARN_ON(offset + count > lseg_end_offset);
174 if (*p_pgbase > PAGE_SIZE) {
175 dprintk("%s: pgbase(0x%x) > PAGE_SIZE\n", __func__, *p_pgbase);
176 *p_pages += *p_pgbase >> PAGE_SHIFT;
177 *p_pgbase &= ~PAGE_MASK;
182 * I/O done common code
185 objlayout_iodone(struct objlayout_io_res *oir)
187 if (likely(oir->status >= 0)) {
188 objio_free_result(oir);
190 struct objlayout *objlay = oir->objlay;
192 spin_lock(&objlay->lock);
193 objlay->delta_space_valid = OBJ_DSU_INVALID;
194 list_add(&objlay->err_list, &oir->err_list);
195 spin_unlock(&objlay->lock);
200 * objlayout_io_set_result - Set an osd_error code on a specific osd comp.
202 * The @index component IO failed (error returned from target). Register
203 * the error for later reporting at layout-return.
206 objlayout_io_set_result(struct objlayout_io_res *oir, unsigned index,
207 struct pnfs_osd_objid *pooid, int osd_error,
208 u64 offset, u64 length, bool is_write)
210 struct pnfs_osd_ioerr *ioerr = &oir->ioerrs[index];
212 BUG_ON(index >= oir->num_comps);
214 ioerr->oer_component = *pooid;
215 ioerr->oer_comp_offset = offset;
216 ioerr->oer_comp_length = length;
217 ioerr->oer_iswrite = is_write;
218 ioerr->oer_errno = osd_error;
220 dprintk("%s: err[%d]: errno=%d is_write=%d dev(%llx:%llx) "
221 "par=0x%llx obj=0x%llx offset=0x%llx length=0x%llx\n",
222 __func__, index, ioerr->oer_errno,
224 _DEVID_LO(&ioerr->oer_component.oid_device_id),
225 _DEVID_HI(&ioerr->oer_component.oid_device_id),
226 ioerr->oer_component.oid_partition_id,
227 ioerr->oer_component.oid_object_id,
228 ioerr->oer_comp_offset,
229 ioerr->oer_comp_length);
231 /* User need not call if no error is reported */
232 ioerr->oer_errno = 0;
236 /* Function scheduled on rpc workqueue to call ->nfs_readlist_complete().
237 * This is because the osd completion is called with ints-off from
240 static void _rpc_read_complete(struct work_struct *work)
242 struct rpc_task *task;
243 struct nfs_read_data *rdata;
245 dprintk("%s enter\n", __func__);
246 task = container_of(work, struct rpc_task, u.tk_work);
247 rdata = container_of(task, struct nfs_read_data, task);
249 pnfs_ld_read_done(rdata);
253 objlayout_read_done(struct objlayout_io_res *oir, ssize_t status, bool sync)
255 struct nfs_read_data *rdata = oir->rpcdata;
257 oir->status = rdata->task.tk_status = status;
259 rdata->res.count = status;
261 rdata->pnfs_error = status;
262 objlayout_iodone(oir);
263 /* must not use oir after this point */
265 dprintk("%s: Return status=%zd eof=%d sync=%d\n", __func__,
266 status, rdata->res.eof, sync);
269 pnfs_ld_read_done(rdata);
271 INIT_WORK(&rdata->task.u.tk_work, _rpc_read_complete);
272 schedule_work(&rdata->task.u.tk_work);
277 * Perform sync or async reads.
280 objlayout_read_pagelist(struct nfs_read_data *rdata)
282 loff_t offset = rdata->args.offset;
283 size_t count = rdata->args.count;
287 eof = i_size_read(rdata->inode);
288 if (unlikely(offset + count > eof)) {
291 rdata->res.count = 0;
293 /*FIXME: do we need to call pnfs_ld_read_done() */
296 count = eof - offset;
299 rdata->res.eof = (offset + count) >= eof;
300 _fix_verify_io_params(rdata->lseg, &rdata->args.pages,
302 rdata->args.offset, rdata->args.count);
304 dprintk("%s: inode(%lx) offset 0x%llx count 0x%Zx eof=%d\n",
305 __func__, rdata->inode->i_ino, offset, count, rdata->res.eof);
307 err = objio_read_pagelist(rdata);
310 rdata->pnfs_error = err;
311 dprintk("%s: Returned Error %d\n", __func__, err);
312 return PNFS_NOT_ATTEMPTED;
314 return PNFS_ATTEMPTED;
317 /* Function scheduled on rpc workqueue to call ->nfs_writelist_complete().
318 * This is because the osd completion is called with ints-off from
321 static void _rpc_write_complete(struct work_struct *work)
323 struct rpc_task *task;
324 struct nfs_write_data *wdata;
326 dprintk("%s enter\n", __func__);
327 task = container_of(work, struct rpc_task, u.tk_work);
328 wdata = container_of(task, struct nfs_write_data, task);
330 pnfs_ld_write_done(wdata);
334 objlayout_write_done(struct objlayout_io_res *oir, ssize_t status, bool sync)
336 struct nfs_write_data *wdata = oir->rpcdata;
338 oir->status = wdata->task.tk_status = status;
340 wdata->res.count = status;
341 wdata->verf.committed = oir->committed;
343 wdata->pnfs_error = status;
345 objlayout_iodone(oir);
346 /* must not use oir after this point */
348 dprintk("%s: Return status %zd committed %d sync=%d\n", __func__,
349 status, wdata->verf.committed, sync);
352 pnfs_ld_write_done(wdata);
354 INIT_WORK(&wdata->task.u.tk_work, _rpc_write_complete);
355 schedule_work(&wdata->task.u.tk_work);
360 * Perform sync or async writes.
363 objlayout_write_pagelist(struct nfs_write_data *wdata,
368 _fix_verify_io_params(wdata->lseg, &wdata->args.pages,
370 wdata->args.offset, wdata->args.count);
372 err = objio_write_pagelist(wdata, how);
374 wdata->pnfs_error = err;
375 dprintk("%s: Returned Error %d\n", __func__, err);
376 return PNFS_NOT_ATTEMPTED;
378 return PNFS_ATTEMPTED;
382 objlayout_encode_layoutcommit(struct pnfs_layout_hdr *pnfslay,
383 struct xdr_stream *xdr,
384 const struct nfs4_layoutcommit_args *args)
386 struct objlayout *objlay = OBJLAYOUT(pnfslay);
387 struct pnfs_osd_layoutupdate lou;
390 dprintk("%s: Begin\n", __func__);
392 spin_lock(&objlay->lock);
393 lou.dsu_valid = (objlay->delta_space_valid == OBJ_DSU_VALID);
394 lou.dsu_delta = objlay->delta_space_used;
395 objlay->delta_space_used = 0;
396 objlay->delta_space_valid = OBJ_DSU_INIT;
397 lou.olu_ioerr_flag = !list_empty(&objlay->err_list);
398 spin_unlock(&objlay->lock);
400 start = xdr_reserve_space(xdr, 4);
402 BUG_ON(pnfs_osd_xdr_encode_layoutupdate(xdr, &lou));
404 *start = cpu_to_be32((xdr->p - start - 1) * 4);
406 dprintk("%s: Return delta_space_used %lld err %d\n", __func__,
407 lou.dsu_delta, lou.olu_ioerr_flag);
411 err_prio(u32 oer_errno)
417 case PNFS_OSD_ERR_RESOURCE:
418 return OSD_ERR_PRI_RESOURCE;
419 case PNFS_OSD_ERR_BAD_CRED:
420 return OSD_ERR_PRI_BAD_CRED;
421 case PNFS_OSD_ERR_NO_ACCESS:
422 return OSD_ERR_PRI_NO_ACCESS;
423 case PNFS_OSD_ERR_UNREACHABLE:
424 return OSD_ERR_PRI_UNREACHABLE;
425 case PNFS_OSD_ERR_NOT_FOUND:
426 return OSD_ERR_PRI_NOT_FOUND;
427 case PNFS_OSD_ERR_NO_SPACE:
428 return OSD_ERR_PRI_NO_SPACE;
432 case PNFS_OSD_ERR_EIO:
433 return OSD_ERR_PRI_EIO;
438 merge_ioerr(struct pnfs_osd_ioerr *dest_err,
439 const struct pnfs_osd_ioerr *src_err)
441 u64 dest_end, src_end;
443 if (!dest_err->oer_errno) {
444 *dest_err = *src_err;
445 /* accumulated device must be blank */
446 memset(&dest_err->oer_component.oid_device_id, 0,
447 sizeof(dest_err->oer_component.oid_device_id));
452 if (dest_err->oer_component.oid_partition_id !=
453 src_err->oer_component.oid_partition_id)
454 dest_err->oer_component.oid_partition_id = 0;
456 if (dest_err->oer_component.oid_object_id !=
457 src_err->oer_component.oid_object_id)
458 dest_err->oer_component.oid_object_id = 0;
460 if (dest_err->oer_comp_offset > src_err->oer_comp_offset)
461 dest_err->oer_comp_offset = src_err->oer_comp_offset;
463 dest_end = end_offset(dest_err->oer_comp_offset,
464 dest_err->oer_comp_length);
465 src_end = end_offset(src_err->oer_comp_offset,
466 src_err->oer_comp_length);
467 if (dest_end < src_end)
470 dest_err->oer_comp_length = dest_end - dest_err->oer_comp_offset;
472 if ((src_err->oer_iswrite == dest_err->oer_iswrite) &&
473 (err_prio(src_err->oer_errno) > err_prio(dest_err->oer_errno))) {
474 dest_err->oer_errno = src_err->oer_errno;
475 } else if (src_err->oer_iswrite) {
476 dest_err->oer_iswrite = true;
477 dest_err->oer_errno = src_err->oer_errno;
482 encode_accumulated_error(struct objlayout *objlay, __be32 *p)
484 struct objlayout_io_res *oir, *tmp;
485 struct pnfs_osd_ioerr accumulated_err = {.oer_errno = 0};
487 list_for_each_entry_safe(oir, tmp, &objlay->err_list, err_list) {
490 for (i = 0; i < oir->num_comps; i++) {
491 struct pnfs_osd_ioerr *ioerr = &oir->ioerrs[i];
493 if (!ioerr->oer_errno)
496 printk(KERN_ERR "NFS: %s: err[%d]: errno=%d "
497 "is_write=%d dev(%llx:%llx) par=0x%llx "
498 "obj=0x%llx offset=0x%llx length=0x%llx\n",
499 __func__, i, ioerr->oer_errno,
501 _DEVID_LO(&ioerr->oer_component.oid_device_id),
502 _DEVID_HI(&ioerr->oer_component.oid_device_id),
503 ioerr->oer_component.oid_partition_id,
504 ioerr->oer_component.oid_object_id,
505 ioerr->oer_comp_offset,
506 ioerr->oer_comp_length);
508 merge_ioerr(&accumulated_err, ioerr);
510 list_del(&oir->err_list);
511 objio_free_result(oir);
514 pnfs_osd_xdr_encode_ioerr(p, &accumulated_err);
518 objlayout_encode_layoutreturn(struct pnfs_layout_hdr *pnfslay,
519 struct xdr_stream *xdr,
520 const struct nfs4_layoutreturn_args *args)
522 struct objlayout *objlay = OBJLAYOUT(pnfslay);
523 struct objlayout_io_res *oir, *tmp;
526 dprintk("%s: Begin\n", __func__);
527 start = xdr_reserve_space(xdr, 4);
530 spin_lock(&objlay->lock);
532 list_for_each_entry_safe(oir, tmp, &objlay->err_list, err_list) {
533 __be32 *last_xdr = NULL, *p;
537 for (i = 0; i < oir->num_comps; i++) {
538 struct pnfs_osd_ioerr *ioerr = &oir->ioerrs[i];
540 if (!ioerr->oer_errno)
543 dprintk("%s: err[%d]: errno=%d is_write=%d "
544 "dev(%llx:%llx) par=0x%llx obj=0x%llx "
545 "offset=0x%llx length=0x%llx\n",
546 __func__, i, ioerr->oer_errno,
548 _DEVID_LO(&ioerr->oer_component.oid_device_id),
549 _DEVID_HI(&ioerr->oer_component.oid_device_id),
550 ioerr->oer_component.oid_partition_id,
551 ioerr->oer_component.oid_object_id,
552 ioerr->oer_comp_offset,
553 ioerr->oer_comp_length);
555 p = pnfs_osd_xdr_ioerr_reserve_space(xdr);
558 break; /* accumulated_error */
562 pnfs_osd_xdr_encode_ioerr(p, &oir->ioerrs[i]);
565 /* TODO: use xdr_write_pages */
567 /* no space for even one error descriptor */
570 /* we've encountered a situation with lots and lots of
571 * errors and no space to encode them all. Use the last
572 * available slot to report the union of all the
575 encode_accumulated_error(objlay, last_xdr);
578 list_del(&oir->err_list);
579 objio_free_result(oir);
582 spin_unlock(&objlay->lock);
584 *start = cpu_to_be32((xdr->p - start - 1) * 4);
585 dprintk("%s: Return\n", __func__);
590 * Get Device Info API for io engines
592 struct objlayout_deviceinfo {
594 struct pnfs_osd_deviceaddr da; /* This must be last */
597 /* Initialize and call nfs_getdeviceinfo, then decode and return a
598 * "struct pnfs_osd_deviceaddr *" Eventually objlayout_put_deviceinfo()
601 int objlayout_get_deviceinfo(struct pnfs_layout_hdr *pnfslay,
602 struct nfs4_deviceid *d_id, struct pnfs_osd_deviceaddr **deviceaddr,
605 struct objlayout_deviceinfo *odi;
606 struct pnfs_device pd;
607 struct super_block *sb;
608 struct page *page, **pages;
612 page = alloc_page(gfp_flags);
619 memcpy(&pd.dev_id, d_id, sizeof(*d_id));
620 pd.layout_type = LAYOUT_OSD2_OBJECTS;
623 pd.pglen = PAGE_SIZE;
626 sb = pnfslay->plh_inode->i_sb;
627 err = nfs4_proc_getdeviceinfo(NFS_SERVER(pnfslay->plh_inode), &pd);
628 dprintk("%s nfs_getdeviceinfo returned %d\n", __func__, err);
632 p = page_address(page);
633 odi = kzalloc(sizeof(*odi), gfp_flags);
638 pnfs_osd_xdr_decode_deviceaddr(&odi->da, p);
640 *deviceaddr = &odi->da;
648 void objlayout_put_deviceinfo(struct pnfs_osd_deviceaddr *deviceaddr)
650 struct objlayout_deviceinfo *odi = container_of(deviceaddr,
651 struct objlayout_deviceinfo,
654 __free_page(odi->page);
659 OBJLAYOUT_MAX_URI_LEN = 256, OBJLAYOUT_MAX_OSDNAME_LEN = 64,
660 OBJLAYOUT_MAX_SYSID_HEX_LEN = OSD_SYSTEMID_LEN * 2 + 1,
661 OSD_LOGIN_UPCALL_PATHLEN = 256
664 static char osd_login_prog[OSD_LOGIN_UPCALL_PATHLEN] = "/sbin/osd_login";
666 module_param_string(osd_login_prog, osd_login_prog, sizeof(osd_login_prog),
668 MODULE_PARM_DESC(osd_login_prog, "Path to the osd_login upcall program");
670 struct __auto_login {
671 char uri[OBJLAYOUT_MAX_URI_LEN];
672 char osdname[OBJLAYOUT_MAX_OSDNAME_LEN];
673 char systemid_hex[OBJLAYOUT_MAX_SYSID_HEX_LEN];
676 static int __objlayout_upcall(struct __auto_login *login)
678 static char *envp[] = { "HOME=/",
680 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
686 if (unlikely(!osd_login_prog[0])) {
687 dprintk("%s: osd_login_prog is disabled\n", __func__);
691 dprintk("%s uri: %s\n", __func__, login->uri);
692 dprintk("%s osdname %s\n", __func__, login->osdname);
693 dprintk("%s systemid_hex %s\n", __func__, login->systemid_hex);
695 argv[0] = (char *)osd_login_prog;
697 argv[2] = login->uri;
699 argv[4] = login->osdname;
701 argv[6] = login->systemid_hex;
704 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
706 * Disable the upcall mechanism if we're getting an ENOENT or
707 * EACCES error. The admin can re-enable it on the fly by using
708 * sysfs to set the objlayoutdriver.osd_login_prog module parameter once
709 * the problem has been fixed.
711 if (ret == -ENOENT || ret == -EACCES) {
712 printk(KERN_ERR "PNFS-OBJ: %s was not found please set "
713 "objlayoutdriver.osd_login_prog kernel parameter!\n",
715 osd_login_prog[0] = '\0';
717 dprintk("%s %s return value: %d\n", __func__, osd_login_prog, ret);
722 /* Assume dest is all zeros */
723 static void __copy_nfsS_and_zero_terminate(struct nfs4_string s,
724 char *dest, int max_len,
725 const char *var_name)
730 if (s.len >= max_len) {
732 "objlayout_autologin: %s: s.len(%d) >= max_len(%d)",
733 var_name, s.len, max_len);
734 s.len = max_len - 1; /* space for null terminator */
737 memcpy(dest, s.data, s.len);
740 /* Assume sysid is all zeros */
741 static void _sysid_2_hex(struct nfs4_string s,
742 char sysid[OBJLAYOUT_MAX_SYSID_HEX_LEN])
750 if (s.len != OSD_SYSTEMID_LEN) {
752 "objlayout_autologin: systemid_len(%d) != OSD_SYSTEMID_LEN",
754 if (s.len > OSD_SYSTEMID_LEN)
755 s.len = OSD_SYSTEMID_LEN;
759 for (i = 0; i < s.len; i++)
760 cur = hex_byte_pack(cur, s.data[i]);
763 int objlayout_autologin(struct pnfs_osd_deviceaddr *deviceaddr)
766 struct __auto_login login;
768 if (!deviceaddr->oda_targetaddr.ota_netaddr.r_addr.len)
771 memset(&login, 0, sizeof(login));
772 __copy_nfsS_and_zero_terminate(
773 deviceaddr->oda_targetaddr.ota_netaddr.r_addr,
774 login.uri, sizeof(login.uri), "URI");
776 __copy_nfsS_and_zero_terminate(
777 deviceaddr->oda_osdname,
778 login.osdname, sizeof(login.osdname), "OSDNAME");
780 _sysid_2_hex(deviceaddr->oda_systemid, login.systemid_hex);
782 rc = __objlayout_upcall(&login);
783 if (rc > 0) /* script returns positive values */