4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_lib.c
38 * Lustre Light Super operations
41 #define DEBUG_SUBSYSTEM S_LLITE
43 #include <linux/module.h>
44 #include <linux/statfs.h>
45 #include <linux/time.h>
46 #include <linux/types.h>
47 #include <linux/version.h>
49 #include <linux/user_namespace.h>
50 #ifdef HAVE_UIDGID_HEADER
51 # include <linux/uidgid.h>
53 #include <linux/security.h>
55 #include <lustre_ioctl.h>
56 #include <lustre_ha.h>
57 #include <lustre_dlm.h>
58 #include <lprocfs_status.h>
59 #include <lustre_disk.h>
60 #include <lustre_param.h>
61 #include <lustre_log.h>
62 #include <cl_object.h>
63 #include <obd_cksum.h>
64 #include "llite_internal.h"
66 struct kmem_cache *ll_file_data_slab;
69 #define log2(n) ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
76 unsigned long lru_page_max;
86 spin_lock_init(&sbi->ll_lock);
87 mutex_init(&sbi->ll_lco.lco_lock);
88 spin_lock_init(&sbi->ll_pp_extent_lock);
89 spin_lock_init(&sbi->ll_process_lock);
90 sbi->ll_rw_stats_on = 0;
93 pages = si.totalram - si.totalhigh;
94 lru_page_max = pages / 2;
96 /* initialize ll_cache data */
97 sbi->ll_cache = cl_cache_init(lru_page_max);
98 if (sbi->ll_cache == NULL) {
99 OBD_FREE(sbi, sizeof(*sbi));
103 sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
104 SBI_DEFAULT_READAHEAD_MAX);
105 sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
106 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
107 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
109 ll_generate_random_uuid(uuid);
110 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
111 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
113 sbi->ll_flags |= LL_SBI_VERBOSE;
114 #ifdef ENABLE_CHECKSUM
115 sbi->ll_flags |= LL_SBI_CHECKSUM;
118 #ifdef HAVE_LRU_RESIZE_SUPPORT
119 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
121 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
123 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
124 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
130 /* metadata statahead is enabled by default */
131 sbi->ll_sa_max = LL_SA_RPC_DEF;
132 atomic_set(&sbi->ll_sa_total, 0);
133 atomic_set(&sbi->ll_sa_wrong, 0);
134 atomic_set(&sbi->ll_sa_running, 0);
135 atomic_set(&sbi->ll_agl_total, 0);
136 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
137 sbi->ll_flags |= LL_SBI_FAST_READ;
140 sbi->ll_squash.rsi_uid = 0;
141 sbi->ll_squash.rsi_gid = 0;
142 INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
143 init_rwsem(&sbi->ll_squash.rsi_sem);
148 static void ll_free_sbi(struct super_block *sb)
150 struct ll_sb_info *sbi = ll_s2sbi(sb);
154 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
155 cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
156 if (sbi->ll_cache != NULL) {
157 cl_cache_decref(sbi->ll_cache);
158 sbi->ll_cache = NULL;
160 OBD_FREE(sbi, sizeof(*sbi));
165 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
167 return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
168 data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
171 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
172 struct vfsmount *mnt)
174 struct inode *root = NULL;
175 struct ll_sb_info *sbi = ll_s2sbi(sb);
176 struct obd_device *obd;
177 struct obd_statfs *osfs = NULL;
178 struct ptlrpc_request *request = NULL;
179 struct obd_connect_data *data = NULL;
180 struct obd_uuid *uuid;
181 struct md_op_data *op_data;
182 struct lustre_md lmd;
184 int size, err, checksum;
187 obd = class_name2obd(md);
189 CERROR("MD %s: not setup or attached\n", md);
203 /* indicate the features supported by this client */
204 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
205 OBD_CONNECT_ATTRFID |
206 OBD_CONNECT_VERSION | OBD_CONNECT_BRW_SIZE |
207 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
208 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
209 OBD_CONNECT_AT | OBD_CONNECT_LOV_V3 |
210 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
211 OBD_CONNECT_64BITHASH |
212 OBD_CONNECT_EINPROGRESS |
213 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
214 OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
215 OBD_CONNECT_MAX_EASIZE |
216 OBD_CONNECT_FLOCK_DEAD |
217 OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
218 OBD_CONNECT_OPEN_BY_FID |
219 OBD_CONNECT_DIR_STRIPE |
220 OBD_CONNECT_BULK_MBITS |
221 OBD_CONNECT_SUBTREE |
224 data->ocd_connect_flags2 = 0;
226 #ifdef HAVE_LRU_RESIZE_SUPPORT
227 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
228 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
230 #ifdef CONFIG_FS_POSIX_ACL
231 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
234 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
235 /* flag mdc connection as lightweight, only used for test
236 * purpose, use with care */
237 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
239 data->ocd_ibits_known = MDS_INODELOCK_FULL;
240 data->ocd_version = LUSTRE_VERSION_CODE;
242 if (sb->s_flags & MS_RDONLY)
243 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
244 if (sbi->ll_flags & LL_SBI_USER_XATTR)
245 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
247 if (sbi->ll_flags & LL_SBI_FLOCK)
248 sbi->ll_fop = &ll_file_operations_flock;
249 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
250 sbi->ll_fop = &ll_file_operations;
252 sbi->ll_fop = &ll_file_operations_noflock;
254 /* always ping even if server suppress_pings */
255 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
256 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
258 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
259 data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
260 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
262 data->ocd_brw_size = MD_MAX_BRW_SIZE;
264 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
266 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
267 "recovery, of which this client is not a "
268 "part. Please wait for recovery to complete,"
269 " abort, or time out.\n", md);
272 CERROR("cannot connect to %s: rc = %d\n", md, err);
276 sbi->ll_md_exp->exp_connect_data = *data;
278 err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
279 LUSTRE_SEQ_METADATA);
281 CERROR("%s: Can't init metadata layer FID infrastructure, "
282 "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
286 /* For mount, we only need fs info from MDT0, and also in DNE, it
287 * can make sure the client can be mounted as long as MDT0 is
289 err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
290 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
291 OBD_STATFS_FOR_MDT0);
293 GOTO(out_md_fid, err);
295 /* This needs to be after statfs to ensure connect has finished.
296 * Note that "data" does NOT contain the valid connect reply.
297 * If connecting to a 1.8 server there will be no LMV device, so
298 * we can access the MDC export directly and exp_connect_flags will
299 * be non-zero, but if accessing an upgraded 2.1 server it will
300 * have the correct flags filled in.
301 * XXX: fill in the LMV exp_connect_flags from MDC(s). */
302 valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
303 if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
304 valid != CLIENT_CONNECT_MDT_REQD) {
307 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
308 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
309 valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
310 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
311 "feature(s) needed for correct operation "
312 "of this client (%s). Please upgrade "
313 "server or downgrade client.\n",
314 sbi->ll_md_exp->exp_obd->obd_name, buf);
315 OBD_FREE(buf, PAGE_CACHE_SIZE);
316 GOTO(out_md_fid, err = -EPROTO);
319 size = sizeof(*data);
320 err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
321 KEY_CONN_DATA, &size, data);
323 CERROR("%s: Get connect data failed: rc = %d\n",
324 sbi->ll_md_exp->exp_obd->obd_name, err);
325 GOTO(out_md_fid, err);
328 LASSERT(osfs->os_bsize);
329 sb->s_blocksize = osfs->os_bsize;
330 sb->s_blocksize_bits = log2(osfs->os_bsize);
331 sb->s_magic = LL_SUPER_MAGIC;
332 sb->s_maxbytes = MAX_LFS_FILESIZE;
333 sbi->ll_namelen = osfs->os_namelen;
335 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
336 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
337 LCONSOLE_INFO("Disabling user_xattr feature because "
338 "it is not supported on the server\n");
339 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
342 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
344 sb->s_flags |= MS_POSIXACL;
346 sbi->ll_flags |= LL_SBI_ACL;
348 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
350 sb->s_flags &= ~MS_POSIXACL;
352 sbi->ll_flags &= ~LL_SBI_ACL;
355 if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
356 sbi->ll_flags |= LL_SBI_64BIT_HASH;
358 if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
359 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
361 sbi->ll_md_brw_pages = 1;
363 if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
364 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
366 if (obd_connect_has_secctx(data))
367 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
369 if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
370 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
371 LCONSOLE_INFO("%s: disabling xattr cache due to "
372 "unknown maximum xattr size.\n", dt);
374 sbi->ll_flags |= LL_SBI_XATTR_CACHE;
375 sbi->ll_xattr_cache_enabled = 1;
379 obd = class_name2obd(dt);
381 CERROR("DT %s: not setup or attached\n", dt);
382 GOTO(out_md_fid, err = -ENODEV);
385 /* pass client page size via ocd_grant_blkbits, the server should report
386 * back its backend blocksize for grant calculation purpose */
387 data->ocd_grant_blkbits = PAGE_SHIFT;
389 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
390 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
391 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
392 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
393 OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
394 OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
395 OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
396 OBD_CONNECT_EINPROGRESS |
397 OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
398 OBD_CONNECT_LAYOUTLOCK |
399 OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
400 OBD_CONNECT_BULK_MBITS;
402 data->ocd_connect_flags2 = 0;
404 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
405 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
407 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
408 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
409 * disabled by default, because it can still be enabled on the
410 * fly via /proc. As a consequence, we still need to come to an
411 * agreement on the supported algorithms at connect time */
412 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
414 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
415 data->ocd_cksum_types = OBD_CKSUM_ADLER;
417 data->ocd_cksum_types = cksum_types_supported_client();
420 #ifdef HAVE_LRU_RESIZE_SUPPORT
421 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
423 /* always ping even if server suppress_pings */
424 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
425 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
427 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
428 "ocd_grant: %d\n", data->ocd_connect_flags,
429 data->ocd_version, data->ocd_grant);
431 obd->obd_upcall.onu_owner = &sbi->ll_lco;
432 obd->obd_upcall.onu_upcall = cl_ocd_update;
434 data->ocd_brw_size = DT_MAX_BRW_SIZE;
436 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
439 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
440 "recovery, of which this client is not a "
441 "part. Please wait for recovery to "
442 "complete, abort, or time out.\n", dt);
445 CERROR("%s: Cannot connect to %s: rc = %d\n",
446 sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
450 sbi->ll_dt_exp->exp_connect_data = *data;
452 err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
453 LUSTRE_SEQ_METADATA);
455 CERROR("%s: Can't init data layer FID infrastructure, "
456 "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
460 mutex_lock(&sbi->ll_lco.lco_lock);
461 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
462 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
463 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
464 mutex_unlock(&sbi->ll_lco.lco_lock);
466 fid_zero(&sbi->ll_root_fid);
467 err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
470 CERROR("cannot mds_connect: rc = %d\n", err);
471 GOTO(out_lock_cn_cb, err);
473 if (!fid_is_sane(&sbi->ll_root_fid)) {
474 CERROR("%s: Invalid root fid "DFID" during mount\n",
475 sbi->ll_md_exp->exp_obd->obd_name,
476 PFID(&sbi->ll_root_fid));
477 GOTO(out_lock_cn_cb, err = -EINVAL);
479 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
481 sb->s_op = &lustre_super_operations;
482 #if THREAD_SIZE >= 8192 /*b=17630*/
483 sb->s_export_op = &lustre_export_operations;
487 * XXX: move this to after cbd setup? */
488 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
489 if (sbi->ll_flags & LL_SBI_ACL)
490 valid |= OBD_MD_FLACL;
492 OBD_ALLOC_PTR(op_data);
494 GOTO(out_lock_cn_cb, err = -ENOMEM);
496 op_data->op_fid1 = sbi->ll_root_fid;
497 op_data->op_mode = 0;
498 op_data->op_valid = valid;
500 err = md_getattr(sbi->ll_md_exp, op_data, &request);
502 OBD_FREE_PTR(op_data);
504 CERROR("%s: md_getattr failed for root: rc = %d\n",
505 sbi->ll_md_exp->exp_obd->obd_name, err);
506 GOTO(out_lock_cn_cb, err);
509 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
510 sbi->ll_md_exp, &lmd);
512 CERROR("failed to understand root inode md: rc = %d\n", err);
513 ptlrpc_req_finished(request);
514 GOTO(out_lock_cn_cb, err);
517 LASSERT(fid_is_sane(&sbi->ll_root_fid));
518 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
519 sbi->ll_flags & LL_SBI_32BIT_API),
521 md_free_lustre_md(sbi->ll_md_exp, &lmd);
522 ptlrpc_req_finished(request);
525 #ifdef CONFIG_FS_POSIX_ACL
527 posix_acl_release(lmd.posix_acl);
528 lmd.posix_acl = NULL;
531 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
533 CERROR("lustre_lite: bad iget4 for root\n");
537 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
538 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
539 KEY_CHECKSUM, sizeof(checksum), &checksum,
542 CERROR("%s: Set checksum failed: rc = %d\n",
543 sbi->ll_dt_exp->exp_obd->obd_name, err);
548 err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
549 KEY_CACHE_SET, sizeof(*sbi->ll_cache),
550 sbi->ll_cache, NULL);
552 CERROR("%s: Set cache_set failed: rc = %d\n",
553 sbi->ll_dt_exp->exp_obd->obd_name, err);
557 sb->s_root = d_make_root(root);
558 if (sb->s_root == NULL) {
559 CERROR("%s: can't make root dentry\n",
560 ll_get_fsname(sb, NULL, 0));
561 GOTO(out_root, err = -ENOMEM);
563 #ifdef HAVE_DCACHE_LOCK
564 sb->s_root->d_op = &ll_d_ops;
567 sbi->ll_sdev_orig = sb->s_dev;
569 /* We set sb->s_dev equal on all lustre clients in order to support
570 * NFS export clustering. NFSD requires that the FSID be the same
572 /* s_dev is also used in lt_compare() to compare two fs, but that is
573 * only a node-local comparison. */
574 uuid = obd_get_uuid(sbi->ll_md_exp);
576 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
582 if (proc_lustre_fs_root != NULL) {
583 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
586 CERROR("%s: could not register mount in lprocfs: "
587 "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
597 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
599 obd_disconnect(sbi->ll_dt_exp);
600 sbi->ll_dt_exp = NULL;
602 obd_fid_fini(sbi->ll_md_exp->exp_obd);
604 obd_disconnect(sbi->ll_md_exp);
605 sbi->ll_md_exp = NULL;
614 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
618 size = sizeof(*lmmsize);
619 rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
620 KEY_MAX_EASIZE, &size, lmmsize);
622 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
623 sbi->ll_dt_exp->exp_obd->obd_name, rc);
628 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
629 KEY_MAX_EASIZE, &size, lmmsize);
631 CERROR("Get max mdsize error rc %d\n", rc);
637 * Get the value of the default_easize parameter.
639 * \see client_obd::cl_default_mds_easize
641 * \param[in] sbi superblock info for this filesystem
642 * \param[out] lmmsize pointer to storage location for value
644 * \retval 0 on success
645 * \retval negative negated errno on failure
647 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
652 rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
653 KEY_DEFAULT_EASIZE, &size, lmmsize);
655 CERROR("Get default mdsize error rc %d\n", rc);
661 * Set the default_easize parameter to the given value.
663 * \see client_obd::cl_default_mds_easize
665 * \param[in] sbi superblock info for this filesystem
666 * \param[in] lmmsize the size to set
668 * \retval 0 on success
669 * \retval negative negated errno on failure
671 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
675 if (lmmsize < sizeof(struct lov_mds_md) ||
676 lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
679 rc = obd_set_info_async(NULL, sbi->ll_md_exp,
680 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
681 sizeof(int), &lmmsize, NULL);
686 static void ll_dump_inode(struct inode *inode)
688 struct ll_d_hlist_node *tmp;
689 int dentry_count = 0;
691 LASSERT(inode != NULL);
693 ll_d_hlist_for_each(tmp, &inode->i_dentry)
696 CERROR("%s: inode %p dump: dev=%s fid="DFID
697 " mode=%o count=%u, %d dentries\n",
698 ll_get_fsname(inode->i_sb, NULL, 0), inode,
699 ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
700 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
703 void lustre_dump_dentry(struct dentry *dentry, int recur)
705 struct list_head *tmp;
708 LASSERT(dentry != NULL);
710 list_for_each(tmp, &dentry->d_subdirs)
713 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
714 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
715 dentry->d_name.len, dentry->d_name.name,
716 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
717 dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
718 dentry->d_flags, dentry->d_fsdata, subdirs);
719 if (dentry->d_inode != NULL)
720 ll_dump_inode(dentry->d_inode);
725 list_for_each(tmp, &dentry->d_subdirs) {
726 struct dentry *d = list_entry(tmp, struct dentry, d_child);
727 lustre_dump_dentry(d, recur - 1);
731 static void client_common_put_super(struct super_block *sb)
733 struct ll_sb_info *sbi = ll_s2sbi(sb);
738 obd_fid_fini(sbi->ll_dt_exp->exp_obd);
739 obd_disconnect(sbi->ll_dt_exp);
740 sbi->ll_dt_exp = NULL;
742 lprocfs_unregister_mountpoint(sbi);
744 obd_fid_fini(sbi->ll_md_exp->exp_obd);
745 obd_disconnect(sbi->ll_md_exp);
746 sbi->ll_md_exp = NULL;
751 void ll_kill_super(struct super_block *sb)
753 struct ll_sb_info *sbi;
757 if (!(sb->s_flags & MS_ACTIVE))
761 /* we need restore s_dev from changed for clustred NFS before put_super
762 * because new kernels have cached s_dev and change sb->s_dev in
763 * put_super not affected real removing devices */
765 sb->s_dev = sbi->ll_sdev_orig;
766 sbi->ll_umounting = 1;
768 /* wait running statahead threads to quit */
769 while (atomic_read(&sbi->ll_sa_running) > 0) {
770 set_current_state(TASK_UNINTERRUPTIBLE);
771 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
778 static inline int ll_set_opt(const char *opt, char *data, int fl)
780 if (strncmp(opt, data, strlen(opt)) != 0)
786 /* non-client-specific mount options are parsed in lmd_parse */
787 static int ll_options(char *options, int *flags)
790 char *s1 = options, *s2;
796 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
799 CDEBUG(D_SUPER, "next opt=%s\n", s1);
800 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
805 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
810 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
815 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
820 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
825 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
830 tmp = ll_set_opt("context", s1, 1);
833 tmp = ll_set_opt("fscontext", s1, 1);
836 tmp = ll_set_opt("defcontext", s1, 1);
839 tmp = ll_set_opt("rootcontext", s1, 1);
842 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
847 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
853 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
858 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
863 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
868 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
873 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
878 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
883 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
888 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
893 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
898 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
903 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
909 s2 = strchr(s1, ',');
917 void ll_lli_init(struct ll_inode_info *lli)
919 lli->lli_inode_magic = LLI_INODE_MAGIC;
921 spin_lock_init(&lli->lli_lock);
922 lli->lli_posix_acl = NULL;
923 /* Do not set lli_fid, it has been initialized already. */
924 fid_zero(&lli->lli_pfid);
925 lli->lli_mds_read_och = NULL;
926 lli->lli_mds_write_och = NULL;
927 lli->lli_mds_exec_och = NULL;
928 lli->lli_open_fd_read_count = 0;
929 lli->lli_open_fd_write_count = 0;
930 lli->lli_open_fd_exec_count = 0;
931 mutex_init(&lli->lli_och_mutex);
932 spin_lock_init(&lli->lli_agl_lock);
933 spin_lock_init(&lli->lli_layout_lock);
934 ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
935 lli->lli_clob = NULL;
937 init_rwsem(&lli->lli_xattrs_list_rwsem);
938 mutex_init(&lli->lli_xattrs_enq_lock);
940 LASSERT(lli->lli_vfs_inode.i_mode != 0);
941 if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
942 mutex_init(&lli->lli_readdir_mutex);
943 lli->lli_opendir_key = NULL;
945 spin_lock_init(&lli->lli_sa_lock);
946 lli->lli_opendir_pid = 0;
947 lli->lli_sa_enabled = 0;
948 lli->lli_def_stripe_offset = -1;
950 mutex_init(&lli->lli_size_mutex);
951 lli->lli_symlink_name = NULL;
952 init_rwsem(&lli->lli_trunc_sem);
953 range_lock_tree_init(&lli->lli_write_tree);
954 init_rwsem(&lli->lli_glimpse_sem);
955 lli->lli_glimpse_time = 0;
956 INIT_LIST_HEAD(&lli->lli_agl_list);
957 lli->lli_agl_index = 0;
958 lli->lli_async_rc = 0;
960 mutex_init(&lli->lli_layout_mutex);
963 static inline int ll_bdi_register(struct backing_dev_info *bdi)
965 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
967 bdi->name = "lustre";
968 return bdi_register(bdi, NULL, "lustre-%d",
969 atomic_inc_return(&ll_bdi_num));
972 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
974 struct lustre_profile *lprof = NULL;
975 struct lustre_sb_info *lsi = s2lsi(sb);
976 struct ll_sb_info *sbi;
977 char *dt = NULL, *md = NULL;
978 char *profilenm = get_profile_name(sb);
979 struct config_llog_instance *cfg;
980 /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
981 const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
985 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
991 try_module_get(THIS_MODULE);
993 /* client additional sb info */
994 lsi->lsi_llsbi = sbi = ll_init_sbi();
996 module_put(THIS_MODULE);
1001 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1003 GOTO(out_free, err);
1005 err = bdi_init(&lsi->lsi_bdi);
1007 GOTO(out_free, err);
1008 lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1009 #ifdef HAVE_BDI_CAP_MAP_COPY
1010 lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1012 lsi->lsi_bdi.capabilities = 0;
1014 err = ll_bdi_register(&lsi->lsi_bdi);
1016 GOTO(out_free, err);
1018 sb->s_bdi = &lsi->lsi_bdi;
1019 #ifndef HAVE_DCACHE_LOCK
1020 /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1021 sb->s_d_op = &ll_d_ops;
1024 /* Generate a string unique to this super, in case some joker tries
1025 to mount the same fs at two mount points.
1026 Use the address of the super itself.*/
1027 cfg->cfg_instance = sb;
1028 cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1029 cfg->cfg_callback = class_config_llog_handler;
1030 /* set up client obds */
1031 err = lustre_process_log(sb, profilenm, cfg);
1033 GOTO(out_free, err);
1035 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1036 lprof = class_get_profile(profilenm);
1037 if (lprof == NULL) {
1038 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1039 " read from the MGS. Does that filesystem "
1040 "exist?\n", profilenm);
1041 GOTO(out_free, err = -EINVAL);
1043 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1044 lprof->lp_md, lprof->lp_dt);
1046 OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1048 GOTO(out_free, err = -ENOMEM);
1049 sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1051 OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1053 GOTO(out_free, err = -ENOMEM);
1054 sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1056 /* connections, registrations, sb setup */
1057 err = client_common_fill_super(sb, md, dt, mnt);
1059 GOTO(out_free, err);
1061 sbi->ll_client_common_fill_super_succeeded = 1;
1065 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1067 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1069 class_put_profile(lprof);
1072 else if (sbi->ll_flags & LL_SBI_VERBOSE)
1073 LCONSOLE_WARN("Mounted %s\n", profilenm);
1077 } /* ll_fill_super */
1079 void ll_put_super(struct super_block *sb)
1081 struct config_llog_instance cfg, params_cfg;
1082 struct obd_device *obd;
1083 struct lustre_sb_info *lsi = s2lsi(sb);
1084 struct ll_sb_info *sbi = ll_s2sbi(sb);
1085 char *profilenm = get_profile_name(sb);
1087 int next, force = 1, rc = 0;
1090 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1092 cfg.cfg_instance = sb;
1093 lustre_end_log(sb, profilenm, &cfg);
1095 params_cfg.cfg_instance = sb;
1096 lustre_end_log(sb, PARAMS_FILENAME, ¶ms_cfg);
1098 if (sbi->ll_md_exp) {
1099 obd = class_exp2obd(sbi->ll_md_exp);
1101 force = obd->obd_force;
1104 /* Wait for unstable pages to be committed to stable storage */
1106 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1107 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1108 atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1112 ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1113 if (force == 0 && rc != -EINTR)
1114 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1117 /* We need to set force before the lov_disconnect in
1118 lustre_common_put_super, since l_d cleans up osc's as well. */
1121 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1123 obd->obd_force = force;
1127 if (sbi->ll_client_common_fill_super_succeeded) {
1128 /* Only if client_common_fill_super succeeded */
1129 client_common_put_super(sb);
1133 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1134 class_manual_cleanup(obd);
1137 if (sbi->ll_flags & LL_SBI_VERBOSE)
1138 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1141 class_del_profile(profilenm);
1143 if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1144 bdi_destroy(&lsi->lsi_bdi);
1145 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1149 lsi->lsi_llsbi = NULL;
1151 lustre_common_put_super(sb);
1153 cl_env_cache_purge(~0);
1155 module_put(THIS_MODULE);
1158 } /* client_put_super */
1160 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1162 struct inode *inode = NULL;
1164 /* NOTE: we depend on atomic igrab() -bzzz */
1165 lock_res_and_lock(lock);
1166 if (lock->l_resource->lr_lvb_inode) {
1167 struct ll_inode_info * lli;
1168 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1169 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1170 inode = igrab(lock->l_resource->lr_lvb_inode);
1172 inode = lock->l_resource->lr_lvb_inode;
1173 LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ? D_INFO :
1174 D_WARNING, lock, "lr_lvb_inode %p is "
1175 "bogus: magic %08x",
1176 lock->l_resource->lr_lvb_inode,
1177 lli->lli_inode_magic);
1181 unlock_res_and_lock(lock);
1185 void ll_dir_clear_lsm_md(struct inode *inode)
1187 struct ll_inode_info *lli = ll_i2info(inode);
1189 LASSERT(S_ISDIR(inode->i_mode));
1191 if (lli->lli_lsm_md != NULL) {
1192 lmv_free_memmd(lli->lli_lsm_md);
1193 lli->lli_lsm_md = NULL;
1197 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1198 const struct lu_fid *fid,
1199 struct lustre_md *md)
1201 struct ll_sb_info *sbi = ll_s2sbi(sb);
1202 struct mdt_body *body = md->body;
1203 struct inode *inode;
1207 ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1208 inode = iget_locked(sb, ino);
1209 if (inode == NULL) {
1210 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1211 ll_get_fsname(sb, NULL, 0), PFID(fid));
1212 RETURN(ERR_PTR(-ENOENT));
1215 if (inode->i_state & I_NEW) {
1216 struct ll_inode_info *lli = ll_i2info(inode);
1217 struct lmv_stripe_md *lsm = md->lmv;
1219 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1220 (body->mbo_mode & S_IFMT);
1221 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1224 LTIME_S(inode->i_mtime) = 0;
1225 LTIME_S(inode->i_atime) = 0;
1226 LTIME_S(inode->i_ctime) = 0;
1229 #ifdef HAVE_BACKING_DEV_INFO
1230 /* initializing backing dev info. */
1231 inode->i_mapping->backing_dev_info =
1232 &s2lsi(inode->i_sb)->lsi_bdi;
1234 inode->i_op = &ll_dir_inode_operations;
1235 inode->i_fop = &ll_dir_operations;
1236 lli->lli_fid = *fid;
1239 LASSERT(lsm != NULL);
1240 /* master object FID */
1241 lli->lli_pfid = body->mbo_fid1;
1242 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1243 lli, PFID(fid), PFID(&lli->lli_pfid));
1244 unlock_new_inode(inode);
1250 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1253 struct lmv_stripe_md *lsm = md->lmv;
1256 LASSERT(lsm != NULL);
1257 /* XXX sigh, this lsm_root initialization should be in
1258 * LMV layer, but it needs ll_iget right now, so we
1259 * put this here right now. */
1260 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1261 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1262 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1263 /* Unfortunately ll_iget will call ll_update_inode,
1264 * where the initialization of slave inode is slightly
1265 * different, so it reset lsm_md to NULL to avoid
1266 * initializing lsm for slave inode. */
1267 /* For migrating inode, master stripe and master object will
1268 * be same, so we only need assign this inode */
1269 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1270 lsm->lsm_md_oinfo[i].lmo_root = inode;
1272 lsm->lsm_md_oinfo[i].lmo_root =
1273 ll_iget_anon_dir(inode->i_sb, fid, md);
1275 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1276 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1278 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1286 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1287 const struct lmv_stripe_md *lsm_md2)
1289 return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1290 lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1291 lsm_md1->lsm_md_master_mdt_index ==
1292 lsm_md2->lsm_md_master_mdt_index &&
1293 lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1294 lsm_md1->lsm_md_layout_version ==
1295 lsm_md2->lsm_md_layout_version &&
1296 strcmp(lsm_md1->lsm_md_pool_name,
1297 lsm_md2->lsm_md_pool_name) == 0;
1300 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1302 struct ll_inode_info *lli = ll_i2info(inode);
1303 struct lmv_stripe_md *lsm = md->lmv;
1307 LASSERT(S_ISDIR(inode->i_mode));
1308 CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1309 PFID(ll_inode2fid(inode)));
1311 /* no striped information from request. */
1313 if (lli->lli_lsm_md == NULL) {
1315 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1316 LMV_HASH_FLAG_MIGRATION) {
1317 /* migration is done, the temporay MIGRATE layout has
1319 CDEBUG(D_INODE, DFID" finish migration.\n",
1320 PFID(ll_inode2fid(inode)));
1321 lmv_free_memmd(lli->lli_lsm_md);
1322 lli->lli_lsm_md = NULL;
1325 /* The lustre_md from req does not include stripeEA,
1326 * see ll_md_setattr */
1331 /* set the directory layout */
1332 if (lli->lli_lsm_md == NULL) {
1333 struct cl_attr *attr;
1335 rc = ll_init_lsm_md(inode, md);
1339 /* set md->lmv to NULL, so the following free lustre_md
1340 * will not free this lsm */
1342 lli->lli_lsm_md = lsm;
1344 OBD_ALLOC_PTR(attr);
1348 /* validate the lsm */
1349 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1350 ll_md_blocking_ast);
1356 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1357 md->body->mbo_nlink = attr->cat_nlink;
1358 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1359 md->body->mbo_size = attr->cat_size;
1360 if (md->body->mbo_valid & OBD_MD_FLATIME)
1361 md->body->mbo_atime = attr->cat_atime;
1362 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1363 md->body->mbo_ctime = attr->cat_ctime;
1364 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1365 md->body->mbo_mtime = attr->cat_mtime;
1369 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1370 lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1374 /* Compare the old and new stripe information */
1375 if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1376 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1379 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1380 "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1381 "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1382 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1383 inode, lsm, old_lsm,
1384 lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1385 lsm->lsm_md_stripe_count,
1386 old_lsm->lsm_md_stripe_count,
1387 lsm->lsm_md_master_mdt_index,
1388 old_lsm->lsm_md_master_mdt_index,
1389 lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1390 lsm->lsm_md_layout_version,
1391 old_lsm->lsm_md_layout_version,
1392 lsm->lsm_md_pool_name,
1393 old_lsm->lsm_md_pool_name);
1395 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1396 CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1397 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1398 PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1401 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1402 CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1403 ll_get_fsname(inode->i_sb, NULL, 0), idx,
1404 PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1413 void ll_clear_inode(struct inode *inode)
1415 struct ll_inode_info *lli = ll_i2info(inode);
1416 struct ll_sb_info *sbi = ll_i2sbi(inode);
1419 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1420 PFID(ll_inode2fid(inode)), inode);
1422 if (S_ISDIR(inode->i_mode)) {
1423 /* these should have been cleared in ll_file_release */
1424 LASSERT(lli->lli_opendir_key == NULL);
1425 LASSERT(lli->lli_sai == NULL);
1426 LASSERT(lli->lli_opendir_pid == 0);
1429 md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1431 LASSERT(!lli->lli_open_fd_write_count);
1432 LASSERT(!lli->lli_open_fd_read_count);
1433 LASSERT(!lli->lli_open_fd_exec_count);
1435 if (lli->lli_mds_write_och)
1436 ll_md_real_close(inode, FMODE_WRITE);
1437 if (lli->lli_mds_exec_och)
1438 ll_md_real_close(inode, FMODE_EXEC);
1439 if (lli->lli_mds_read_och)
1440 ll_md_real_close(inode, FMODE_READ);
1442 if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1443 OBD_FREE(lli->lli_symlink_name,
1444 strlen(lli->lli_symlink_name) + 1);
1445 lli->lli_symlink_name = NULL;
1448 ll_xattr_cache_destroy(inode);
1450 #ifdef CONFIG_FS_POSIX_ACL
1451 if (lli->lli_posix_acl) {
1452 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1453 posix_acl_release(lli->lli_posix_acl);
1454 lli->lli_posix_acl = NULL;
1457 lli->lli_inode_magic = LLI_INODE_DEAD;
1459 if (S_ISDIR(inode->i_mode))
1460 ll_dir_clear_lsm_md(inode);
1461 else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1462 LASSERT(list_empty(&lli->lli_agl_list));
1465 * XXX This has to be done before lsm is freed below, because
1466 * cl_object still uses inode lsm.
1468 cl_inode_fini(inode);
1473 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1475 struct lustre_md md;
1476 struct inode *inode = dentry->d_inode;
1477 struct ll_sb_info *sbi = ll_i2sbi(inode);
1478 struct ptlrpc_request *request = NULL;
1482 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1483 LUSTRE_OPC_ANY, NULL);
1484 if (IS_ERR(op_data))
1485 RETURN(PTR_ERR(op_data));
1487 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1489 ptlrpc_req_finished(request);
1490 if (rc == -ENOENT) {
1492 /* Unlinked special device node? Or just a race?
1493 * Pretend we done everything. */
1494 if (!S_ISREG(inode->i_mode) &&
1495 !S_ISDIR(inode->i_mode)) {
1496 ia_valid = op_data->op_attr.ia_valid;
1497 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1498 rc = simple_setattr(dentry, &op_data->op_attr);
1499 op_data->op_attr.ia_valid = ia_valid;
1501 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1502 CERROR("md_setattr fails: rc = %d\n", rc);
1507 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1508 sbi->ll_md_exp, &md);
1510 ptlrpc_req_finished(request);
1514 ia_valid = op_data->op_attr.ia_valid;
1515 /* inode size will be in ll_setattr_ost, can't do it now since dirty
1516 * cache is not cleared yet. */
1517 op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1518 if (S_ISREG(inode->i_mode))
1520 rc = simple_setattr(dentry, &op_data->op_attr);
1521 if (S_ISREG(inode->i_mode))
1522 inode_unlock(inode);
1523 op_data->op_attr.ia_valid = ia_valid;
1525 rc = ll_update_inode(inode, &md);
1526 ptlrpc_req_finished(request);
1531 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1532 * object(s) determine the file size and mtime. Otherwise, the MDS will
1533 * keep these values until such a time that objects are allocated for it.
1534 * We do the MDS operations first, as it is checking permissions for us.
1535 * We don't to the MDS RPC if there is nothing that we want to store there,
1536 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1537 * going to do an RPC anyways.
1539 * If we are doing a truncate, we will send the mtime and ctime updates
1540 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1541 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1544 * In case of HSMimport, we only set attr on MDS.
1546 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1548 struct inode *inode = dentry->d_inode;
1549 struct ll_inode_info *lli = ll_i2info(inode);
1550 struct md_op_data *op_data = NULL;
1554 CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1555 "valid %x, hsm_import %d\n",
1556 ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1557 inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1560 if (attr->ia_valid & ATTR_SIZE) {
1561 /* Check new size against VFS/VM file size limit and rlimit */
1562 rc = inode_newsize_ok(inode, attr->ia_size);
1566 /* The maximum Lustre file size is variable, based on the
1567 * OST maximum object size and number of stripes. This
1568 * needs another check in addition to the VFS check above. */
1569 if (attr->ia_size > ll_file_maxbytes(inode)) {
1570 CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1571 PFID(&lli->lli_fid), attr->ia_size,
1572 ll_file_maxbytes(inode));
1576 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1579 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1580 if (attr->ia_valid & TIMES_SET_FLAGS) {
1581 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1582 !cfs_capable(CFS_CAP_FOWNER))
1586 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1587 if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1588 (attr->ia_valid & ATTR_CTIME)) {
1589 attr->ia_ctime = CURRENT_TIME;
1590 attr->ia_valid |= ATTR_CTIME_SET;
1592 if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1593 (attr->ia_valid & ATTR_ATIME)) {
1594 attr->ia_atime = CURRENT_TIME;
1595 attr->ia_valid |= ATTR_ATIME_SET;
1597 if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1598 (attr->ia_valid & ATTR_MTIME)) {
1599 attr->ia_mtime = CURRENT_TIME;
1600 attr->ia_valid |= ATTR_MTIME_SET;
1603 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1604 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1605 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1606 cfs_time_current_sec());
1608 if (S_ISREG(inode->i_mode)) {
1609 if (attr->ia_valid & ATTR_SIZE)
1610 inode_dio_write_done(inode);
1611 inode_unlock(inode);
1614 /* We always do an MDS RPC, even if we're only changing the size;
1615 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1617 OBD_ALLOC_PTR(op_data);
1618 if (op_data == NULL)
1619 GOTO(out, rc = -ENOMEM);
1621 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1622 /* If we are changing file size, file content is
1623 * modified, flag it. */
1624 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1625 op_data->op_bias |= MDS_DATA_MODIFIED;
1626 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1629 op_data->op_attr = *attr;
1631 rc = ll_md_setattr(dentry, op_data);
1635 if (!S_ISREG(inode->i_mode) || hsm_import)
1638 if (attr->ia_valid & (ATTR_SIZE |
1639 ATTR_ATIME | ATTR_ATIME_SET |
1640 ATTR_MTIME | ATTR_MTIME_SET |
1641 ATTR_CTIME | ATTR_CTIME_SET)) {
1642 /* For truncate and utimes sending attributes to OSTs, setting
1643 * mtime/atime to the past will be performed under PW [0:EOF]
1644 * extent lock (new_size:EOF for truncate). It may seem
1645 * excessive to send mtime/atime updates to OSTs when not
1646 * setting times to past, but it is necessary due to possible
1647 * time de-synchronization between MDT inode and OST objects */
1648 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1651 /* If the file was restored, it needs to set dirty flag.
1653 * We've already sent MDS_DATA_MODIFIED flag in
1654 * ll_md_setattr() for truncate. However, the MDT refuses to
1655 * set the HS_DIRTY flag on released files, so we have to set
1656 * it again if the file has been restored. Please check how
1657 * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1659 * Please notice that if the file is not released, the previous
1660 * MDS_DATA_MODIFIED has taken effect and usually
1661 * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1662 * This way we can save an RPC for common open + trunc
1664 if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1665 struct hsm_state_set hss = {
1666 .hss_valid = HSS_SETMASK,
1667 .hss_setmask = HS_DIRTY,
1671 rc2 = ll_hsm_state_set(inode, &hss);
1672 /* truncate and write can happen at the same time, so that
1673 * the file can be set modified even though the file is not
1674 * restored from released state, and ll_hsm_state_set() is
1675 * not applicable for the file, and rc2 < 0 is normal in this
1678 CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1679 PFID(ll_inode2fid(inode)), rc2);
1684 if (op_data != NULL)
1685 ll_finish_md_op_data(op_data);
1687 if (S_ISREG(inode->i_mode)) {
1689 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1690 inode_dio_wait(inode);
1693 ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1694 LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1699 int ll_setattr(struct dentry *de, struct iattr *attr)
1701 int mode = de->d_inode->i_mode;
1703 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1704 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1705 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1707 if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1708 (ATTR_SIZE|ATTR_MODE)) &&
1709 (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1710 (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1711 !(attr->ia_mode & S_ISGID))))
1712 attr->ia_valid |= ATTR_FORCE;
1714 if ((attr->ia_valid & ATTR_MODE) &&
1716 !(attr->ia_mode & S_ISUID) &&
1717 !(attr->ia_valid & ATTR_KILL_SUID))
1718 attr->ia_valid |= ATTR_KILL_SUID;
1720 if ((attr->ia_valid & ATTR_MODE) &&
1721 ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1722 !(attr->ia_mode & S_ISGID) &&
1723 !(attr->ia_valid & ATTR_KILL_SGID))
1724 attr->ia_valid |= ATTR_KILL_SGID;
1726 return ll_setattr_raw(de, attr, false);
1729 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1730 __u64 max_age, __u32 flags)
1732 struct ll_sb_info *sbi = ll_s2sbi(sb);
1733 struct obd_statfs obd_osfs;
1737 rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1739 CERROR("md_statfs fails: rc = %d\n", rc);
1743 osfs->os_type = sb->s_magic;
1745 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1746 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1748 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1749 flags |= OBD_STATFS_NODELAY;
1751 rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1753 CERROR("obd_statfs fails: rc = %d\n", rc);
1757 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1758 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1761 osfs->os_bsize = obd_osfs.os_bsize;
1762 osfs->os_blocks = obd_osfs.os_blocks;
1763 osfs->os_bfree = obd_osfs.os_bfree;
1764 osfs->os_bavail = obd_osfs.os_bavail;
1766 /* If we don't have as many objects free on the OST as inodes
1767 * on the MDS, we reduce the total number of inodes to
1768 * compensate, so that the "inodes in use" number is correct.
1770 if (obd_osfs.os_ffree < osfs->os_ffree) {
1771 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1773 osfs->os_ffree = obd_osfs.os_ffree;
1778 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1780 struct super_block *sb = de->d_sb;
1781 struct obd_statfs osfs;
1782 __u64 fsid = huge_encode_dev(sb->s_dev);
1785 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1786 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1788 /* Some amount of caching on the client is allowed */
1789 rc = ll_statfs_internal(sb, &osfs,
1790 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1795 statfs_unpack(sfs, &osfs);
1797 /* We need to downshift for all 32-bit kernels, because we can't
1798 * tell if the kernel is being called via sys_statfs64() or not.
1799 * Stop before overflowing f_bsize - in which case it is better
1800 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1801 if (sizeof(long) < 8) {
1802 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1805 osfs.os_blocks >>= 1;
1806 osfs.os_bfree >>= 1;
1807 osfs.os_bavail >>= 1;
1811 sfs->f_blocks = osfs.os_blocks;
1812 sfs->f_bfree = osfs.os_bfree;
1813 sfs->f_bavail = osfs.os_bavail;
1814 sfs->f_fsid.val[0] = (__u32)fsid;
1815 sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1819 void ll_inode_size_lock(struct inode *inode)
1821 struct ll_inode_info *lli;
1823 LASSERT(!S_ISDIR(inode->i_mode));
1825 lli = ll_i2info(inode);
1826 mutex_lock(&lli->lli_size_mutex);
1829 void ll_inode_size_unlock(struct inode *inode)
1831 struct ll_inode_info *lli;
1833 lli = ll_i2info(inode);
1834 mutex_unlock(&lli->lli_size_mutex);
1837 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1839 struct ll_inode_info *lli = ll_i2info(inode);
1840 struct mdt_body *body = md->body;
1841 struct ll_sb_info *sbi = ll_i2sbi(inode);
1843 if (body->mbo_valid & OBD_MD_FLEASIZE)
1844 cl_file_inode_init(inode, md);
1846 if (S_ISDIR(inode->i_mode)) {
1849 rc = ll_update_lsm_md(inode, md);
1854 #ifdef CONFIG_FS_POSIX_ACL
1855 if (body->mbo_valid & OBD_MD_FLACL) {
1856 spin_lock(&lli->lli_lock);
1857 if (lli->lli_posix_acl)
1858 posix_acl_release(lli->lli_posix_acl);
1859 lli->lli_posix_acl = md->posix_acl;
1860 spin_unlock(&lli->lli_lock);
1863 inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1864 sbi->ll_flags & LL_SBI_32BIT_API);
1865 inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1867 if (body->mbo_valid & OBD_MD_FLATIME) {
1868 if (body->mbo_atime > LTIME_S(inode->i_atime))
1869 LTIME_S(inode->i_atime) = body->mbo_atime;
1870 lli->lli_atime = body->mbo_atime;
1873 if (body->mbo_valid & OBD_MD_FLMTIME) {
1874 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1875 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1876 "to "LPU64"\n", inode->i_ino,
1877 LTIME_S(inode->i_mtime), body->mbo_mtime);
1878 LTIME_S(inode->i_mtime) = body->mbo_mtime;
1880 lli->lli_mtime = body->mbo_mtime;
1883 if (body->mbo_valid & OBD_MD_FLCTIME) {
1884 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1885 LTIME_S(inode->i_ctime) = body->mbo_ctime;
1886 lli->lli_ctime = body->mbo_ctime;
1889 if (body->mbo_valid & OBD_MD_FLMODE)
1890 inode->i_mode = (inode->i_mode & S_IFMT) |
1891 (body->mbo_mode & ~S_IFMT);
1893 if (body->mbo_valid & OBD_MD_FLTYPE)
1894 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1895 (body->mbo_mode & S_IFMT);
1897 LASSERT(inode->i_mode != 0);
1898 if (S_ISREG(inode->i_mode))
1899 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1900 LL_MAX_BLKSIZE_BITS);
1902 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1904 if (body->mbo_valid & OBD_MD_FLUID)
1905 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1906 if (body->mbo_valid & OBD_MD_FLGID)
1907 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1908 if (body->mbo_valid & OBD_MD_FLFLAGS)
1909 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1910 if (body->mbo_valid & OBD_MD_FLNLINK)
1911 set_nlink(inode, body->mbo_nlink);
1912 if (body->mbo_valid & OBD_MD_FLRDEV)
1913 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1915 if (body->mbo_valid & OBD_MD_FLID) {
1916 /* FID shouldn't be changed! */
1917 if (fid_is_sane(&lli->lli_fid)) {
1918 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1919 "Trying to change FID "DFID
1920 " to the "DFID", inode "DFID"(%p)\n",
1921 PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1922 PFID(ll_inode2fid(inode)), inode);
1924 lli->lli_fid = body->mbo_fid1;
1928 LASSERT(fid_seq(&lli->lli_fid) != 0);
1930 if (body->mbo_valid & OBD_MD_FLSIZE) {
1931 i_size_write(inode, body->mbo_size);
1933 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1934 PFID(ll_inode2fid(inode)),
1935 (unsigned long long)body->mbo_size);
1937 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1938 inode->i_blocks = body->mbo_blocks;
1941 if (body->mbo_valid & OBD_MD_TSTATE) {
1942 /* Set LLIF_FILE_RESTORING if restore ongoing and
1943 * clear it when done to ensure to start again
1944 * glimpsing updated attrs
1946 if (body->mbo_t_state & MS_RESTORE)
1947 ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1949 ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1955 int ll_read_inode2(struct inode *inode, void *opaque)
1957 struct lustre_md *md = opaque;
1958 struct ll_inode_info *lli = ll_i2info(inode);
1962 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1963 PFID(&lli->lli_fid), inode);
1965 /* Core attributes from the MDS first. This is a new inode, and
1966 * the VFS doesn't zero times in the core inode so we have to do
1967 * it ourselves. They will be overwritten by either MDS or OST
1968 * attributes - we just need to make sure they aren't newer. */
1969 LTIME_S(inode->i_mtime) = 0;
1970 LTIME_S(inode->i_atime) = 0;
1971 LTIME_S(inode->i_ctime) = 0;
1973 rc = ll_update_inode(inode, md);
1977 /* OIDEBUG(inode); */
1979 #ifdef HAVE_BACKING_DEV_INFO
1980 /* initializing backing dev info. */
1981 inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1983 if (S_ISREG(inode->i_mode)) {
1984 struct ll_sb_info *sbi = ll_i2sbi(inode);
1985 inode->i_op = &ll_file_inode_operations;
1986 inode->i_fop = sbi->ll_fop;
1987 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1989 } else if (S_ISDIR(inode->i_mode)) {
1990 inode->i_op = &ll_dir_inode_operations;
1991 inode->i_fop = &ll_dir_operations;
1993 } else if (S_ISLNK(inode->i_mode)) {
1994 inode->i_op = &ll_fast_symlink_inode_operations;
1997 inode->i_op = &ll_special_inode_operations;
1999 init_special_inode(inode, inode->i_mode,
2008 void ll_delete_inode(struct inode *inode)
2010 struct ll_inode_info *lli = ll_i2info(inode);
2013 if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2014 /* It is last chance to write out dirty pages,
2015 * otherwise we may lose data while umount */
2016 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2018 truncate_inode_pages_final(&inode->i_data);
2020 LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2021 "see https://jira.hpdd.intel.com/browse/LU-118\n",
2022 PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2024 #ifdef HAVE_SBOPS_EVICT_INODE
2025 ll_clear_inode(inode);
2032 int ll_iocontrol(struct inode *inode, struct file *file,
2033 unsigned int cmd, unsigned long arg)
2035 struct ll_sb_info *sbi = ll_i2sbi(inode);
2036 struct ptlrpc_request *req = NULL;
2041 case FSFILT_IOC_GETFLAGS: {
2042 struct mdt_body *body;
2043 struct md_op_data *op_data;
2045 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2046 0, 0, LUSTRE_OPC_ANY,
2048 if (IS_ERR(op_data))
2049 RETURN(PTR_ERR(op_data));
2051 op_data->op_valid = OBD_MD_FLFLAGS;
2052 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2053 ll_finish_md_op_data(op_data);
2055 CERROR("%s: failure inode "DFID": rc = %d\n",
2056 sbi->ll_md_exp->exp_obd->obd_name,
2057 PFID(ll_inode2fid(inode)), rc);
2061 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2063 flags = body->mbo_flags;
2065 ptlrpc_req_finished(req);
2067 RETURN(put_user(flags, (int __user *)arg));
2069 case FSFILT_IOC_SETFLAGS: {
2071 struct md_op_data *op_data;
2072 struct cl_object *obj;
2074 if (get_user(flags, (int __user *)arg))
2077 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2078 LUSTRE_OPC_ANY, NULL);
2079 if (IS_ERR(op_data))
2080 RETURN(PTR_ERR(op_data));
2082 op_data->op_attr_flags = flags;
2083 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2084 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2085 ll_finish_md_op_data(op_data);
2086 ptlrpc_req_finished(req);
2090 inode->i_flags = ll_ext_to_inode_flags(flags);
2092 obj = ll_i2info(inode)->lli_clob;
2096 OBD_ALLOC_PTR(attr);
2100 attr->ia_valid = ATTR_ATTR_FLAG;
2101 rc = cl_setattr_ost(obj, attr, flags);
2113 int ll_flush_ctx(struct inode *inode)
2115 struct ll_sb_info *sbi = ll_i2sbi(inode);
2117 CDEBUG(D_SEC, "flush context for user %d\n",
2118 from_kuid(&init_user_ns, current_uid()));
2120 obd_set_info_async(NULL, sbi->ll_md_exp,
2121 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2123 obd_set_info_async(NULL, sbi->ll_dt_exp,
2124 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2129 /* umount -f client means force down, don't save state */
2130 void ll_umount_begin(struct super_block *sb)
2132 struct ll_sb_info *sbi = ll_s2sbi(sb);
2133 struct obd_device *obd;
2134 struct obd_ioctl_data *ioc_data;
2137 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2138 sb->s_count, atomic_read(&sb->s_active));
2140 obd = class_exp2obd(sbi->ll_md_exp);
2142 CERROR("Invalid MDC connection handle "LPX64"\n",
2143 sbi->ll_md_exp->exp_handle.h_cookie);
2149 obd = class_exp2obd(sbi->ll_dt_exp);
2151 CERROR("Invalid LOV connection handle "LPX64"\n",
2152 sbi->ll_dt_exp->exp_handle.h_cookie);
2158 OBD_ALLOC_PTR(ioc_data);
2160 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2161 sizeof *ioc_data, ioc_data, NULL);
2163 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2164 sizeof *ioc_data, ioc_data, NULL);
2166 OBD_FREE_PTR(ioc_data);
2169 /* Really, we'd like to wait until there are no requests outstanding,
2170 * and then continue. For now, we just invalidate the requests,
2171 * schedule() and sleep one second if needed, and hope.
2177 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2179 struct ll_sb_info *sbi = ll_s2sbi(sb);
2180 char *profilenm = get_profile_name(sb);
2184 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2185 read_only = *flags & MS_RDONLY;
2186 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2187 sizeof(KEY_READ_ONLY),
2188 KEY_READ_ONLY, sizeof(read_only),
2191 LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2192 profilenm, read_only ?
2193 "read-only" : "read-write", err);
2198 sb->s_flags |= MS_RDONLY;
2200 sb->s_flags &= ~MS_RDONLY;
2202 if (sbi->ll_flags & LL_SBI_VERBOSE)
2203 LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2204 read_only ? "read-only" : "read-write");
2210 * Cleanup the open handle that is cached on MDT-side.
2212 * For open case, the client side open handling thread may hit error
2213 * after the MDT grant the open. Under such case, the client should
2214 * send close RPC to the MDT as cleanup; otherwise, the open handle
2215 * on the MDT will be leaked there until the client umount or evicted.
2217 * In further, if someone unlinked the file, because the open handle
2218 * holds the reference on such file/object, then it will block the
2219 * subsequent threads that want to locate such object via FID.
2221 * \param[in] sb super block for this file-system
2222 * \param[in] open_req pointer to the original open request
2224 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2226 struct mdt_body *body;
2227 struct md_op_data *op_data;
2228 struct ptlrpc_request *close_req = NULL;
2229 struct obd_export *exp = ll_s2sbi(sb)->ll_md_exp;
2232 body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2233 OBD_ALLOC_PTR(op_data);
2234 if (op_data == NULL) {
2235 CWARN("%s: cannot allocate op_data to release open handle for "
2237 ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2242 op_data->op_fid1 = body->mbo_fid1;
2243 op_data->op_handle = body->mbo_handle;
2244 op_data->op_mod_time = cfs_time_current_sec();
2245 md_close(exp, op_data, NULL, &close_req);
2246 ptlrpc_req_finished(close_req);
2247 ll_finish_md_op_data(op_data);
2252 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2253 struct super_block *sb, struct lookup_intent *it)
2255 struct ll_sb_info *sbi = NULL;
2256 struct lustre_md md = { NULL };
2260 LASSERT(*inode || sb);
2261 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2262 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2263 sbi->ll_md_exp, &md);
2268 rc = ll_update_inode(*inode, &md);
2272 LASSERT(sb != NULL);
2275 * At this point server returns to client's same fid as client
2276 * generated for creating. So using ->fid1 is okay here.
2278 if (!fid_is_sane(&md.body->mbo_fid1)) {
2279 CERROR("%s: Fid is insane "DFID"\n",
2280 ll_get_fsname(sb, NULL, 0),
2281 PFID(&md.body->mbo_fid1));
2282 GOTO(out, rc = -EINVAL);
2285 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2286 sbi->ll_flags & LL_SBI_32BIT_API),
2288 if (IS_ERR(*inode)) {
2289 #ifdef CONFIG_FS_POSIX_ACL
2291 posix_acl_release(md.posix_acl);
2292 md.posix_acl = NULL;
2295 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2297 CERROR("new_inode -fatal: rc %d\n", rc);
2302 /* Handling piggyback layout lock.
2303 * Layout lock can be piggybacked by getattr and open request.
2304 * The lsm can be applied to inode only if it comes with a layout lock
2305 * otherwise correct layout may be overwritten, for example:
2306 * 1. proc1: mdt returns a lsm but not granting layout
2307 * 2. layout was changed by another client
2308 * 3. proc2: refresh layout and layout lock granted
2309 * 4. proc1: to apply a stale layout */
2310 if (it != NULL && it->it_lock_mode != 0) {
2311 struct lustre_handle lockh;
2312 struct ldlm_lock *lock;
2314 lockh.cookie = it->it_lock_handle;
2315 lock = ldlm_handle2lock(&lockh);
2316 LASSERT(lock != NULL);
2317 if (ldlm_has_layout(lock)) {
2318 struct cl_object_conf conf;
2320 memset(&conf, 0, sizeof(conf));
2321 conf.coc_opc = OBJECT_CONF_SET;
2322 conf.coc_inode = *inode;
2323 conf.coc_lock = lock;
2324 conf.u.coc_layout = md.layout;
2325 (void)ll_layout_conf(*inode, &conf);
2327 LDLM_LOCK_PUT(lock);
2333 md_free_lustre_md(sbi->ll_md_exp, &md);
2336 if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2337 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2342 int ll_obd_statfs(struct inode *inode, void __user *arg)
2344 struct ll_sb_info *sbi = NULL;
2345 struct obd_export *exp;
2347 struct obd_ioctl_data *data = NULL;
2351 if (!inode || !(sbi = ll_i2sbi(inode)))
2352 GOTO(out_statfs, rc = -EINVAL);
2354 rc = obd_ioctl_getdata(&buf, &len, arg);
2356 GOTO(out_statfs, rc);
2359 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2360 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2361 GOTO(out_statfs, rc = -EINVAL);
2363 if (data->ioc_inllen1 != sizeof(__u32) ||
2364 data->ioc_inllen2 != sizeof(__u32) ||
2365 data->ioc_plen1 != sizeof(struct obd_statfs) ||
2366 data->ioc_plen2 != sizeof(struct obd_uuid))
2367 GOTO(out_statfs, rc = -EINVAL);
2369 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2370 if (type & LL_STATFS_LMV)
2371 exp = sbi->ll_md_exp;
2372 else if (type & LL_STATFS_LOV)
2373 exp = sbi->ll_dt_exp;
2375 GOTO(out_statfs, rc = -ENODEV);
2377 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2379 GOTO(out_statfs, rc);
2382 obd_ioctl_freedata(buf, len);
2386 int ll_process_config(struct lustre_cfg *lcfg)
2388 struct super_block *sb;
2393 /* The instance name contains the sb: lustre-client-aacfe000 */
2394 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2395 if (!ptr || !*(++ptr))
2397 if (sscanf(ptr, "%lx", &x) != 1)
2399 sb = (struct super_block *)x;
2400 /* This better be a real Lustre superblock! */
2401 LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2403 /* Note we have not called client_common_fill_super yet, so
2404 proc fns must be able to handle that! */
2405 rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2412 /* this function prepares md_op_data hint for passing it down to MD stack. */
2413 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2414 struct inode *i1, struct inode *i2,
2415 const char *name, size_t namelen,
2416 __u32 mode, __u32 opc, void *data)
2418 LASSERT(i1 != NULL);
2421 /* Do not reuse namelen for something else. */
2423 return ERR_PTR(-EINVAL);
2425 if (namelen > ll_i2sbi(i1)->ll_namelen)
2426 return ERR_PTR(-ENAMETOOLONG);
2428 if (!lu_name_is_valid_2(name, namelen))
2429 return ERR_PTR(-EINVAL);
2432 if (op_data == NULL)
2433 OBD_ALLOC_PTR(op_data);
2435 if (op_data == NULL)
2436 return ERR_PTR(-ENOMEM);
2438 ll_i2gids(op_data->op_suppgids, i1, i2);
2439 op_data->op_fid1 = *ll_inode2fid(i1);
2440 op_data->op_default_stripe_offset = -1;
2441 if (S_ISDIR(i1->i_mode)) {
2442 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2443 if (opc == LUSTRE_OPC_MKDIR)
2444 op_data->op_default_stripe_offset =
2445 ll_i2info(i1)->lli_def_stripe_offset;
2449 op_data->op_fid2 = *ll_inode2fid(i2);
2450 if (S_ISDIR(i2->i_mode))
2451 op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2453 fid_zero(&op_data->op_fid2);
2456 if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2457 op_data->op_cli_flags |= CLI_HASH64;
2459 if (ll_need_32bit_api(ll_i2sbi(i1)))
2460 op_data->op_cli_flags |= CLI_API32;
2462 op_data->op_name = name;
2463 op_data->op_namelen = namelen;
2464 op_data->op_mode = mode;
2465 op_data->op_mod_time = cfs_time_current_sec();
2466 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2467 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2468 op_data->op_cap = cfs_curproc_cap_pack();
2469 if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2470 filename_is_volatile(name, namelen, &op_data->op_mds)) {
2471 op_data->op_bias |= MDS_CREATE_VOLATILE;
2473 op_data->op_mds = 0;
2475 op_data->op_data = data;
2480 void ll_finish_md_op_data(struct md_op_data *op_data)
2482 security_release_secctx(op_data->op_file_secctx,
2483 op_data->op_file_secctx_size);
2484 OBD_FREE_PTR(op_data);
2487 #ifdef HAVE_SUPEROPS_USE_DENTRY
2488 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2490 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2493 struct ll_sb_info *sbi;
2495 #ifdef HAVE_SUPEROPS_USE_DENTRY
2496 LASSERT((seq != NULL) && (dentry != NULL));
2497 sbi = ll_s2sbi(dentry->d_sb);
2499 LASSERT((seq != NULL) && (vfs != NULL));
2500 sbi = ll_s2sbi(vfs->mnt_sb);
2503 if (sbi->ll_flags & LL_SBI_NOLCK)
2504 seq_puts(seq, ",nolock");
2506 if (sbi->ll_flags & LL_SBI_FLOCK)
2507 seq_puts(seq, ",flock");
2509 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2510 seq_puts(seq, ",localflock");
2512 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2513 seq_puts(seq, ",user_xattr");
2515 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2516 seq_puts(seq, ",lazystatfs");
2518 if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2519 seq_puts(seq, ",user_fid2path");
2521 if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2522 seq_puts(seq, ",always_ping");
2528 * Get obd name by cmd, and copy out to user space
2530 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2532 struct ll_sb_info *sbi = ll_i2sbi(inode);
2533 struct obd_device *obd;
2536 if (cmd == OBD_IOC_GETDTNAME)
2537 obd = class_exp2obd(sbi->ll_dt_exp);
2538 else if (cmd == OBD_IOC_GETMDNAME)
2539 obd = class_exp2obd(sbi->ll_md_exp);
2546 if (copy_to_user((void __user *)arg, obd->obd_name,
2547 strlen(obd->obd_name) + 1))
2554 * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2555 * fsname will be returned in this buffer; otherwise, a static buffer will be
2556 * used to store the fsname and returned to caller.
2558 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2560 static char fsname_static[MTI_NAME_MAXLEN];
2561 struct lustre_sb_info *lsi = s2lsi(sb);
2566 /* this means the caller wants to use static buffer
2567 * and it doesn't care about race. Usually this is
2568 * in error reporting path */
2569 buf = fsname_static;
2570 buflen = sizeof(fsname_static);
2573 len = strlen(lsi->lsi_lmd->lmd_profile);
2574 ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2575 if (ptr && (strcmp(ptr, "-client") == 0))
2578 if (unlikely(len >= buflen))
2580 strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2586 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2593 p.mnt = current->fs->root.mnt;
2595 path = d_path(&p, buf, bufsize);
2600 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2602 char *buf, *path = NULL;
2603 struct dentry *dentry = NULL;
2604 struct inode *inode = page->mapping->host;
2606 /* this can be called inside spin lock so use GFP_ATOMIC. */
2607 buf = (char *)__get_free_page(GFP_ATOMIC);
2609 dentry = d_find_alias(page->mapping->host);
2611 path = ll_d_path(dentry, buf, PAGE_SIZE);
2615 "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2616 "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2617 s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2618 PFID(ll_inode2fid(inode)),
2619 (path && !IS_ERR(path)) ? path : "", ioret);
2625 free_page((unsigned long)buf);
2628 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2629 struct lov_user_md **kbuf)
2631 struct lov_user_md lum;
2635 if (copy_from_user(&lum, md, sizeof(lum)))
2638 lum_size = ll_lov_user_md_size(&lum);
2642 OBD_ALLOC(*kbuf, lum_size);
2646 if (copy_from_user(*kbuf, md, lum_size) != 0) {
2647 OBD_FREE(*kbuf, lum_size);
2655 * Compute llite root squash state after a change of root squash
2656 * configuration setting or add/remove of a lnet nid
2658 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2660 struct root_squash_info *squash = &sbi->ll_squash;
2663 lnet_process_id_t id;
2665 /* Update norootsquash flag */
2666 down_write(&squash->rsi_sem);
2667 if (list_empty(&squash->rsi_nosquash_nids))
2668 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2670 /* Do not apply root squash as soon as one of our NIDs is
2671 * in the nosquash_nids list */
2674 while (LNetGetId(i++, &id) != -ENOENT) {
2675 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2677 if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2683 sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2685 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2687 up_write(&squash->rsi_sem);
2691 * Parse linkea content to extract information about a given hardlink
2693 * \param[in] ldata - Initialized linkea data
2694 * \param[in] linkno - Link identifier
2695 * \param[out] parent_fid - The entry's parent FID
2696 * \param[out] ln - Entry name destination buffer
2698 * \retval 0 on success
2699 * \retval Appropriate negative error code on failure
2701 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2702 struct lu_fid *parent_fid, struct lu_name *ln)
2708 rc = linkea_init(ldata);
2712 if (linkno >= ldata->ld_leh->leh_reccount)
2713 /* beyond last link */
2716 linkea_first_entry(ldata);
2717 for (idx = 0; ldata->ld_lee != NULL; idx++) {
2718 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2723 linkea_next_entry(ldata);
2733 * Get parent FID and name of an identified link. Operation is performed for
2734 * a given link number, letting the caller iterate over linkno to list one or
2735 * all links of an entry.
2737 * \param[in] file - File descriptor against which to perform the operation
2738 * \param[in,out] arg - User-filled structure containing the linkno to operate
2739 * on and the available size. It is eventually filled with
2740 * the requested information or left untouched on error
2742 * \retval - 0 on success
2743 * \retval - Appropriate negative error code on failure
2745 int ll_getparent(struct file *file, struct getparent __user *arg)
2747 struct dentry *dentry = file_dentry(file);
2748 struct inode *inode = file_inode(file);
2749 struct linkea_data *ldata;
2750 struct lu_buf buf = LU_BUF_NULL;
2752 struct lu_fid parent_fid;
2759 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2760 !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2763 if (get_user(name_size, &arg->gp_name_size))
2766 if (get_user(linkno, &arg->gp_linkno))
2769 if (name_size > PATH_MAX)
2772 OBD_ALLOC(ldata, sizeof(*ldata));
2776 rc = linkea_data_new(ldata, &buf);
2778 GOTO(ldata_free, rc);
2780 rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2784 rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2788 if (ln.ln_namelen >= name_size)
2789 GOTO(lb_free, rc = -EOVERFLOW);
2791 if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2792 GOTO(lb_free, rc = -EFAULT);
2794 if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2795 GOTO(lb_free, rc = -EFAULT);
2797 if (put_user('\0', arg->gp_name + ln.ln_namelen))
2798 GOTO(lb_free, rc = -EFAULT);
2803 OBD_FREE(ldata, sizeof(*ldata));