1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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/types.h>
45 #include <linux/random.h>
46 #include <linux/version.h>
49 #include <lustre_lite.h>
50 #include <lustre_ha.h>
51 #include <lustre_dlm.h>
52 #include <lprocfs_status.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
55 #include <lustre_log.h>
56 #include <cl_object.h>
57 #include <obd_cksum.h>
58 #include "llite_internal.h"
60 cfs_mem_cache_t *ll_file_data_slab;
62 CFS_LIST_HEAD(ll_super_blocks);
63 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
65 extern struct address_space_operations ll_aops;
66 extern struct address_space_operations ll_dir_aops;
69 #define log2(n) cfs_ffz(~(n))
72 static struct ll_sb_info *ll_init_sbi(void)
74 struct ll_sb_info *sbi = NULL;
81 OBD_ALLOC(sbi, sizeof(*sbi));
85 cfs_spin_lock_init(&sbi->ll_lock);
86 cfs_init_mutex(&sbi->ll_lco.lco_lock);
87 cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
88 cfs_spin_lock_init(&sbi->ll_process_lock);
89 sbi->ll_rw_stats_on = 0;
92 pages = si.totalram - si.totalhigh;
93 if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
94 #ifdef HAVE_BGL_SUPPORT
95 sbi->ll_async_page_max = pages / 4;
97 sbi->ll_async_page_max = pages / 2;
100 sbi->ll_async_page_max = (pages / 4) * 3;
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;
108 CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
109 CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
111 ll_generate_random_uuid(uuid);
112 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
113 CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
115 cfs_spin_lock(&ll_sb_lock);
116 cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
117 cfs_spin_unlock(&ll_sb_lock);
119 #ifdef ENABLE_LLITE_CHECKSUM
120 sbi->ll_flags |= LL_SBI_CHECKSUM;
123 #ifdef HAVE_LRU_RESIZE_SUPPORT
124 sbi->ll_flags |= LL_SBI_LRU_RESIZE;
127 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
128 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
130 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
134 /* metadata statahead is enabled by default */
135 sbi->ll_sa_max = LL_SA_RPC_DEF;
136 atomic_set(&sbi->ll_sa_total, 0);
137 atomic_set(&sbi->ll_sa_wrong, 0);
142 void ll_free_sbi(struct super_block *sb)
144 struct ll_sb_info *sbi = ll_s2sbi(sb);
148 cfs_spin_lock(&ll_sb_lock);
149 cfs_list_del(&sbi->ll_list);
150 cfs_spin_unlock(&ll_sb_lock);
151 OBD_FREE(sbi, sizeof(*sbi));
156 static struct dentry_operations ll_d_root_ops = {
157 .d_compare = ll_dcompare,
158 .d_revalidate = ll_revalidate_nd,
161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
163 struct inode *root = 0;
164 struct ll_sb_info *sbi = ll_s2sbi(sb);
165 struct obd_device *obd;
166 struct obd_capa *oc = NULL;
167 struct obd_statfs osfs;
168 struct ptlrpc_request *request = NULL;
169 struct obd_connect_data *data = NULL;
170 struct obd_uuid *uuid;
171 struct md_op_data *op_data;
172 struct lustre_md lmd;
174 int size, err, checksum;
177 obd = class_name2obd(md);
179 CERROR("MD %s: not setup or attached\n", md);
187 if (proc_lustre_fs_root) {
188 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
191 CERROR("could not register mount in /proc/fs/lustre\n");
194 /* indicate the features supported by this client */
195 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
196 OBD_CONNECT_JOIN | OBD_CONNECT_ATTRFID |
197 OBD_CONNECT_VERSION | OBD_CONNECT_MDS_CAPA |
198 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
199 OBD_CONNECT_FID | OBD_CONNECT_AT |
200 OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
201 OBD_CONNECT_VBR | OBD_CONNECT_FULL20;
203 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
204 data->ocd_connect_flags |= OBD_CONNECT_SOM;
206 #ifdef HAVE_LRU_RESIZE_SUPPORT
207 if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
208 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
210 #ifdef CONFIG_FS_POSIX_ACL
211 data->ocd_connect_flags |= OBD_CONNECT_ACL;
213 data->ocd_ibits_known = MDS_INODELOCK_FULL;
214 data->ocd_version = LUSTRE_VERSION_CODE;
216 if (sb->s_flags & MS_RDONLY)
217 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
218 if (sbi->ll_flags & LL_SBI_USER_XATTR)
219 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
221 #ifdef HAVE_MS_FLOCK_LOCK
222 /* force vfs to use lustre handler for flock() calls - bug 10743 */
223 sb->s_flags |= MS_FLOCK_LOCK;
226 if (sbi->ll_flags & LL_SBI_FLOCK)
227 sbi->ll_fop = &ll_file_operations_flock;
228 else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
229 sbi->ll_fop = &ll_file_operations;
231 sbi->ll_fop = &ll_file_operations_noflock;
234 data->ocd_connect_flags |= OBD_CONNECT_REAL;
235 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
236 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
238 err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
240 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
241 "recovery, of which this client is not a "
242 "part. Please wait for recovery to complete,"
243 " abort, or time out.\n", md);
246 CERROR("cannot connect to %s: rc = %d\n", md, err);
250 err = obd_fid_init(sbi->ll_md_exp);
252 CERROR("Can't init metadata layer FID infrastructure, "
257 err = obd_statfs(obd, &osfs, cfs_time_current_64() - CFS_HZ, 0);
259 GOTO(out_md_fid, err);
261 size = sizeof(*data);
262 err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
263 KEY_CONN_DATA, &size, data, NULL);
265 CERROR("Get connect data failed: %d \n", err);
269 LASSERT(osfs.os_bsize);
270 sb->s_blocksize = osfs.os_bsize;
271 sb->s_blocksize_bits = log2(osfs.os_bsize);
272 sb->s_magic = LL_SUPER_MAGIC;
274 /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
275 * retval = in_file->f_op->sendfile(...);
277 * retval = -EOVERFLOW;
279 * it will check if *ppos is greater than max. However, max equals to
280 * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
281 * has been defined as a signed long long integer in linux kernel. */
282 #if BITS_PER_LONG == 64
283 sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
285 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
287 sbi->ll_namelen = osfs.os_namelen;
288 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
290 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
291 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
292 LCONSOLE_INFO("Disabling user_xattr feature because "
293 "it is not supported on the server\n");
294 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
297 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
299 sb->s_flags |= MS_POSIXACL;
301 sbi->ll_flags |= LL_SBI_ACL;
303 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
305 sb->s_flags &= ~MS_POSIXACL;
307 sbi->ll_flags &= ~LL_SBI_ACL;
310 if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
311 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
312 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
313 LCONSOLE_INFO("client is set as remote by default.\n");
316 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
317 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
318 LCONSOLE_INFO("client claims to be remote, but server "
319 "rejected, forced to be local.\n");
323 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
324 LCONSOLE_INFO("client enabled MDS capability!\n");
325 sbi->ll_flags |= LL_SBI_MDS_CAPA;
328 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
329 LCONSOLE_INFO("client enabled OSS capability!\n");
330 sbi->ll_flags |= LL_SBI_OSS_CAPA;
333 obd = class_name2obd(dt);
335 CERROR("DT %s: not setup or attached\n", dt);
336 GOTO(out_md_fid, err = -ENODEV);
339 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
340 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
341 OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
342 OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
343 OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
344 OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
347 if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
348 data->ocd_connect_flags |= OBD_CONNECT_SOM;
350 if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
351 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
352 * disabled by default, because it can still be enabled on the
353 * fly via /proc. As a consequence, we still need to come to an
354 * agreement on the supported algorithms at connect time */
355 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
357 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
358 data->ocd_cksum_types = OBD_CKSUM_ADLER;
360 /* send the list of supported checksum types */
361 data->ocd_cksum_types = OBD_CKSUM_ALL;
364 #ifdef HAVE_LRU_RESIZE_SUPPORT
365 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
367 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
368 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
370 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
371 "ocd_grant: %d\n", data->ocd_connect_flags,
372 data->ocd_version, data->ocd_grant);
374 obd->obd_upcall.onu_owner = &sbi->ll_lco;
375 obd->obd_upcall.onu_upcall = cl_ocd_update;
376 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
378 err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
380 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
381 "recovery, of which this client is not a "
382 "part. Please wait for recovery to "
383 "complete, abort, or time out.\n", dt);
384 GOTO(out_md_fid, err);
386 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
387 GOTO(out_md_fid, err);
390 err = obd_fid_init(sbi->ll_dt_exp);
392 CERROR("Can't init data layer FID infrastructure, "
397 cfs_mutex_down(&sbi->ll_lco.lco_lock);
398 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
399 sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
400 sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
401 cfs_mutex_up(&sbi->ll_lco.lco_lock);
403 fid_zero(&sbi->ll_root_fid);
404 err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
406 CERROR("cannot mds_connect: rc = %d\n", err);
407 GOTO(out_lock_cn_cb, err);
409 if (!fid_is_sane(&sbi->ll_root_fid)) {
410 CERROR("Invalid root fid during mount\n");
411 GOTO(out_lock_cn_cb, err = -EINVAL);
413 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
415 sb->s_op = &lustre_super_operations;
416 sb->s_export_op = &lustre_export_operations;
419 * XXX: move this to after cbd setup? */
420 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
421 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
422 valid |= OBD_MD_FLRMTPERM;
423 else if (sbi->ll_flags & LL_SBI_ACL)
424 valid |= OBD_MD_FLACL;
426 OBD_ALLOC_PTR(op_data);
428 GOTO(out_lock_cn_cb, err = -ENOMEM);
430 op_data->op_fid1 = sbi->ll_root_fid;
431 op_data->op_mode = 0;
432 op_data->op_capa1 = oc;
433 op_data->op_valid = valid;
435 err = md_getattr(sbi->ll_md_exp, op_data, &request);
438 OBD_FREE_PTR(op_data);
440 CERROR("md_getattr failed for root: rc = %d\n", err);
441 GOTO(out_lock_cn_cb, err);
443 memset(&lmd, 0, sizeof(lmd));
444 err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
445 sbi->ll_md_exp, &lmd);
447 CERROR("failed to understand root inode md: rc = %d\n", err);
448 ptlrpc_req_finished (request);
449 GOTO(out_lock_cn_cb, err);
452 LASSERT(fid_is_sane(&sbi->ll_root_fid));
453 root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
454 md_free_lustre_md(sbi->ll_md_exp, &lmd);
455 ptlrpc_req_finished(request);
457 if (root == NULL || IS_ERR(root)) {
459 obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
460 #ifdef CONFIG_FS_POSIX_ACL
462 posix_acl_release(lmd.posix_acl);
463 lmd.posix_acl = NULL;
466 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
468 CERROR("lustre_lite: bad iget4 for root\n");
472 err = ll_close_thread_start(&sbi->ll_lcq);
474 CERROR("cannot start close thread: rc %d\n", err);
478 #ifdef CONFIG_FS_POSIX_ACL
479 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
480 rct_init(&sbi->ll_rct);
481 et_init(&sbi->ll_et);
485 checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
486 err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
487 KEY_CHECKSUM, sizeof(checksum), &checksum,
491 sb->s_root = d_alloc_root(root);
493 OBD_FREE(data, sizeof(*data));
495 sb->s_root->d_op = &ll_d_root_ops;
497 sbi->ll_sdev_orig = sb->s_dev;
499 /* We set sb->s_dev equal on all lustre clients in order to support
500 * NFS export clustering. NFSD requires that the FSID be the same
502 /* s_dev is also used in lt_compare() to compare two fs, but that is
503 * only a node-local comparison. */
504 uuid = obd_get_uuid(sbi->ll_md_exp);
506 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
513 obd_fid_fini(sbi->ll_dt_exp);
515 obd_disconnect(sbi->ll_dt_exp);
516 sbi->ll_dt_exp = NULL;
518 obd_fid_fini(sbi->ll_md_exp);
520 obd_disconnect(sbi->ll_md_exp);
521 sbi->ll_md_exp = NULL;
525 lprocfs_unregister_mountpoint(sbi);
529 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
533 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
535 rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
536 KEY_MAX_EASIZE, &size, lmmsize, NULL);
538 CERROR("Get max mdsize error rc %d \n", rc);
543 void ll_dump_inode(struct inode *inode)
545 struct list_head *tmp;
546 int dentry_count = 0;
548 LASSERT(inode != NULL);
550 list_for_each(tmp, &inode->i_dentry)
553 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
554 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
555 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
558 void lustre_dump_dentry(struct dentry *dentry, int recur)
560 struct list_head *tmp;
563 LASSERT(dentry != NULL);
565 list_for_each(tmp, &dentry->d_subdirs)
568 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
569 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
570 dentry->d_name.len, dentry->d_name.name,
571 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
572 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
573 dentry->d_flags, dentry->d_fsdata, subdirs);
574 if (dentry->d_inode != NULL)
575 ll_dump_inode(dentry->d_inode);
580 list_for_each(tmp, &dentry->d_subdirs) {
581 struct dentry *d = list_entry(tmp, struct dentry, d_child);
582 lustre_dump_dentry(d, recur - 1);
586 void client_common_put_super(struct super_block *sb)
588 struct ll_sb_info *sbi = ll_s2sbi(sb);
591 #ifdef CONFIG_FS_POSIX_ACL
592 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
593 et_fini(&sbi->ll_et);
594 rct_fini(&sbi->ll_rct);
598 obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
600 ll_close_thread_shutdown(sbi->ll_lcq);
604 cfs_list_del(&sbi->ll_conn_chain);
606 obd_fid_fini(sbi->ll_dt_exp);
607 obd_disconnect(sbi->ll_dt_exp);
608 sbi->ll_dt_exp = NULL;
610 lprocfs_unregister_mountpoint(sbi);
612 obd_fid_fini(sbi->ll_md_exp);
613 obd_disconnect(sbi->ll_md_exp);
614 sbi->ll_md_exp = NULL;
619 void ll_kill_super(struct super_block *sb)
621 struct ll_sb_info *sbi;
626 if (!(sb->s_flags & MS_ACTIVE))
630 /* we need restore s_dev from changed for clustred NFS before put_super
631 * because new kernels have cached s_dev and change sb->s_dev in
632 * put_super not affected real removing devices */
634 sb->s_dev = sbi->ll_sdev_orig;
638 char *ll_read_opt(const char *opt, char *data)
644 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
645 if (strncmp(opt, data, strlen(opt)))
647 if ((value = strchr(data, '=')) == NULL)
651 OBD_ALLOC(retval, strlen(value) + 1);
653 CERROR("out of memory!\n");
657 memcpy(retval, value, strlen(value)+1);
658 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
662 static inline int ll_set_opt(const char *opt, char *data, int fl)
664 if (strncmp(opt, data, strlen(opt)) != 0)
670 /* non-client-specific mount options are parsed in lmd_parse */
671 static int ll_options(char *options, int *flags)
674 char *s1 = options, *s2;
680 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
683 CDEBUG(D_SUPER, "next opt=%s\n", s1);
684 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
689 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
694 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
699 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
704 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
709 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
714 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
716 /* Ignore deprecated mount option. The client will
717 * always try to mount with ACL support, whether this
718 * is used depends on whether server supports it. */
721 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
725 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
731 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
736 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
741 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
746 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
751 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
756 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
761 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
767 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
773 s2 = strchr(s1, ',');
781 void ll_lli_init(struct ll_inode_info *lli)
783 lli->lli_inode_magic = LLI_INODE_MAGIC;
784 cfs_sema_init(&lli->lli_size_sem, 1);
785 cfs_sema_init(&lli->lli_write_sem, 1);
786 cfs_sema_init(&lli->lli_trunc_sem, 1);
788 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
789 cfs_spin_lock_init(&lli->lli_lock);
790 CFS_INIT_LIST_HEAD(&lli->lli_close_list);
791 lli->lli_inode_magic = LLI_INODE_MAGIC;
792 cfs_sema_init(&lli->lli_och_sem, 1);
793 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
794 lli->lli_mds_exec_och = NULL;
795 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
796 lli->lli_open_fd_exec_count = 0;
797 CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
798 lli->lli_remote_perms = NULL;
799 lli->lli_rmtperm_utime = 0;
800 cfs_sema_init(&lli->lli_rmtperm_sem, 1);
801 CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
802 cfs_spin_lock_init(&lli->lli_sa_lock);
803 CFS_INIT_LIST_HEAD(&lli->lli_sa_dentry);
806 int ll_fill_super(struct super_block *sb)
808 struct lustre_profile *lprof;
809 struct lustre_sb_info *lsi = s2lsi(sb);
810 struct ll_sb_info *sbi;
811 char *dt = NULL, *md = NULL;
812 char *profilenm = get_profile_name(sb);
813 struct config_llog_instance cfg = {0, };
814 char ll_instance[sizeof(sb) * 2 + 1];
818 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
822 /* client additional sb info */
823 lsi->lsi_llsbi = sbi = ll_init_sbi();
825 cfs_module_put(THIS_MODULE);
829 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
833 /* Generate a string unique to this super, in case some joker tries
834 to mount the same fs at two mount points.
835 Use the address of the super itself.*/
836 sprintf(ll_instance, "%p", sb);
837 cfg.cfg_instance = ll_instance;
838 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
840 /* set up client obds */
841 err = lustre_process_log(sb, profilenm, &cfg);
843 CERROR("Unable to process log: %d\n", err);
847 /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
848 lprof = class_get_profile(profilenm);
850 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
851 " read from the MGS. Does that filesystem "
852 "exist?\n", profilenm);
853 GOTO(out_free, err = -EINVAL);
855 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
856 lprof->lp_md, lprof->lp_dt);
858 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
859 strlen(ll_instance) + 2);
861 GOTO(out_free, err = -ENOMEM);
862 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
864 OBD_ALLOC(md, strlen(lprof->lp_md) +
865 strlen(ll_instance) + 2);
867 GOTO(out_free, err = -ENOMEM);
868 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
870 /* connections, registrations, sb setup */
871 err = client_common_fill_super(sb, md, dt);
875 OBD_FREE(md, strlen(md) + 1);
877 OBD_FREE(dt, strlen(dt) + 1);
881 LCONSOLE_WARN("Client %s has started\n", profilenm);
884 } /* ll_fill_super */
887 void lu_context_keys_dump(void);
889 void ll_put_super(struct super_block *sb)
891 struct config_llog_instance cfg;
892 char ll_instance[sizeof(sb) * 2 + 1];
893 struct obd_device *obd;
894 struct lustre_sb_info *lsi = s2lsi(sb);
895 struct ll_sb_info *sbi = ll_s2sbi(sb);
896 char *profilenm = get_profile_name(sb);
900 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
902 ll_print_capa_stat(sbi);
904 sprintf(ll_instance, "%p", sb);
905 cfg.cfg_instance = ll_instance;
906 lustre_end_log(sb, NULL, &cfg);
908 if (sbi->ll_md_exp) {
909 obd = class_exp2obd(sbi->ll_md_exp);
911 force = obd->obd_force;
914 /* We need to set force before the lov_disconnect in
915 lustre_common_put_super, since l_d cleans up osc's as well. */
918 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
920 obd->obd_force = force;
925 /* Only if client_common_fill_super succeeded */
926 client_common_put_super(sb);
930 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
931 class_manual_cleanup(obd);
935 class_del_profile(profilenm);
938 lsi->lsi_llsbi = NULL;
940 lustre_common_put_super(sb);
942 cl_env_cache_purge(~0);
944 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
946 cfs_module_put(THIS_MODULE);
949 } /* client_put_super */
951 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
953 struct inode *inode = NULL;
954 /* NOTE: we depend on atomic igrab() -bzzz */
955 lock_res_and_lock(lock);
956 if (lock->l_ast_data) {
957 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
958 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
959 inode = igrab(lock->l_ast_data);
961 inode = lock->l_ast_data;
962 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
964 lock, __FILE__, __func__, __LINE__,
965 "l_ast_data %p is bogus: magic %08x",
966 lock->l_ast_data, lli->lli_inode_magic);
970 unlock_res_and_lock(lock);
974 static int null_if_equal(struct ldlm_lock *lock, void *data)
976 if (data == lock->l_ast_data) {
977 lock->l_ast_data = NULL;
979 if (lock->l_req_mode != lock->l_granted_mode)
980 LDLM_ERROR(lock,"clearing inode with ungranted lock");
983 return LDLM_ITER_CONTINUE;
986 void ll_clear_inode(struct inode *inode)
988 struct ll_inode_info *lli = ll_i2info(inode);
989 struct ll_sb_info *sbi = ll_i2sbi(inode);
992 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
993 inode->i_generation, inode);
995 if (S_ISDIR(inode->i_mode)) {
996 /* these should have been cleared in ll_file_release */
997 LASSERT(lli->lli_sai == NULL);
998 LASSERT(lli->lli_opendir_key == NULL);
999 LASSERT(lli->lli_opendir_pid == 0);
1002 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1003 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1004 null_if_equal, inode);
1006 LASSERT(!lli->lli_open_fd_write_count);
1007 LASSERT(!lli->lli_open_fd_read_count);
1008 LASSERT(!lli->lli_open_fd_exec_count);
1010 if (lli->lli_mds_write_och)
1011 ll_md_real_close(inode, FMODE_WRITE);
1012 if (lli->lli_mds_exec_och)
1013 ll_md_real_close(inode, FMODE_EXEC);
1014 if (lli->lli_mds_read_och)
1015 ll_md_real_close(inode, FMODE_READ);
1017 if (lli->lli_symlink_name) {
1018 OBD_FREE(lli->lli_symlink_name,
1019 strlen(lli->lli_symlink_name) + 1);
1020 lli->lli_symlink_name = NULL;
1023 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1024 LASSERT(lli->lli_posix_acl == NULL);
1025 if (lli->lli_remote_perms) {
1026 free_rmtperm_hash(lli->lli_remote_perms);
1027 lli->lli_remote_perms = NULL;
1030 #ifdef CONFIG_FS_POSIX_ACL
1031 else if (lli->lli_posix_acl) {
1032 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1033 LASSERT(lli->lli_remote_perms == NULL);
1034 posix_acl_release(lli->lli_posix_acl);
1035 lli->lli_posix_acl = NULL;
1038 lli->lli_inode_magic = LLI_INODE_DEAD;
1040 ll_clear_inode_capas(inode);
1042 * XXX This has to be done before lsm is freed below, because
1043 * cl_object still uses inode lsm.
1045 cl_inode_fini(inode);
1048 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1049 lli->lli_smd = NULL;
1056 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1057 struct md_open_data **mod)
1059 struct lustre_md md;
1060 struct ll_sb_info *sbi = ll_i2sbi(inode);
1061 struct ptlrpc_request *request = NULL;
1065 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1066 LUSTRE_OPC_ANY, NULL);
1067 if (IS_ERR(op_data))
1068 RETURN(PTR_ERR(op_data));
1070 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1073 ptlrpc_req_finished(request);
1074 if (rc == -ENOENT) {
1076 /* Unlinked special device node? Or just a race?
1077 * Pretend we done everything. */
1078 if (!S_ISREG(inode->i_mode) &&
1079 !S_ISDIR(inode->i_mode))
1080 rc = inode_setattr(inode, &op_data->op_attr);
1081 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1082 CERROR("md_setattr fails: rc = %d\n", rc);
1087 rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1088 sbi->ll_md_exp, &md);
1090 ptlrpc_req_finished(request);
1094 /* We call inode_setattr to adjust timestamps.
1095 * If there is at least some data in file, we cleared ATTR_SIZE
1096 * above to avoid invoking vmtruncate, otherwise it is important
1097 * to call vmtruncate in inode_setattr to update inode->i_size
1099 rc = inode_setattr(inode, &op_data->op_attr);
1101 /* Extract epoch data if obtained. */
1102 op_data->op_handle = md.body->handle;
1103 op_data->op_ioepoch = md.body->ioepoch;
1105 ll_update_inode(inode, &md);
1106 ptlrpc_req_finished(request);
1111 /* Close IO epoch and send Size-on-MDS attribute update. */
1112 static int ll_setattr_done_writing(struct inode *inode,
1113 struct md_op_data *op_data,
1114 struct md_open_data *mod)
1116 struct ll_inode_info *lli = ll_i2info(inode);
1120 LASSERT(op_data != NULL);
1121 if (!S_ISREG(inode->i_mode))
1124 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1125 op_data->op_ioepoch, PFID(&lli->lli_fid));
1127 op_data->op_flags = MF_EPOCH_CLOSE;
1128 ll_done_writing_attr(inode, op_data);
1129 ll_pack_inode2opdata(inode, op_data, NULL);
1131 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1132 if (rc == -EAGAIN) {
1133 /* MDS has instructed us to obtain Size-on-MDS attribute
1134 * from OSTs and send setattr to back to MDS. */
1135 rc = ll_som_update(inode, op_data);
1137 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1143 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1145 struct obd_capa *capa;
1148 if (attr->ia_valid & ATTR_SIZE)
1149 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1151 capa = ll_mdscapa_get(inode);
1153 rc = cl_setattr_ost(inode, attr, capa);
1155 if (attr->ia_valid & ATTR_SIZE)
1156 ll_truncate_free_capa(capa);
1163 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1164 * object(s) determine the file size and mtime. Otherwise, the MDS will
1165 * keep these values until such a time that objects are allocated for it.
1166 * We do the MDS operations first, as it is checking permissions for us.
1167 * We don't to the MDS RPC if there is nothing that we want to store there,
1168 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1169 * going to do an RPC anyways.
1171 * If we are doing a truncate, we will send the mtime and ctime updates
1172 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1173 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1176 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1178 struct ll_inode_info *lli = ll_i2info(inode);
1179 struct lov_stripe_md *lsm = lli->lli_smd;
1180 struct md_op_data *op_data = NULL;
1181 struct md_open_data *mod = NULL;
1182 int ia_valid = attr->ia_valid;
1183 int rc = 0, rc1 = 0;
1186 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1188 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1190 if (ia_valid & ATTR_SIZE) {
1191 if (attr->ia_size > ll_file_maxbytes(inode)) {
1192 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1193 attr->ia_size, ll_file_maxbytes(inode));
1197 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1200 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1201 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1202 if (cfs_curproc_fsuid() != inode->i_uid &&
1203 !cfs_capable(CFS_CAP_FOWNER))
1207 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1208 if (attr->ia_valid & ATTR_CTIME) {
1209 attr->ia_ctime = CFS_CURRENT_TIME;
1210 attr->ia_valid |= ATTR_CTIME_SET;
1212 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1213 attr->ia_atime = CFS_CURRENT_TIME;
1214 attr->ia_valid |= ATTR_ATIME_SET;
1216 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1217 attr->ia_mtime = CFS_CURRENT_TIME;
1218 attr->ia_valid |= ATTR_MTIME_SET;
1221 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1222 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1223 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1224 cfs_time_current_sec());
1226 /* NB: ATTR_SIZE will only be set after this point if the size
1227 * resides on the MDS, ie, this file has no objects. */
1229 attr->ia_valid &= ~ATTR_SIZE;
1231 /* We always do an MDS RPC, even if we're only changing the size;
1232 * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1234 OBD_ALLOC_PTR(op_data);
1235 if (op_data == NULL)
1238 UNLOCK_INODE_MUTEX(inode);
1239 if (ia_valid & ATTR_SIZE)
1240 UP_WRITE_I_ALLOC_SEM(inode);
1241 cfs_down(&lli->lli_trunc_sem);
1242 LOCK_INODE_MUTEX(inode);
1243 if (ia_valid & ATTR_SIZE)
1244 DOWN_WRITE_I_ALLOC_SEM(inode);
1246 memcpy(&op_data->op_attr, attr, sizeof(*attr));
1248 /* Open epoch for truncate. */
1249 if (exp_connect_som(ll_i2mdexp(inode)) &&
1250 (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1251 op_data->op_flags = MF_EPOCH_OPEN;
1253 rc = ll_md_setattr(inode, op_data, &mod);
1257 ll_ioepoch_open(lli, op_data->op_ioepoch);
1258 if (!lsm || !S_ISREG(inode->i_mode)) {
1259 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1263 if (ia_valid & ATTR_SIZE)
1264 attr->ia_valid |= ATTR_SIZE;
1265 if ((ia_valid & ATTR_SIZE) |
1266 ((ia_valid | ATTR_ATIME | ATTR_ATIME_SET) &&
1267 LTIME_S(attr->ia_atime) < LTIME_S(attr->ia_ctime)) ||
1268 ((ia_valid | ATTR_MTIME | ATTR_MTIME_SET) &&
1269 LTIME_S(attr->ia_mtime) < LTIME_S(attr->ia_ctime)))
1270 /* perform truncate and setting mtime/atime to past under PW
1271 * 0:EOF extent lock (new_size:EOF for truncate) */
1272 rc = ll_setattr_ost(inode, attr);
1276 if (op_data->op_ioepoch)
1277 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1278 ll_finish_md_op_data(op_data);
1280 cfs_up(&lli->lli_trunc_sem);
1281 return rc ? rc : rc1;
1284 int ll_setattr(struct dentry *de, struct iattr *attr)
1286 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1287 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1288 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1290 if ((de->d_inode->i_mode & S_ISUID) &&
1291 !(attr->ia_mode & S_ISUID) &&
1292 !(attr->ia_valid & ATTR_KILL_SUID))
1293 attr->ia_valid |= ATTR_KILL_SUID;
1295 if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1296 !(attr->ia_mode & S_ISGID) &&
1297 !(attr->ia_valid & ATTR_KILL_SGID))
1298 attr->ia_valid |= ATTR_KILL_SGID;
1300 return ll_setattr_raw(de->d_inode, attr);
1303 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1304 __u64 max_age, __u32 flags)
1306 struct ll_sb_info *sbi = ll_s2sbi(sb);
1307 struct obd_statfs obd_osfs;
1311 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1313 CERROR("md_statfs fails: rc = %d\n", rc);
1317 osfs->os_type = sb->s_magic;
1319 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1320 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1322 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1323 flags |= OBD_STATFS_NODELAY;
1325 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1326 &obd_osfs, max_age, flags);
1328 CERROR("obd_statfs fails: rc = %d\n", rc);
1332 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1333 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1336 osfs->os_bsize = obd_osfs.os_bsize;
1337 osfs->os_blocks = obd_osfs.os_blocks;
1338 osfs->os_bfree = obd_osfs.os_bfree;
1339 osfs->os_bavail = obd_osfs.os_bavail;
1341 /* If we don't have as many objects free on the OST as inodes
1342 * on the MDS, we reduce the total number of inodes to
1343 * compensate, so that the "inodes in use" number is correct.
1345 if (obd_osfs.os_ffree < osfs->os_ffree) {
1346 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1348 osfs->os_ffree = obd_osfs.os_ffree;
1353 #ifndef HAVE_STATFS_DENTRY_PARAM
1354 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1357 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1359 struct super_block *sb = de->d_sb;
1361 struct obd_statfs osfs;
1364 CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1365 ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1367 /* For now we will always get up-to-date statfs values, but in the
1368 * future we may allow some amount of caching on the client (e.g.
1369 * from QOS or lprocfs updates). */
1370 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1374 statfs_unpack(sfs, &osfs);
1376 /* We need to downshift for all 32-bit kernels, because we can't
1377 * tell if the kernel is being called via sys_statfs64() or not.
1378 * Stop before overflowing f_bsize - in which case it is better
1379 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1380 if (sizeof(long) < 8) {
1381 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1384 osfs.os_blocks >>= 1;
1385 osfs.os_bfree >>= 1;
1386 osfs.os_bavail >>= 1;
1390 sfs->f_blocks = osfs.os_blocks;
1391 sfs->f_bfree = osfs.os_bfree;
1392 sfs->f_bavail = osfs.os_bavail;
1397 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1399 struct ll_inode_info *lli;
1400 struct lov_stripe_md *lsm;
1402 lli = ll_i2info(inode);
1403 LASSERT(lli->lli_size_sem_owner != current);
1404 cfs_down(&lli->lli_size_sem);
1405 LASSERT(lli->lli_size_sem_owner == NULL);
1406 lli->lli_size_sem_owner = current;
1408 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1411 lov_stripe_lock(lsm);
1414 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1416 struct ll_inode_info *lli;
1417 struct lov_stripe_md *lsm;
1419 lli = ll_i2info(inode);
1421 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1424 lov_stripe_unlock(lsm);
1425 LASSERT(lli->lli_size_sem_owner == current);
1426 lli->lli_size_sem_owner = NULL;
1427 cfs_up(&lli->lli_size_sem);
1430 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1432 struct ll_inode_info *lli = ll_i2info(inode);
1433 struct mdt_body *body = md->body;
1434 struct lov_stripe_md *lsm = md->lsm;
1435 struct ll_sb_info *sbi = ll_i2sbi(inode);
1437 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1439 cfs_down(&lli->lli_och_sem);
1440 if (lli->lli_smd == NULL) {
1441 if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1442 lsm->lsm_magic != LOV_MAGIC_V3) {
1443 dump_lsm(D_ERROR, lsm);
1446 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1447 lsm, inode->i_ino, inode->i_generation, inode);
1448 /* cl_inode_init must go before lli_smd or a race is
1449 * possible where client thinks the file has stripes,
1450 * but lov raid0 is not setup yet and parallel e.g.
1451 * glimpse would try to use uninitialized lov */
1452 cl_inode_init(inode, md);
1454 cfs_up(&lli->lli_och_sem);
1455 lli->lli_maxbytes = lsm->lsm_maxbytes;
1456 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1457 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1459 cfs_up(&lli->lli_och_sem);
1460 LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1461 lli->lli_smd->lsm_stripe_count ==
1462 lsm->lsm_stripe_count);
1463 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1464 CERROR("lsm mismatch for inode %ld\n",
1466 CERROR("lli_smd:\n");
1467 dump_lsm(D_ERROR, lli->lli_smd);
1469 dump_lsm(D_ERROR, lsm);
1473 if (lli->lli_smd != lsm)
1474 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1477 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1478 if (body->valid & OBD_MD_FLRMTPERM)
1479 ll_update_remote_perm(inode, md->remote_perm);
1481 #ifdef CONFIG_FS_POSIX_ACL
1482 else if (body->valid & OBD_MD_FLACL) {
1483 cfs_spin_lock(&lli->lli_lock);
1484 if (lli->lli_posix_acl)
1485 posix_acl_release(lli->lli_posix_acl);
1486 lli->lli_posix_acl = md->posix_acl;
1487 cfs_spin_unlock(&lli->lli_lock);
1490 inode->i_ino = cl_fid_build_ino(&body->fid1);
1491 inode->i_generation = cl_fid_build_gen(&body->fid1);
1493 if (body->valid & OBD_MD_FLATIME) {
1494 if (body->atime > LTIME_S(inode->i_atime))
1495 LTIME_S(inode->i_atime) = body->atime;
1496 lli->lli_lvb.lvb_atime = body->atime;
1498 if (body->valid & OBD_MD_FLMTIME) {
1499 if (body->mtime > LTIME_S(inode->i_mtime)) {
1500 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1501 "to "LPU64"\n", inode->i_ino,
1502 LTIME_S(inode->i_mtime), body->mtime);
1503 LTIME_S(inode->i_mtime) = body->mtime;
1505 lli->lli_lvb.lvb_mtime = body->mtime;
1507 if (body->valid & OBD_MD_FLCTIME) {
1508 if (body->ctime > LTIME_S(inode->i_ctime))
1509 LTIME_S(inode->i_ctime) = body->ctime;
1510 lli->lli_lvb.lvb_ctime = body->ctime;
1512 if (body->valid & OBD_MD_FLMODE)
1513 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1514 if (body->valid & OBD_MD_FLTYPE)
1515 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1516 LASSERT(inode->i_mode != 0);
1517 if (S_ISREG(inode->i_mode)) {
1518 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1520 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1522 #ifdef HAVE_INODE_BLKSIZE
1523 inode->i_blksize = 1<<inode->i_blkbits;
1525 if (body->valid & OBD_MD_FLUID)
1526 inode->i_uid = body->uid;
1527 if (body->valid & OBD_MD_FLGID)
1528 inode->i_gid = body->gid;
1529 if (body->valid & OBD_MD_FLFLAGS)
1530 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1531 if (body->valid & OBD_MD_FLNLINK)
1532 inode->i_nlink = body->nlink;
1533 if (body->valid & OBD_MD_FLRDEV)
1534 inode->i_rdev = old_decode_dev(body->rdev);
1536 if (body->valid & OBD_MD_FLID) {
1537 /* FID shouldn't be changed! */
1538 if (fid_is_sane(&lli->lli_fid)) {
1539 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1540 "Trying to change FID "DFID
1541 " to the "DFID", inode %lu/%u(%p)\n",
1542 PFID(&lli->lli_fid), PFID(&body->fid1),
1543 inode->i_ino, inode->i_generation, inode);
1545 lli->lli_fid = body->fid1;
1548 LASSERT(fid_seq(&lli->lli_fid) != 0);
1550 if (body->valid & OBD_MD_FLSIZE) {
1551 if (exp_connect_som(ll_i2mdexp(inode)) &&
1552 S_ISREG(inode->i_mode) && lli->lli_smd) {
1553 struct lustre_handle lockh;
1556 /* As it is possible a blocking ast has been processed
1557 * by this time, we need to check there is an UPDATE
1558 * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1560 mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1563 if (lli->lli_flags & (LLIF_DONE_WRITING |
1564 LLIF_EPOCH_PENDING |
1566 CERROR("ino %lu flags %lu still has "
1567 "size authority! do not trust "
1568 "the size got from MDS\n",
1569 inode->i_ino, lli->lli_flags);
1571 /* Use old size assignment to avoid
1572 * deadlock bz14138 & bz14326 */
1573 inode->i_size = body->size;
1574 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1576 ldlm_lock_decref(&lockh, mode);
1579 /* Use old size assignment to avoid
1580 * deadlock bz14138 & bz14326 */
1581 inode->i_size = body->size;
1584 if (body->valid & OBD_MD_FLBLOCKS)
1585 inode->i_blocks = body->blocks;
1588 if (body->valid & OBD_MD_FLMDSCAPA) {
1589 LASSERT(md->mds_capa);
1590 ll_add_capa(inode, md->mds_capa);
1592 if (body->valid & OBD_MD_FLOSSCAPA) {
1593 LASSERT(md->oss_capa);
1594 ll_add_capa(inode, md->oss_capa);
1598 static struct backing_dev_info ll_backing_dev_info = {
1599 .ra_pages = 0, /* No readahead */
1600 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1601 .capabilities = 0, /* Does contribute to dirty memory */
1603 .memory_backed = 0, /* Does contribute to dirty memory */
1607 void ll_read_inode2(struct inode *inode, void *opaque)
1609 struct lustre_md *md = opaque;
1610 struct ll_inode_info *lli = ll_i2info(inode);
1613 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1614 inode->i_ino, inode->i_generation, inode);
1618 LASSERT(!lli->lli_smd);
1620 /* Core attributes from the MDS first. This is a new inode, and
1621 * the VFS doesn't zero times in the core inode so we have to do
1622 * it ourselves. They will be overwritten by either MDS or OST
1623 * attributes - we just need to make sure they aren't newer. */
1624 LTIME_S(inode->i_mtime) = 0;
1625 LTIME_S(inode->i_atime) = 0;
1626 LTIME_S(inode->i_ctime) = 0;
1628 ll_update_inode(inode, md);
1630 /* OIDEBUG(inode); */
1632 if (S_ISREG(inode->i_mode)) {
1633 struct ll_sb_info *sbi = ll_i2sbi(inode);
1634 inode->i_op = &ll_file_inode_operations;
1635 inode->i_fop = sbi->ll_fop;
1636 inode->i_mapping->a_ops = &ll_aops;
1638 } else if (S_ISDIR(inode->i_mode)) {
1639 inode->i_op = &ll_dir_inode_operations;
1640 inode->i_fop = &ll_dir_operations;
1641 inode->i_mapping->a_ops = &ll_dir_aops;
1643 } else if (S_ISLNK(inode->i_mode)) {
1644 inode->i_op = &ll_fast_symlink_inode_operations;
1647 inode->i_op = &ll_special_inode_operations;
1649 init_special_inode(inode, inode->i_mode,
1650 kdev_t_to_nr(inode->i_rdev));
1652 /* initializing backing dev info. */
1653 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1659 void ll_delete_inode(struct inode *inode)
1661 struct ll_sb_info *sbi = ll_i2sbi(inode);
1665 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1667 CERROR("fid_delete() failed, rc %d\n", rc);
1669 truncate_inode_pages(&inode->i_data, 0);
1675 int ll_iocontrol(struct inode *inode, struct file *file,
1676 unsigned int cmd, unsigned long arg)
1678 struct ll_sb_info *sbi = ll_i2sbi(inode);
1679 struct ptlrpc_request *req = NULL;
1684 case FSFILT_IOC_GETFLAGS: {
1685 struct mdt_body *body;
1686 struct md_op_data *op_data;
1688 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1689 0, 0, LUSTRE_OPC_ANY,
1691 if (op_data == NULL)
1694 op_data->op_valid = OBD_MD_FLFLAGS;
1695 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1696 ll_finish_md_op_data(op_data);
1698 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1702 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1704 flags = body->flags;
1706 ptlrpc_req_finished(req);
1708 RETURN(put_user(flags, (int *)arg));
1710 case FSFILT_IOC_SETFLAGS: {
1711 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1712 struct obd_info oinfo = { { { 0 } } };
1713 struct md_op_data *op_data;
1715 if (get_user(flags, (int *)arg))
1719 OBDO_ALLOC(oinfo.oi_oa);
1723 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1724 LUSTRE_OPC_ANY, NULL);
1725 if (IS_ERR(op_data))
1726 RETURN(PTR_ERR(op_data));
1728 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1729 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1730 rc = md_setattr(sbi->ll_md_exp, op_data,
1731 NULL, 0, NULL, 0, &req, NULL);
1732 ll_finish_md_op_data(op_data);
1733 ptlrpc_req_finished(req);
1735 OBDO_FREE(oinfo.oi_oa);
1740 OBDO_FREE(oinfo.oi_oa);
1741 GOTO(update_cache, rc);
1744 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1745 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1746 oinfo.oi_oa->o_flags = flags;
1747 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1749 oinfo.oi_capa = ll_mdscapa_get(inode);
1750 obdo_from_inode(oinfo.oi_oa, inode,
1751 &ll_i2info(inode)->lli_fid, 0);
1752 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1753 capa_put(oinfo.oi_capa);
1754 OBDO_FREE(oinfo.oi_oa);
1756 if (rc != -EPERM && rc != -EACCES)
1757 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1763 inode->i_flags = ll_ext_to_inode_flags(flags);
1773 int ll_flush_ctx(struct inode *inode)
1775 struct ll_sb_info *sbi = ll_i2sbi(inode);
1777 CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1779 obd_set_info_async(sbi->ll_md_exp,
1780 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1782 obd_set_info_async(sbi->ll_dt_exp,
1783 sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1788 /* umount -f client means force down, don't save state */
1789 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1790 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1792 struct super_block *sb = vfsmnt->mnt_sb;
1794 void ll_umount_begin(struct super_block *sb)
1797 struct lustre_sb_info *lsi = s2lsi(sb);
1798 struct ll_sb_info *sbi = ll_s2sbi(sb);
1799 struct obd_device *obd;
1800 struct obd_ioctl_data *ioc_data;
1803 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1804 if (!(flags & MNT_FORCE)) {
1810 /* Tell the MGC we got umount -f */
1811 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1813 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1814 sb->s_count, atomic_read(&sb->s_active));
1816 obd = class_exp2obd(sbi->ll_md_exp);
1818 CERROR("Invalid MDC connection handle "LPX64"\n",
1819 sbi->ll_md_exp->exp_handle.h_cookie);
1825 obd = class_exp2obd(sbi->ll_dt_exp);
1827 CERROR("Invalid LOV connection handle "LPX64"\n",
1828 sbi->ll_dt_exp->exp_handle.h_cookie);
1834 OBD_ALLOC_PTR(ioc_data);
1836 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1837 sizeof ioc_data, ioc_data, NULL);
1839 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1840 sizeof ioc_data, ioc_data, NULL);
1842 OBD_FREE_PTR(ioc_data);
1846 /* Really, we'd like to wait until there are no requests outstanding,
1847 * and then continue. For now, we just invalidate the requests,
1848 * schedule() and sleep one second if needed, and hope.
1851 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1852 if (atomic_read(&vfsmnt->mnt_count) > 2) {
1853 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1854 cfs_time_seconds(1));
1855 if (atomic_read(&vfsmnt->mnt_count) > 2)
1856 LCONSOLE_WARN("Mount still busy with %d refs! You "
1857 "may try to umount it a bit later\n",
1858 atomic_read(&vfsmnt->mnt_count));
1865 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1867 struct ll_sb_info *sbi = ll_s2sbi(sb);
1871 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1872 read_only = *flags & MS_RDONLY;
1873 err = obd_set_info_async(sbi->ll_md_exp,
1874 sizeof(KEY_READ_ONLY),
1875 KEY_READ_ONLY, sizeof(read_only),
1878 CERROR("Failed to change the read-only flag during "
1879 "remount: %d\n", err);
1884 sb->s_flags |= MS_RDONLY;
1886 sb->s_flags &= ~MS_RDONLY;
1891 int ll_prep_inode(struct inode **inode,
1892 struct ptlrpc_request *req,
1893 struct super_block *sb)
1895 struct ll_sb_info *sbi = NULL;
1896 struct lustre_md md;
1900 LASSERT(*inode || sb);
1901 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1902 memset(&md, 0, sizeof(struct lustre_md));
1904 rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1905 sbi->ll_md_exp, &md);
1910 ll_update_inode(*inode, &md);
1912 LASSERT(sb != NULL);
1915 * At this point server returns to client's same fid as client
1916 * generated for creating. So using ->fid1 is okay here.
1918 LASSERT(fid_is_sane(&md.body->fid1));
1920 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1921 if (*inode == NULL || IS_ERR(*inode)) {
1923 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1924 #ifdef CONFIG_FS_POSIX_ACL
1926 posix_acl_release(md.posix_acl);
1927 md.posix_acl = NULL;
1930 rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1932 CERROR("new_inode -fatal: rc %d\n", rc);
1938 md_free_lustre_md(sbi->ll_md_exp, &md);
1942 int ll_obd_statfs(struct inode *inode, void *arg)
1944 struct ll_sb_info *sbi = NULL;
1945 struct obd_export *exp;
1947 struct obd_ioctl_data *data = NULL;
1951 if (!inode || !(sbi = ll_i2sbi(inode)))
1952 GOTO(out_statfs, rc = -EINVAL);
1954 rc = obd_ioctl_getdata(&buf, &len, arg);
1956 GOTO(out_statfs, rc);
1959 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1960 !data->ioc_pbuf1 || !data->ioc_pbuf2)
1961 GOTO(out_statfs, rc = -EINVAL);
1963 if (data->ioc_inllen1 != sizeof(__u32) ||
1964 data->ioc_inllen2 != sizeof(__u32) ||
1965 data->ioc_plen1 != sizeof(struct obd_statfs) ||
1966 data->ioc_plen2 != sizeof(struct obd_uuid))
1967 GOTO(out_statfs, rc = -EINVAL);
1969 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1970 if (type == LL_STATFS_MDC)
1971 exp = sbi->ll_md_exp;
1972 else if (type == LL_STATFS_LOV)
1973 exp = sbi->ll_dt_exp;
1975 GOTO(out_statfs, rc = -ENODEV);
1977 rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
1979 GOTO(out_statfs, rc);
1982 obd_ioctl_freedata(buf, len);
1986 int ll_process_config(struct lustre_cfg *lcfg)
1990 struct lprocfs_static_vars lvars;
1994 lprocfs_llite_init_vars(&lvars);
1996 /* The instance name contains the sb: lustre-client-aacfe000 */
1997 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
1998 if (!ptr || !*(++ptr))
2000 if (sscanf(ptr, "%lx", &x) != 1)
2003 /* This better be a real Lustre superblock! */
2004 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2006 /* Note we have not called client_common_fill_super yet, so
2007 proc fns must be able to handle that! */
2008 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2015 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2016 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2017 struct inode *i1, struct inode *i2,
2018 const char *name, int namelen,
2019 int mode, __u32 opc, void *data)
2021 LASSERT(i1 != NULL);
2023 if (namelen > ll_i2sbi(i1)->ll_namelen)
2024 return ERR_PTR(-ENAMETOOLONG);
2026 if (op_data == NULL)
2027 OBD_ALLOC_PTR(op_data);
2029 if (op_data == NULL)
2030 return ERR_PTR(-ENOMEM);
2032 ll_i2gids(op_data->op_suppgids, i1, i2);
2033 op_data->op_fid1 = *ll_inode2fid(i1);
2034 op_data->op_capa1 = ll_mdscapa_get(i1);
2037 op_data->op_fid2 = *ll_inode2fid(i2);
2038 op_data->op_capa2 = ll_mdscapa_get(i2);
2040 fid_zero(&op_data->op_fid2);
2041 op_data->op_capa2 = NULL;
2044 op_data->op_name = name;
2045 op_data->op_namelen = namelen;
2046 op_data->op_mode = mode;
2047 op_data->op_mod_time = cfs_time_current_sec();
2048 op_data->op_fsuid = cfs_curproc_fsuid();
2049 op_data->op_fsgid = cfs_curproc_fsgid();
2050 op_data->op_cap = cfs_curproc_cap_pack();
2051 op_data->op_bias = MDS_CHECK_SPLIT;
2052 op_data->op_opc = opc;
2053 op_data->op_mds = 0;
2054 op_data->op_data = data;
2059 void ll_finish_md_op_data(struct md_op_data *op_data)
2061 capa_put(op_data->op_capa1);
2062 capa_put(op_data->op_capa2);
2063 OBD_FREE_PTR(op_data);
2066 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2068 struct ll_sb_info *sbi;
2070 LASSERT((seq != NULL) && (vfs != NULL));
2071 sbi = ll_s2sbi(vfs->mnt_sb);
2073 if (sbi->ll_flags & LL_SBI_NOLCK)
2074 seq_puts(seq, ",nolock");
2076 if (sbi->ll_flags & LL_SBI_FLOCK)
2077 seq_puts(seq, ",flock");
2079 if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2080 seq_puts(seq, ",localflock");
2082 if (sbi->ll_flags & LL_SBI_USER_XATTR)
2083 seq_puts(seq, ",user_xattr");
2085 if (sbi->ll_flags & LL_SBI_ACL)
2086 seq_puts(seq, ",acl");
2088 if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2089 seq_puts(seq, ",lazystatfs");