4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 * Directory code for lustre client.
38 #include <linux/pagemap.h>
40 #include <linux/version.h>
41 #include <linux/security.h>
42 #include <linux/user_namespace.h>
43 #ifdef HAVE_UIDGID_HEADER
44 # include <linux/uidgid.h>
46 #include <linux/uaccess.h>
47 #include <linux/buffer_head.h> // for wait_on_buffer
48 #include <linux/pagevec.h>
50 #define DEBUG_SUBSYSTEM S_LLITE
52 #include <obd_support.h>
53 #include <obd_class.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #include <lustre_lib.h>
56 #include <lustre_dlm.h>
57 #include <lustre_fid.h>
58 #include <lustre_kernelcomm.h>
59 #include <lustre_swab.h>
61 #include "llite_internal.h"
64 * (new) readdir implementation overview.
66 * Original lustre readdir implementation cached exact copy of raw directory
67 * pages on the client. These pages were indexed in client page cache by
68 * logical offset in the directory file. This design, while very simple and
69 * intuitive had some inherent problems:
71 * . it implies that byte offset to the directory entry serves as a
72 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
73 * ext3/htree directory entries may move due to splits, and more
76 * . it is incompatible with the design of split directories for cmd3,
77 * that assumes that names are distributed across nodes based on their
78 * hash, and so readdir should be done in hash order.
80 * New readdir implementation does readdir in hash order, and uses hash of a
81 * file name as a telldir/seekdir cookie. This led to number of complications:
83 * . hash is not unique, so it cannot be used to index cached directory
84 * pages on the client (note, that it requires a whole pageful of hash
85 * collided entries to cause two pages to have identical hashes);
87 * . hash is not unique, so it cannot, strictly speaking, be used as an
88 * entry cookie. ext3/htree has the same problem and lustre implementation
89 * mimics their solution: seekdir(hash) positions directory at the first
90 * entry with the given hash.
96 * Client caches directory pages using hash of the first entry as an index. As
97 * noted above hash is not unique, so this solution doesn't work as is:
98 * special processing is needed for "page hash chains" (i.e., sequences of
99 * pages filled with entries all having the same hash value).
101 * First, such chains have to be detected. To this end, server returns to the
102 * client the hash of the first entry on the page next to one returned. When
103 * client detects that this hash is the same as hash of the first entry on the
104 * returned page, page hash collision has to be handled. Pages in the
105 * hash chain, except first one, are termed "overflow pages".
107 * Solution to index uniqueness problem is to not cache overflow
108 * pages. Instead, when page hash collision is detected, all overflow pages
109 * from emerging chain are immediately requested from the server and placed in
110 * a special data structure (struct ll_dir_chain). This data structure is used
111 * by ll_readdir() to process entries from overflow pages. When readdir
112 * invocation finishes, overflow pages are discarded. If page hash collision
113 * chain weren't completely processed, next call to readdir will again detect
114 * page hash collision, again read overflow pages in, process next portion of
115 * entries and again discard the pages. This is not as wasteful as it looks,
116 * because, given reasonable hash, page hash collisions are extremely rare.
118 * 1. directory positioning
120 * When seekdir(hash) is called, original
131 * identification of and access to overflow pages
135 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
136 * a header lu_dirpage which describes the start/end hash, and whether this
137 * page is empty (contains no dir entry) or hash collide with next page.
138 * After client receives reply, several pages will be integrated into dir page
139 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
140 * lu_dirpage for this integrated page will be adjusted. See
141 * mdc_adjust_dirpages().
144 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
145 __u64 offset, struct ll_dir_chain *chain)
147 struct md_callback cb_op;
151 cb_op.md_blocking_ast = ll_md_blocking_ast;
152 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
159 void ll_release_page(struct inode *inode, struct page *page,
164 /* Always remove the page for striped dir, because the page is
165 * built from temporarily in LMV layer */
166 if (inode != NULL && S_ISDIR(inode->i_mode) &&
167 ll_i2info(inode)->lli_lsm_md != NULL) {
174 if (likely(page->mapping != NULL))
175 truncate_complete_page(page->mapping, page);
182 * return IF_* type for given lu_dirent entry.
183 * IF_* flag shld be converted to particular OS file type in
184 * platform llite module.
186 static u16 ll_dirent_type_get(struct lu_dirent *ent)
189 struct luda_type *lt;
192 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
193 const unsigned align = sizeof(struct luda_type) - 1;
195 len = le16_to_cpu(ent->lde_namelen);
196 len = (len + align) & ~align;
197 lt = (void *)ent->lde_name + len;
198 type = IFTODT(le16_to_cpu(lt->lt_type));
204 #ifdef HAVE_DIR_CONTEXT
205 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
206 struct dir_context *ctx)
209 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
210 void *cookie, filldir_t filldir)
213 struct ll_sb_info *sbi = ll_i2sbi(inode);
215 bool is_api32 = ll_need_32bit_api(sbi);
216 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
218 struct ll_dir_chain chain;
223 ll_dir_chain_init(&chain);
225 page = ll_get_dir_page(inode, op_data, pos, &chain);
227 while (rc == 0 && !done) {
228 struct lu_dirpage *dp;
229 struct lu_dirent *ent;
238 hash = MDS_DIR_END_OFF;
239 dp = page_address(page);
240 for (ent = lu_dirent_start(dp); ent != NULL && !done;
241 ent = lu_dirent_next(ent)) {
248 hash = le64_to_cpu(ent->lde_hash);
249 if (hash < pos) /* Skip until we find target hash */
252 namelen = le16_to_cpu(ent->lde_namelen);
253 if (namelen == 0) /* Skip dummy record */
256 if (is_api32 && is_hash64)
260 fid_le_to_cpu(&fid, &ent->lde_fid);
261 ino = cl_fid_build_ino(&fid, is_api32);
262 type = ll_dirent_type_get(ent);
263 /* For ll_nfs_get_name_filldir(), it will try to access
264 * 'ent' through 'lde_name', so the parameter 'name'
265 * for 'filldir()' must be part of the 'ent'. */
266 #ifdef HAVE_DIR_CONTEXT
268 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
271 done = filldir(cookie, ent->lde_name, namelen, lhash,
278 ll_release_page(inode, page, false);
282 next = le64_to_cpu(dp->ldp_hash_end);
284 if (pos == MDS_DIR_END_OFF) {
286 * End of directory reached.
289 ll_release_page(inode, page, false);
292 * Normal case: continue to the next
295 ll_release_page(inode, page,
296 le32_to_cpu(dp->ldp_flags) &
299 page = ll_get_dir_page(inode, op_data, pos,
303 #ifdef HAVE_DIR_CONTEXT
308 ll_dir_chain_fini(&chain);
312 #ifdef HAVE_DIR_CONTEXT
313 static int ll_iterate(struct file *filp, struct dir_context *ctx)
315 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
318 struct inode *inode = file_inode(filp);
319 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
320 struct ll_sb_info *sbi = ll_i2sbi(inode);
321 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
322 int api32 = ll_need_32bit_api(sbi);
323 struct md_op_data *op_data;
324 struct lu_fid pfid = { 0 };
334 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
335 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
336 inode, (unsigned long)pos, i_size_read(inode), api32);
338 if (pos == MDS_DIR_END_OFF)
344 if (unlikely(ll_i2info(inode)->lli_lsm_md != NULL)) {
345 /* This is only needed for striped dir to fill ..,
346 * see lmv_read_entry */
347 if (file_dentry(filp)->d_parent != NULL &&
348 file_dentry(filp)->d_parent->d_inode != NULL) {
349 __u64 ibits = MDS_INODELOCK_UPDATE;
350 struct inode *parent =
351 file_dentry(filp)->d_parent->d_inode;
353 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
354 pfid = *ll_inode2fid(parent);
357 /* If it can not find in cache, do lookup .. on the master
359 if (fid_is_zero(&pfid)) {
360 rc = ll_dir_get_parent_fid(inode, &pfid);
366 op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
367 LUSTRE_OPC_ANY, inode);
369 GOTO(out, rc = PTR_ERR(op_data));
371 /* foreign dirs are browsed out of Lustre */
372 if (unlikely(op_data->op_mea1 != NULL &&
373 op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
374 ll_finish_md_op_data(op_data);
378 op_data->op_fid3 = pfid;
380 #ifdef HAVE_DIR_CONTEXT
382 rc = ll_dir_read(inode, &pos, op_data, ctx);
385 rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
390 if (pos == MDS_DIR_END_OFF) {
392 pos = LL_DIR_END_OFF_32BIT;
394 pos = LL_DIR_END_OFF;
399 #ifdef HAVE_DIR_CONTEXT
404 ll_finish_md_op_data(op_data);
408 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
413 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
414 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
416 struct mgs_send_param *msp;
423 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
424 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
425 sizeof(struct mgs_send_param), msp, NULL);
427 CERROR("Failed to set parameter: %d\n", rc);
435 * Create striped directory with specified stripe(@lump)
437 * \param[in] dparent the parent of the directory.
438 * \param[in] lump the specified stripes.
439 * \param[in] dirname the name of the directory.
440 * \param[in] mode the specified mode of the directory.
442 * \retval =0 if striped directory is being created successfully.
443 * <0 if the creation is failed.
445 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
446 size_t len, const char *dirname, umode_t mode)
448 struct inode *parent = dparent->d_inode;
449 struct ptlrpc_request *request = NULL;
450 struct md_op_data *op_data;
451 struct ll_sb_info *sbi = ll_i2sbi(parent);
452 struct inode *inode = NULL;
453 struct dentry dentry = {
457 .len = strlen(dirname),
458 .hash = ll_full_name_hash(dparent, dirname,
465 if (unlikely(!lmv_magic_supported(lump->lum_magic)))
468 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
470 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
471 PFID(ll_inode2fid(parent)), parent, dirname,
472 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
474 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
477 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
478 PFID(ll_inode2fid(parent)), parent, dirname,
479 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
482 if (lump->lum_stripe_count > 1 &&
483 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
486 if (IS_DEADDIR(parent) &&
487 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
490 if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
491 lustre_swab_lmv_user_md(lump);
493 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
494 mode &= ~current_umask();
495 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
496 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
497 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
500 RETURN(PTR_ERR(op_data));
502 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
503 /* selinux_dentry_init_security() uses dentry->d_parent and name
504 * to determine the security context for the file. So our fake
505 * dentry should be real enough for this purpose. */
506 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
507 &op_data->op_file_secctx_name,
508 &op_data->op_file_secctx,
509 &op_data->op_file_secctx_size);
511 GOTO(out_op_data, err);
514 op_data->op_cli_flags |= CLI_SET_MEA;
515 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
516 from_kuid(&init_user_ns, current_fsuid()),
517 from_kgid(&init_user_ns, current_fsgid()),
518 cfs_curproc_cap_pack(), 0, &request);
520 GOTO(out_request, err);
522 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
524 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
526 GOTO(out_inode, err);
528 dentry.d_inode = inode;
530 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
532 err = security_inode_notifysecctx(inode,
533 op_data->op_file_secctx,
534 op_data->op_file_secctx_size);
537 err = ll_inode_init_security(&dentry, inode, parent);
540 GOTO(out_inode, err);
546 ptlrpc_req_finished(request);
548 ll_finish_md_op_data(op_data);
553 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
556 struct ll_sb_info *sbi = ll_i2sbi(inode);
557 struct md_op_data *op_data;
558 struct ptlrpc_request *req = NULL;
560 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
561 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
562 struct obd_device *mgc = lsi->lsi_mgc;
569 * This is coming from userspace, so should be in
570 * local endian. But the MDS would like it in little
571 * endian, so we swab it before we send it.
573 switch (lump->lmm_magic) {
574 case LOV_USER_MAGIC_V1: {
575 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
576 lustre_swab_lov_user_md_v1(lump);
577 lum_size = sizeof(struct lov_user_md_v1);
580 case LOV_USER_MAGIC_V3: {
581 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
582 lustre_swab_lov_user_md_v3(
583 (struct lov_user_md_v3 *)lump);
584 lum_size = sizeof(struct lov_user_md_v3);
587 case LOV_USER_MAGIC_COMP_V1: {
588 if (lump->lmm_magic !=
589 cpu_to_le32(LOV_USER_MAGIC_COMP_V1))
590 lustre_swab_lov_comp_md_v1(
591 (struct lov_comp_md_v1 *)lump);
592 lum_size = le32_to_cpu(
593 ((struct lov_comp_md_v1 *)lump)->lcm_size);
596 case LMV_USER_MAGIC: {
597 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
598 lustre_swab_lmv_user_md(
599 (struct lmv_user_md *)lump);
600 lum_size = sizeof(struct lmv_user_md);
603 case LOV_USER_MAGIC_SPECIFIC: {
604 struct lov_user_md_v3 *v3 =
605 (struct lov_user_md_v3 *)lump;
606 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
608 if (lump->lmm_magic !=
609 cpu_to_le32(LOV_USER_MAGIC_SPECIFIC)) {
610 lustre_swab_lov_user_md_v3(v3);
611 lustre_swab_lov_user_md_objects(v3->lmm_objects,
612 v3->lmm_stripe_count);
614 lum_size = lov_user_md_size(v3->lmm_stripe_count,
615 LOV_USER_MAGIC_SPECIFIC);
619 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
620 " %#08x != %#08x nor %#08x\n",
621 lump->lmm_magic, LOV_USER_MAGIC_V1,
627 lum_size = sizeof(struct lov_user_md_v1);
630 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
631 LUSTRE_OPC_ANY, NULL);
633 RETURN(PTR_ERR(op_data));
635 /* swabbing is done in lov_setstripe() on server side */
636 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
637 ll_finish_md_op_data(op_data);
638 ptlrpc_req_finished(req);
642 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
644 * 2.9 server has stored filesystem default stripe in ROOT xattr,
645 * and it's stored into system config for backward compatibility.
647 * In the following we use the fact that LOV_USER_MAGIC_V1 and
648 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
649 * need the make the distiction between the 2 versions
651 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
653 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
654 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
658 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
660 GOTO(end, rc = -ENOMEM);
663 /* Get fsname and assume devname to be -MDT0000. */
664 snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
668 /* Set root stripesize */
669 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
670 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
671 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
675 /* Set root stripecount */
676 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
677 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
678 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
682 /* Set root stripeoffset */
683 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
684 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
685 (typeof(lump->lmm_stripe_offset))(-1));
686 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
690 OBD_FREE(param, MGS_PARAM_MAXLEN);
697 * This function will be used to get default LOV/LMV/Default LMV
698 * @valid will be used to indicate which stripe it will retrieve
699 * OBD_MD_MEA LMV stripe EA
700 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
701 * otherwise Default LOV EA.
702 * Each time, it can only retrieve 1 stripe EA
704 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
705 struct ptlrpc_request **request, u64 valid)
707 struct ll_sb_info *sbi = ll_i2sbi(inode);
708 struct mdt_body *body;
709 struct lov_mds_md *lmm = NULL;
710 struct ptlrpc_request *req = NULL;
712 struct md_op_data *op_data;
715 rc = ll_get_default_mdsize(sbi, &lmm_size);
719 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
720 0, lmm_size, LUSTRE_OPC_ANY,
723 RETURN(PTR_ERR(op_data));
725 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
726 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
727 ll_finish_md_op_data(op_data);
729 CDEBUG(D_INFO, "md_getattr failed on inode "
730 DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
734 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
735 LASSERT(body != NULL);
737 lmm_size = body->mbo_eadatasize;
739 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
741 GOTO(out, rc = -ENODATA);
744 lmm = req_capsule_server_sized_get(&req->rq_pill,
745 &RMF_MDT_MD, lmm_size);
746 LASSERT(lmm != NULL);
749 * This is coming from the MDS, so is probably in
750 * little endian. We convert it to host endian before
751 * passing it to userspace.
753 /* We don't swab objects for directories */
754 switch (le32_to_cpu(lmm->lmm_magic)) {
756 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
757 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
760 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
761 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
763 case LOV_MAGIC_COMP_V1:
764 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
765 lustre_swab_lov_comp_md_v1(
766 (struct lov_comp_md_v1 *)lmm);
769 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
770 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
773 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
774 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
776 case LOV_USER_MAGIC_SPECIFIC: {
777 struct lov_user_md_v3 *v3 = (struct lov_user_md_v3 *)lmm;
779 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
780 lustre_swab_lov_user_md_v3(v3);
781 lustre_swab_lov_user_md_objects(v3->lmm_objects,
782 v3->lmm_stripe_count);
786 case LMV_MAGIC_FOREIGN: {
787 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
789 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
790 __swab32s(&lfm->lfm_magic);
791 __swab32s(&lfm->lfm_length);
792 __swab32s(&lfm->lfm_type);
793 __swab32s(&lfm->lfm_flags);
798 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
803 *plmm_size = lmm_size;
808 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
810 struct md_op_data *op_data;
815 OBD_ALLOC_PTR(op_data);
819 op_data->op_flags |= MF_GET_MDT_IDX;
820 op_data->op_fid1 = *fid;
821 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
822 mdt_index = op_data->op_mds;
823 OBD_FREE_PTR(op_data);
831 * Get MDT index for the inode.
833 int ll_get_mdt_idx(struct inode *inode)
835 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
839 * Generic handler to do any pre-copy work.
841 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
842 * first information for it that real work has started.
844 * Moreover, for a ARCHIVE request, it will sample the file data version and
845 * store it in \a copy.
847 * \return 0 on success.
849 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
851 struct ll_sb_info *sbi = ll_s2sbi(sb);
852 struct hsm_progress_kernel hpk;
857 /* Forge a hsm_progress based on data from copy. */
858 hpk.hpk_fid = copy->hc_hai.hai_fid;
859 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
860 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
861 hpk.hpk_extent.length = 0;
864 hpk.hpk_data_version = 0;
867 /* For archive request, we need to read the current file version. */
868 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
870 __u64 data_version = 0;
872 /* Get inode for this fid */
873 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
875 hpk.hpk_flags |= HP_FLAG_RETRY;
876 /* hpk_errval is >= 0 */
877 hpk.hpk_errval = -PTR_ERR(inode);
878 GOTO(progress, rc = PTR_ERR(inode));
881 /* Read current file data version */
882 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
885 CDEBUG(D_HSM, "Could not read file data version of "
886 DFID" (rc = %d). Archive request ("
887 "%#llx) could not be done.\n",
888 PFID(©->hc_hai.hai_fid), rc,
889 copy->hc_hai.hai_cookie);
890 hpk.hpk_flags |= HP_FLAG_RETRY;
891 /* hpk_errval must be >= 0 */
892 hpk.hpk_errval = -rc;
896 /* Store in the hsm_copy for later copytool use.
897 * Always modified even if no lsm. */
898 copy->hc_data_version = data_version;
902 /* On error, the request should be considered as completed */
903 if (hpk.hpk_errval > 0)
904 hpk.hpk_flags |= HP_FLAG_COMPLETED;
906 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
909 /* Return first error */
910 RETURN(rc != 0 ? rc : rc2);
914 * Generic handler to do any post-copy work.
916 * It will send the last hsm_progress update to coordinator to inform it
917 * that copy is finished and whether it was successful or not.
920 * - for ARCHIVE request, it will sample the file data version and compare it
921 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
922 * will be considered as failed.
923 * - for RESTORE request, it will sample the file data version and send it to
924 * coordinator which is useful if the file was imported as 'released'.
926 * \return 0 on success.
928 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
930 struct ll_sb_info *sbi = ll_s2sbi(sb);
931 struct hsm_progress_kernel hpk;
936 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
937 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
938 * initialized if copy_end was called with copy == NULL.
941 /* Forge a hsm_progress based on data from copy. */
942 hpk.hpk_fid = copy->hc_hai.hai_fid;
943 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
944 hpk.hpk_extent = copy->hc_hai.hai_extent;
945 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
946 hpk.hpk_errval = copy->hc_errval;
947 hpk.hpk_data_version = 0;
949 /* For archive request, we need to check the file data was not changed.
951 * For restore request, we need to send the file data version, this is
952 * useful when the file was created using hsm_import.
954 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
955 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
956 (copy->hc_errval == 0)) {
958 __u64 data_version = 0;
960 /* Get lsm for this fid */
961 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
963 hpk.hpk_flags |= HP_FLAG_RETRY;
964 /* hpk_errval must be >= 0 */
965 hpk.hpk_errval = -PTR_ERR(inode);
966 GOTO(progress, rc = PTR_ERR(inode));
969 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
972 CDEBUG(D_HSM, "Could not read file data version. "
973 "Request could not be confirmed.\n");
974 if (hpk.hpk_errval == 0)
975 hpk.hpk_errval = -rc;
979 /* Store in the hsm_copy for later copytool use.
980 * Always modified even if no lsm. */
981 hpk.hpk_data_version = data_version;
983 /* File could have been stripped during archiving, so we need
984 * to check anyway. */
985 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
986 (copy->hc_data_version != data_version)) {
987 CDEBUG(D_HSM, "File data version mismatched. "
988 "File content was changed during archiving. "
989 DFID", start:%#llx current:%#llx\n",
990 PFID(©->hc_hai.hai_fid),
991 copy->hc_data_version, data_version);
992 /* File was changed, send error to cdt. Do not ask for
993 * retry because if a file is modified frequently,
994 * the cdt will loop on retried archive requests.
995 * The policy engine will ask for a new archive later
996 * when the file will not be modified for some tunable
998 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1000 /* hpk_errval must be >= 0 */
1001 hpk.hpk_errval = -rc;
1008 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1011 /* Return first error */
1012 RETURN(rc != 0 ? rc : rc2);
1016 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1017 const struct lustre_kernelcomm __user *data)
1019 struct lustre_kernelcomm *lk;
1020 struct lustre_kernelcomm *tmp;
1021 size_t size = sizeof(*lk);
1026 /* copy data from userspace to get numbers of archive_id */
1027 OBD_ALLOC(lk, size);
1031 if (copy_from_user(lk, data, size))
1032 GOTO(out_lk, rc = -EFAULT);
1034 if (lk->lk_flags & LK_FLG_STOP)
1037 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1038 __u32 archive_mask = lk->lk_data_count;
1041 /* old hsm agent to old MDS */
1042 if (!exp_connect_archive_id_array(exp))
1045 /* old hsm agent to new MDS */
1046 lk->lk_flags |= LK_FLG_DATANR;
1048 if (archive_mask == 0)
1051 count = hweight32(archive_mask);
1052 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1053 OBD_ALLOC(tmp, new_size);
1055 GOTO(out_lk, rc = -ENOMEM);
1057 memcpy(tmp, lk, size);
1058 tmp->lk_data_count = count;
1064 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1065 if ((1 << i) & archive_mask) {
1066 lk->lk_data[count] = i + 1;
1073 /* new hsm agent to new mds */
1074 if (lk->lk_data_count > 0) {
1075 new_size = offsetof(struct lustre_kernelcomm,
1076 lk_data[lk->lk_data_count]);
1077 OBD_ALLOC(tmp, new_size);
1079 GOTO(out_lk, rc = -ENOMEM);
1085 if (copy_from_user(lk, data, size))
1086 GOTO(out_lk, rc = -EFAULT);
1089 /* new hsm agent to old MDS */
1090 if (!exp_connect_archive_id_array(exp)) {
1093 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1094 GOTO(out_lk, rc = -EINVAL);
1096 for (i = 0; i < lk->lk_data_count; i++) {
1097 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1099 CERROR("%s: archive id %d requested but only "
1100 "[0 - %zu] supported: rc = %d\n",
1101 exp->exp_obd->obd_name, lk->lk_data[i],
1102 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1106 if (lk->lk_data[i] == 0) {
1111 archives |= (1 << (lk->lk_data[i] - 1));
1113 lk->lk_flags &= ~LK_FLG_DATANR;
1114 lk->lk_data_count = archives;
1117 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1123 static int check_owner(int type, int id)
1127 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1131 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1140 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1142 int cmd = qctl->qc_cmd;
1143 int type = qctl->qc_type;
1144 int id = qctl->qc_id;
1145 int valid = qctl->qc_valid;
1152 case LUSTRE_Q_SETDEFAULT:
1153 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1157 case LUSTRE_Q_GETDEFAULT:
1158 if (check_owner(type, id) &&
1159 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1165 CERROR("unsupported quotactl op: %#x\n", cmd);
1169 if (valid != QC_GENERAL) {
1170 if (cmd == Q_GETINFO)
1171 qctl->qc_cmd = Q_GETOINFO;
1172 else if (cmd == Q_GETQUOTA)
1173 qctl->qc_cmd = Q_GETOQUOTA;
1179 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1180 sizeof(*qctl), qctl, NULL);
1183 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1184 sizeof(*qctl), qctl, NULL);
1187 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1188 sizeof(*qctl), qctl, NULL);
1190 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1192 sizeof(*qctl), qctl, NULL);
1204 struct obd_quotactl *oqctl;
1206 OBD_ALLOC_PTR(oqctl);
1210 QCTL_COPY(oqctl, qctl);
1211 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1213 OBD_FREE_PTR(oqctl);
1216 /* If QIF_SPACE is not set, client should collect the
1217 * space usage from OSSs by itself */
1218 if (cmd == Q_GETQUOTA &&
1219 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1220 !oqctl->qc_dqblk.dqb_curspace) {
1221 struct obd_quotactl *oqctl_tmp;
1223 OBD_ALLOC_PTR(oqctl_tmp);
1224 if (oqctl_tmp == NULL)
1225 GOTO(out, rc = -ENOMEM);
1227 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1228 oqctl_tmp->qc_id = oqctl->qc_id;
1229 oqctl_tmp->qc_type = oqctl->qc_type;
1231 /* collect space usage from OSTs */
1232 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1233 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1234 if (!rc || rc == -EREMOTEIO) {
1235 oqctl->qc_dqblk.dqb_curspace =
1236 oqctl_tmp->qc_dqblk.dqb_curspace;
1237 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1240 /* collect space & inode usage from MDTs */
1241 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1242 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1243 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1244 if (!rc || rc == -EREMOTEIO) {
1245 oqctl->qc_dqblk.dqb_curspace +=
1246 oqctl_tmp->qc_dqblk.dqb_curspace;
1247 oqctl->qc_dqblk.dqb_curinodes =
1248 oqctl_tmp->qc_dqblk.dqb_curinodes;
1249 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1251 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1254 OBD_FREE_PTR(oqctl_tmp);
1257 QCTL_COPY(qctl, oqctl);
1258 OBD_FREE_PTR(oqctl);
1264 /* This function tries to get a single name component,
1265 * to send to the server. No actual path traversal involved,
1266 * so we limit to NAME_MAX */
1267 static char *ll_getname(const char __user *filename)
1272 OBD_ALLOC(tmp, NAME_MAX + 1);
1275 return ERR_PTR(-ENOMEM);
1277 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1280 else if (len > NAME_MAX)
1281 ret = -ENAMETOOLONG;
1284 OBD_FREE(tmp, NAME_MAX + 1);
1290 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1292 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1294 struct dentry *dentry = file_dentry(file);
1295 struct inode *inode = file_inode(file);
1296 struct ll_sb_info *sbi = ll_i2sbi(inode);
1297 struct obd_ioctl_data *data;
1301 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1302 PFID(ll_inode2fid(inode)), inode, cmd);
1304 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1305 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1308 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1310 case FS_IOC_GETFLAGS:
1311 case FS_IOC_SETFLAGS:
1312 RETURN(ll_iocontrol(inode, file, cmd, arg));
1313 case FSFILT_IOC_GETVERSION:
1314 case FS_IOC_GETVERSION:
1315 RETURN(put_user(inode->i_generation, (int __user *)arg));
1316 /* We need to special case any other ioctls we want to handle,
1317 * to send them to the MDS/OST as appropriate and to properly
1318 * network encode the arg field. */
1319 case FS_IOC_SETVERSION:
1322 case LL_IOC_GET_MDTIDX: {
1325 mdtidx = ll_get_mdt_idx(inode);
1329 if (put_user((int)mdtidx, (int __user *)arg))
1334 case IOC_MDC_LOOKUP: {
1335 int namelen, len = 0;
1339 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1344 filename = data->ioc_inlbuf1;
1345 namelen = strlen(filename);
1347 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1348 GOTO(out_free, rc = -EINVAL);
1351 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1353 CERROR("%s: lookup %.*s failed: rc = %d\n",
1354 sbi->ll_fsname, namelen, filename, rc);
1358 OBD_FREE_LARGE(buf, len);
1361 case LL_IOC_LMV_SETSTRIPE: {
1362 struct lmv_user_md *lum;
1371 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1376 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1377 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1378 GOTO(lmv_out_free, rc = -EINVAL);
1380 filename = data->ioc_inlbuf1;
1381 namelen = data->ioc_inllen1;
1384 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1385 GOTO(lmv_out_free, rc = -EINVAL);
1387 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1388 lumlen = data->ioc_inllen2;
1390 if (!lmv_magic_supported(lum->lum_magic)) {
1391 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1392 lum->lum_magic, -EINVAL);
1393 GOTO(lmv_out_free, rc = -EINVAL);
1396 if ((lum->lum_magic == LMV_USER_MAGIC ||
1397 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1398 lumlen < sizeof(*lum)) {
1399 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1400 filename, lumlen, lum->lum_magic, -EINVAL);
1401 GOTO(lmv_out_free, rc = -EINVAL);
1404 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1405 lumlen < sizeof(struct lmv_foreign_md)) {
1406 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1407 filename, lum->lum_magic, lumlen, -EFAULT);
1408 GOTO(lmv_out_free, rc = -EINVAL);
1411 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1412 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1414 mode = data->ioc_type;
1416 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1418 OBD_FREE_LARGE(buf, len);
1422 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1423 struct lmv_user_md lum;
1424 struct lmv_user_md __user *ulump =
1425 (struct lmv_user_md __user *)arg;
1428 if (copy_from_user(&lum, ulump, sizeof(lum)))
1431 if (lum.lum_magic != LMV_USER_MAGIC)
1434 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1438 case LL_IOC_LOV_SETSTRIPE_NEW:
1439 case LL_IOC_LOV_SETSTRIPE: {
1440 struct lov_user_md_v3 *lumv3 = NULL;
1441 struct lov_user_md_v1 lumv1;
1442 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1443 struct lov_user_md_v1 __user *lumv1p =
1444 (struct lov_user_md_v1 __user *)arg;
1445 struct lov_user_md_v3 __user *lumv3p =
1446 (struct lov_user_md_v3 __user *)arg;
1449 int set_default = 0;
1451 CLASSERT(sizeof(struct lov_user_md_v3) >
1452 sizeof(struct lov_comp_md_v1));
1453 CLASSERT(sizeof(*lumv3) == sizeof(*lumv3p));
1454 /* first try with v1 which is smaller than v3 */
1455 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1458 if (inode->i_sb->s_root == file_dentry(file))
1461 switch (lumv1.lmm_magic) {
1462 case LOV_USER_MAGIC_V3:
1463 case LOV_USER_MAGIC_SPECIFIC:
1464 lum_size = ll_lov_user_md_size(&lumv1);
1467 OBD_ALLOC(lumv3, lum_size);
1470 if (copy_from_user(lumv3, lumv3p, lum_size))
1471 GOTO(out, rc = -EFAULT);
1472 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1474 case LOV_USER_MAGIC_V1:
1477 GOTO(out, rc = -ENOTSUPP);
1480 /* in v1 and v3 cases lumv1 points to data */
1481 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1484 OBD_FREE(lumv3, lum_size);
1487 case LL_IOC_LMV_GETSTRIPE: {
1488 struct lmv_user_md __user *ulmv =
1489 (struct lmv_user_md __user *)arg;
1490 struct lmv_user_md lum;
1491 struct ptlrpc_request *request = NULL;
1492 union lmv_mds_md *lmm = NULL;
1495 struct lmv_user_md *tmp = NULL;
1499 int max_stripe_count;
1503 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1506 max_stripe_count = lum.lum_stripe_count;
1507 /* lum_magic will indicate which stripe the ioctl will like
1508 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1509 * is for default LMV stripe */
1510 if (lum.lum_magic == LMV_MAGIC_V1)
1511 valid |= OBD_MD_MEA;
1512 else if (lum.lum_magic == LMV_USER_MAGIC)
1513 valid |= OBD_MD_DEFAULT_MEA;
1517 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1520 GOTO(finish_req, rc);
1522 /* Get default LMV EA */
1523 if (lum.lum_magic == LMV_USER_MAGIC) {
1524 if (lmmsize > sizeof(*ulmv))
1525 GOTO(finish_req, rc = -EINVAL);
1527 if (copy_to_user(ulmv, lmm, lmmsize))
1528 GOTO(finish_req, rc = -EFAULT);
1530 GOTO(finish_req, rc);
1533 /* if foreign LMV case, fake stripes number */
1534 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1535 struct lmv_foreign_md *lfm;
1537 lfm = (struct lmv_foreign_md *)lmm;
1538 if (lfm->lfm_length < XATTR_SIZE_MAX -
1539 offsetof(typeof(*lfm), lfm_value)) {
1540 __u32 size = lfm->lfm_length +
1541 offsetof(typeof(*lfm), lfm_value);
1543 stripe_count = lmv_foreign_to_md_stripes(size);
1545 CERROR("invalid %d foreign size returned\n",
1550 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1552 if (max_stripe_count < stripe_count) {
1553 lum.lum_stripe_count = stripe_count;
1554 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1555 GOTO(finish_req, rc = -EFAULT);
1556 GOTO(finish_req, rc = -E2BIG);
1559 /* enough room on user side and foreign case */
1560 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1561 struct lmv_foreign_md *lfm;
1564 lfm = (struct lmv_foreign_md *)lmm;
1565 size = lfm->lfm_length +
1566 offsetof(struct lmv_foreign_md, lfm_value);
1567 if (copy_to_user(ulmv, lfm, size))
1568 GOTO(finish_req, rc = -EFAULT);
1569 GOTO(finish_req, rc);
1572 lum_size = lmv_user_md_size(stripe_count,
1573 LMV_USER_MAGIC_SPECIFIC);
1574 OBD_ALLOC(tmp, lum_size);
1576 GOTO(finish_req, rc = -ENOMEM);
1578 mdt_index = ll_get_mdt_idx(inode);
1580 GOTO(out_tmp, rc = -ENOMEM);
1582 tmp->lum_magic = LMV_MAGIC_V1;
1583 tmp->lum_stripe_count = 0;
1584 tmp->lum_stripe_offset = mdt_index;
1585 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1586 for (i = 0; i < stripe_count; i++) {
1589 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1590 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1592 GOTO(out_tmp, rc = mdt_index);
1594 tmp->lum_objects[i].lum_mds = mdt_index;
1595 tmp->lum_objects[i].lum_fid = fid;
1596 tmp->lum_stripe_count++;
1599 if (copy_to_user(ulmv, tmp, lum_size))
1600 GOTO(out_tmp, rc = -EFAULT);
1602 OBD_FREE(tmp, lum_size);
1604 ptlrpc_req_finished(request);
1608 case LL_IOC_REMOVE_ENTRY: {
1609 char *filename = NULL;
1613 /* Here is a little hack to avoid sending REINT_RMENTRY to
1614 * unsupported server, which might crash the server(LU-2730),
1615 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1616 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1617 * server will support REINT_RMENTRY XXX*/
1618 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1621 filename = ll_getname((const char __user *)arg);
1622 if (IS_ERR(filename))
1623 RETURN(PTR_ERR(filename));
1625 namelen = strlen(filename);
1627 GOTO(out_rmdir, rc = -EINVAL);
1629 rc = ll_rmdir_entry(inode, filename, namelen);
1632 ll_putname(filename);
1635 case LL_IOC_LOV_SWAP_LAYOUTS:
1637 case IOC_OBD_STATFS:
1638 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1639 case LL_IOC_LOV_GETSTRIPE:
1640 case LL_IOC_LOV_GETSTRIPE_NEW:
1641 case LL_IOC_MDC_GETINFO:
1642 case IOC_MDC_GETFILEINFO:
1643 case IOC_MDC_GETFILESTRIPE: {
1644 struct ptlrpc_request *request = NULL;
1645 struct lov_user_md __user *lump;
1646 struct lov_mds_md *lmm = NULL;
1647 struct mdt_body *body;
1648 char *filename = NULL;
1651 if (cmd == IOC_MDC_GETFILEINFO ||
1652 cmd == IOC_MDC_GETFILESTRIPE) {
1653 filename = ll_getname((const char __user *)arg);
1654 if (IS_ERR(filename))
1655 RETURN(PTR_ERR(filename));
1657 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1658 &lmmsize, &request);
1660 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1665 body = req_capsule_server_get(&request->rq_pill,
1667 LASSERT(body != NULL);
1672 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1673 cmd == LL_IOC_MDC_GETINFO)) {
1681 if (cmd == IOC_MDC_GETFILESTRIPE ||
1682 cmd == LL_IOC_LOV_GETSTRIPE ||
1683 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1684 lump = (struct lov_user_md __user *)arg;
1686 struct lov_user_mds_data __user *lmdp;
1687 lmdp = (struct lov_user_mds_data __user *)arg;
1688 lump = &lmdp->lmd_lmm;
1692 /* If the file has no striping then zero out *lump so
1693 * that the caller isn't confused by garbage. */
1694 if (clear_user(lump, sizeof(*lump)))
1695 GOTO(out_req, rc = -EFAULT);
1696 } else if (copy_to_user(lump, lmm, lmmsize)) {
1697 if (copy_to_user(lump, lmm, sizeof(*lump)))
1698 GOTO(out_req, rc = -EFAULT);
1702 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1703 struct lov_user_mds_data __user *lmdp;
1706 st.st_dev = inode->i_sb->s_dev;
1707 st.st_mode = body->mbo_mode;
1708 st.st_nlink = body->mbo_nlink;
1709 st.st_uid = body->mbo_uid;
1710 st.st_gid = body->mbo_gid;
1711 st.st_rdev = body->mbo_rdev;
1712 st.st_size = body->mbo_size;
1713 st.st_blksize = PAGE_SIZE;
1714 st.st_blocks = body->mbo_blocks;
1715 st.st_atime = body->mbo_atime;
1716 st.st_mtime = body->mbo_mtime;
1717 st.st_ctime = body->mbo_ctime;
1718 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1722 lmdp = (struct lov_user_mds_data __user *)arg;
1723 if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1724 GOTO(out_req, rc = -EFAULT);
1729 ptlrpc_req_finished(request);
1731 ll_putname(filename);
1734 case OBD_IOC_QUOTACTL: {
1735 struct if_quotactl *qctl;
1737 OBD_ALLOC_PTR(qctl);
1741 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1742 GOTO(out_quotactl, rc = -EFAULT);
1744 rc = quotactl_ioctl(sbi, qctl);
1747 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1754 case OBD_IOC_GETDTNAME:
1755 case OBD_IOC_GETMDNAME:
1756 RETURN(ll_get_obd_name(inode, cmd, arg));
1757 case LL_IOC_FLUSHCTX:
1758 RETURN(ll_flush_ctx(inode));
1759 case LL_IOC_GETOBDCOUNT: {
1761 struct obd_export *exp;
1763 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1766 /* get ost count when count is zero, get mdt count otherwise */
1767 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1768 vallen = sizeof(count);
1769 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1770 KEY_TGT_COUNT, &vallen, &count);
1772 CERROR("get target count failed: %d\n", rc);
1776 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1781 case LL_IOC_PATH2FID:
1782 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1783 sizeof(struct lu_fid)))
1786 case LL_IOC_GET_CONNECT_FLAGS: {
1787 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1788 (void __user *)arg));
1790 case OBD_IOC_FID2PATH:
1791 RETURN(ll_fid2path(inode, (void __user *)arg));
1792 case LL_IOC_GETPARENT:
1793 RETURN(ll_getparent(file, (void __user *)arg));
1794 case LL_IOC_FID2MDTIDX: {
1795 struct obd_export *exp = ll_i2mdexp(inode);
1799 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1803 /* Call mdc_iocontrol */
1804 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1805 (__u32 __user *)&index);
1811 case LL_IOC_HSM_REQUEST: {
1812 struct hsm_user_request *hur;
1819 /* We don't know the true size yet; copy the fixed-size part */
1820 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1825 /* Compute the whole struct size */
1826 totalsize = hur_len(hur);
1831 /* Final size will be more than double totalsize */
1832 if (totalsize >= MDS_MAXREQSIZE / 3)
1835 OBD_ALLOC_LARGE(hur, totalsize);
1839 /* Copy the whole struct */
1840 if (copy_from_user(hur, (void __user *)arg, totalsize))
1841 GOTO(out_hur, rc = -EFAULT);
1843 if (hur->hur_request.hr_action == HUA_RELEASE) {
1844 const struct lu_fid *fid;
1848 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1849 fid = &hur->hur_user_item[i].hui_fid;
1850 f = search_inode_for_lustre(inode->i_sb, fid);
1856 rc = ll_hsm_release(f);
1862 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1867 OBD_FREE_LARGE(hur, totalsize);
1871 case LL_IOC_HSM_PROGRESS: {
1872 struct hsm_progress_kernel hpk;
1873 struct hsm_progress hp;
1875 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1878 hpk.hpk_fid = hp.hp_fid;
1879 hpk.hpk_cookie = hp.hp_cookie;
1880 hpk.hpk_extent = hp.hp_extent;
1881 hpk.hpk_flags = hp.hp_flags;
1882 hpk.hpk_errval = hp.hp_errval;
1883 hpk.hpk_data_version = 0;
1885 /* File may not exist in Lustre; all progress
1886 * reported to Lustre root */
1887 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1891 case LL_IOC_HSM_CT_START:
1892 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1895 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
1896 (struct lustre_kernelcomm __user *)arg);
1899 case LL_IOC_HSM_COPY_START: {
1900 struct hsm_copy *copy;
1903 OBD_ALLOC_PTR(copy);
1906 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1911 rc = ll_ioc_copy_start(inode->i_sb, copy);
1912 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1918 case LL_IOC_HSM_COPY_END: {
1919 struct hsm_copy *copy;
1922 OBD_ALLOC_PTR(copy);
1925 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1930 rc = ll_ioc_copy_end(inode->i_sb, copy);
1931 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1937 case LL_IOC_MIGRATE: {
1938 struct lmv_user_md *lum;
1945 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1949 data = (struct obd_ioctl_data *)buf;
1950 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1951 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1952 GOTO(migrate_free, rc = -EINVAL);
1954 filename = data->ioc_inlbuf1;
1955 namelen = data->ioc_inllen1;
1957 if (namelen < 1 || namelen != strlen(filename) + 1) {
1958 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1959 GOTO(migrate_free, rc = -EINVAL);
1962 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1963 if (lum->lum_magic != LMV_USER_MAGIC &&
1964 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
1966 CERROR("%s: wrong lum magic %x: rc = %d\n",
1967 filename, lum->lum_magic, rc);
1968 GOTO(migrate_free, rc);
1971 rc = ll_migrate(inode, file, lum, filename);
1973 OBD_FREE_LARGE(buf, len);
1977 case LL_IOC_FSGETXATTR:
1978 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
1979 case LL_IOC_FSSETXATTR:
1980 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
1982 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1983 (void __user *)arg));
1987 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1989 struct inode *inode = file->f_mapping->host;
1990 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1991 struct ll_sb_info *sbi = ll_i2sbi(inode);
1992 int api32 = ll_need_32bit_api(sbi);
1993 loff_t ret = -EINVAL;
2001 offset += file->f_pos;
2007 offset += LL_DIR_END_OFF_32BIT;
2009 offset += LL_DIR_END_OFF;
2016 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2017 (!api32 && offset <= LL_DIR_END_OFF))) {
2018 if (offset != file->f_pos) {
2019 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2020 (!api32 && offset == LL_DIR_END_OFF))
2021 fd->lfd_pos = MDS_DIR_END_OFF;
2022 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2023 fd->lfd_pos = offset << 32;
2025 fd->lfd_pos = offset;
2026 file->f_pos = offset;
2027 file->f_version = 0;
2034 inode_unlock(inode);
2038 static int ll_dir_open(struct inode *inode, struct file *file)
2041 RETURN(ll_file_open(inode, file));
2044 static int ll_dir_release(struct inode *inode, struct file *file)
2047 RETURN(ll_file_release(inode, file));
2050 const struct file_operations ll_dir_operations = {
2051 .llseek = ll_dir_seek,
2052 .open = ll_dir_open,
2053 .release = ll_dir_release,
2054 .read = generic_read_dir,
2055 #ifdef HAVE_DIR_CONTEXT
2056 .iterate = ll_iterate,
2058 .readdir = ll_readdir,
2060 .unlocked_ioctl = ll_dir_ioctl,