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 <scsi/osd_initiator.h>
41 #include "objlayout.h"
43 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
45 * Create a objlayout layout structure for the given inode and return it.
47 struct pnfs_layout_hdr *
48 objlayout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
50 struct objlayout *objlay;
52 objlay = kzalloc(sizeof(struct objlayout), gfp_flags);
54 spin_lock_init(&objlay->lock);
55 INIT_LIST_HEAD(&objlay->err_list);
57 dprintk("%s: Return %p\n", __func__, objlay);
58 return &objlay->pnfs_layout;
62 * Free an objlayout layout structure
65 objlayout_free_layout_hdr(struct pnfs_layout_hdr *lo)
67 struct objlayout *objlay = OBJLAYOUT(lo);
69 dprintk("%s: objlay %p\n", __func__, objlay);
71 WARN_ON(!list_empty(&objlay->err_list));
76 * Unmarshall layout and store it in pnfslay.
78 struct pnfs_layout_segment *
79 objlayout_alloc_lseg(struct pnfs_layout_hdr *pnfslay,
80 struct nfs4_layoutget_res *lgr,
84 struct xdr_stream stream;
85 struct xdr_buf buf = {
86 .pages = lgr->layoutp->pages,
87 .page_len = lgr->layoutp->len,
88 .buflen = lgr->layoutp->len,
89 .len = lgr->layoutp->len,
92 struct pnfs_layout_segment *lseg;
94 dprintk("%s: Begin pnfslay %p\n", __func__, pnfslay);
96 scratch = alloc_page(gfp_flags);
100 xdr_init_decode(&stream, &buf, NULL);
101 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
103 status = objio_alloc_lseg(&lseg, pnfslay, &lgr->range, &stream, gfp_flags);
104 if (unlikely(status)) {
105 dprintk("%s: objio_alloc_lseg Return err %d\n", __func__,
110 __free_page(scratch);
112 dprintk("%s: Return %p\n", __func__, lseg);
116 __free_page(scratch);
118 dprintk("%s: Err Return=>%d\n", __func__, status);
119 return ERR_PTR(status);
123 * Free a layout segement
126 objlayout_free_lseg(struct pnfs_layout_segment *lseg)
128 dprintk("%s: freeing layout segment %p\n", __func__, lseg);
133 objio_free_lseg(lseg);
140 end_offset(u64 start, u64 len)
145 return end >= start ? end : NFS4_MAX_UINT64;
148 /* last octet in a range */
150 last_byte_offset(u64 start, u64 len)
156 return end > start ? end - 1 : NFS4_MAX_UINT64;
159 static struct objlayout_io_state *
160 objlayout_alloc_io_state(struct pnfs_layout_hdr *pnfs_layout_type,
165 struct pnfs_layout_segment *lseg,
169 struct objlayout_io_state *state;
172 dprintk("%s: allocating io_state\n", __func__);
173 if (objio_alloc_io_state(lseg, &state, gfp_flags))
176 BUG_ON(offset < lseg->pls_range.offset);
177 lseg_end_offset = end_offset(lseg->pls_range.offset,
178 lseg->pls_range.length);
179 BUG_ON(offset >= lseg_end_offset);
180 if (offset + count > lseg_end_offset) {
181 count = lseg->pls_range.length -
182 (offset - lseg->pls_range.offset);
183 dprintk("%s: truncated count %Zd\n", __func__, count);
186 if (pgbase > PAGE_SIZE) {
187 pages += pgbase >> PAGE_SHIFT;
188 pgbase &= ~PAGE_MASK;
191 INIT_LIST_HEAD(&state->err_list);
193 state->rpcdata = rpcdata;
194 state->pages = pages;
195 state->pgbase = pgbase;
196 state->nr_pages = (pgbase + count + PAGE_SIZE - 1) >> PAGE_SHIFT;
197 state->offset = offset;
198 state->count = count;
205 objlayout_free_io_state(struct objlayout_io_state *state)
207 dprintk("%s: freeing io_state\n", __func__);
208 if (unlikely(!state))
211 objio_free_io_state(state);
215 * I/O done common code
218 objlayout_iodone(struct objlayout_io_state *state)
220 dprintk("%s: state %p status\n", __func__, state);
222 if (likely(state->status >= 0)) {
223 objlayout_free_io_state(state);
225 struct objlayout *objlay = OBJLAYOUT(state->lseg->pls_layout);
227 spin_lock(&objlay->lock);
228 list_add(&objlay->err_list, &state->err_list);
229 spin_unlock(&objlay->lock);
234 * objlayout_io_set_result - Set an osd_error code on a specific osd comp.
236 * The @index component IO failed (error returned from target). Register
237 * the error for later reporting at layout-return.
240 objlayout_io_set_result(struct objlayout_io_state *state, unsigned index,
241 struct pnfs_osd_objid *pooid, int osd_error,
242 u64 offset, u64 length, bool is_write)
244 struct pnfs_osd_ioerr *ioerr = &state->ioerrs[index];
246 BUG_ON(index >= state->num_comps);
248 ioerr->oer_component = *pooid;
249 ioerr->oer_comp_offset = offset;
250 ioerr->oer_comp_length = length;
251 ioerr->oer_iswrite = is_write;
252 ioerr->oer_errno = osd_error;
254 dprintk("%s: err[%d]: errno=%d is_write=%d dev(%llx:%llx) "
255 "par=0x%llx obj=0x%llx offset=0x%llx length=0x%llx\n",
256 __func__, index, ioerr->oer_errno,
258 _DEVID_LO(&ioerr->oer_component.oid_device_id),
259 _DEVID_HI(&ioerr->oer_component.oid_device_id),
260 ioerr->oer_component.oid_partition_id,
261 ioerr->oer_component.oid_object_id,
262 ioerr->oer_comp_offset,
263 ioerr->oer_comp_length);
265 /* User need not call if no error is reported */
266 ioerr->oer_errno = 0;
270 /* Function scheduled on rpc workqueue to call ->nfs_readlist_complete().
271 * This is because the osd completion is called with ints-off from
274 static void _rpc_read_complete(struct work_struct *work)
276 struct rpc_task *task;
277 struct nfs_read_data *rdata;
279 dprintk("%s enter\n", __func__);
280 task = container_of(work, struct rpc_task, u.tk_work);
281 rdata = container_of(task, struct nfs_read_data, task);
283 pnfs_ld_read_done(rdata);
287 objlayout_read_done(struct objlayout_io_state *state, ssize_t status, bool sync)
289 int eof = state->eof;
290 struct nfs_read_data *rdata;
292 state->status = status;
293 dprintk("%s: Begin status=%ld eof=%d\n", __func__, status, eof);
294 rdata = state->rpcdata;
295 rdata->task.tk_status = status;
297 rdata->res.count = status;
298 rdata->res.eof = eof;
300 objlayout_iodone(state);
301 /* must not use state after this point */
304 pnfs_ld_read_done(rdata);
306 INIT_WORK(&rdata->task.u.tk_work, _rpc_read_complete);
307 schedule_work(&rdata->task.u.tk_work);
312 * Perform sync or async reads.
315 objlayout_read_pagelist(struct nfs_read_data *rdata)
317 loff_t offset = rdata->args.offset;
318 size_t count = rdata->args.count;
319 struct objlayout_io_state *state;
323 dprintk("%s: Begin inode %p offset %llu count %d\n",
324 __func__, rdata->inode, offset, (int)count);
326 eof = i_size_read(rdata->inode);
327 if (unlikely(offset + count > eof)) {
330 rdata->res.count = 0;
334 count = eof - offset;
337 state = objlayout_alloc_io_state(NFS_I(rdata->inode)->layout,
338 rdata->args.pages, rdata->args.pgbase,
342 if (unlikely(!state)) {
347 state->eof = state->offset + state->count >= eof;
349 status = objio_read_pagelist(state);
351 dprintk("%s: Return status %Zd\n", __func__, status);
352 rdata->pnfs_error = status;
353 return PNFS_ATTEMPTED;
356 /* Function scheduled on rpc workqueue to call ->nfs_writelist_complete().
357 * This is because the osd completion is called with ints-off from
360 static void _rpc_write_complete(struct work_struct *work)
362 struct rpc_task *task;
363 struct nfs_write_data *wdata;
365 dprintk("%s enter\n", __func__);
366 task = container_of(work, struct rpc_task, u.tk_work);
367 wdata = container_of(task, struct nfs_write_data, task);
369 pnfs_ld_write_done(wdata);
373 objlayout_write_done(struct objlayout_io_state *state, ssize_t status,
376 struct nfs_write_data *wdata;
378 dprintk("%s: Begin\n", __func__);
379 wdata = state->rpcdata;
380 state->status = status;
381 wdata->task.tk_status = status;
383 wdata->res.count = status;
384 wdata->verf.committed = state->committed;
385 dprintk("%s: Return status %d committed %d\n",
386 __func__, wdata->task.tk_status,
387 wdata->verf.committed);
389 dprintk("%s: Return status %d\n",
390 __func__, wdata->task.tk_status);
391 objlayout_iodone(state);
392 /* must not use state after this point */
395 pnfs_ld_write_done(wdata);
397 INIT_WORK(&wdata->task.u.tk_work, _rpc_write_complete);
398 schedule_work(&wdata->task.u.tk_work);
403 * Perform sync or async writes.
406 objlayout_write_pagelist(struct nfs_write_data *wdata,
409 struct objlayout_io_state *state;
412 dprintk("%s: Begin inode %p offset %llu count %u\n",
413 __func__, wdata->inode, wdata->args.offset, wdata->args.count);
415 state = objlayout_alloc_io_state(NFS_I(wdata->inode)->layout,
422 if (unlikely(!state)) {
427 state->sync = how & FLUSH_SYNC;
429 status = objio_write_pagelist(state, how & FLUSH_STABLE);
431 dprintk("%s: Return status %Zd\n", __func__, status);
432 wdata->pnfs_error = status;
433 return PNFS_ATTEMPTED;
437 err_prio(u32 oer_errno)
443 case PNFS_OSD_ERR_RESOURCE:
444 return OSD_ERR_PRI_RESOURCE;
445 case PNFS_OSD_ERR_BAD_CRED:
446 return OSD_ERR_PRI_BAD_CRED;
447 case PNFS_OSD_ERR_NO_ACCESS:
448 return OSD_ERR_PRI_NO_ACCESS;
449 case PNFS_OSD_ERR_UNREACHABLE:
450 return OSD_ERR_PRI_UNREACHABLE;
451 case PNFS_OSD_ERR_NOT_FOUND:
452 return OSD_ERR_PRI_NOT_FOUND;
453 case PNFS_OSD_ERR_NO_SPACE:
454 return OSD_ERR_PRI_NO_SPACE;
458 case PNFS_OSD_ERR_EIO:
459 return OSD_ERR_PRI_EIO;
464 merge_ioerr(struct pnfs_osd_ioerr *dest_err,
465 const struct pnfs_osd_ioerr *src_err)
467 u64 dest_end, src_end;
469 if (!dest_err->oer_errno) {
470 *dest_err = *src_err;
471 /* accumulated device must be blank */
472 memset(&dest_err->oer_component.oid_device_id, 0,
473 sizeof(dest_err->oer_component.oid_device_id));
478 if (dest_err->oer_component.oid_partition_id !=
479 src_err->oer_component.oid_partition_id)
480 dest_err->oer_component.oid_partition_id = 0;
482 if (dest_err->oer_component.oid_object_id !=
483 src_err->oer_component.oid_object_id)
484 dest_err->oer_component.oid_object_id = 0;
486 if (dest_err->oer_comp_offset > src_err->oer_comp_offset)
487 dest_err->oer_comp_offset = src_err->oer_comp_offset;
489 dest_end = end_offset(dest_err->oer_comp_offset,
490 dest_err->oer_comp_length);
491 src_end = end_offset(src_err->oer_comp_offset,
492 src_err->oer_comp_length);
493 if (dest_end < src_end)
496 dest_err->oer_comp_length = dest_end - dest_err->oer_comp_offset;
498 if ((src_err->oer_iswrite == dest_err->oer_iswrite) &&
499 (err_prio(src_err->oer_errno) > err_prio(dest_err->oer_errno))) {
500 dest_err->oer_errno = src_err->oer_errno;
501 } else if (src_err->oer_iswrite) {
502 dest_err->oer_iswrite = true;
503 dest_err->oer_errno = src_err->oer_errno;
508 encode_accumulated_error(struct objlayout *objlay, __be32 *p)
510 struct objlayout_io_state *state, *tmp;
511 struct pnfs_osd_ioerr accumulated_err = {.oer_errno = 0};
513 list_for_each_entry_safe(state, tmp, &objlay->err_list, err_list) {
516 for (i = 0; i < state->num_comps; i++) {
517 struct pnfs_osd_ioerr *ioerr = &state->ioerrs[i];
519 if (!ioerr->oer_errno)
522 printk(KERN_ERR "%s: err[%d]: errno=%d is_write=%d "
523 "dev(%llx:%llx) par=0x%llx obj=0x%llx "
524 "offset=0x%llx length=0x%llx\n",
525 __func__, i, ioerr->oer_errno,
527 _DEVID_LO(&ioerr->oer_component.oid_device_id),
528 _DEVID_HI(&ioerr->oer_component.oid_device_id),
529 ioerr->oer_component.oid_partition_id,
530 ioerr->oer_component.oid_object_id,
531 ioerr->oer_comp_offset,
532 ioerr->oer_comp_length);
534 merge_ioerr(&accumulated_err, ioerr);
536 list_del(&state->err_list);
537 objlayout_free_io_state(state);
540 pnfs_osd_xdr_encode_ioerr(p, &accumulated_err);
544 objlayout_encode_layoutreturn(struct pnfs_layout_hdr *pnfslay,
545 struct xdr_stream *xdr,
546 const struct nfs4_layoutreturn_args *args)
548 struct objlayout *objlay = OBJLAYOUT(pnfslay);
549 struct objlayout_io_state *state, *tmp;
552 dprintk("%s: Begin\n", __func__);
553 start = xdr_reserve_space(xdr, 4);
556 spin_lock(&objlay->lock);
558 list_for_each_entry_safe(state, tmp, &objlay->err_list, err_list) {
559 __be32 *last_xdr = NULL, *p;
563 for (i = 0; i < state->num_comps; i++) {
564 struct pnfs_osd_ioerr *ioerr = &state->ioerrs[i];
566 if (!ioerr->oer_errno)
569 dprintk("%s: err[%d]: errno=%d is_write=%d "
570 "dev(%llx:%llx) par=0x%llx obj=0x%llx "
571 "offset=0x%llx length=0x%llx\n",
572 __func__, i, ioerr->oer_errno,
574 _DEVID_LO(&ioerr->oer_component.oid_device_id),
575 _DEVID_HI(&ioerr->oer_component.oid_device_id),
576 ioerr->oer_component.oid_partition_id,
577 ioerr->oer_component.oid_object_id,
578 ioerr->oer_comp_offset,
579 ioerr->oer_comp_length);
581 p = pnfs_osd_xdr_ioerr_reserve_space(xdr);
584 break; /* accumulated_error */
588 pnfs_osd_xdr_encode_ioerr(p, &state->ioerrs[i]);
591 /* TODO: use xdr_write_pages */
593 /* no space for even one error descriptor */
596 /* we've encountered a situation with lots and lots of
597 * errors and no space to encode them all. Use the last
598 * available slot to report the union of all the
601 encode_accumulated_error(objlay, last_xdr);
604 list_del(&state->err_list);
605 objlayout_free_io_state(state);
608 spin_unlock(&objlay->lock);
610 *start = cpu_to_be32((xdr->p - start - 1) * 4);
611 dprintk("%s: Return\n", __func__);
616 * Get Device Info API for io engines
618 struct objlayout_deviceinfo {
620 struct pnfs_osd_deviceaddr da; /* This must be last */
623 /* Initialize and call nfs_getdeviceinfo, then decode and return a
624 * "struct pnfs_osd_deviceaddr *" Eventually objlayout_put_deviceinfo()
627 int objlayout_get_deviceinfo(struct pnfs_layout_hdr *pnfslay,
628 struct nfs4_deviceid *d_id, struct pnfs_osd_deviceaddr **deviceaddr,
631 struct objlayout_deviceinfo *odi;
632 struct pnfs_device pd;
633 struct super_block *sb;
634 struct page *page, **pages;
638 page = alloc_page(gfp_flags);
645 memcpy(&pd.dev_id, d_id, sizeof(*d_id));
646 pd.layout_type = LAYOUT_OSD2_OBJECTS;
649 pd.pglen = PAGE_SIZE;
652 sb = pnfslay->plh_inode->i_sb;
653 err = nfs4_proc_getdeviceinfo(NFS_SERVER(pnfslay->plh_inode), &pd);
654 dprintk("%s nfs_getdeviceinfo returned %d\n", __func__, err);
658 p = page_address(page);
659 odi = kzalloc(sizeof(*odi), gfp_flags);
664 pnfs_osd_xdr_decode_deviceaddr(&odi->da, p);
666 *deviceaddr = &odi->da;
674 void objlayout_put_deviceinfo(struct pnfs_osd_deviceaddr *deviceaddr)
676 struct objlayout_deviceinfo *odi = container_of(deviceaddr,
677 struct objlayout_deviceinfo,
680 __free_page(odi->page);