1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre Light Super operations
6 * Copyright (c) 2002-2005 Cluster File Systems, Inc.
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
31 #include <lustre_lite.h>
32 #include <lustre_ha.h>
33 #include <lustre_dlm.h>
34 #include <lprocfs_status.h>
35 #include <lustre_disk.h>
36 #include <lustre_param.h>
37 #include <lustre_log.h>
38 #include "llite_internal.h"
40 kmem_cache_t *ll_file_data_slab;
42 LIST_HEAD(ll_super_blocks);
43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
45 extern struct address_space_operations ll_aops;
46 extern struct address_space_operations ll_dir_aops;
49 #define log2(n) ffz(~(n))
53 static struct ll_sb_info *ll_init_sbi(void)
55 struct ll_sb_info *sbi = NULL;
60 OBD_ALLOC(sbi, sizeof(*sbi));
64 spin_lock_init(&sbi->ll_lock);
65 spin_lock_init(&sbi->ll_lco.lco_lock);
66 INIT_LIST_HEAD(&sbi->ll_pglist);
67 if (num_physpages >> (20 - PAGE_SHIFT) < 512)
68 sbi->ll_async_page_max = num_physpages / 2;
70 sbi->ll_async_page_max = (num_physpages / 4) * 3;
71 sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
72 SBI_DEFAULT_READAHEAD_MAX);
73 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
74 SBI_DEFAULT_READAHEAD_WHOLE_MAX;
76 INIT_LIST_HEAD(&sbi->ll_conn_chain);
77 INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
79 class_generate_random_uuid(uuid);
80 class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
81 CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
83 spin_lock(&ll_sb_lock);
84 list_add_tail(&sbi->ll_list, &ll_super_blocks);
85 spin_unlock(&ll_sb_lock);
87 INIT_LIST_HEAD(&sbi->ll_deathrow);
88 spin_lock_init(&sbi->ll_deathrow_lock);
89 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
90 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
91 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
97 void ll_free_sbi(struct super_block *sb)
99 struct ll_sb_info *sbi = ll_s2sbi(sb);
103 spin_lock(&ll_sb_lock);
104 list_del(&sbi->ll_list);
105 spin_unlock(&ll_sb_lock);
106 OBD_FREE(sbi, sizeof(*sbi));
111 static struct dentry_operations ll_d_root_ops = {
112 #ifdef LUSTRE_KERNEL_VERSION
113 .d_compare = ll_dcompare,
117 /* Initialize the default and maximum LOV EA and cookie sizes. This allows
118 * us to make MDS RPCs with large enough reply buffers to hold the
119 * maximum-sized (= maximum striped) EA and cookie without having to
120 * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
121 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
123 struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
124 __u32 valsize = sizeof(struct lov_desc);
125 int rc, easize, def_easize, cookiesize;
126 struct lov_desc desc;
130 rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
135 stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
136 lsm.lsm_stripe_count = stripes;
137 easize = obd_size_diskmd(dt_exp, &lsm);
139 lsm.lsm_stripe_count = desc.ld_default_stripe_count;
140 def_easize = obd_size_diskmd(dt_exp, &lsm);
142 cookiesize = stripes * sizeof(struct llog_cookie);
144 CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
147 rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
151 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
152 int mdt_pag, uid_t nllu, gid_t nllg)
154 struct inode *root = 0;
155 struct ll_sb_info *sbi = ll_s2sbi(sb);
156 struct obd_device *obd;
157 struct lu_fid rootfid;
158 struct obd_capa *oc = NULL;
159 struct obd_statfs osfs;
160 struct ptlrpc_request *request = NULL;
161 struct lustre_handle dt_conn = {0, };
162 struct lustre_handle md_conn = {0, };
163 struct obd_connect_data *data = NULL;
164 struct lustre_md lmd;
169 obd = class_name2obd(md);
171 CERROR("MD %s: not setup or attached\n", md);
179 if (proc_lustre_fs_root) {
180 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
183 CERROR("could not register mount in /proc/lustre");
186 /* indicate the features supported by this client */
187 data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
188 OBD_CONNECT_ACL | OBD_CONNECT_JOIN |
189 OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
190 OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA;
191 data->ocd_ibits_known = MDS_INODELOCK_FULL;
192 data->ocd_version = LUSTRE_VERSION_CODE;
194 if (sb->s_flags & MS_RDONLY)
195 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
196 if (sbi->ll_flags & LL_SBI_USER_XATTR)
197 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
199 if (sbi->ll_flags & LL_SBI_FLOCK)
200 sbi->ll_fop = &ll_file_operations_flock;
202 sbi->ll_fop = &ll_file_operations;
205 data->ocd_connect_flags |= OBD_CONNECT_REAL;
206 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
207 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
208 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
209 data->ocd_nllu = nllu;
210 data->ocd_nllg = nllg;
212 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
213 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
217 obd_set_info_async(obd->obd_self_export, 3, "pag",
220 err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
222 LCONSOLE_ERROR("An MDT (md %s) is performing recovery, of "
223 "which this client is not a part. Please wait "
224 "for recovery to complete, abort, or "
228 CERROR("cannot connect to %s: rc = %d\n", md, err);
231 sbi->ll_md_exp = class_conn2export(&md_conn);
233 err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
237 size = sizeof(*data);
238 err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA), KEY_CONN_DATA,
241 CERROR("Get connect data failed: %d \n", err);
245 LASSERT(osfs.os_bsize);
246 sb->s_blocksize = osfs.os_bsize;
247 sb->s_blocksize_bits = log2(osfs.os_bsize);
248 sb->s_magic = LL_SUPER_MAGIC;
249 sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
250 sbi->ll_namelen = osfs.os_namelen;
251 sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
253 if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
254 !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
255 LCONSOLE_INFO("Disabling user_xattr feature because "
256 "it is not supported on the server\n");
257 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
260 if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
262 sb->s_flags |= MS_POSIXACL;
264 sbi->ll_flags |= LL_SBI_ACL;
266 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
268 sb->s_flags &= ~MS_POSIXACL;
270 sbi->ll_flags &= ~LL_SBI_ACL;
273 if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
274 sbi->ll_flags |= LL_SBI_JOIN;
276 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
277 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
278 /* sometimes local client claims to be remote, but mdt
279 * will disagree when client gss not applied. */
280 LCONSOLE_INFO("client claims to be remote, but server "
281 "rejected, forced to be local.\n");
282 sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
285 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
286 /* with gss applied, remote client can not claim to be
287 * local, so mdt maybe force client to be remote. */
288 LCONSOLE_INFO("client claims to be local, but server "
289 "rejected, forced to be remote.\n");
290 sbi->ll_flags |= LL_SBI_RMT_CLIENT;
294 if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
295 LCONSOLE_INFO("client enabled MDS capability!\n");
296 sbi->ll_flags |= LL_SBI_MDS_CAPA;
299 if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
300 LCONSOLE_INFO("client enabled OSS capability!\n");
301 sbi->ll_flags |= LL_SBI_OSS_CAPA;
304 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
305 /* We set sb->s_dev equal on all lustre clients in order to support
306 * NFS export clustering. NFSD requires that the FSID be the same
308 /* s_dev is also used in lt_compare() to compare two fs, but that is
309 * only a node-local comparison. */
311 /* XXX: this will not work with LMV */
312 sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
313 strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
316 obd = class_name2obd(dt);
318 CERROR("DT %s: not setup or attached\n", dt);
319 GOTO(out_md, err = -ENODEV);
322 data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
323 OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE;
324 if (sbi->ll_flags & LL_SBI_OSS_CAPA)
325 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
327 CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
328 "ocd_grant: %d\n", data->ocd_connect_flags,
329 data->ocd_version, data->ocd_grant);
331 obd->obd_upcall.onu_owner = &sbi->ll_lco;
332 obd->obd_upcall.onu_upcall = ll_ocd_update;
333 data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << PAGE_SHIFT;
335 err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
337 LCONSOLE_ERROR("An OST (dt %s) is performing recovery, of which this"
338 " client is not a part. Please wait for recovery to "
339 "complete, abort, or time out.\n", dt);
342 CERROR("cannot connect to %s: rc = %d\n", dt, err);
346 sbi->ll_dt_exp = class_conn2export(&dt_conn);
348 spin_lock(&sbi->ll_lco.lco_lock);
349 sbi->ll_lco.lco_flags = data->ocd_connect_flags;
350 spin_unlock(&sbi->ll_lco.lco_lock);
352 ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
354 err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
355 0, NULL, NULL, NULL);
357 LCONSOLE_ERROR("There are no OST's in this filesystem. "
358 "There must be at least one active OST for "
359 "a client to start.\n");
363 if (!ll_async_page_slab) {
364 ll_async_page_slab_size =
365 size_round(sizeof(struct ll_async_page)) + err;
366 ll_async_page_slab = kmem_cache_create("ll_async_page",
367 ll_async_page_slab_size,
369 if (!ll_async_page_slab)
370 GOTO(out_dt, err = -ENOMEM);
373 err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
375 CERROR("cannot mds_connect: rc = %d\n", err);
378 CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
379 sbi->ll_root_fid = rootfid;
381 sb->s_op = &lustre_super_operations;
382 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
383 sb->s_export_op = &lustre_export_operations;
387 * XXX: move this to after cbd setup? */
388 valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
389 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
390 valid |= OBD_MD_FLRMTPERM;
391 else if (sbi->ll_flags & LL_SBI_ACL)
392 valid |= OBD_MD_FLACL;
394 err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
398 CERROR("md_getattr failed for root: rc = %d\n", err);
402 err = md_get_lustre_md(sbi->ll_md_exp, request,
403 REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
406 CERROR("failed to understand root inode md: rc = %d\n", err);
407 ptlrpc_req_finished (request);
411 LASSERT(fid_is_sane(&sbi->ll_root_fid));
412 root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
413 ptlrpc_req_finished(request);
415 if (root == NULL || is_bad_inode(root)) {
416 md_free_lustre_md(sbi->ll_dt_exp, &lmd);
417 CERROR("lustre_lite: bad iget4 for root\n");
418 GOTO(out_root, err = -EBADF);
421 err = ll_close_thread_start(&sbi->ll_lcq);
423 CERROR("cannot start close thread: rc %d\n", err);
427 /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
428 backing dev info assigned to inode mapping is used for
429 determining maximal readahead. */
430 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
431 !defined(KERNEL_HAS_AS_MAX_READAHEAD)
432 /* bug 2805 - set VM readahead to zero */
433 vm_max_readahead = vm_min_readahead = 0;
436 sb->s_root = d_alloc_root(root);
438 OBD_FREE(data, sizeof(*data));
439 sb->s_root->d_op = &ll_d_root_ops;
446 obd_disconnect(sbi->ll_dt_exp);
447 sbi->ll_dt_exp = NULL;
449 obd_disconnect(sbi->ll_md_exp);
450 sbi->ll_md_exp = NULL;
454 lprocfs_unregister_mountpoint(sbi);
458 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
462 *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
464 rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
467 CERROR("Get max mdsize error rc %d \n", rc);
472 void ll_dump_inode(struct inode *inode)
474 struct list_head *tmp;
475 int dentry_count = 0;
477 LASSERT(inode != NULL);
479 list_for_each(tmp, &inode->i_dentry)
482 CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
483 inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
484 inode->i_mode, atomic_read(&inode->i_count), dentry_count);
487 void lustre_dump_dentry(struct dentry *dentry, int recur)
489 struct list_head *tmp;
492 LASSERT(dentry != NULL);
494 list_for_each(tmp, &dentry->d_subdirs)
497 CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
498 " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
499 dentry->d_name.len, dentry->d_name.name,
500 dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
501 dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
502 dentry->d_flags, dentry->d_fsdata, subdirs);
503 if (dentry->d_inode != NULL)
504 ll_dump_inode(dentry->d_inode);
509 list_for_each(tmp, &dentry->d_subdirs) {
510 struct dentry *d = list_entry(tmp, struct dentry, d_child);
511 lustre_dump_dentry(d, recur - 1);
515 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
516 void lustre_throw_orphan_dentries(struct super_block *sb)
518 struct dentry *dentry, *next;
519 struct ll_sb_info *sbi = ll_s2sbi(sb);
521 /* Do this to get rid of orphaned dentries. That is not really trw. */
522 list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
524 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
525 "before and after shrink_dcache_parent\n",
526 dentry->d_name.len, dentry->d_name.name, dentry, next);
527 lustre_dump_dentry(dentry, 1);
528 shrink_dcache_parent(dentry);
529 lustre_dump_dentry(dentry, 1);
533 #define lustre_throw_orphan_dentries(sb)
536 static void prune_dir_dentries(struct inode *inode)
538 struct dentry *dentry, *prev = NULL;
540 /* due to lustre specific logic, a directory
541 * can have few dentries - a bug from VFS POV */
543 spin_lock(&dcache_lock);
544 if (!list_empty(&inode->i_dentry)) {
545 dentry = list_entry(inode->i_dentry.prev,
546 struct dentry, d_alias);
547 /* in order to prevent infinite loops we
548 * break if previous dentry is busy */
549 if (dentry != prev) {
552 spin_unlock(&dcache_lock);
554 /* try to kill all child dentries */
556 shrink_dcache_parent(dentry);
557 unlock_dentry(dentry);
560 /* now try to get rid of current dentry */
561 d_prune_aliases(inode);
565 spin_unlock(&dcache_lock);
568 static void prune_deathrow_one(struct ll_inode_info *lli)
570 struct inode *inode = ll_info2i(lli);
572 /* first, try to drop any dentries - they hold a ref on the inode */
573 if (S_ISDIR(inode->i_mode))
574 prune_dir_dentries(inode);
576 d_prune_aliases(inode);
579 /* if somebody still uses it, leave it */
580 LASSERT(atomic_read(&inode->i_count) > 0);
581 if (atomic_read(&inode->i_count) > 1)
584 CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
585 inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
587 /* seems nobody uses it anymore */
595 static void prune_deathrow(struct ll_sb_info *sbi, int try)
597 struct ll_inode_info *lli;
601 if (need_resched() && try)
605 if (!spin_trylock(&sbi->ll_deathrow_lock))
608 spin_lock(&sbi->ll_deathrow_lock);
613 if (!list_empty(&sbi->ll_deathrow)) {
614 lli = list_entry(sbi->ll_deathrow.next,
615 struct ll_inode_info,
617 list_del_init(&lli->lli_dead_list);
618 if (!list_empty(&sbi->ll_deathrow))
621 spin_unlock(&sbi->ll_deathrow_lock);
624 prune_deathrow_one(lli);
626 } while (empty == 0);
629 void client_common_put_super(struct super_block *sb)
631 struct ll_sb_info *sbi = ll_s2sbi(sb);
634 ll_close_thread_shutdown(sbi->ll_lcq);
636 /* destroy inodes in deathrow */
637 prune_deathrow(sbi, 0);
639 list_del(&sbi->ll_conn_chain);
640 obd_disconnect(sbi->ll_dt_exp);
641 sbi->ll_dt_exp = NULL;
643 lprocfs_unregister_mountpoint(sbi);
645 obd_disconnect(sbi->ll_md_exp);
646 sbi->ll_md_exp = NULL;
648 lustre_throw_orphan_dentries(sb);
652 char *ll_read_opt(const char *opt, char *data)
658 CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
659 if (strncmp(opt, data, strlen(opt)))
661 if ((value = strchr(data, '=')) == NULL)
665 OBD_ALLOC(retval, strlen(value) + 1);
667 CERROR("out of memory!\n");
671 memcpy(retval, value, strlen(value)+1);
672 CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
676 static inline int ll_set_opt(const char *opt, char *data, int fl)
678 if (strncmp(opt, data, strlen(opt)) != 0)
684 /* non-client-specific mount options are parsed in lmd_parse */
685 static int ll_options(char *options, int *flags)
688 char *s1 = options, *s2;
694 CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
697 CDEBUG(D_SUPER, "next opt=%s\n", s1);
698 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
703 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
708 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK);
713 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
718 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
723 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
725 /* Ignore deprecated mount option. The client will
726 * always try to mount with ACL support, whether this
727 * is used depends on whether server supports it. */
730 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
734 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
740 LCONSOLE_ERROR("Unknown option '%s', won't mount.\n", s1);
745 s2 = strchr(s1, ',');
753 void ll_lli_init(struct ll_inode_info *lli)
755 sema_init(&lli->lli_open_sem, 1);
756 sema_init(&lli->lli_size_sem, 1);
757 sema_init(&lli->lli_write_sem, 1);
759 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
760 spin_lock_init(&lli->lli_lock);
761 INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
762 INIT_LIST_HEAD(&lli->lli_close_list);
763 lli->lli_inode_magic = LLI_INODE_MAGIC;
764 sema_init(&lli->lli_och_sem, 1);
765 lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
766 lli->lli_mds_exec_och = NULL;
767 lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
768 lli->lli_open_fd_exec_count = 0;
769 INIT_LIST_HEAD(&lli->lli_dead_list);
770 lli->lli_remote_perms = NULL;
771 lli->lli_rmtperm_utime = 0;
772 sema_init(&lli->lli_rmtperm_sem, 1);
773 INIT_LIST_HEAD(&lli->lli_oss_capas);
777 #define MDCDEV "mdc_dev"
778 static int old_lustre_process_log(struct super_block *sb, char *newprofile,
779 struct config_llog_instance *cfg)
781 struct lustre_sb_info *lsi = s2lsi(sb);
782 struct obd_device *obd;
783 struct lustre_handle mdc_conn = {0, };
784 struct obd_export *exp;
785 char *ptr, *mdt, *profile;
786 char niduuid[10] = "mdtnid0";
788 struct obd_uuid mdc_uuid;
789 struct llog_ctxt *ctxt;
790 struct obd_connect_data ocd = { 0 };
792 int i, rc = 0, recov_bk = 1, failnodes = 0;
795 class_generate_random_uuid(uuid);
796 class_uuid_unparse(uuid, &mdc_uuid);
797 CDEBUG(D_HA, "generated uuid: %s\n", mdc_uuid.uuid);
799 /* Figure out the old mdt and profile name from new-style profile
800 ("lustre" from "mds/lustre-client") */
802 profile = strchr(mdt, '/');
803 if (profile == NULL) {
804 CDEBUG(D_CONFIG, "Can't find MDT name in %s\n", newprofile);
805 GOTO(out, rc = -EINVAL);
809 ptr = strrchr(profile, '-');
811 CDEBUG(D_CONFIG, "Can't find client name in %s\n", newprofile);
812 GOTO(out, rc = -EINVAL);
816 LCONSOLE_WARN("This looks like an old mount command; I will try to "
817 "contact MDT '%s' for profile '%s'\n", mdt, profile);
819 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
821 ptr = lsi->lsi_lmd->lmd_dev;
822 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
823 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid, 0,0,0);
825 /* Stop at the first failover nid */
830 CERROR("No valid MDT nids found.\n");
831 GOTO(out, rc = -EINVAL);
835 rc = do_lcfg(MDCDEV, 0, LCFG_ATTACH, LUSTRE_MDC_NAME, mdc_uuid.uuid, 0, 0);
837 GOTO(out_del_uuid, rc);
839 rc = do_lcfg(MDCDEV, 0, LCFG_SETUP, mdt, niduuid, 0, 0);
841 LCONSOLE_ERROR("I couldn't establish a connection with the MDT."
842 " Check that the MDT host NID is correct and the"
843 " networks are up.\n");
844 GOTO(out_detach, rc);
847 obd = class_name2obd(MDCDEV);
849 GOTO(out_cleanup, rc = -EINVAL);
851 /* Add any failover nids */
852 while (*ptr == ':') {
853 /* New failover node */
854 sprintf(niduuid, "mdtnid%d", failnodes);
856 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
858 rc = do_lcfg(MDCDEV, nid, LCFG_ADD_UUID, niduuid,0,0,0);
860 CERROR("Add uuid for %s failed %d\n",
861 libcfs_nid2str(nid), rc);
866 rc = do_lcfg(MDCDEV, 0, LCFG_ADD_CONN, niduuid, 0, 0,0);
868 CERROR("Add conn for %s failed %d\n",
869 libcfs_nid2str(nid), rc);
877 /* Try all connections, but only once. */
878 rc = obd_set_info_async(obd->obd_self_export,
879 strlen("init_recov_bk"), "init_recov_bk",
880 sizeof(recov_bk), &recov_bk, NULL);
882 GOTO(out_cleanup, rc);
884 /* If we don't have this then an ACL MDS will refuse the connection */
885 ocd.ocd_connect_flags = OBD_CONNECT_ACL;
887 rc = obd_connect(NULL, &mdc_conn, obd, &mdc_uuid, &ocd);
889 CERROR("cannot connect to %s: rc = %d\n", mdt, rc);
890 GOTO(out_cleanup, rc);
893 exp = class_conn2export(&mdc_conn);
895 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
897 cfg->cfg_flags |= CFG_F_COMPAT146;
900 rc = class_config_parse_llog(ctxt, profile, cfg);
903 * For debugging, it's useful to just dump the log
905 rc = class_config_dump_llog(ctxt, profile, cfg);
909 /* Set the caller's profile name to the old-style */
910 memcpy(newprofile, profile, strlen(profile) + 1);
914 LCONSOLE_ERROR("%s: The configuration '%s' could not be read "
915 "from the MDT '%s'. Make sure this client and "
916 "the MDT are running compatible versions of "
918 obd->obd_name, profile, mdt);
921 LCONSOLE_ERROR("%s: The configuration '%s' could not be read "
922 "from the MDT '%s'. This may be the result of "
923 "communication errors between the client and "
924 "the MDT, or if the MDT is not running.\n",
925 obd->obd_name, profile, mdt);
929 /* We don't so much care about errors in cleaning up the config llog
930 * connection, as we have already read the config by this point. */
934 do_lcfg(MDCDEV, 0, LCFG_CLEANUP, 0, 0, 0, 0);
937 do_lcfg(MDCDEV, 0, LCFG_DETACH, 0, 0, 0, 0);
940 /* class_add_uuid adds a nid even if the same uuid exists; we might
941 delete any copy here. So they all better match. */
942 for (i = 0; i < failnodes; i++) {
943 sprintf(niduuid, "mdtnid%d", i);
944 do_lcfg(MDCDEV, 0, LCFG_DEL_UUID, niduuid, 0, 0, 0);
946 /* class_import_put will get rid of the additional connections */
952 int ll_fill_super(struct super_block *sb)
954 struct lustre_profile *lprof;
955 struct lustre_sb_info *lsi = s2lsi(sb);
956 struct ll_sb_info *sbi;
957 char *dt = NULL, *md = NULL;
958 char *profilenm = get_profile_name(sb);
959 struct config_llog_instance cfg;
960 char ll_instance[sizeof(sb) * 2 + 1];
964 CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
968 /* client additional sb info */
969 lsi->lsi_llsbi = sbi = ll_init_sbi();
975 err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
979 /* Generate a string unique to this super, in case some joker tries
980 to mount the same fs at two mount points.
981 Use the address of the super itself.*/
982 sprintf(ll_instance, "%p", sb);
983 cfg.cfg_instance = ll_instance;
984 cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
985 cfg.cfg_last_idx = 0;
987 /* set up client obds */
988 err = lustre_process_log(sb, profilenm, &cfg);
993 oldnamelen = strlen(profilenm) + 1;
994 /* Temp storage for 1.4.6 profile name */
995 OBD_ALLOC(oldname, oldnamelen);
997 memcpy(oldname, profilenm, oldnamelen);
998 rc = old_lustre_process_log(sb, oldname, &cfg);
1000 /* That worked - update the profile name
1003 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1004 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1005 OBD_ALLOC(lsi->lsi_lmd->lmd_profile,
1006 strlen(oldname) + 1);
1007 if (!lsi->lsi_lmd->lmd_profile) {
1008 OBD_FREE(oldname, oldnamelen);
1009 GOTO(out_free, err = -ENOMEM);
1011 memcpy(lsi->lsi_lmd->lmd_profile, oldname,
1012 strlen(oldname) + 1);
1013 profilenm = get_profile_name(sb);
1015 OBD_FREE(oldname, oldnamelen);
1018 /* end COMPAT_146 */
1020 CERROR("Unable to process log: %d\n", err);
1021 GOTO(out_free, err);
1024 lprof = class_get_profile(profilenm);
1025 if (lprof == NULL) {
1026 LCONSOLE_ERROR("The client profile '%s' could not be read "
1027 "from the MGS. Does that filesystem exist?\n",
1029 GOTO(out_free, err = -EINVAL);
1031 CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1032 lprof->lp_md, lprof->lp_dt);
1034 OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1035 strlen(ll_instance) + 2);
1037 GOTO(out_free, err = -ENOMEM);
1038 sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1040 OBD_ALLOC(md, strlen(lprof->lp_md) +
1041 strlen(ll_instance) + 2);
1043 GOTO(out_free, err = -ENOMEM);
1044 sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1046 /* connections, registrations, sb setup */
1047 err = client_common_fill_super(sb, md, dt,
1048 lsi->lsi_lmd->lmd_pag,
1049 lsi->lsi_lmd->lmd_nllu,
1050 lsi->lsi_lmd->lmd_nllg);
1054 OBD_FREE(md, strlen(md) + 1);
1056 OBD_FREE(dt, strlen(dt) + 1);
1060 } /* ll_fill_super */
1063 void ll_put_super(struct super_block *sb)
1065 struct config_llog_instance cfg;
1066 char ll_instance[sizeof(sb) * 2 + 1];
1067 struct obd_device *obd;
1068 struct lustre_sb_info *lsi = s2lsi(sb);
1069 struct ll_sb_info *sbi = ll_s2sbi(sb);
1070 char *profilenm = get_profile_name(sb);
1071 int force = 1, next;
1074 CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1076 sprintf(ll_instance, "%p", sb);
1077 cfg.cfg_instance = ll_instance;
1078 lustre_end_log(sb, NULL, &cfg);
1080 if (sbi->ll_md_exp) {
1081 obd = class_exp2obd(sbi->ll_md_exp);
1083 force = obd->obd_no_recov;
1086 /* We need to set force before the lov_disconnect in
1087 lustre_common_put_super, since l_d cleans up osc's as well. */
1090 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1092 obd->obd_force = force;
1097 /* Only if client_common_fill_super succeeded */
1098 client_common_put_super(sb);
1101 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1102 class_manual_cleanup(obd);
1106 class_del_profile(profilenm);
1109 lsi->lsi_llsbi = NULL;
1111 lustre_common_put_super(sb);
1113 LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1118 } /* client_put_super */
1120 #ifdef HAVE_REGISTER_CACHE
1121 #include <linux/cache_def.h>
1122 #ifdef HAVE_CACHE_RETURN_INT
1127 ll_shrink_cache(int priority, unsigned int gfp_mask)
1129 struct ll_sb_info *sbi;
1132 list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1133 count += llap_shrink_cache(sbi, priority);
1135 #ifdef HAVE_CACHE_RETURN_INT
1140 struct cache_definition ll_cache_definition = {
1141 .name = "llap_cache",
1142 .shrink = ll_shrink_cache
1144 #endif /* HAVE_REGISTER_CACHE */
1146 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1148 struct inode *inode = NULL;
1149 /* NOTE: we depend on atomic igrab() -bzzz */
1150 lock_res_and_lock(lock);
1151 if (lock->l_ast_data) {
1152 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1153 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1154 inode = igrab(lock->l_ast_data);
1156 inode = lock->l_ast_data;
1157 ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1159 lock, __FILE__, __func__, __LINE__,
1160 "l_ast_data %p is bogus: magic %08x",
1161 lock->l_ast_data, lli->lli_inode_magic);
1165 unlock_res_and_lock(lock);
1169 static int null_if_equal(struct ldlm_lock *lock, void *data)
1171 if (data == lock->l_ast_data) {
1172 lock->l_ast_data = NULL;
1174 if (lock->l_req_mode != lock->l_granted_mode)
1175 LDLM_ERROR(lock,"clearing inode with ungranted lock");
1178 return LDLM_ITER_CONTINUE;
1181 void ll_clear_inode(struct inode *inode)
1183 struct ll_inode_info *lli = ll_i2info(inode);
1184 struct ll_sb_info *sbi = ll_i2sbi(inode);
1187 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1188 inode->i_generation, inode);
1190 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1191 md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1192 null_if_equal, inode);
1194 LASSERT(!lli->lli_open_fd_write_count);
1195 LASSERT(!lli->lli_open_fd_read_count);
1196 LASSERT(!lli->lli_open_fd_exec_count);
1198 if (lli->lli_mds_write_och)
1199 ll_md_real_close(inode, FMODE_WRITE);
1200 if (lli->lli_mds_exec_och) {
1202 CERROR("No FMODE exec, bug exec och is present for "
1203 "inode %ld\n", inode->i_ino);
1204 ll_md_real_close(inode, FMODE_EXEC);
1206 if (lli->lli_mds_read_och)
1207 ll_md_real_close(inode, FMODE_READ);
1210 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1211 null_if_equal, inode);
1213 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1214 lli->lli_smd = NULL;
1217 if (lli->lli_symlink_name) {
1218 OBD_FREE(lli->lli_symlink_name,
1219 strlen(lli->lli_symlink_name) + 1);
1220 lli->lli_symlink_name = NULL;
1223 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1224 LASSERT(lli->lli_posix_acl == NULL);
1225 if (lli->lli_remote_perms) {
1226 free_rmtperm_hash(lli->lli_remote_perms);
1227 lli->lli_remote_perms = NULL;
1230 #ifdef CONFIG_FS_POSIX_ACL
1231 else if (lli->lli_posix_acl) {
1232 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1233 LASSERT(lli->lli_remote_perms == NULL);
1234 posix_acl_release(lli->lli_posix_acl);
1235 lli->lli_posix_acl = NULL;
1238 lli->lli_inode_magic = LLI_INODE_DEAD;
1240 spin_lock(&sbi->ll_deathrow_lock);
1241 list_del_init(&lli->lli_dead_list);
1242 spin_unlock(&sbi->ll_deathrow_lock);
1244 ll_clear_inode_capas(inode);
1249 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data)
1251 struct lustre_md md;
1252 struct ll_sb_info *sbi = ll_i2sbi(inode);
1253 struct ptlrpc_request *request = NULL;
1257 op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0);
1258 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0, &request);
1260 ptlrpc_req_finished(request);
1261 if (rc == -ENOENT) {
1263 /* Unlinked special device node? Or just a race?
1264 * Pretend we done everything. */
1265 if (!S_ISREG(inode->i_mode) &&
1266 !S_ISDIR(inode->i_mode))
1267 rc = inode_setattr(inode, &op_data->attr);
1268 } else if (rc != -EPERM && rc != -EACCES) {
1269 CERROR("md_setattr fails: rc = %d\n", rc);
1274 rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1275 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1277 ptlrpc_req_finished(request);
1281 /* We call inode_setattr to adjust timestamps.
1282 * If there is at least some data in file, we cleared ATTR_SIZE
1283 * above to avoid invoking vmtruncate, otherwise it is important
1284 * to call vmtruncate in inode_setattr to update inode->i_size
1286 rc = inode_setattr(inode, &op_data->attr);
1288 /* Extract epoch data if obtained. */
1289 memcpy(&op_data->handle, &md.body->handle, sizeof(op_data->handle));
1290 op_data->ioepoch = md.body->ioepoch;
1292 ll_update_inode(inode, &md);
1293 ptlrpc_req_finished(request);
1298 /* Close IO epoch and send Size-on-MDS attribute update. */
1299 static int ll_setattr_done_writing(struct inode *inode,
1300 struct md_op_data *op_data)
1302 struct ll_inode_info *lli = ll_i2info(inode);
1306 LASSERT(op_data != NULL);
1307 if (!S_ISREG(inode->i_mode))
1310 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1311 op_data->ioepoch, PFID(&lli->lli_fid));
1313 op_data->flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1314 /* XXX: pass och here for the recovery purpose. */
1315 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
1316 if (rc == -EAGAIN) {
1317 /* MDS has instructed us to obtain Size-on-MDS attribute
1318 * from OSTs and send setattr to back to MDS. */
1319 rc = ll_sizeonmds_update(inode, &op_data->handle);
1321 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1327 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1328 * object(s) determine the file size and mtime. Otherwise, the MDS will
1329 * keep these values until such a time that objects are allocated for it.
1330 * We do the MDS operations first, as it is checking permissions for us.
1331 * We don't to the MDS RPC if there is nothing that we want to store there,
1332 * otherwise there is no harm in updating mtime/atime on the MDS if we are
1333 * going to do an RPC anyways.
1335 * If we are doing a truncate, we will send the mtime and ctime updates
1336 * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1337 * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1340 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1342 struct ll_inode_info *lli = ll_i2info(inode);
1343 struct lov_stripe_md *lsm = lli->lli_smd;
1344 struct ll_sb_info *sbi = ll_i2sbi(inode);
1345 struct md_op_data *op_data = NULL;
1346 int ia_valid = attr->ia_valid;
1347 int rc = 0, rc1 = 0;
1350 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1352 lprocfs_counter_incr(sbi->ll_stats, LPROC_LL_SETATTR);
1354 if (ia_valid & ATTR_SIZE) {
1355 if (attr->ia_size > ll_file_maxbytes(inode)) {
1356 CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1357 attr->ia_size, ll_file_maxbytes(inode));
1361 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1364 /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1365 if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1366 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1370 /* We mark all of the fields "set" so MDS/OST does not re-set them */
1371 if (attr->ia_valid & ATTR_CTIME) {
1372 attr->ia_ctime = CURRENT_TIME;
1373 attr->ia_valid |= ATTR_CTIME_SET;
1375 if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1376 attr->ia_atime = CURRENT_TIME;
1377 attr->ia_valid |= ATTR_ATIME_SET;
1379 if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1380 attr->ia_mtime = CURRENT_TIME;
1381 attr->ia_valid |= ATTR_MTIME_SET;
1383 if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1384 /* To avoid stale mtime on mds, obtain it from ost and send
1386 rc = ll_glimpse_size(inode, 0);
1390 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1391 attr->ia_mtime = inode->i_mtime;
1394 if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1395 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1396 LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1399 /* NB: ATTR_SIZE will only be set after this point if the size
1400 * resides on the MDS, ie, this file has no objects. */
1402 attr->ia_valid &= ~ATTR_SIZE;
1404 /* If only OST attributes being set on objects, don't do MDS RPC.
1405 * In that case, we need to check permissions and update the local
1406 * inode ourselves so we can call obdo_from_inode() always. */
1407 if (ia_valid & (lsm ? ~(ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
1408 OBD_ALLOC_PTR(op_data);
1409 if (op_data == NULL)
1412 memcpy(&op_data->attr, attr, sizeof(*attr));
1414 /* Open epoch for truncate. */
1415 if (ia_valid & ATTR_SIZE)
1416 op_data->flags = MF_EPOCH_OPEN;
1417 rc = ll_md_setattr(inode, op_data);
1421 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
1422 op_data->ioepoch, PFID(&lli->lli_fid));
1424 if (!lsm || !S_ISREG(inode->i_mode)) {
1425 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1429 /* The OST doesn't check permissions, but the alternative is
1430 * a gratuitous RPC to the MDS. We already rely on the client
1431 * to do read/write/truncate permission checks, so is mtime OK?
1433 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1434 /* from sys_utime() */
1435 if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1436 if (current->fsuid != inode->i_uid &&
1437 (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1440 /* from inode_change_ok() */
1441 if (current->fsuid != inode->i_uid &&
1442 !capable(CAP_FOWNER))
1447 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1448 rc = inode_setattr(inode, attr);
1451 /* We really need to get our PW lock before we change inode->i_size.
1452 * If we don't we can race with other i_size updaters on our node, like
1453 * ll_file_read. We can also race with i_size propogation to other
1454 * nodes through dirtying and writeback of final cached pages. This
1455 * last one is especially bad for racing o_append users on other
1457 if (ia_valid & ATTR_SIZE) {
1458 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1460 struct lustre_handle lockh = { 0 };
1461 int err, ast_flags = 0;
1462 /* XXX when we fix the AST intents to pass the discard-range
1463 * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1465 if (attr->ia_size == 0)
1466 ast_flags = LDLM_AST_DISCARD_DATA;
1468 UNLOCK_INODE_MUTEX(inode);
1469 UP_WRITE_I_ALLOC_SEM(inode);
1470 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1472 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1473 DOWN_WRITE_I_ALLOC_SEM(inode);
1474 LOCK_INODE_MUTEX(inode);
1476 LOCK_INODE_MUTEX(inode);
1477 DOWN_WRITE_I_ALLOC_SEM(inode);
1482 /* Only ll_inode_size_lock is taken at this level.
1483 * lov_stripe_lock() is grabbed by ll_truncate() only over
1484 * call to obd_adjust_kms(). If vmtruncate returns 0, then
1485 * ll_truncate dropped ll_inode_size_lock() */
1486 ll_inode_size_lock(inode, 0);
1487 rc = vmtruncate(inode, attr->ia_size);
1489 LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1490 ll_inode_size_unlock(inode, 0);
1493 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1495 CERROR("ll_extent_unlock failed: %d\n", err);
1499 } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1501 struct obd_info oinfo = { { { 0 } } };
1502 struct obdo *oa = obdo_alloc();
1504 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1505 inode->i_ino, LTIME_S(attr->ia_mtime));
1508 oa->o_id = lsm->lsm_object_id;
1509 oa->o_gr = lsm->lsm_object_gr;
1510 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1512 flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1513 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1514 OBD_MD_FLFID | OBD_MD_FLGENER |
1517 obdo_from_inode(oa, inode, flags);
1521 oinfo.oi_capa = ll_mdscapa_get(inode);
1523 /* XXX: this looks unnecessary now. */
1524 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1525 capa_put(oinfo.oi_capa);
1527 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1536 if (op_data->ioepoch) {
1537 rc1 = ll_setattr_done_writing(inode, op_data);
1539 ll_finish_md_op_data(op_data);
1541 return rc ? rc : rc1;
1544 int ll_setattr(struct dentry *de, struct iattr *attr)
1546 if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1547 (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1548 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1550 return ll_setattr_raw(de->d_inode, attr);
1553 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1556 struct ll_sb_info *sbi = ll_s2sbi(sb);
1557 struct obd_statfs obd_osfs;
1561 rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1563 CERROR("md_statfs fails: rc = %d\n", rc);
1567 osfs->os_type = sb->s_magic;
1569 CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1570 osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1572 rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1573 &obd_osfs, max_age);
1575 CERROR("obd_statfs fails: rc = %d\n", rc);
1579 CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1580 obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1583 osfs->os_blocks = obd_osfs.os_blocks;
1584 osfs->os_bfree = obd_osfs.os_bfree;
1585 osfs->os_bavail = obd_osfs.os_bavail;
1587 /* If we don't have as many objects free on the OST as inodes
1588 * on the MDS, we reduce the total number of inodes to
1589 * compensate, so that the "inodes in use" number is correct.
1591 if (obd_osfs.os_ffree < osfs->os_ffree) {
1592 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1594 osfs->os_ffree = obd_osfs.os_ffree;
1600 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1602 struct obd_statfs osfs;
1605 CDEBUG(D_VFSTRACE, "VFS Op:\n");
1606 lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1608 /* For now we will always get up-to-date statfs values, but in the
1609 * future we may allow some amount of caching on the client (e.g.
1610 * from QOS or lprocfs updates). */
1611 rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1615 statfs_unpack(sfs, &osfs);
1617 /* We need to downshift for all 32-bit kernels, because we can't
1618 * tell if the kernel is being called via sys_statfs64() or not.
1619 * Stop before overflowing f_bsize - in which case it is better
1620 * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1621 if (sizeof(long) < 8) {
1622 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1625 osfs.os_blocks >>= 1;
1626 osfs.os_bfree >>= 1;
1627 osfs.os_bavail >>= 1;
1631 sfs->f_blocks = osfs.os_blocks;
1632 sfs->f_bfree = osfs.os_bfree;
1633 sfs->f_bavail = osfs.os_bavail;
1638 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1640 struct ll_inode_info *lli;
1641 struct lov_stripe_md *lsm;
1643 lli = ll_i2info(inode);
1644 LASSERT(lli->lli_size_sem_owner != current);
1645 down(&lli->lli_size_sem);
1646 LASSERT(lli->lli_size_sem_owner == NULL);
1647 lli->lli_size_sem_owner = current;
1649 LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1652 lov_stripe_lock(lsm);
1655 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1657 struct ll_inode_info *lli;
1658 struct lov_stripe_md *lsm;
1660 lli = ll_i2info(inode);
1662 LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1665 lov_stripe_unlock(lsm);
1666 LASSERT(lli->lli_size_sem_owner == current);
1667 lli->lli_size_sem_owner = NULL;
1668 up(&lli->lli_size_sem);
1671 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1673 struct ll_inode_info *lli = ll_i2info(inode);
1675 dump_lsm(D_INODE, lsm);
1676 dump_lsm(D_INODE, lli->lli_smd);
1677 LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1678 "lsm must be joined lsm %p\n", lsm);
1679 obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1680 CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1681 lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1683 lli->lli_maxbytes = lsm->lsm_maxbytes;
1684 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1685 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1688 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1690 struct ll_inode_info *lli = ll_i2info(inode);
1691 struct mdt_body *body = md->body;
1692 struct lov_stripe_md *lsm = md->lsm;
1693 struct ll_sb_info *sbi = ll_i2sbi(inode);
1695 LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1697 if (lli->lli_smd == NULL) {
1698 if (lsm->lsm_magic != LOV_MAGIC &&
1699 lsm->lsm_magic != LOV_MAGIC_JOIN) {
1700 dump_lsm(D_ERROR, lsm);
1703 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1704 lsm, inode->i_ino, inode->i_generation, inode);
1705 /* ll_inode_size_lock() requires it is only called
1706 * with lli_smd != NULL or lock_lsm == 0 or we can
1707 * race between lock/unlock. bug 9547 */
1709 lli->lli_maxbytes = lsm->lsm_maxbytes;
1710 if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1711 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1713 if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1714 lli->lli_smd->lsm_stripe_count ==
1715 lsm->lsm_stripe_count) {
1716 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1717 CERROR("lsm mismatch for inode %ld\n",
1719 CERROR("lli_smd:\n");
1720 dump_lsm(D_ERROR, lli->lli_smd);
1722 dump_lsm(D_ERROR, lsm);
1726 ll_replace_lsm(inode, lsm);
1728 if (lli->lli_smd != lsm)
1729 obd_free_memmd(ll_i2dtexp(inode), &lsm);
1732 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1733 if (body->valid & OBD_MD_FLRMTPERM)
1734 ll_update_remote_perm(inode, md->remote_perm);
1736 #ifdef CONFIG_FS_POSIX_ACL
1737 else if (body->valid & OBD_MD_FLACL) {
1738 spin_lock(&lli->lli_lock);
1739 if (lli->lli_posix_acl)
1740 posix_acl_release(lli->lli_posix_acl);
1741 lli->lli_posix_acl = md->posix_acl;
1742 spin_unlock(&lli->lli_lock);
1745 if (body->valid & OBD_MD_FLATIME &&
1746 body->atime > LTIME_S(inode->i_atime))
1747 LTIME_S(inode->i_atime) = body->atime;
1749 /* mtime is always updated with ctime, but can be set in past.
1750 As write and utime(2) may happen within 1 second, and utime's
1751 mtime has a priority over write's one, so take mtime from mds
1752 for the same ctimes. */
1753 if (body->valid & OBD_MD_FLCTIME &&
1754 body->ctime >= LTIME_S(inode->i_ctime)) {
1755 LTIME_S(inode->i_ctime) = body->ctime;
1756 if (body->valid & OBD_MD_FLMTIME) {
1757 CDEBUG(D_INODE, "setting ino %lu mtime "
1758 "from %lu to "LPU64"\n", inode->i_ino,
1759 LTIME_S(inode->i_mtime), body->mtime);
1760 LTIME_S(inode->i_mtime) = body->mtime;
1763 if (body->valid & OBD_MD_FLMODE)
1764 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1765 if (body->valid & OBD_MD_FLTYPE)
1766 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1767 if (S_ISREG(inode->i_mode))
1768 inode->i_blksize = min(2UL*PTLRPC_MAX_BRW_SIZE, LL_MAX_BLKSIZE);
1770 inode->i_blksize = inode->i_sb->s_blocksize;
1771 if (body->valid & OBD_MD_FLUID)
1772 inode->i_uid = body->uid;
1773 if (body->valid & OBD_MD_FLGID)
1774 inode->i_gid = body->gid;
1775 if (body->valid & OBD_MD_FLFLAGS)
1776 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1777 if (body->valid & OBD_MD_FLNLINK)
1778 inode->i_nlink = body->nlink;
1779 if (body->valid & OBD_MD_FLRDEV)
1780 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1781 inode->i_rdev = body->rdev;
1783 inode->i_rdev = old_decode_dev(body->rdev);
1785 if (body->valid & OBD_MD_FLSIZE) {
1786 inode->i_size = body->size;
1788 if (body->valid & OBD_MD_FLBLOCKS)
1789 inode->i_blocks = body->blocks;
1791 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1794 if (body->valid & OBD_MD_FLID) {
1795 /* FID shouldn't be changed! */
1796 if (fid_is_sane(&lli->lli_fid)) {
1797 LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1798 "Trying to change FID "DFID
1799 " to the "DFID", inode %lu/%u(%p)\n",
1800 PFID(&lli->lli_fid), PFID(&body->fid1),
1801 inode->i_ino, inode->i_generation, inode);
1803 lli->lli_fid = body->fid1;
1806 LASSERT(fid_seq(&lli->lli_fid) != 0);
1808 if (body->valid & OBD_MD_FLMDSCAPA) {
1809 LASSERT(md->mds_capa);
1810 ll_add_capa(inode, md->mds_capa);
1812 if (body->valid & OBD_MD_FLOSSCAPA) {
1813 LASSERT(md->oss_capa);
1814 ll_add_capa(inode, md->oss_capa);
1818 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1819 static struct backing_dev_info ll_backing_dev_info = {
1820 .ra_pages = 0, /* No readahead */
1821 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1822 .capabilities = 0, /* Does contribute to dirty memory */
1824 .memory_backed = 0, /* Does contribute to dirty memory */
1829 void ll_read_inode2(struct inode *inode, void *opaque)
1831 struct lustre_md *md = opaque;
1832 struct ll_inode_info *lli = ll_i2info(inode);
1835 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1836 inode->i_ino, inode->i_generation, inode);
1840 LASSERT(!lli->lli_smd);
1842 /* Core attributes from the MDS first. This is a new inode, and
1843 * the VFS doesn't zero times in the core inode so we have to do
1844 * it ourselves. They will be overwritten by either MDS or OST
1845 * attributes - we just need to make sure they aren't newer. */
1846 LTIME_S(inode->i_mtime) = 0;
1847 LTIME_S(inode->i_atime) = 0;
1848 LTIME_S(inode->i_ctime) = 0;
1850 ll_update_inode(inode, md);
1852 /* OIDEBUG(inode); */
1854 if (S_ISREG(inode->i_mode)) {
1855 struct ll_sb_info *sbi = ll_i2sbi(inode);
1856 inode->i_op = &ll_file_inode_operations;
1857 inode->i_fop = sbi->ll_fop;
1858 inode->i_mapping->a_ops = &ll_aops;
1860 } else if (S_ISDIR(inode->i_mode)) {
1861 inode->i_op = &ll_dir_inode_operations;
1862 inode->i_fop = &ll_dir_operations;
1863 inode->i_mapping->a_ops = &ll_dir_aops;
1865 } else if (S_ISLNK(inode->i_mode)) {
1866 inode->i_op = &ll_fast_symlink_inode_operations;
1869 inode->i_op = &ll_special_inode_operations;
1871 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1872 init_special_inode(inode, inode->i_mode,
1873 kdev_t_to_nr(inode->i_rdev));
1875 /* initializing backing dev info. */
1876 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1878 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1884 void ll_delete_inode(struct inode *inode)
1886 struct ll_sb_info *sbi = ll_i2sbi(inode);
1890 rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1892 CERROR("fid_delete() failed, rc %d\n", rc);
1899 int ll_iocontrol(struct inode *inode, struct file *file,
1900 unsigned int cmd, unsigned long arg)
1902 struct ll_sb_info *sbi = ll_i2sbi(inode);
1903 struct ptlrpc_request *req = NULL;
1908 case EXT3_IOC_GETFLAGS: {
1909 struct mdt_body *body;
1910 struct obd_capa *oc;
1912 oc = ll_mdscapa_get(inode);
1913 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1914 OBD_MD_FLFLAGS, 0, &req);
1917 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1921 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1924 /*Now the ext3 will be packed directly back to client,
1925 *no need convert here*/
1926 flags = body->flags;
1928 ptlrpc_req_finished (req);
1930 RETURN(put_user(flags, (int *)arg));
1932 case EXT3_IOC_SETFLAGS: {
1933 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1934 struct obd_info oinfo = { { { 0 } } };
1935 struct md_op_data *op_data;
1937 if (get_user(flags, (int *)arg))
1941 oinfo.oi_oa = obdo_alloc();
1945 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0);
1946 if (op_data == NULL)
1949 ((struct ll_iattr *)&op_data->attr)->ia_attr_flags = flags;
1950 op_data->attr.ia_valid |= ATTR_ATTR_FLAG;
1951 rc = md_setattr(sbi->ll_md_exp, op_data,
1952 NULL, 0, NULL, 0, &req);
1953 ll_finish_md_op_data(op_data);
1954 ptlrpc_req_finished(req);
1955 if (rc || lsm == NULL) {
1956 obdo_free(oinfo.oi_oa);
1960 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1961 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1962 oinfo.oi_oa->o_flags = flags;
1963 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1966 obdo_from_inode(oinfo.oi_oa, inode,
1967 OBD_MD_FLFID | OBD_MD_FLGENER);
1968 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1969 obdo_free(oinfo.oi_oa);
1971 if (rc != -EPERM && rc != -EACCES)
1972 CERROR("md_setattr_async fails: rc = %d\n", rc);
1976 inode->i_flags = ll_ext_to_inode_flags(flags |
1977 MDS_BFLAG_EXT_FLAGS);
1987 int ll_flush_ctx(struct inode *inode)
1989 struct ll_sb_info *sbi = ll_i2sbi(inode);
1991 CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1993 obd_set_info_async(sbi->ll_md_exp,
1994 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1996 obd_set_info_async(sbi->ll_dt_exp,
1997 sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
2002 /* umount -f client means force down, don't save state */
2003 void ll_umount_begin(struct super_block *sb)
2005 struct lustre_sb_info *lsi = s2lsi(sb);
2006 struct ll_sb_info *sbi = ll_s2sbi(sb);
2007 struct obd_device *obd;
2008 struct obd_ioctl_data ioc_data = { 0 };
2011 /* Tell the MGC we got umount -f */
2012 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2014 CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2015 sb->s_count, atomic_read(&sb->s_active));
2017 obd = class_exp2obd(sbi->ll_md_exp);
2019 CERROR("Invalid MDC connection handle "LPX64"\n",
2020 sbi->ll_md_exp->exp_handle.h_cookie);
2024 obd->obd_no_recov = 1;
2025 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2028 obd = class_exp2obd(sbi->ll_dt_exp);
2030 CERROR("Invalid LOV connection handle "LPX64"\n",
2031 sbi->ll_dt_exp->exp_handle.h_cookie);
2036 obd->obd_no_recov = 1;
2037 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2040 /* Really, we'd like to wait until there are no requests outstanding,
2041 * and then continue. For now, we just invalidate the requests,
2042 * schedule, and hope.
2049 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2051 struct ll_sb_info *sbi = ll_s2sbi(sb);
2055 if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2056 read_only = *flags & MS_RDONLY;
2057 err = obd_set_info_async(sbi->ll_md_exp, strlen("read-only"),
2058 "read-only", sizeof(read_only),
2061 CERROR("Failed to change the read-only flag during "
2062 "remount: %d\n", err);
2067 sb->s_flags |= MS_RDONLY;
2069 sb->s_flags &= ~MS_RDONLY;
2074 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2075 int offset, struct super_block *sb)
2077 struct ll_sb_info *sbi = NULL;
2078 struct lustre_md md;
2082 LASSERT(*inode || sb);
2083 sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2084 prune_deathrow(sbi, 1);
2086 rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
2087 sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2092 ll_update_inode(*inode, &md);
2094 LASSERT(sb != NULL);
2097 * At this point server returns to client's same fid as client
2098 * generated for creating. So using ->fid1 is okay here.
2100 LASSERT(fid_is_sane(&md.body->fid1));
2102 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2103 if (*inode == NULL || is_bad_inode(*inode)) {
2104 md_free_lustre_md(sbi->ll_dt_exp, &md);
2106 CERROR("new_inode -fatal: rc %d\n", rc);
2111 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2112 ll_i2info(*inode)->lli_smd);
2117 char *llap_origins[] = {
2118 [LLAP_ORIGIN_UNKNOWN] = "--",
2119 [LLAP_ORIGIN_READPAGE] = "rp",
2120 [LLAP_ORIGIN_READAHEAD] = "ra",
2121 [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2122 [LLAP_ORIGIN_WRITEPAGE] = "wp",
2125 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2126 struct list_head *list)
2128 struct ll_async_page *llap;
2129 struct list_head *pos;
2131 list_for_each(pos, list) {
2132 if (pos == &sbi->ll_pglist)
2134 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2135 if (llap->llap_page == NULL)
2143 int ll_obd_statfs(struct inode *inode, void *arg)
2145 struct ll_sb_info *sbi = NULL;
2146 struct obd_device *client_obd = NULL, *lov_obd = NULL;
2147 struct lov_obd *lov = NULL;
2148 struct obd_statfs stat_buf = {0};
2150 struct obd_ioctl_data *data = NULL;
2154 if (!inode || !(sbi = ll_i2sbi(inode)))
2155 GOTO(out_statfs, rc = -EINVAL);
2157 rc = obd_ioctl_getdata(&buf, &len, arg);
2159 GOTO(out_statfs, rc);
2162 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2163 !data->ioc_pbuf1 || !data->ioc_pbuf2)
2164 GOTO(out_statfs, rc = -EINVAL);
2166 memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2167 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2169 if (type == LL_STATFS_MDC) {
2171 GOTO(out_statfs, rc = -ENODEV);
2172 client_obd = class_exp2obd(sbi->ll_md_exp);
2173 } else if (type == LL_STATFS_LOV) {
2174 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2175 lov = &lov_obd->u.lov;
2177 if ((index >= lov->desc.ld_tgt_count) ||
2178 !lov->lov_tgts[index])
2179 GOTO(out_statfs, rc = -ENODEV);
2181 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2182 if (!lov->lov_tgts[index]->ltd_active)
2183 GOTO(out_uuid, rc = -ENODATA);
2187 GOTO(out_statfs, rc = -EINVAL);
2189 rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2191 GOTO(out_statfs, rc);
2193 if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2194 GOTO(out_statfs, rc = -EFAULT);
2197 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2203 obd_ioctl_freedata(buf, len);
2207 int ll_process_config(struct lustre_cfg *lcfg)
2211 struct lprocfs_static_vars lvars;
2215 lprocfs_init_vars(llite, &lvars);
2217 /* The instance name contains the sb: lustre-client-aacfe000 */
2218 ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2219 if (!ptr || !*(++ptr))
2221 if (sscanf(ptr, "%lx", &x) != 1)
2224 /* This better be a real Lustre superblock! */
2225 LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2227 /* Note we have not called client_common_fill_super yet, so
2228 proc fns must be able to handle that! */
2229 rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2234 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2236 ll_prep_md_op_data(struct md_op_data *op_data, struct inode *i1,
2237 struct inode *i2, const char *name, int namelen, int mode)
2239 LASSERT(i1 != NULL);
2241 if (op_data == NULL)
2242 OBD_ALLOC_PTR(op_data);
2244 if (op_data == NULL)
2247 ll_i2gids(op_data->suppgids, i1, i2);
2248 op_data->fid1 = *ll_inode2fid(i1);
2249 op_data->mod_capa1 = ll_mdscapa_get(i1);
2251 /* @i2 may be NULL. In this case caller itself has to initialize ->fid2
2254 op_data->fid2 = *ll_inode2fid(i2);
2255 op_data->mod_capa2 = ll_mdscapa_get(i2);
2258 op_data->name = name;
2259 op_data->namelen = namelen;
2260 op_data->mode = mode;
2261 op_data->mod_time = CURRENT_SECONDS;
2262 op_data->fsuid = current->fsuid;
2263 op_data->fsgid = current->fsgid;
2264 op_data->cap = current->cap_effective;
2269 void ll_finish_md_op_data(struct md_op_data *op_data)
2271 capa_put(op_data->mod_capa1);
2272 capa_put(op_data->mod_capa2);
2273 OBD_FREE_PTR(op_data);
2276 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2278 struct ll_sb_info *sbi = ll_i2sbi(inode);
2279 struct ptlrpc_request *req = NULL;
2280 struct mdt_body *body;
2282 struct obd_capa *oc;
2286 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2289 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2291 OBD_ALLOC(cmd, ioc->cmd_len);
2294 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2295 GOTO(out, rc = -EFAULT);
2297 oc = ll_mdscapa_get(inode);
2298 rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2299 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2300 ioc->cmd_len, ioc->res_len, 0, &req);
2303 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2304 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2308 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2311 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2312 LASSERT(buflen <= ioc->res_len);
2313 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2315 if (copy_to_user(ioc->res, buf, buflen))
2316 GOTO(out, rc = -EFAULT);
2320 ptlrpc_req_finished(req);
2321 OBD_FREE(cmd, ioc->cmd_len);
2325 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2327 struct ll_sb_info *sbi = ll_i2sbi(inode);
2328 struct ptlrpc_request *req = NULL;
2330 struct obd_capa *oc;
2334 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2337 if (!(sbi->ll_flags & LL_SBI_ACL))
2338 RETURN(-EOPNOTSUPP);
2340 LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2342 OBD_ALLOC(cmd, ioc->cmd_len);
2345 if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2346 GOTO(out, rc = -EFAULT);
2348 oc = ll_mdscapa_get(inode);
2349 rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2350 OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2351 ioc->cmd_len, ioc->res_len, 0, &req);
2354 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2355 XATTR_NAME_LUSTRE_ACL, cmd, rc);
2359 buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2360 LASSERT(buflen <= ioc->res_len);
2361 buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2363 if (copy_to_user(ioc->res, buf, buflen))
2364 GOTO(out, rc = -EFAULT);
2368 ptlrpc_req_finished(req);
2369 OBD_FREE(cmd, ioc->cmd_len);