Whamcloud - gitweb
LU-10308 misc: update Intel copyright messages for 2017
[fs/lustre-release.git] / lustre / lov / lov_pack.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/lov/lov_pack.c
33  *
34  * (Un)packing of OST/MDS requests
35  *
36  * Author: Andreas Dilger <adilger@clusterfs.com>
37  */
38
39 #define DEBUG_SUBSYSTEM S_LOV
40
41 #include <lustre_net.h>
42 #include <lustre_swab.h>
43 #include <obd.h>
44 #include <obd_class.h>
45 #include <obd_support.h>
46
47 #include "lov_cl_internal.h"
48 #include "lov_internal.h"
49
50 void lov_dump_lmm_common(int level, void *lmmp)
51 {
52         struct lov_mds_md *lmm = lmmp;
53         struct ost_id   oi;
54
55         lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
56         CDEBUG(level, "objid "DOSTID", magic 0x%08x, pattern %#x\n",
57                POSTID(&oi), le32_to_cpu(lmm->lmm_magic),
58                le32_to_cpu(lmm->lmm_pattern));
59         CDEBUG(level, "stripe_size %u, stripe_count %u, layout_gen %u\n",
60                le32_to_cpu(lmm->lmm_stripe_size),
61                le16_to_cpu(lmm->lmm_stripe_count),
62                le16_to_cpu(lmm->lmm_layout_gen));
63 }
64
65 static void lov_dump_lmm_objects(int level, struct lov_ost_data *lod,
66                                  int stripe_count)
67 {
68         int i;
69
70         if (stripe_count > LOV_V1_INSANE_STRIPE_COUNT) {
71                 CDEBUG(level, "bad stripe_count %u > max_stripe_count %u\n",
72                        stripe_count, LOV_V1_INSANE_STRIPE_COUNT);
73                 return;
74         }
75
76         for (i = 0; i < stripe_count; ++i, ++lod) {
77                 struct ost_id oi;
78
79                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
80                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n", i,
81                        le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
82         }
83 }
84
85 void lov_dump_lmm_v1(int level, struct lov_mds_md_v1 *lmm)
86 {
87         lov_dump_lmm_common(level, lmm);
88         lov_dump_lmm_objects(level, lmm->lmm_objects,
89                              le16_to_cpu(lmm->lmm_stripe_count));
90 }
91
92 void lov_dump_lmm_v3(int level, struct lov_mds_md_v3 *lmm)
93 {
94         lov_dump_lmm_common(level, lmm);
95         CDEBUG(level, "pool_name "LOV_POOLNAMEF"\n", lmm->lmm_pool_name);
96         lov_dump_lmm_objects(level, lmm->lmm_objects,
97                              le16_to_cpu(lmm->lmm_stripe_count));
98 }
99
100 void lov_dump_lmm(int level, void *lmm)
101 {
102         int magic;
103
104         magic = le32_to_cpu(((struct lov_mds_md *)lmm)->lmm_magic);
105         switch (magic) {
106         case LOV_MAGIC_V1:
107                 lov_dump_lmm_v1(level, (struct lov_mds_md_v1 *)lmm);
108                 break;
109         case LOV_MAGIC_V3:
110                 lov_dump_lmm_v3(level, (struct lov_mds_md_v3 *)lmm);
111                 break;
112         default:
113                 CDEBUG(level, "unrecognized lmm_magic %x, assuming %x\n",
114                        magic, LOV_MAGIC_V1);
115                 lov_dump_lmm_common(level, lmm);
116                 break;
117         }
118 }
119
120 /**
121  * Pack LOV striping metadata for disk storage format (in little
122  * endian byte order).
123  *
124  * This follows the getxattr() conventions. If \a buf_size is zero
125  * then return the size needed. If \a buf_size is too small then
126  * return -ERANGE. Otherwise return the size of the result.
127  */
128 ssize_t lov_lsm_pack_v1v3(const struct lov_stripe_md *lsm, void *buf,
129                           size_t buf_size)
130 {
131         struct lov_mds_md_v1 *lmmv1 = buf;
132         struct lov_mds_md_v3 *lmmv3 = buf;
133         struct lov_ost_data_v1 *lmm_objects;
134         size_t lmm_size;
135         unsigned int i;
136         ENTRY;
137
138         lmm_size = lov_mds_md_size(lsm->lsm_entries[0]->lsme_stripe_count,
139                                    lsm->lsm_magic);
140         if (buf_size == 0)
141                 RETURN(lmm_size);
142
143         if (buf_size < lmm_size)
144                 RETURN(-ERANGE);
145
146         /* lmmv1 and lmmv3 point to the same struct and have the
147          * same first fields
148          */
149         lmmv1->lmm_magic = cpu_to_le32(lsm->lsm_magic);
150         lmm_oi_cpu_to_le(&lmmv1->lmm_oi, &lsm->lsm_oi);
151         lmmv1->lmm_stripe_size = cpu_to_le32(
152                                 lsm->lsm_entries[0]->lsme_stripe_size);
153         lmmv1->lmm_stripe_count = cpu_to_le16(
154                                 lsm->lsm_entries[0]->lsme_stripe_count);
155         lmmv1->lmm_pattern = cpu_to_le32(lsm->lsm_entries[0]->lsme_pattern);
156         lmmv1->lmm_layout_gen = cpu_to_le16(lsm->lsm_layout_gen);
157
158         if (lsm->lsm_magic == LOV_MAGIC_V3) {
159                 CLASSERT(sizeof(lsm->lsm_entries[0]->lsme_pool_name) ==
160                          sizeof(lmmv3->lmm_pool_name));
161                 strlcpy(lmmv3->lmm_pool_name,
162                         lsm->lsm_entries[0]->lsme_pool_name,
163                         sizeof(lmmv3->lmm_pool_name));
164                 lmm_objects = lmmv3->lmm_objects;
165         } else {
166                 lmm_objects = lmmv1->lmm_objects;
167         }
168
169         if (lsm->lsm_is_released)
170                 RETURN(lmm_size);
171
172         for (i = 0; i < lsm->lsm_entries[0]->lsme_stripe_count; i++) {
173                 struct lov_oinfo *loi = lsm->lsm_entries[0]->lsme_oinfo[i];
174
175                 ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
176                 lmm_objects[i].l_ost_gen = cpu_to_le32(loi->loi_ost_gen);
177                 lmm_objects[i].l_ost_idx = cpu_to_le32(loi->loi_ost_idx);
178         }
179
180         RETURN(lmm_size);
181 }
182
183 ssize_t lov_lsm_pack(const struct lov_stripe_md *lsm, void *buf,
184                      size_t buf_size)
185 {
186         struct lov_comp_md_v1 *lcmv1 = buf;
187         struct lov_comp_md_entry_v1 *lcme;
188         struct lov_ost_data_v1 *lmm_objects;
189         size_t lmm_size;
190         unsigned int entry;
191         unsigned int offset;
192         unsigned int size;
193         unsigned int i;
194         ENTRY;
195
196         if (lsm->lsm_magic == LOV_MAGIC_V1 || lsm->lsm_magic == LOV_MAGIC_V3)
197                 return lov_lsm_pack_v1v3(lsm, buf, buf_size);
198
199         lmm_size = lov_comp_md_size(lsm);
200         if (buf_size == 0)
201                 RETURN(lmm_size);
202
203         if (buf_size < lmm_size)
204                 RETURN(-ERANGE);
205
206         lcmv1->lcm_magic = cpu_to_le32(lsm->lsm_magic);
207         lcmv1->lcm_size = cpu_to_le32(lmm_size);
208         lcmv1->lcm_layout_gen = cpu_to_le32(lsm->lsm_layout_gen);
209         lcmv1->lcm_flags = cpu_to_le16(lsm->lsm_flags);
210         lcmv1->lcm_mirror_count = cpu_to_le16(lsm->lsm_mirror_count);
211         lcmv1->lcm_entry_count = cpu_to_le16(lsm->lsm_entry_count);
212
213         offset = sizeof(*lcmv1) + sizeof(*lcme) * lsm->lsm_entry_count;
214
215         for (entry = 0; entry < lsm->lsm_entry_count; entry++) {
216                 struct lov_stripe_md_entry *lsme;
217                 struct lov_mds_md *lmm;
218                 __u16 stripe_count;
219
220                 lsme = lsm->lsm_entries[entry];
221                 lcme = &lcmv1->lcm_entries[entry];
222
223                 lcme->lcme_id = cpu_to_le32(lsme->lsme_id);
224                 lcme->lcme_flags = cpu_to_le32(lsme->lsme_flags);
225                 lcme->lcme_extent.e_start =
226                         cpu_to_le64(lsme->lsme_extent.e_start);
227                 lcme->lcme_extent.e_end =
228                         cpu_to_le64(lsme->lsme_extent.e_end);
229                 lcme->lcme_offset = cpu_to_le32(offset);
230
231                 lmm = (struct lov_mds_md *)((char *)lcmv1 + offset);
232                 lmm->lmm_magic = cpu_to_le32(lsme->lsme_magic);
233                 /* lmm->lmm_oi not set */
234                 lmm->lmm_pattern = cpu_to_le32(lsme->lsme_pattern);
235                 lmm->lmm_stripe_size = cpu_to_le32(lsme->lsme_stripe_size);
236                 lmm->lmm_stripe_count = cpu_to_le16(lsme->lsme_stripe_count);
237                 lmm->lmm_layout_gen = cpu_to_le16(lsme->lsme_layout_gen);
238
239                 if (lsme->lsme_magic == LOV_MAGIC_V3) {
240                         struct lov_mds_md_v3 *lmmv3 =
241                                                 (struct lov_mds_md_v3 *)lmm;
242
243                         strlcpy(lmmv3->lmm_pool_name, lsme->lsme_pool_name,
244                                 sizeof(lmmv3->lmm_pool_name));
245                         lmm_objects = lmmv3->lmm_objects;
246                 } else {
247                         lmm_objects =
248                                 ((struct lov_mds_md_v1 *)lmm)->lmm_objects;
249                 }
250
251                 if (lsme_inited(lsme) &&
252                     !(lsme->lsme_pattern & LOV_PATTERN_F_RELEASED))
253                         stripe_count = lsme->lsme_stripe_count;
254                 else
255                         stripe_count = 0;
256
257                 for (i = 0; i < stripe_count; i++) {
258                         struct lov_oinfo *loi = lsme->lsme_oinfo[i];
259
260                         ostid_cpu_to_le(&loi->loi_oi, &lmm_objects[i].l_ost_oi);
261                         lmm_objects[i].l_ost_gen =
262                                         cpu_to_le32(loi->loi_ost_gen);
263                         lmm_objects[i].l_ost_idx =
264                                         cpu_to_le32(loi->loi_ost_idx);
265                 }
266
267                 size = lov_mds_md_size(stripe_count, lsme->lsme_magic);
268                 lcme->lcme_size = cpu_to_le32(size);
269                 offset += size;
270         } /* for each layout component */
271
272         RETURN(lmm_size);
273 }
274
275 /* Find the max stripecount we should use */
276 __u16 lov_get_stripe_count(struct lov_obd *lov, __u32 magic, __u16 stripe_count)
277 {
278         __u32 max_stripes = LOV_MAX_STRIPE_COUNT_OLD;
279
280         if (!stripe_count)
281                 stripe_count = lov->desc.ld_default_stripe_count;
282         if (stripe_count > lov->desc.ld_active_tgt_count)
283                 stripe_count = lov->desc.ld_active_tgt_count;
284         if (!stripe_count)
285                 stripe_count = 1;
286
287         /* stripe count is based on whether ldiskfs can handle
288          * larger EA sizes */
289         if (lov->lov_ocd.ocd_connect_flags & OBD_CONNECT_MAX_EASIZE &&
290             lov->lov_ocd.ocd_max_easize)
291                 max_stripes = lov_mds_md_max_stripe_count(
292                         lov->lov_ocd.ocd_max_easize, magic);
293
294         if (stripe_count > max_stripes)
295                 stripe_count = max_stripes;
296
297         return stripe_count;
298 }
299
300 int lov_free_memmd(struct lov_stripe_md **lsmp)
301 {
302         struct lov_stripe_md *lsm = *lsmp;
303         int refc;
304
305         *lsmp = NULL;
306         refc = atomic_dec_return(&lsm->lsm_refc);
307         LASSERT(refc >= 0);
308         if (refc == 0)
309                 lsm_free(lsm);
310
311         return refc;
312 }
313
314 /* Unpack LOV object metadata from disk storage.  It is packed in LE byte
315  * order and is opaque to the networking layer.
316  */
317 struct lov_stripe_md *lov_unpackmd(struct lov_obd *lov, void *buf,
318                                    size_t buf_size)
319 {
320         const struct lsm_operations *op;
321         struct lov_stripe_md *lsm;
322         u32 magic;
323         ENTRY;
324
325         if (buf_size < sizeof(magic))
326                 RETURN(ERR_PTR(-EINVAL));
327
328         magic = le32_to_cpu(*(u32 *)buf);
329         op = lsm_op_find(magic);
330         if (op == NULL)
331                 RETURN(ERR_PTR(-EINVAL));
332
333         lsm = op->lsm_unpackmd(lov, buf, buf_size);
334
335         RETURN(lsm);
336 }
337
338 /* Retrieve object striping information.
339  *
340  * @lump is a pointer to an in-core struct with lmm_ost_count indicating
341  * the maximum number of OST indices which will fit in the user buffer.
342  * lmm_magic must be LOV_USER_MAGIC.
343  *
344  * If @size > 0, User specified limited buffer size, usually the buffer is from
345  * ll_lov_setstripe(), and the buffer can only hold basic layout template info.
346  */
347 int lov_getstripe(const struct lu_env *env, struct lov_object *obj,
348                   struct lov_stripe_md *lsm, struct lov_user_md __user *lump,
349                   size_t size)
350 {
351         /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
352         struct lov_mds_md *lmmk, *lmm;
353         struct lov_user_md_v1 lum;
354         size_t  lmmk_size;
355         ssize_t lmm_size, lum_size = 0;
356         static bool printed;
357         int     rc = 0;
358         ENTRY;
359
360         if (lsm->lsm_magic != LOV_MAGIC_V1 && lsm->lsm_magic != LOV_MAGIC_V3 &&
361             lsm->lsm_magic != LOV_MAGIC_COMP_V1) {
362                 CERROR("bad LSM MAGIC: 0x%08X != 0x%08X nor 0x%08X\n",
363                        lsm->lsm_magic, LOV_MAGIC_V1, LOV_MAGIC_V3);
364                 GOTO(out, rc = -EIO);
365         }
366
367         if (!printed) {
368                 LCONSOLE_WARN("%s: using old ioctl(LL_IOC_LOV_GETSTRIPE) on "
369                               DFID", use llapi_layout_get_by_path()\n",
370                               current->comm,
371                               PFID(&obj->lo_cl.co_lu.lo_header->loh_fid));
372                 printed = true;
373         }
374
375         lmmk_size = lov_comp_md_size(lsm);
376
377         OBD_ALLOC_LARGE(lmmk, lmmk_size);
378         if (lmmk == NULL)
379                 GOTO(out, rc = -ENOMEM);
380
381         lmm_size = lov_lsm_pack(lsm, lmmk, lmmk_size);
382         if (lmm_size < 0)
383                 GOTO(out_free, rc = lmm_size);
384
385         if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC) {
386                 if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V1) ||
387                     lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
388                         lustre_swab_lov_mds_md(lmmk);
389                         lustre_swab_lov_user_md_objects(
390                                 (struct lov_user_ost_data *)lmmk->lmm_objects,
391                                 lmmk->lmm_stripe_count);
392                 } else if (lmmk->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1)) {
393                         lustre_swab_lov_comp_md_v1(
394                                         (struct lov_comp_md_v1 *)lmmk);
395                 }
396         }
397
398         /* Legacy appication passes limited buffer, we need to figure out
399          * the user buffer size by the passed in lmm_stripe_count. */
400         if (copy_from_user(&lum, lump, sizeof(struct lov_user_md_v1)))
401                 GOTO(out_free, rc = -EFAULT);
402
403         if (lum.lmm_magic == LOV_USER_MAGIC_V1 ||
404             lum.lmm_magic == LOV_USER_MAGIC_V3)
405                 lum_size = lov_user_md_size(lum.lmm_stripe_count,
406                                             lum.lmm_magic);
407
408         if (lum_size != 0) {
409                 struct lov_mds_md *comp_md = lmmk;
410
411                 /* Legacy app (ADIO for instance) treats the layout as V1/V3
412                  * blindly, we'd return a reasonable V1/V3 for them. */
413                 if (lmmk->lmm_magic == LOV_MAGIC_COMP_V1) {
414                         struct lov_comp_md_v1 *comp_v1;
415                         struct cl_object *cl_obj;
416                         struct cl_attr attr;
417                         int i;
418
419                         attr.cat_size = 0;
420                         cl_obj = cl_object_top(&obj->lo_cl);
421                         cl_object_attr_lock(cl_obj);
422                         cl_object_attr_get(env, cl_obj, &attr);
423                         cl_object_attr_unlock(cl_obj);
424
425                         /* return the last instantiated component if file size
426                          * is non-zero, otherwise, return the last component.*/
427                         comp_v1 = (struct lov_comp_md_v1 *)lmmk;
428                         i = attr.cat_size == 0 ? comp_v1->lcm_entry_count : 0;
429                         for (; i < comp_v1->lcm_entry_count; i++) {
430                                 if (!(comp_v1->lcm_entries[i].lcme_flags &
431                                                 LCME_FL_INIT))
432                                         break;
433                         }
434                         if (i > 0)
435                                 i--;
436                         comp_md = (struct lov_mds_md *)((char *)comp_v1 +
437                                         comp_v1->lcm_entries[i].lcme_offset);
438                 }
439
440                 lmm = comp_md;
441                 lmm_size = lum_size;
442         } else {
443                 lmm = lmmk;
444                 lmm_size = lmmk_size;
445         }
446         /**
447          * User specified limited buffer size, usually the buffer is
448          * from ll_lov_setstripe(), and the buffer can only hold basic
449          * layout template info.
450          */
451         if (size == 0 || size > lmm_size)
452                 size = lmm_size;
453         if (copy_to_user(lump, lmm, size))
454                 GOTO(out_free, rc = -EFAULT);
455
456 out_free:
457         OBD_FREE_LARGE(lmmk, lmmk_size);
458 out:
459         RETURN(rc);
460 }