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 #include <linux/uidgid.h>
44 #include <linux/uaccess.h>
45 #include <linux/buffer_head.h> // for wait_on_buffer
46 #include <linux/pagevec.h>
48 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <obd_support.h>
51 #include <obd_class.h>
52 #include <uapi/linux/lustre/lustre_ioctl.h>
53 #include <lustre_lib.h>
54 #include <lustre_dlm.h>
55 #include <lustre_fid.h>
56 #include <lustre_kernelcomm.h>
57 #include <lustre_swab.h>
59 #include "llite_internal.h"
62 * (new) readdir implementation overview.
64 * Original lustre readdir implementation cached exact copy of raw directory
65 * pages on the client. These pages were indexed in client page cache by
66 * logical offset in the directory file. This design, while very simple and
67 * intuitive had some inherent problems:
69 * . it implies that byte offset to the directory entry serves as a
70 * telldir(3)/seekdir(3) cookie, but that offset is not stable: in
71 * ext3/htree directory entries may move due to splits, and more
74 * . it is incompatible with the design of split directories for cmd3,
75 * that assumes that names are distributed across nodes based on their
76 * hash, and so readdir should be done in hash order.
78 * New readdir implementation does readdir in hash order, and uses hash of a
79 * file name as a telldir/seekdir cookie. This led to number of complications:
81 * . hash is not unique, so it cannot be used to index cached directory
82 * pages on the client (note, that it requires a whole pageful of hash
83 * collided entries to cause two pages to have identical hashes);
85 * . hash is not unique, so it cannot, strictly speaking, be used as an
86 * entry cookie. ext3/htree has the same problem and lustre implementation
87 * mimics their solution: seekdir(hash) positions directory at the first
88 * entry with the given hash.
94 * Client caches directory pages using hash of the first entry as an index. As
95 * noted above hash is not unique, so this solution doesn't work as is:
96 * special processing is needed for "page hash chains" (i.e., sequences of
97 * pages filled with entries all having the same hash value).
99 * First, such chains have to be detected. To this end, server returns to the
100 * client the hash of the first entry on the page next to one returned. When
101 * client detects that this hash is the same as hash of the first entry on the
102 * returned page, page hash collision has to be handled. Pages in the
103 * hash chain, except first one, are termed "overflow pages".
105 * Solution to index uniqueness problem is to not cache overflow
106 * pages. Instead, when page hash collision is detected, all overflow pages
107 * from emerging chain are immediately requested from the server and placed in
108 * a special data structure (struct ll_dir_chain). This data structure is used
109 * by ll_readdir() to process entries from overflow pages. When readdir
110 * invocation finishes, overflow pages are discarded. If page hash collision
111 * chain weren't completely processed, next call to readdir will again detect
112 * page hash collision, again read overflow pages in, process next portion of
113 * entries and again discard the pages. This is not as wasteful as it looks,
114 * because, given reasonable hash, page hash collisions are extremely rare.
116 * 1. directory positioning
118 * When seekdir(hash) is called, original
129 * identification of and access to overflow pages
133 * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
134 * a header lu_dirpage which describes the start/end hash, and whether this
135 * page is empty (contains no dir entry) or hash collide with next page.
136 * After client receives reply, several pages will be integrated into dir page
137 * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
138 * lu_dirpage for this integrated page will be adjusted. See
139 * mdc_adjust_dirpages().
142 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
143 __u64 offset, struct ll_dir_chain *chain)
145 struct md_callback cb_op;
149 cb_op.md_blocking_ast = ll_md_blocking_ast;
150 rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
157 void ll_release_page(struct inode *inode, struct page *page,
162 /* Always remove the page for striped dir, because the page is
163 * built from temporarily in LMV layer */
164 if (inode && ll_dir_striped(inode)) {
171 if (likely(page->mapping != NULL))
172 delete_from_page_cache(page);
179 * return IF_* type for given lu_dirent entry.
180 * IF_* flag shld be converted to particular OS file type in
181 * platform llite module.
183 static u16 ll_dirent_type_get(struct lu_dirent *ent)
186 struct luda_type *lt;
189 if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
190 const unsigned align = sizeof(struct luda_type) - 1;
192 len = le16_to_cpu(ent->lde_namelen);
193 len = (len + align) & ~align;
194 lt = (void *)ent->lde_name + len;
195 type = IFTODT(le16_to_cpu(lt->lt_type));
201 #ifdef HAVE_DIR_CONTEXT
202 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
203 struct dir_context *ctx)
206 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
207 void *cookie, filldir_t filldir)
210 struct ll_sb_info *sbi = ll_i2sbi(inode);
212 bool is_api32 = ll_need_32bit_api(sbi);
213 bool is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
215 struct ll_dir_chain chain;
220 ll_dir_chain_init(&chain);
222 page = ll_get_dir_page(inode, op_data, pos, &chain);
224 while (rc == 0 && !done) {
225 struct lu_dirpage *dp;
226 struct lu_dirent *ent;
235 hash = MDS_DIR_END_OFF;
236 dp = page_address(page);
237 for (ent = lu_dirent_start(dp); ent != NULL && !done;
238 ent = lu_dirent_next(ent)) {
245 hash = le64_to_cpu(ent->lde_hash);
246 if (hash < pos) /* Skip until we find target hash */
249 namelen = le16_to_cpu(ent->lde_namelen);
250 if (namelen == 0) /* Skip dummy record */
253 if (is_api32 && is_hash64)
257 fid_le_to_cpu(&fid, &ent->lde_fid);
258 ino = cl_fid_build_ino(&fid, is_api32);
259 type = ll_dirent_type_get(ent);
260 /* For ll_nfs_get_name_filldir(), it will try to access
261 * 'ent' through 'lde_name', so the parameter 'name'
262 * for 'filldir()' must be part of the 'ent'. */
263 #ifdef HAVE_DIR_CONTEXT
265 done = !dir_emit(ctx, ent->lde_name, namelen, ino,
268 done = filldir(cookie, ent->lde_name, namelen, lhash,
275 ll_release_page(inode, page, false);
279 next = le64_to_cpu(dp->ldp_hash_end);
281 if (pos == MDS_DIR_END_OFF) {
283 * End of directory reached.
286 ll_release_page(inode, page, false);
289 * Normal case: continue to the next
292 ll_release_page(inode, page,
293 le32_to_cpu(dp->ldp_flags) &
296 page = ll_get_dir_page(inode, op_data, pos,
300 #ifdef HAVE_DIR_CONTEXT
305 ll_dir_chain_fini(&chain);
309 #ifdef HAVE_DIR_CONTEXT
310 static int ll_iterate(struct file *filp, struct dir_context *ctx)
312 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
315 struct inode *inode = file_inode(filp);
316 struct ll_file_data *lfd = LUSTRE_FPRIVATE(filp);
317 struct ll_sb_info *sbi = ll_i2sbi(inode);
318 int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
319 int api32 = ll_need_32bit_api(sbi);
320 struct md_op_data *op_data;
321 struct lu_fid pfid = { 0 };
322 ktime_t kstart = ktime_get();
332 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
333 "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
334 inode, (unsigned long)pos, i_size_read(inode), api32);
336 if (pos == MDS_DIR_END_OFF)
342 if (unlikely(ll_dir_striped(inode))) {
344 * This is only needed for striped dir to fill ..,
345 * see lmv_read_page()
347 if (file_dentry(filp)->d_parent != NULL &&
348 file_dentry(filp)->d_parent->d_inode != NULL) {
349 __u64 ibits = MDS_INODELOCK_LOOKUP;
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,
409 ktime_us_delta(ktime_get(), kstart));
414 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
415 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
417 struct mgs_send_param *msp;
424 strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
425 rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
426 sizeof(struct mgs_send_param), msp, NULL);
428 CERROR("Failed to set parameter: %d\n", rc);
436 * Create striped directory with specified stripe(@lump)
438 * \param[in] dparent the parent of the directory.
439 * \param[in] lump the specified stripes.
440 * \param[in] dirname the name of the directory.
441 * \param[in] mode the specified mode of the directory.
443 * \retval =0 if striped directory is being created successfully.
444 * <0 if the creation is failed.
446 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
447 size_t len, const char *dirname, umode_t mode)
449 struct inode *parent = dparent->d_inode;
450 struct ptlrpc_request *request = NULL;
451 struct md_op_data *op_data;
452 struct ll_sb_info *sbi = ll_i2sbi(parent);
453 struct inode *inode = NULL;
454 struct dentry dentry = {
458 .len = strlen(dirname),
459 .hash = ll_full_name_hash(dparent, dirname,
466 if (unlikely(!lmv_magic_supported(lump->lum_magic)))
469 if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
471 "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
472 PFID(ll_inode2fid(parent)), parent, dirname,
473 (int)lump->lum_stripe_offset, lump->lum_stripe_count);
475 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
478 "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
479 PFID(ll_inode2fid(parent)), parent, dirname,
480 lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
483 if (lump->lum_stripe_count > 1 &&
484 !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
487 if (IS_DEADDIR(parent) &&
488 !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
491 if (unlikely(!lmv_magic_supported(cpu_to_le32(lump->lum_magic))))
492 lustre_swab_lmv_user_md(lump);
494 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
495 mode &= ~current_umask();
496 mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
497 op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
498 strlen(dirname), mode, LUSTRE_OPC_MKDIR,
501 RETURN(PTR_ERR(op_data));
503 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
504 /* selinux_dentry_init_security() uses dentry->d_parent and name
505 * to determine the security context for the file. So our fake
506 * dentry should be real enough for this purpose. */
507 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
508 &op_data->op_file_secctx_name,
509 &op_data->op_file_secctx,
510 &op_data->op_file_secctx_size);
512 GOTO(out_op_data, err);
515 op_data->op_cli_flags |= CLI_SET_MEA;
516 err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
517 from_kuid(&init_user_ns, current_fsuid()),
518 from_kgid(&init_user_ns, current_fsgid()),
519 cfs_curproc_cap_pack(), 0, &request);
521 GOTO(out_request, err);
523 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
525 err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
527 GOTO(out_inode, err);
529 dentry.d_inode = inode;
531 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
533 err = security_inode_notifysecctx(inode,
534 op_data->op_file_secctx,
535 op_data->op_file_secctx_size);
538 err = ll_inode_init_security(&dentry, inode, parent);
541 GOTO(out_inode, err);
547 ptlrpc_req_finished(request);
549 ll_finish_md_op_data(op_data);
554 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
557 struct ll_sb_info *sbi = ll_i2sbi(inode);
558 struct md_op_data *op_data;
559 struct ptlrpc_request *req = NULL;
561 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
562 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
563 struct obd_device *mgc = lsi->lsi_mgc;
569 switch (lump->lmm_magic) {
570 case LOV_USER_MAGIC_V1:
571 lum_size = sizeof(struct lov_user_md_v1);
573 case LOV_USER_MAGIC_V3:
574 lum_size = sizeof(struct lov_user_md_v3);
576 case LOV_USER_MAGIC_COMP_V1:
577 lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
580 if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
581 lustre_swab_lmv_user_md(
582 (struct lmv_user_md *)lump);
583 lum_size = sizeof(struct lmv_user_md);
585 case LOV_USER_MAGIC_SPECIFIC: {
586 struct lov_user_md_v3 *v3 =
587 (struct lov_user_md_v3 *)lump;
588 if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
590 lum_size = lov_user_md_size(v3->lmm_stripe_count,
591 LOV_USER_MAGIC_SPECIFIC);
595 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
596 " %#08x != %#08x nor %#08x\n",
597 lump->lmm_magic, LOV_USER_MAGIC_V1,
603 * This is coming from userspace, so should be in
604 * local endian. But the MDS would like it in little
605 * endian, so we swab it before we send it.
607 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
608 le32_to_cpu(LOV_MAGIC_MAGIC))
609 lustre_swab_lov_user_md(lump, 0);
611 lum_size = sizeof(struct lov_user_md_v1);
614 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
615 LUSTRE_OPC_ANY, NULL);
617 RETURN(PTR_ERR(op_data));
619 /* swabbing is done in lov_setstripe() on server side */
620 rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
621 ll_finish_md_op_data(op_data);
622 ptlrpc_req_finished(req);
626 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 13, 53, 0)
628 * 2.9 server has stored filesystem default stripe in ROOT xattr,
629 * and it's stored into system config for backward compatibility.
631 * In the following we use the fact that LOV_USER_MAGIC_V1 and
632 * LOV_USER_MAGIC_V3 have the same initial fields so we do not
633 * need the make the distiction between the 2 versions
635 if (set_default && mgc->u.cli.cl_mgc_mgsexp &&
637 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V1 ||
638 le32_to_cpu(lump->lmm_magic) == LOV_USER_MAGIC_V3)) {
642 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
644 GOTO(end, rc = -ENOMEM);
647 /* Get fsname and assume devname to be -MDT0000. */
648 snprintf(buf, MGS_PARAM_MAXLEN, "%s-MDT0000.lov",
652 /* Set root stripesize */
653 snprintf(buf, MGS_PARAM_MAXLEN, ".stripesize=%u",
654 lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
655 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
659 /* Set root stripecount */
660 snprintf(buf, MGS_PARAM_MAXLEN, ".stripecount=%hd",
661 lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
662 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
666 /* Set root stripeoffset */
667 snprintf(buf, MGS_PARAM_MAXLEN, ".stripeoffset=%hd",
668 lump ? le16_to_cpu(lump->lmm_stripe_offset) :
669 (typeof(lump->lmm_stripe_offset))(-1));
670 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
674 OBD_FREE(param, MGS_PARAM_MAXLEN);
680 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
682 struct ptlrpc_request **request, u64 valid,
683 enum get_default_layout_type type)
685 struct ll_sb_info *sbi = ll_i2sbi(inode);
686 struct mdt_body *body;
687 struct lov_mds_md *lmm = NULL;
688 struct ptlrpc_request *req = NULL;
690 struct md_op_data *op_data;
694 rc = ll_get_default_mdsize(sbi, &lmm_size);
698 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
699 0, lmm_size, LUSTRE_OPC_ANY,
702 RETURN(PTR_ERR(op_data));
704 op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
706 if (type == GET_DEFAULT_LAYOUT_ROOT) {
707 lu_root_fid(&op_data->op_fid1);
708 fid = op_data->op_fid1;
710 fid = *ll_inode2fid(inode);
713 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
714 ll_finish_md_op_data(op_data);
716 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
721 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
722 LASSERT(body != NULL);
724 lmm_size = body->mbo_eadatasize;
726 if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
728 GOTO(out, rc = -ENODATA);
731 lmm = req_capsule_server_sized_get(&req->rq_pill,
732 &RMF_MDT_MD, lmm_size);
733 LASSERT(lmm != NULL);
736 * This is coming from the MDS, so is probably in
737 * little endian. We convert it to host endian before
738 * passing it to userspace.
740 /* We don't swab objects for directories */
741 switch (le32_to_cpu(lmm->lmm_magic)) {
744 case LOV_MAGIC_COMP_V1:
745 case LOV_USER_MAGIC_SPECIFIC:
746 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
747 lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
750 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
751 lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
754 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
755 lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
757 case LMV_MAGIC_FOREIGN: {
758 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
760 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
761 __swab32s(&lfm->lfm_magic);
762 __swab32s(&lfm->lfm_length);
763 __swab32s(&lfm->lfm_type);
764 __swab32s(&lfm->lfm_flags);
769 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
774 *plmm_size = lmm_size;
780 * This function will be used to get default LOV/LMV/Default LMV
781 * @valid will be used to indicate which stripe it will retrieve.
782 * If the directory does not have its own default layout, then the
783 * function will request the default layout from root FID.
784 * OBD_MD_MEA LMV stripe EA
785 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
786 * otherwise Default LOV EA.
787 * Each time, it can only retrieve 1 stripe EA
789 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
790 struct ptlrpc_request **request,
791 struct ptlrpc_request **root_request,
794 struct ptlrpc_request *req = NULL;
795 struct ptlrpc_request *root_req = NULL;
796 struct lov_mds_md *lmm = NULL;
801 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
803 if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
804 !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL)
805 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
807 GET_DEFAULT_LAYOUT_ROOT);
810 *plmm_size = lmm_size;
812 if (root_request != NULL)
813 *root_request = root_req;
819 * This function will be used to get default LOV/LMV/Default LMV
820 * @valid will be used to indicate which stripe it will retrieve
821 * OBD_MD_MEA LMV stripe EA
822 * OBD_MD_DEFAULT_MEA Default LMV stripe EA
823 * otherwise Default LOV EA.
824 * Each time, it can only retrieve 1 stripe EA
826 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
827 struct ptlrpc_request **request, u64 valid)
829 struct ptlrpc_request *req = NULL;
830 struct lov_mds_md *lmm = NULL;
835 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
839 *plmm_size = lmm_size;
845 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
847 struct md_op_data *op_data;
852 OBD_ALLOC_PTR(op_data);
856 op_data->op_flags |= MF_GET_MDT_IDX;
857 op_data->op_fid1 = *fid;
858 rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
859 mdt_index = op_data->op_mds;
860 OBD_FREE_PTR(op_data);
868 * Get MDT index for the inode.
870 int ll_get_mdt_idx(struct inode *inode)
872 return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
876 * Generic handler to do any pre-copy work.
878 * It sends a first hsm_progress (with extent length == 0) to coordinator as a
879 * first information for it that real work has started.
881 * Moreover, for a ARCHIVE request, it will sample the file data version and
882 * store it in \a copy.
884 * \return 0 on success.
886 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
888 struct ll_sb_info *sbi = ll_s2sbi(sb);
889 struct hsm_progress_kernel hpk;
894 /* Forge a hsm_progress based on data from copy. */
895 hpk.hpk_fid = copy->hc_hai.hai_fid;
896 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
897 hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
898 hpk.hpk_extent.length = 0;
901 hpk.hpk_data_version = 0;
904 /* For archive request, we need to read the current file version. */
905 if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
907 __u64 data_version = 0;
909 /* Get inode for this fid */
910 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
912 hpk.hpk_flags |= HP_FLAG_RETRY;
913 /* hpk_errval is >= 0 */
914 hpk.hpk_errval = -PTR_ERR(inode);
915 GOTO(progress, rc = PTR_ERR(inode));
918 /* Read current file data version */
919 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
922 CDEBUG(D_HSM, "Could not read file data version of "
923 DFID" (rc = %d). Archive request ("
924 "%#llx) could not be done.\n",
925 PFID(©->hc_hai.hai_fid), rc,
926 copy->hc_hai.hai_cookie);
927 hpk.hpk_flags |= HP_FLAG_RETRY;
928 /* hpk_errval must be >= 0 */
929 hpk.hpk_errval = -rc;
933 /* Store in the hsm_copy for later copytool use.
934 * Always modified even if no lsm. */
935 copy->hc_data_version = data_version;
939 /* On error, the request should be considered as completed */
940 if (hpk.hpk_errval > 0)
941 hpk.hpk_flags |= HP_FLAG_COMPLETED;
943 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
946 /* Return first error */
947 RETURN(rc != 0 ? rc : rc2);
951 * Generic handler to do any post-copy work.
953 * It will send the last hsm_progress update to coordinator to inform it
954 * that copy is finished and whether it was successful or not.
957 * - for ARCHIVE request, it will sample the file data version and compare it
958 * with the version saved in ll_ioc_copy_start(). If they do not match, copy
959 * will be considered as failed.
960 * - for RESTORE request, it will sample the file data version and send it to
961 * coordinator which is useful if the file was imported as 'released'.
963 * \return 0 on success.
965 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
967 struct ll_sb_info *sbi = ll_s2sbi(sb);
968 struct hsm_progress_kernel hpk;
973 /* If you modify the logic here, also check llapi_hsm_copy_end(). */
974 /* Take care: copy->hc_hai.hai_action, len, gid and data are not
975 * initialized if copy_end was called with copy == NULL.
978 /* Forge a hsm_progress based on data from copy. */
979 hpk.hpk_fid = copy->hc_hai.hai_fid;
980 hpk.hpk_cookie = copy->hc_hai.hai_cookie;
981 hpk.hpk_extent = copy->hc_hai.hai_extent;
982 hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
983 hpk.hpk_errval = copy->hc_errval;
984 hpk.hpk_data_version = 0;
986 /* For archive request, we need to check the file data was not changed.
988 * For restore request, we need to send the file data version, this is
989 * useful when the file was created using hsm_import.
991 if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
992 (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
993 (copy->hc_errval == 0)) {
995 __u64 data_version = 0;
997 /* Get lsm for this fid */
998 inode = search_inode_for_lustre(sb, ©->hc_hai.hai_fid);
1000 hpk.hpk_flags |= HP_FLAG_RETRY;
1001 /* hpk_errval must be >= 0 */
1002 hpk.hpk_errval = -PTR_ERR(inode);
1003 GOTO(progress, rc = PTR_ERR(inode));
1006 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
1009 CDEBUG(D_HSM, "Could not read file data version. "
1010 "Request could not be confirmed.\n");
1011 if (hpk.hpk_errval == 0)
1012 hpk.hpk_errval = -rc;
1016 /* Store in the hsm_copy for later copytool use.
1017 * Always modified even if no lsm. */
1018 hpk.hpk_data_version = data_version;
1020 /* File could have been stripped during archiving, so we need
1021 * to check anyway. */
1022 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1023 (copy->hc_data_version != data_version)) {
1024 CDEBUG(D_HSM, "File data version mismatched. "
1025 "File content was changed during archiving. "
1026 DFID", start:%#llx current:%#llx\n",
1027 PFID(©->hc_hai.hai_fid),
1028 copy->hc_data_version, data_version);
1029 /* File was changed, send error to cdt. Do not ask for
1030 * retry because if a file is modified frequently,
1031 * the cdt will loop on retried archive requests.
1032 * The policy engine will ask for a new archive later
1033 * when the file will not be modified for some tunable
1035 hpk.hpk_flags &= ~HP_FLAG_RETRY;
1037 /* hpk_errval must be >= 0 */
1038 hpk.hpk_errval = -rc;
1045 rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1048 /* Return first error */
1049 RETURN(rc != 0 ? rc : rc2);
1053 static int copy_and_ct_start(int cmd, struct obd_export *exp,
1054 const struct lustre_kernelcomm __user *data)
1056 struct lustre_kernelcomm *lk;
1057 struct lustre_kernelcomm *tmp;
1058 size_t size = sizeof(*lk);
1063 /* copy data from userspace to get numbers of archive_id */
1064 OBD_ALLOC(lk, size);
1068 if (copy_from_user(lk, data, size))
1069 GOTO(out_lk, rc = -EFAULT);
1071 if (lk->lk_flags & LK_FLG_STOP)
1074 if (!(lk->lk_flags & LK_FLG_DATANR)) {
1075 __u32 archive_mask = lk->lk_data_count;
1078 /* old hsm agent to old MDS */
1079 if (!exp_connect_archive_id_array(exp))
1082 /* old hsm agent to new MDS */
1083 lk->lk_flags |= LK_FLG_DATANR;
1085 if (archive_mask == 0)
1088 count = hweight32(archive_mask);
1089 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1090 OBD_ALLOC(tmp, new_size);
1092 GOTO(out_lk, rc = -ENOMEM);
1094 memcpy(tmp, lk, size);
1095 tmp->lk_data_count = count;
1101 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1102 if ((1 << i) & archive_mask) {
1103 lk->lk_data[count] = i + 1;
1110 /* new hsm agent to new mds */
1111 if (lk->lk_data_count > 0) {
1112 new_size = offsetof(struct lustre_kernelcomm,
1113 lk_data[lk->lk_data_count]);
1114 OBD_ALLOC(tmp, new_size);
1116 GOTO(out_lk, rc = -ENOMEM);
1122 if (copy_from_user(lk, data, size))
1123 GOTO(out_lk, rc = -EFAULT);
1126 /* new hsm agent to old MDS */
1127 if (!exp_connect_archive_id_array(exp)) {
1130 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1131 GOTO(out_lk, rc = -EINVAL);
1133 for (i = 0; i < lk->lk_data_count; i++) {
1134 if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1136 CERROR("%s: archive id %d requested but only "
1137 "[0 - %zu] supported: rc = %d\n",
1138 exp->exp_obd->obd_name, lk->lk_data[i],
1139 LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1143 if (lk->lk_data[i] == 0) {
1148 archives |= (1 << (lk->lk_data[i] - 1));
1150 lk->lk_flags &= ~LK_FLG_DATANR;
1151 lk->lk_data_count = archives;
1154 rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1160 static int check_owner(int type, int id)
1164 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1168 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1177 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1179 int cmd = qctl->qc_cmd;
1180 int type = qctl->qc_type;
1181 int id = qctl->qc_id;
1182 int valid = qctl->qc_valid;
1189 case LUSTRE_Q_SETDEFAULT:
1190 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1194 case LUSTRE_Q_GETDEFAULT:
1195 if (check_owner(type, id) &&
1196 (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1202 CERROR("unsupported quotactl op: %#x\n", cmd);
1206 if (valid != QC_GENERAL) {
1207 if (cmd == Q_GETINFO)
1208 qctl->qc_cmd = Q_GETOINFO;
1209 else if (cmd == Q_GETQUOTA)
1210 qctl->qc_cmd = Q_GETOQUOTA;
1216 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1217 sizeof(*qctl), qctl, NULL);
1220 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1221 sizeof(*qctl), qctl, NULL);
1224 rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1225 sizeof(*qctl), qctl, NULL);
1227 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1229 sizeof(*qctl), qctl, NULL);
1241 struct obd_quotactl *oqctl;
1243 OBD_ALLOC_PTR(oqctl);
1247 QCTL_COPY(oqctl, qctl);
1248 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1250 OBD_FREE_PTR(oqctl);
1253 /* If QIF_SPACE is not set, client should collect the
1254 * space usage from OSSs by itself */
1255 if (cmd == Q_GETQUOTA &&
1256 !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1257 !oqctl->qc_dqblk.dqb_curspace) {
1258 struct obd_quotactl *oqctl_tmp;
1260 OBD_ALLOC_PTR(oqctl_tmp);
1261 if (oqctl_tmp == NULL)
1262 GOTO(out, rc = -ENOMEM);
1264 oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1265 oqctl_tmp->qc_id = oqctl->qc_id;
1266 oqctl_tmp->qc_type = oqctl->qc_type;
1268 /* collect space usage from OSTs */
1269 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1270 rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1271 if (!rc || rc == -EREMOTEIO) {
1272 oqctl->qc_dqblk.dqb_curspace =
1273 oqctl_tmp->qc_dqblk.dqb_curspace;
1274 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1277 /* collect space & inode usage from MDTs */
1278 oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1279 oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1280 rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1281 if (!rc || rc == -EREMOTEIO) {
1282 oqctl->qc_dqblk.dqb_curspace +=
1283 oqctl_tmp->qc_dqblk.dqb_curspace;
1284 oqctl->qc_dqblk.dqb_curinodes =
1285 oqctl_tmp->qc_dqblk.dqb_curinodes;
1286 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1288 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1291 OBD_FREE_PTR(oqctl_tmp);
1294 QCTL_COPY(qctl, oqctl);
1295 OBD_FREE_PTR(oqctl);
1301 int ll_rmfid(struct file *file, void __user *arg)
1303 const struct fid_array __user *ufa = arg;
1304 struct fid_array *lfa = NULL;
1307 int i, rc, *rcs = NULL;
1310 if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1311 !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1313 /* Only need to get the buflen */
1314 if (get_user(nr, &ufa->fa_nr))
1316 /* DoS protection */
1317 if (nr > OBD_MAX_FIDS_IN_ARRAY)
1320 size = offsetof(struct fid_array, fa_fids[nr]);
1321 OBD_ALLOC(lfa, size);
1324 OBD_ALLOC(rcs, sizeof(int) * nr);
1326 GOTO(free_lfa, rc = -ENOMEM);
1328 if (copy_from_user(lfa, arg, size))
1329 GOTO(free_rcs, rc = -EFAULT);
1331 /* Call mdc_iocontrol */
1332 rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1334 for (i = 0; i < nr; i++)
1336 lfa->fa_fids[i].f_ver = rcs[i];
1337 if (copy_to_user(arg, lfa, size))
1342 OBD_FREE(rcs, sizeof(int) * nr);
1344 OBD_FREE(lfa, size);
1349 /* This function tries to get a single name component,
1350 * to send to the server. No actual path traversal involved,
1351 * so we limit to NAME_MAX */
1352 static char *ll_getname(const char __user *filename)
1357 OBD_ALLOC(tmp, NAME_MAX + 1);
1360 return ERR_PTR(-ENOMEM);
1362 len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1365 else if (len > NAME_MAX)
1366 ret = -ENAMETOOLONG;
1369 OBD_FREE(tmp, NAME_MAX + 1);
1375 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1377 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1379 struct dentry *dentry = file_dentry(file);
1380 struct inode *inode = file_inode(file);
1381 struct ll_sb_info *sbi = ll_i2sbi(inode);
1382 struct obd_ioctl_data *data;
1386 CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1387 PFID(ll_inode2fid(inode)), inode, cmd);
1389 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1390 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1393 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1395 case FS_IOC_GETFLAGS:
1396 case FS_IOC_SETFLAGS:
1397 RETURN(ll_iocontrol(inode, file, cmd, arg));
1398 case FSFILT_IOC_GETVERSION:
1399 case FS_IOC_GETVERSION:
1400 RETURN(put_user(inode->i_generation, (int __user *)arg));
1401 /* We need to special case any other ioctls we want to handle,
1402 * to send them to the MDS/OST as appropriate and to properly
1403 * network encode the arg field. */
1404 case FS_IOC_SETVERSION:
1407 case LL_IOC_GET_MDTIDX: {
1410 mdtidx = ll_get_mdt_idx(inode);
1414 if (put_user((int)mdtidx, (int __user *)arg))
1419 case IOC_MDC_LOOKUP: {
1420 int namelen, len = 0;
1424 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1429 filename = data->ioc_inlbuf1;
1430 namelen = strlen(filename);
1432 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1433 GOTO(out_free, rc = -EINVAL);
1436 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1438 CERROR("%s: lookup %.*s failed: rc = %d\n",
1439 sbi->ll_fsname, namelen, filename, rc);
1443 OBD_FREE_LARGE(buf, len);
1446 case LL_IOC_LMV_SETSTRIPE: {
1447 struct lmv_user_md *lum;
1456 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1461 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1462 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1463 GOTO(lmv_out_free, rc = -EINVAL);
1465 filename = data->ioc_inlbuf1;
1466 namelen = data->ioc_inllen1;
1469 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1470 GOTO(lmv_out_free, rc = -EINVAL);
1472 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1473 lumlen = data->ioc_inllen2;
1475 if (!lmv_magic_supported(lum->lum_magic)) {
1476 CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1477 lum->lum_magic, -EINVAL);
1478 GOTO(lmv_out_free, rc = -EINVAL);
1481 if ((lum->lum_magic == LMV_USER_MAGIC ||
1482 lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1483 lumlen < sizeof(*lum)) {
1484 CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1485 filename, lumlen, lum->lum_magic, -EINVAL);
1486 GOTO(lmv_out_free, rc = -EINVAL);
1489 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1490 lumlen < sizeof(struct lmv_foreign_md)) {
1491 CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1492 filename, lum->lum_magic, lumlen, -EFAULT);
1493 GOTO(lmv_out_free, rc = -EINVAL);
1496 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1497 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1499 mode = data->ioc_type;
1501 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1503 OBD_FREE_LARGE(buf, len);
1507 case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1508 struct lmv_user_md lum;
1509 struct lmv_user_md __user *ulump =
1510 (struct lmv_user_md __user *)arg;
1513 if (copy_from_user(&lum, ulump, sizeof(lum)))
1516 if (lum.lum_magic != LMV_USER_MAGIC)
1519 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1523 case LL_IOC_LOV_SETSTRIPE_NEW:
1524 case LL_IOC_LOV_SETSTRIPE: {
1525 struct lov_user_md_v3 *lumv3 = NULL;
1526 struct lov_user_md_v1 lumv1;
1527 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1528 struct lov_user_md_v1 __user *lumv1p =
1529 (struct lov_user_md_v1 __user *)arg;
1530 struct lov_user_md_v3 __user *lumv3p =
1531 (struct lov_user_md_v3 __user *)arg;
1534 int set_default = 0;
1536 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1537 sizeof(struct lov_comp_md_v1));
1538 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1539 /* first try with v1 which is smaller than v3 */
1540 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1543 if (inode->i_sb->s_root == file_dentry(file))
1546 switch (lumv1.lmm_magic) {
1547 case LOV_USER_MAGIC_V3:
1548 case LOV_USER_MAGIC_SPECIFIC:
1549 lum_size = ll_lov_user_md_size(&lumv1);
1552 OBD_ALLOC(lumv3, lum_size);
1555 if (copy_from_user(lumv3, lumv3p, lum_size))
1556 GOTO(out, rc = -EFAULT);
1557 lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1559 case LOV_USER_MAGIC_V1:
1562 GOTO(out, rc = -ENOTSUPP);
1565 /* in v1 and v3 cases lumv1 points to data */
1566 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1569 OBD_FREE(lumv3, lum_size);
1572 case LL_IOC_LMV_GETSTRIPE: {
1573 struct lmv_user_md __user *ulmv =
1574 (struct lmv_user_md __user *)arg;
1575 struct lmv_user_md lum;
1576 struct ptlrpc_request *request = NULL;
1577 struct ptlrpc_request *root_request = NULL;
1578 union lmv_mds_md *lmm = NULL;
1581 struct lmv_user_md *tmp = NULL;
1585 int max_stripe_count;
1589 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1592 max_stripe_count = lum.lum_stripe_count;
1593 /* lum_magic will indicate which stripe the ioctl will like
1594 * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1595 * is for default LMV stripe */
1596 if (lum.lum_magic == LMV_MAGIC_V1)
1597 valid |= OBD_MD_MEA;
1598 else if (lum.lum_magic == LMV_USER_MAGIC)
1599 valid |= OBD_MD_DEFAULT_MEA;
1603 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1604 &request, &root_request, valid);
1606 GOTO(finish_req, rc);
1608 /* Get default LMV EA */
1609 if (lum.lum_magic == LMV_USER_MAGIC) {
1610 if (lmmsize > sizeof(*ulmv))
1611 GOTO(finish_req, rc = -EINVAL);
1613 if (copy_to_user(ulmv, lmm, lmmsize))
1614 GOTO(finish_req, rc = -EFAULT);
1616 GOTO(finish_req, rc);
1619 /* if foreign LMV case, fake stripes number */
1620 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1621 struct lmv_foreign_md *lfm;
1623 lfm = (struct lmv_foreign_md *)lmm;
1624 if (lfm->lfm_length < XATTR_SIZE_MAX -
1625 offsetof(typeof(*lfm), lfm_value)) {
1626 __u32 size = lfm->lfm_length +
1627 offsetof(typeof(*lfm), lfm_value);
1629 stripe_count = lmv_foreign_to_md_stripes(size);
1631 CERROR("invalid %d foreign size returned\n",
1636 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1638 if (max_stripe_count < stripe_count) {
1639 lum.lum_stripe_count = stripe_count;
1640 if (copy_to_user(ulmv, &lum, sizeof(lum)))
1641 GOTO(finish_req, rc = -EFAULT);
1642 GOTO(finish_req, rc = -E2BIG);
1645 /* enough room on user side and foreign case */
1646 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1647 struct lmv_foreign_md *lfm;
1650 lfm = (struct lmv_foreign_md *)lmm;
1651 size = lfm->lfm_length +
1652 offsetof(struct lmv_foreign_md, lfm_value);
1653 if (copy_to_user(ulmv, lfm, size))
1654 GOTO(finish_req, rc = -EFAULT);
1655 GOTO(finish_req, rc);
1658 lum_size = lmv_user_md_size(stripe_count,
1659 LMV_USER_MAGIC_SPECIFIC);
1660 OBD_ALLOC(tmp, lum_size);
1662 GOTO(finish_req, rc = -ENOMEM);
1664 mdt_index = ll_get_mdt_idx(inode);
1666 GOTO(out_tmp, rc = -ENOMEM);
1668 tmp->lum_magic = LMV_MAGIC_V1;
1669 tmp->lum_stripe_count = 0;
1670 tmp->lum_stripe_offset = mdt_index;
1671 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1672 for (i = 0; i < stripe_count; i++) {
1675 fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1676 if (fid_is_sane(&fid)) {
1677 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1679 GOTO(out_tmp, rc = mdt_index);
1681 tmp->lum_objects[i].lum_mds = mdt_index;
1682 tmp->lum_objects[i].lum_fid = fid;
1685 tmp->lum_stripe_count++;
1688 if (copy_to_user(ulmv, tmp, lum_size))
1689 GOTO(out_tmp, rc = -EFAULT);
1691 OBD_FREE(tmp, lum_size);
1693 ptlrpc_req_finished(request);
1694 ptlrpc_req_finished(root_request);
1698 case LL_IOC_REMOVE_ENTRY: {
1699 char *filename = NULL;
1703 /* Here is a little hack to avoid sending REINT_RMENTRY to
1704 * unsupported server, which might crash the server(LU-2730),
1705 * Because both LVB_TYPE and REINT_RMENTRY will be supported
1706 * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1707 * server will support REINT_RMENTRY XXX*/
1708 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1711 filename = ll_getname((const char __user *)arg);
1712 if (IS_ERR(filename))
1713 RETURN(PTR_ERR(filename));
1715 namelen = strlen(filename);
1717 GOTO(out_rmdir, rc = -EINVAL);
1719 rc = ll_rmdir_entry(inode, filename, namelen);
1722 ll_putname(filename);
1726 RETURN(ll_rmfid(file, (void __user *)arg));
1727 case LL_IOC_LOV_SWAP_LAYOUTS:
1729 case IOC_OBD_STATFS:
1730 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1731 case LL_IOC_LOV_GETSTRIPE:
1732 case LL_IOC_LOV_GETSTRIPE_NEW:
1733 case LL_IOC_MDC_GETINFO:
1734 case LL_IOC_MDC_GETINFO_OLD:
1735 case IOC_MDC_GETFILEINFO:
1736 case IOC_MDC_GETFILEINFO_OLD:
1737 case IOC_MDC_GETFILESTRIPE: {
1738 struct ptlrpc_request *request = NULL;
1739 struct ptlrpc_request *root_request = NULL;
1740 struct lov_user_md __user *lump;
1741 struct lov_mds_md *lmm = NULL;
1742 struct mdt_body *body;
1743 char *filename = NULL;
1744 lstat_t __user *statp = NULL;
1745 lstatx_t __user *stxp = NULL;
1746 __u64 __user *flagsp = NULL;
1747 __u32 __user *lmmsizep = NULL;
1748 struct lu_fid __user *fidp = NULL;
1751 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1752 cmd == IOC_MDC_GETFILEINFO ||
1753 cmd == IOC_MDC_GETFILESTRIPE) {
1754 filename = ll_getname((const char __user *)arg);
1755 if (IS_ERR(filename))
1756 RETURN(PTR_ERR(filename));
1758 rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1759 &lmmsize, &request);
1761 rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1767 body = req_capsule_server_get(&request->rq_pill,
1769 LASSERT(body != NULL);
1774 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1775 cmd == LL_IOC_MDC_GETINFO ||
1776 cmd == IOC_MDC_GETFILEINFO_OLD ||
1777 cmd == LL_IOC_MDC_GETINFO_OLD)) {
1785 if (cmd == IOC_MDC_GETFILESTRIPE ||
1786 cmd == LL_IOC_LOV_GETSTRIPE ||
1787 cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1788 lump = (struct lov_user_md __user *)arg;
1789 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1790 cmd == LL_IOC_MDC_GETINFO_OLD){
1791 struct lov_user_mds_data_v1 __user *lmdp;
1793 lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1794 statp = &lmdp->lmd_st;
1795 lump = &lmdp->lmd_lmm;
1797 struct lov_user_mds_data __user *lmdp;
1799 lmdp = (struct lov_user_mds_data __user *)arg;
1800 fidp = &lmdp->lmd_fid;
1801 stxp = &lmdp->lmd_stx;
1802 flagsp = &lmdp->lmd_flags;
1803 lmmsizep = &lmdp->lmd_lmmsize;
1804 lump = &lmdp->lmd_lmm;
1808 /* If the file has no striping then zero out *lump so
1809 * that the caller isn't confused by garbage. */
1810 if (clear_user(lump, sizeof(*lump)))
1811 GOTO(out_req, rc = -EFAULT);
1812 } else if (copy_to_user(lump, lmm, lmmsize)) {
1813 if (copy_to_user(lump, lmm, sizeof(*lump)))
1814 GOTO(out_req, rc = -EFAULT);
1818 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1819 cmd == LL_IOC_MDC_GETINFO_OLD) {
1822 st.st_dev = inode->i_sb->s_dev;
1823 st.st_mode = body->mbo_mode;
1824 st.st_nlink = body->mbo_nlink;
1825 st.st_uid = body->mbo_uid;
1826 st.st_gid = body->mbo_gid;
1827 st.st_rdev = body->mbo_rdev;
1828 st.st_size = body->mbo_size;
1829 st.st_blksize = PAGE_SIZE;
1830 st.st_blocks = body->mbo_blocks;
1831 st.st_atime = body->mbo_atime;
1832 st.st_mtime = body->mbo_mtime;
1833 st.st_ctime = body->mbo_ctime;
1834 st.st_ino = cl_fid_build_ino(&body->mbo_fid1,
1838 if (copy_to_user(statp, &st, sizeof(st)))
1839 GOTO(out_req, rc = -EFAULT);
1840 } else if (cmd == IOC_MDC_GETFILEINFO ||
1841 cmd == LL_IOC_MDC_GETINFO) {
1842 lstatx_t stx = { 0 };
1843 __u64 valid = body->mbo_valid;
1845 stx.stx_blksize = PAGE_SIZE;
1846 stx.stx_nlink = body->mbo_nlink;
1847 stx.stx_uid = body->mbo_uid;
1848 stx.stx_gid = body->mbo_gid;
1849 stx.stx_mode = body->mbo_mode;
1850 stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1853 stx.stx_size = body->mbo_size;
1854 stx.stx_blocks = body->mbo_blocks;
1855 stx.stx_atime.tv_sec = body->mbo_atime;
1856 stx.stx_ctime.tv_sec = body->mbo_ctime;
1857 stx.stx_mtime.tv_sec = body->mbo_mtime;
1858 stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1859 stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1860 stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1861 stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1862 stx.stx_mask |= STATX_BASIC_STATS;
1865 * For a striped directory, the size and blocks returned
1866 * from MDT is not correct.
1867 * The size and blocks are aggregated by client across
1869 * Thus for a striped directory, do not return the valid
1870 * FLSIZE and FLBLOCKS flags to the caller.
1871 * However, this whould be better decided by the MDS
1872 * instead of the client.
1874 if (cmd == LL_IOC_MDC_GETINFO &&
1875 ll_i2info(inode)->lli_lsm_md != NULL)
1876 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1878 if (flagsp && copy_to_user(flagsp, &valid,
1880 GOTO(out_req, rc = -EFAULT);
1882 if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1884 GOTO(out_req, rc = -EFAULT);
1886 if (!(valid & OBD_MD_FLSIZE))
1887 stx.stx_mask &= ~STATX_SIZE;
1888 if (!(valid & OBD_MD_FLBLOCKS))
1889 stx.stx_mask &= ~STATX_BLOCKS;
1891 if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1892 GOTO(out_req, rc = -EFAULT);
1894 if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1896 GOTO(out_req, rc = -EFAULT);
1901 ptlrpc_req_finished(request);
1902 ptlrpc_req_finished(root_request);
1904 ll_putname(filename);
1907 case OBD_IOC_QUOTACTL: {
1908 struct if_quotactl *qctl;
1910 OBD_ALLOC_PTR(qctl);
1914 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1915 GOTO(out_quotactl, rc = -EFAULT);
1917 rc = quotactl_ioctl(sbi, qctl);
1920 copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1927 case OBD_IOC_GETDTNAME:
1928 case OBD_IOC_GETMDNAME:
1929 RETURN(ll_get_obd_name(inode, cmd, arg));
1930 case LL_IOC_FLUSHCTX:
1931 RETURN(ll_flush_ctx(inode));
1932 case LL_IOC_GETOBDCOUNT: {
1934 struct obd_export *exp;
1936 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1939 /* get ost count when count is zero, get mdt count otherwise */
1940 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1941 vallen = sizeof(count);
1942 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1943 KEY_TGT_COUNT, &vallen, &count);
1945 CERROR("get target count failed: %d\n", rc);
1949 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1954 case LL_IOC_PATH2FID:
1955 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1956 sizeof(struct lu_fid)))
1959 case LL_IOC_GET_CONNECT_FLAGS: {
1960 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1961 (void __user *)arg));
1963 case OBD_IOC_FID2PATH:
1964 RETURN(ll_fid2path(inode, (void __user *)arg));
1965 case LL_IOC_GETPARENT:
1966 RETURN(ll_getparent(file, (void __user *)arg));
1967 case LL_IOC_FID2MDTIDX: {
1968 struct obd_export *exp = ll_i2mdexp(inode);
1972 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1976 /* Call mdc_iocontrol */
1977 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1978 (__u32 __user *)&index);
1984 case LL_IOC_HSM_REQUEST: {
1985 struct hsm_user_request *hur;
1992 /* We don't know the true size yet; copy the fixed-size part */
1993 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1998 /* Compute the whole struct size */
1999 totalsize = hur_len(hur);
2004 /* Final size will be more than double totalsize */
2005 if (totalsize >= MDS_MAXREQSIZE / 3)
2008 OBD_ALLOC_LARGE(hur, totalsize);
2012 /* Copy the whole struct */
2013 if (copy_from_user(hur, (void __user *)arg, totalsize))
2014 GOTO(out_hur, rc = -EFAULT);
2016 if (hur->hur_request.hr_action == HUA_RELEASE) {
2017 const struct lu_fid *fid;
2021 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
2022 fid = &hur->hur_user_item[i].hui_fid;
2023 f = search_inode_for_lustre(inode->i_sb, fid);
2029 rc = ll_hsm_release(f);
2035 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
2040 OBD_FREE_LARGE(hur, totalsize);
2044 case LL_IOC_HSM_PROGRESS: {
2045 struct hsm_progress_kernel hpk;
2046 struct hsm_progress hp;
2048 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2051 hpk.hpk_fid = hp.hp_fid;
2052 hpk.hpk_cookie = hp.hp_cookie;
2053 hpk.hpk_extent = hp.hp_extent;
2054 hpk.hpk_flags = hp.hp_flags;
2055 hpk.hpk_errval = hp.hp_errval;
2056 hpk.hpk_data_version = 0;
2058 /* File may not exist in Lustre; all progress
2059 * reported to Lustre root */
2060 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2064 case LL_IOC_HSM_CT_START:
2065 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2068 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2069 (struct lustre_kernelcomm __user *)arg);
2072 case LL_IOC_HSM_COPY_START: {
2073 struct hsm_copy *copy;
2076 OBD_ALLOC_PTR(copy);
2079 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2084 rc = ll_ioc_copy_start(inode->i_sb, copy);
2085 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2091 case LL_IOC_HSM_COPY_END: {
2092 struct hsm_copy *copy;
2095 OBD_ALLOC_PTR(copy);
2098 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2103 rc = ll_ioc_copy_end(inode->i_sb, copy);
2104 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2110 case LL_IOC_MIGRATE: {
2111 struct lmv_user_md *lum;
2118 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2122 data = (struct obd_ioctl_data *)buf;
2123 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2124 data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2125 GOTO(migrate_free, rc = -EINVAL);
2127 filename = data->ioc_inlbuf1;
2128 namelen = data->ioc_inllen1;
2130 if (namelen < 1 || namelen != strlen(filename) + 1) {
2131 CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2132 GOTO(migrate_free, rc = -EINVAL);
2135 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2136 if (lum->lum_magic != LMV_USER_MAGIC &&
2137 lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2139 CERROR("%s: wrong lum magic %x: rc = %d\n",
2140 filename, lum->lum_magic, rc);
2141 GOTO(migrate_free, rc);
2144 rc = ll_migrate(inode, file, lum, filename);
2146 OBD_FREE_LARGE(buf, len);
2150 case LL_IOC_FSGETXATTR:
2151 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2152 case LL_IOC_FSSETXATTR:
2153 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2154 case LL_IOC_PCC_DETACH_BY_FID: {
2155 struct lu_pcc_detach_fid *detach;
2157 struct inode *inode2;
2160 OBD_ALLOC_PTR(detach);
2164 if (copy_from_user(detach,
2165 (const struct lu_pcc_detach_fid __user *)arg,
2167 GOTO(out_detach, rc = -EFAULT);
2169 fid = &detach->pccd_fid;
2170 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2171 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2173 /* Target inode is not in inode cache, and PCC file
2174 * has aleady released, return immdiately.
2176 GOTO(out_detach, rc = 0);
2178 if (!S_ISREG(inode2->i_mode))
2179 GOTO(out_iput, rc = -EINVAL);
2181 if (!inode_owner_or_capable(inode2))
2182 GOTO(out_iput, rc = -EPERM);
2184 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2188 OBD_FREE_PTR(detach);
2192 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2193 (void __user *)arg));
2197 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2199 struct inode *inode = file->f_mapping->host;
2200 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2201 struct ll_sb_info *sbi = ll_i2sbi(inode);
2202 int api32 = ll_need_32bit_api(sbi);
2203 loff_t ret = -EINVAL;
2211 offset += file->f_pos;
2217 offset += LL_DIR_END_OFF_32BIT;
2219 offset += LL_DIR_END_OFF;
2226 ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2227 (!api32 && offset <= LL_DIR_END_OFF))) {
2228 if (offset != file->f_pos) {
2229 if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2230 (!api32 && offset == LL_DIR_END_OFF))
2231 fd->lfd_pos = MDS_DIR_END_OFF;
2232 else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2233 fd->lfd_pos = offset << 32;
2235 fd->lfd_pos = offset;
2236 file->f_pos = offset;
2237 file->f_version = 0;
2244 inode_unlock(inode);
2248 static int ll_dir_open(struct inode *inode, struct file *file)
2251 RETURN(ll_file_open(inode, file));
2254 static int ll_dir_release(struct inode *inode, struct file *file)
2257 RETURN(ll_file_release(inode, file));
2260 const struct file_operations ll_dir_operations = {
2261 .llseek = ll_dir_seek,
2262 .open = ll_dir_open,
2263 .release = ll_dir_release,
2264 .read = generic_read_dir,
2265 #ifdef HAVE_DIR_CONTEXT
2266 .iterate_shared = ll_iterate,
2268 .readdir = ll_readdir,
2270 .unlocked_ioctl = ll_dir_ioctl,