X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Ffile.c;h=d2c7bb2a8dc78a7f3bcd85eea20fb1791635bb65;hp=416fe42d70c9b10c208805f9e15618a087a28e36;hb=999a8f1f0f9ab1a47843b4fa24ff88e5afd894b2;hpb=6869932b552ac705f411de3362f01bd50c1f6f7d diff --git a/lustre/llite/file.c b/lustre/llite/file.c index 416fe42..d2c7bb2 100644 --- a/lustre/llite/file.c +++ b/lustre/llite/file.c @@ -26,7 +26,7 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* @@ -47,13 +47,15 @@ #include #include #include "llite_internal.h" +#include + +#include "cl_object.h" -/* also used by llite/special.c:ll_special_open() */ struct ll_file_data *ll_file_data_get(void) { struct ll_file_data *fd; - OBD_SLAB_ALLOC_PTR(fd, ll_file_data_slab); + OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO); return fd; } @@ -73,12 +75,18 @@ void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data, op_data->op_attr.ia_ctime = inode->i_ctime; op_data->op_attr.ia_size = i_size_read(inode); op_data->op_attr_blocks = inode->i_blocks; - ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = inode->i_flags; + ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = + ll_inode_to_ext_flags(inode->i_flags); op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch; - memcpy(&op_data->op_handle, fh, sizeof(op_data->op_handle)); + if (fh) + op_data->op_handle = *fh; op_data->op_capa1 = ll_mdscapa_get(inode); } +/** + * Closes the IO epoch and packs all the attributes into @op_data for + * the CLOSE rpc. + */ static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data, struct obd_client_handle *och) { @@ -90,14 +98,15 @@ static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data, if (!(och->och_flags & FMODE_WRITE)) goto out; - if (!(ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) || - !S_ISREG(inode->i_mode)) + if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode)) op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS; else - ll_epoch_close(inode, op_data, &och, 0); + ll_ioepoch_close(inode, op_data, &och, 0); out: ll_pack_inode2opdata(inode, op_data, &och->och_fh); + ll_prep_md_op_data(op_data, inode, NULL, NULL, + 0, 0, LUSTRE_OPC_ANY, NULL); EXIT; } @@ -110,7 +119,7 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp, struct ptlrpc_request *req = NULL; struct obd_device *obd = class_exp2obd(exp); int epoch_close = 1; - int seq_end = 0, rc; + int rc; ENTRY; if (obd == NULL) { @@ -123,14 +132,6 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp, GOTO(out, rc = 0); } - /* - * here we check if this is forced umount. If so this is called on - * canceling "open lock" and we do not call md_close() in this case, as - * it will not be successful, as import is already deactivated. - */ - if (obd->obd_force) - GOTO(out, rc = 0); - OBD_ALLOC_PTR(op_data); if (op_data == NULL) GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here. @@ -138,17 +139,12 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp, ll_prepare_close(inode, op_data, och); epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE); rc = md_close(md_exp, op_data, och->och_mod, &req); - if (rc != -EAGAIN) - seq_end = 1; - if (rc == -EAGAIN) { /* This close must have the epoch closed. */ - LASSERT(exp->exp_connect_flags & OBD_CONNECT_SOM); LASSERT(epoch_close); /* MDS has instructed us to obtain Size-on-MDS attribute from * OSTs and send setattr to back to MDS. */ - rc = ll_sizeonmds_update(inode, och->och_mod, - &och->och_fh, op_data->op_ioepoch); + rc = ll_som_update(inode, op_data); if (rc) { CERROR("inode %lu mdc Size-on-MDS update failed: " "rc = %d\n", inode->i_ino, rc); @@ -169,13 +165,11 @@ static int ll_close_inode_openhandle(struct obd_export *md_exp, EXIT; out: - - if ((exp->exp_connect_flags & OBD_CONNECT_SOM) && !epoch_close && + + if (exp_connect_som(exp) && !epoch_close && S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) { ll_queue_done_writing(inode, LLIF_DONE_WRITING); } else { - if (seq_end) - ptlrpc_close_replay_seq(req); md_clear_open_replay_data(md_exp, och); /* Free @och if it is not waiting for DONE_WRITING. */ och->och_fh.cookie = DEAD_HANDLE_MAGIC; @@ -207,15 +201,15 @@ int ll_md_real_close(struct inode *inode, int flags) och_usecount = &lli->lli_open_fd_read_count; } - down(&lli->lli_och_sem); + cfs_down(&lli->lli_och_sem); if (*och_usecount) { /* There are still users of this handle, so skip freeing it. */ - up(&lli->lli_och_sem); + cfs_up(&lli->lli_och_sem); RETURN(0); } och=*och_p; *och_p = NULL; - up(&lli->lli_och_sem); + cfs_up(&lli->lli_och_sem); if (och) { /* There might be a race and somebody have freed this och already */ @@ -235,12 +229,8 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, ENTRY; /* clear group lock, if present */ - if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; - fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK); - rc = ll_extent_unlock(fd, inode, lsm, LCK_GROUP, - &fd->fd_cwlockh); - } + if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) + ll_put_grouplock(inode, file, fd->fd_grouplock.cg_gid); /* Let's see if we have good enough OPEN lock on the file and if we can skip talking to MDS */ @@ -251,7 +241,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, struct inode *inode = file->f_dentry->d_inode; ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}}; - down(&lli->lli_och_sem); + cfs_down(&lli->lli_och_sem); if (fd->fd_omode & FMODE_WRITE) { lockmode = LCK_CW; LASSERT(lli->lli_open_fd_write_count); @@ -265,7 +255,7 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, LASSERT(lli->lli_open_fd_read_count); lli->lli_open_fd_read_count--; } - up(&lli->lli_och_sem); + cfs_up(&lli->lli_och_sem); if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), LDLM_IBITS, &policy, lockmode, @@ -299,8 +289,8 @@ int ll_file_release(struct inode *inode, struct file *file) struct ll_inode_info *lli = ll_i2info(inode); struct lov_stripe_md *lsm = lli->lli_smd; int rc; - ENTRY; + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, inode->i_generation, inode); @@ -326,20 +316,24 @@ int ll_file_release(struct inode *inode, struct file *file) /* The last ref on @file, maybe not the the owner pid of statahead. * Different processes can open the same dir, "ll_opendir_key" means: * it is me that should stop the statahead thread. */ - if (lli->lli_opendir_key == fd) - ll_stop_statahead(inode, fd); + if (lli->lli_opendir_key == fd && lli->lli_opendir_pid != 0) + ll_stop_statahead(inode, lli->lli_opendir_key); if (inode->i_sb->s_root == file->f_dentry) { LUSTRE_FPRIVATE(file) = NULL; ll_file_data_put(fd); RETURN(0); } - + if (lsm) lov_test_and_clear_async_rc(lsm); lli->lli_async_rc = 0; rc = ll_md_close(sbi->ll_md_exp, inode, file); + + if (OBD_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, obd_fail_val)) + libcfs_debug_dumplog(); + RETURN(rc); } @@ -383,11 +377,11 @@ static int ll_intent_file_open(struct file *file, void *lmm, /* reason for keep own exit path - don`t flood log * with messages with -ESTALE errors. */ - if (!it_disposition(itp, DISP_OPEN_OPEN) || + if (!it_disposition(itp, DISP_OPEN_OPEN) || it_open_error(DISP_OPEN_OPEN, itp)) GOTO(out, rc); ll_release_openhandle(file->f_dentry, itp); - GOTO(out_stale, rc); + GOTO(out, rc); } if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { @@ -396,22 +390,34 @@ static int ll_intent_file_open(struct file *file, void *lmm, GOTO(out, rc); } - if (itp->d.lustre.it_lock_mode) + rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL); + if (!rc && itp->d.lustre.it_lock_mode) md_set_lock_data(sbi->ll_md_exp, - &itp->d.lustre.it_lock_handle, - file->f_dentry->d_inode); + &itp->d.lustre.it_lock_handle, + file->f_dentry->d_inode, NULL); - rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL); out: ptlrpc_req_finished(itp->d.lustre.it_data); - -out_stale: it_clear_disposition(itp, DISP_ENQ_COMPLETE); ll_intent_drop_lock(itp); RETURN(rc); } +/** + * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does + * not believe attributes if a few ioepoch holders exist. Attributes for + * previous ioepoch if new one is opened are also skipped by MDS. + */ +void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch) +{ + if (ioepoch && lli->lli_ioepoch != ioepoch) { + lli->lli_ioepoch = ioepoch; + CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n", + ioepoch, PFID(&lli->lli_fid)); + } +} + static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli, struct lookup_intent *it, struct obd_client_handle *och) { @@ -427,7 +433,7 @@ static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli, och->och_magic = OBD_CLIENT_HANDLE_MAGIC; och->och_fid = lli->lli_fid; och->och_flags = it->it_flags; - lli->lli_ioepoch = body->ioepoch; + ll_ioepoch_open(lli, body->ioepoch); return md_set_open_replay_data(md_exp, och, req); } @@ -507,27 +513,16 @@ int ll_file_open(struct inode *inode, struct file *file) if (fd == NULL) RETURN(-ENOMEM); + fd->fd_file = file; if (S_ISDIR(inode->i_mode)) { - spin_lock(&lli->lli_lock); - /* "lli->lli_opendir_pid != 0" means someone has set it. - * "lli->lli_sai != NULL" means the previous statahead has not - * been cleanup. */ - if (lli->lli_opendir_pid == 0 && lli->lli_sai == NULL) { - opendir_set = 1; - lli->lli_opendir_pid = cfs_curproc_pid(); - lli->lli_opendir_key = fd; - } else if (unlikely(lli->lli_opendir_pid == cfs_curproc_pid())) { - /* Two cases for this: - * (1) The same process open such directory many times. - * (2) The old process opened the directory, and exited - * before its children processes. Then new process - * with the same pid opens such directory before the - * old process's children processes exit. - * Change the owner to the latest one. */ - opendir_set = 2; + cfs_spin_lock(&lli->lli_sa_lock); + if (lli->lli_opendir_key == NULL && lli->lli_opendir_pid == 0) { + LASSERT(lli->lli_sai == NULL); lli->lli_opendir_key = fd; + lli->lli_opendir_pid = cfs_curproc_pid(); + opendir_set = 1; } - spin_unlock(&lli->lli_lock); + cfs_spin_unlock(&lli->lli_sa_lock); } if (inode->i_sb->s_root == file->f_dentry) { @@ -555,6 +550,12 @@ int ll_file_open(struct inode *inode, struct file *file) * already? XXX - NFS implications? */ oit.it_flags &= ~O_EXCL; + /* bug20584, if "it_flags" contains O_CREAT, the file will be + * created if necessary, then "IT_CREAT" should be set to keep + * consistent with it */ + if (oit.it_flags & O_CREAT) + oit.it_op |= IT_CREAT; + it = &oit; } @@ -570,28 +571,30 @@ restart: och_p = &lli->lli_mds_read_och; och_usecount = &lli->lli_open_fd_read_count; } - - down(&lli->lli_och_sem); + + cfs_down(&lli->lli_och_sem); if (*och_p) { /* Open handle is present */ if (it_disposition(it, DISP_OPEN_OPEN)) { /* Well, there's extra open request that we do not need, let's close it somehow. This will decref request. */ rc = it_open_error(DISP_OPEN_OPEN, it); if (rc) { + cfs_up(&lli->lli_och_sem); ll_file_data_put(fd); - GOTO(out_och_free, rc); - } + GOTO(out_openerr, rc); + } ll_release_openhandle(file->f_dentry, it); - lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, + lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_OPEN); } (*och_usecount)++; rc = ll_local_open(file, it, fd, NULL); if (rc) { - up(&lli->lli_och_sem); + (*och_usecount)--; + cfs_up(&lli->lli_och_sem); ll_file_data_put(fd); - RETURN(rc); + GOTO(out_openerr, rc); } } else { LASSERT(*och_usecount == 0); @@ -601,10 +604,10 @@ restart: could be cancelled, and since blocking ast handler would attempt to grab och_sem as well, that would result in a deadlock */ - up(&lli->lli_och_sem); - it->it_flags |= O_CHECK_STALE; + cfs_up(&lli->lli_och_sem); + it->it_create_mode |= M_CHECK_STALE; rc = ll_intent_file_open(file, NULL, 0, it); - it->it_flags &= ~O_CHECK_STALE; + it->it_create_mode &= ~M_CHECK_STALE; if (rc) { ll_file_data_put(fd); GOTO(out_openerr, rc); @@ -615,9 +618,6 @@ restart: req = it->d.lustre.it_data; ptlrpc_req_finished(req); } - md_set_lock_data(ll_i2sbi(inode)->ll_md_exp, - &it->d.lustre.it_lock_handle, - file->f_dentry->d_inode); goto restart; } OBD_ALLOC(*och_p, sizeof (struct obd_client_handle)); @@ -642,12 +642,11 @@ restart: ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); rc = ll_local_open(file, it, fd, *och_p); if (rc) { - up(&lli->lli_och_sem); ll_file_data_put(fd); GOTO(out_och_free, rc); } } - up(&lli->lli_och_sem); + cfs_up(&lli->lli_och_sem); /* Must do this outside lli_och_sem lock to prevent deadlock where different kind of OPEN lock for this same inode gets cancelled @@ -678,28 +677,24 @@ out_och_free: *och_p = NULL; /* OBD_FREE writes some magic there */ (*och_usecount)--; } - up(&lli->lli_och_sem); + cfs_up(&lli->lli_och_sem); out_openerr: - if (opendir_set == 1) { - lli->lli_opendir_key = NULL; - lli->lli_opendir_pid = 0; - } else if (unlikely(opendir_set == 2)) { - ll_stop_statahead(inode, fd); - } + if (opendir_set != 0) + ll_stop_statahead(inode, lli->lli_opendir_key); } return rc; } -/* Fills the obdo with the attributes for the inode defined by lsm */ -int ll_inode_getattr(struct inode *inode, struct obdo *obdo) +/* Fills the obdo with the attributes for the lsm */ +static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, + struct obd_capa *capa, struct obdo *obdo, + __u64 ioepoch, int sync) { struct ptlrpc_request_set *set; - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; + struct obd_info oinfo = { { { 0 } } }; + int rc; - struct obd_info oinfo = { { { 0 } } }; - int rc; ENTRY; LASSERT(lsm != NULL); @@ -707,1325 +702,600 @@ int ll_inode_getattr(struct inode *inode, struct obdo *obdo) oinfo.oi_md = lsm; oinfo.oi_oa = obdo; oinfo.oi_oa->o_id = lsm->lsm_object_id; - oinfo.oi_oa->o_gr = lsm->lsm_object_gr; + oinfo.oi_oa->o_seq = lsm->lsm_object_seq; oinfo.oi_oa->o_mode = S_IFREG; + oinfo.oi_oa->o_ioepoch = ioepoch; oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME | - OBD_MD_FLGROUP; - oinfo.oi_capa = ll_mdscapa_get(inode); + OBD_MD_FLGROUP | OBD_MD_FLEPOCH; + oinfo.oi_capa = capa; + if (sync) { + oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; + oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK; + } set = ptlrpc_prep_set(); if (set == NULL) { CERROR("can't allocate ptlrpc set\n"); rc = -ENOMEM; } else { - rc = obd_getattr_async(ll_i2dtexp(inode), &oinfo, set); + rc = obd_getattr_async(exp, &oinfo, set); if (rc == 0) rc = ptlrpc_set_wait(set); ptlrpc_set_destroy(set); } - capa_put(oinfo.oi_capa); - if (rc) - RETURN(rc); + if (rc == 0) + oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | + OBD_MD_FLATIME | OBD_MD_FLMTIME | + OBD_MD_FLCTIME | OBD_MD_FLSIZE); + RETURN(rc); +} - oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | - OBD_MD_FLATIME | OBD_MD_FLMTIME | - OBD_MD_FLCTIME | OBD_MD_FLSIZE); +/** + * Performs the getattr on the inode and updates its fields. + * If @sync != 0, perform the getattr under the server-side lock. + */ +int ll_inode_getattr(struct inode *inode, struct obdo *obdo, + __u64 ioepoch, int sync) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct obd_capa *capa = ll_mdscapa_get(inode); + int rc; + ENTRY; - obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid); - CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %llu, blksize %lu\n", - lli->lli_smd->lsm_object_id, i_size_read(inode), - (unsigned long long)inode->i_blocks, - (unsigned long)ll_inode_blksize(inode)); - RETURN(0); + rc = ll_lsm_getattr(lli->lli_smd, ll_i2dtexp(inode), + capa, obdo, ioepoch, sync); + capa_put(capa); + if (rc == 0) { + obdo_refresh_inode(inode, obdo, obdo->o_valid); + CDEBUG(D_INODE, + "objid "LPX64" size %Lu, blocks %llu, blksize %lu\n", + lli->lli_smd->lsm_object_id, i_size_read(inode), + (unsigned long long)inode->i_blocks, + (unsigned long)ll_inode_blksize(inode)); + } + RETURN(rc); } -static int ll_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock) +int ll_merge_lvb(struct inode *inode) { struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; - struct obd_export *exp = ll_i2dtexp(inode); - struct { - char name[16]; - struct ldlm_lock *lock; - } key = { .name = KEY_LOCK_TO_STRIPE, .lock = lock }; - __u32 stripe, vallen = sizeof(stripe); - struct lov_oinfo *loinfo; + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct ost_lvb lvb; int rc; + ENTRY; - if (lsm->lsm_stripe_count == 1) - GOTO(check, stripe = 0); + ll_inode_size_lock(inode, 1); + inode_init_lvb(inode, &lvb); - /* get our offset in the lov */ - rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe, lsm); - if (rc != 0) { - CERROR("obd_get_info: rc = %d\n", rc); - RETURN(rc); - } - LASSERT(stripe < lsm->lsm_stripe_count); + /* merge timestamps the most resently obtained from mds with + timestamps obtained from osts */ + lvb.lvb_atime = lli->lli_lvb.lvb_atime; + lvb.lvb_mtime = lli->lli_lvb.lvb_mtime; + lvb.lvb_ctime = lli->lli_lvb.lvb_ctime; + rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0); + cl_isize_write_nolock(inode, lvb.lvb_size); -check: - loinfo = lsm->lsm_oinfo[stripe]; - if (!osc_res_name_eq(loinfo->loi_id, loinfo->loi_gr, - &lock->l_resource->lr_name)){ - LDLM_ERROR(lock, "resource doesn't match object "LPU64"/"LPU64, - loinfo->loi_id, loinfo->loi_gr); - RETURN(-ELDLM_NO_LOCK_DATA); - } + CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n", + PFID(&lli->lli_fid), lvb.lvb_size); + inode->i_blocks = lvb.lvb_blocks; + + LTIME_S(inode->i_mtime) = lvb.lvb_mtime; + LTIME_S(inode->i_atime) = lvb.lvb_atime; + LTIME_S(inode->i_ctime) = lvb.lvb_ctime; + ll_inode_size_unlock(inode, 1); - RETURN(stripe); + RETURN(rc); } -/* Get extra page reference to ensure it is not going away */ -void ll_pin_extent_cb(void *data) +int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, + lstat_t *st) { - struct page *page = data; - - page_cache_get(page); + struct obdo obdo = { 0 }; + int rc; - return; + rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0); + if (rc == 0) { + st->st_size = obdo.o_size; + st->st_blocks = obdo.o_blocks; + st->st_mtime = obdo.o_mtime; + st->st_atime = obdo.o_atime; + st->st_ctime = obdo.o_ctime; + } + return rc; } -/* Flush the page from page cache for an extent as its canceled. - * Page to remove is delivered as @data. - * - * No one can dirty the extent until we've finished our work and they cannot - * enqueue another lock. The DLM protects us from ll_file_read/write here, - * but other kernel actors could have pages locked. - * - * If @discard is set, there is no need to write the page if it is dirty. - * - * Called with the DLM lock held. */ -int ll_page_removal_cb(void *data, int discard) +void ll_io_init(struct cl_io *io, const struct file *file, int write) { - int rc; - struct page *page = data; - struct address_space *mapping; - - ENTRY; + struct inode *inode = file->f_dentry->d_inode; - /* We have page reference already from ll_pin_page */ - lock_page(page); + memset(io, 0, sizeof *io); + io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK; + if (write) + io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); + io->ci_obj = ll_i2info(inode)->lli_clob; + io->ci_lockreq = CILR_MAYBE; + if (ll_file_nolock(file)) { + io->ci_lockreq = CILR_NEVER; + io->ci_no_srvlock = 1; + } else if (file->f_flags & O_APPEND) { + io->ci_lockreq = CILR_MANDATORY; + } +} - /* Already truncated by somebody */ - if (!page->mapping) - GOTO(out, rc = 0); - mapping = page->mapping; - - ll_teardown_mmaps(mapping, - (__u64)page->index << PAGE_CACHE_SHIFT, - ((__u64)page->index<mapping); - rc = ll_call_writepage(page->mapping->host, page); - /* either waiting for io to complete or reacquiring - * the lock that the failed writepage released */ - lock_page(page); - wait_on_page_writeback(page); - if (rc != 0) { - CERROR("writepage inode %lu(%p) of page %p " - "failed: %d\n", mapping->host->i_ino, - mapping->host, page, rc); - if (rc == -ENOSPC) - set_bit(AS_ENOSPC, &mapping->flags); - else - set_bit(AS_EIO, &mapping->flags); +static ssize_t ll_file_io_generic(const struct lu_env *env, + struct vvp_io_args *args, struct file *file, + enum cl_io_type iot, loff_t *ppos, size_t count) +{ + struct cl_io *io; + ssize_t result; + ENTRY; + + io = &ccc_env_info(env)->cti_io; + ll_io_init(io, file, iot == CIT_WRITE); + + if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) { + struct vvp_io *vio = vvp_env_io(env); + struct ccc_io *cio = ccc_env_io(env); + struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode); + int write_sem_locked = 0; + + cio->cui_fd = LUSTRE_FPRIVATE(file); + vio->cui_io_subtype = args->via_io_subtype; + + switch (vio->cui_io_subtype) { + case IO_NORMAL: + cio->cui_iov = args->u.normal.via_iov; + cio->cui_nrsegs = args->u.normal.via_nrsegs; + cio->cui_tot_nrsegs = cio->cui_nrsegs; +#ifndef HAVE_FILE_WRITEV + cio->cui_iocb = args->u.normal.via_iocb; +#endif + if ((iot == CIT_WRITE) && + !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { + if(cfs_down_interruptible(&lli->lli_write_sem)) + GOTO(out, result = -ERESTARTSYS); + write_sem_locked = 1; + } else if (iot == CIT_READ) { + cfs_down_read(&lli->lli_trunc_sem); + } + break; + case IO_SENDFILE: + vio->u.sendfile.cui_actor = args->u.sendfile.via_actor; + vio->u.sendfile.cui_target = args->u.sendfile.via_target; + break; + case IO_SPLICE: + vio->u.splice.cui_pipe = args->u.splice.via_pipe; + vio->u.splice.cui_flags = args->u.splice.via_flags; + break; + default: + CERROR("Unknow IO type - %u\n", vio->cui_io_subtype); + LBUG(); } - set_bit(AS_EIO, &mapping->flags); + result = cl_io_loop(env, io); + if (write_sem_locked) + cfs_up(&lli->lli_write_sem); + else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ) + cfs_up_read(&lli->lli_trunc_sem); + } else { + /* cl_io_rw_init() handled IO */ + result = io->ci_result; } - if (page->mapping != NULL) { - struct ll_async_page *llap = llap_cast_private(page); - /* checking again to account for writeback's lock_page() */ - LL_CDEBUG_PAGE(D_PAGE, page, "truncating\n"); - if (llap) - ll_ra_accounting(llap, page->mapping); - ll_truncate_complete_page(page); + + if (io->ci_nob > 0) { + result = io->ci_nob; + *ppos = io->u.ci_wr.wr.crw_pos; } - EXIT; + GOTO(out, result); out: - LASSERT(!PageWriteback(page)); - unlock_page(page); - page_cache_release(page); + cl_io_fini(env, io); + return result; +} + + +/* + * XXX: exact copy from kernel code (__generic_file_aio_write_nolock) + */ +static int ll_file_get_iov_count(const struct iovec *iov, + unsigned long *nr_segs, size_t *count) +{ + size_t cnt = 0; + unsigned long seg; + + for (seg = 0; seg < *nr_segs; seg++) { + const struct iovec *iv = &iov[seg]; + /* + * If any segment has a negative length, or the cumulative + * length ever wraps negative then return -EINVAL. + */ + cnt += iv->iov_len; + if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) + return -EINVAL; + if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len)) + continue; + if (seg == 0) + return -EFAULT; + *nr_segs = seg; + cnt -= iv->iov_len; /* This segment is no good */ + break; + } + *count = cnt; return 0; } -int ll_extent_lock_cancel_cb(struct ldlm_lock *lock, struct ldlm_lock_desc *new, - void *data, int flag) +#ifdef HAVE_FILE_READV +static ssize_t ll_file_readv(struct file *file, const struct iovec *iov, + unsigned long nr_segs, loff_t *ppos) { - struct inode *inode; - struct ll_inode_info *lli; - struct lov_stripe_md *lsm; - int stripe; - __u64 kms; - + struct lu_env *env; + struct vvp_io_args *args; + size_t count; + ssize_t result; + int refcheck; ENTRY; - if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) { - LDLM_ERROR(lock, "cancelling lock with bad data %p", data); - LBUG(); - } + result = ll_file_get_iov_count(iov, &nr_segs, &count); + if (result) + RETURN(result); - inode = ll_inode_from_lock(lock); - if (inode == NULL) - RETURN(0); - lli = ll_i2info(inode); - if (lli == NULL) - GOTO(iput, 0); - if (lli->lli_smd == NULL) - GOTO(iput, 0); - lsm = lli->lli_smd; + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - stripe = ll_lock_to_stripe_offset(inode, lock); - if (stripe < 0) - GOTO(iput, 0); - - lov_stripe_lock(lsm); - lock_res_and_lock(lock); - kms = ldlm_extent_shift_kms(lock, - lsm->lsm_oinfo[stripe]->loi_kms); - - if (lsm->lsm_oinfo[stripe]->loi_kms != kms) - LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64, - lsm->lsm_oinfo[stripe]->loi_kms, kms); - lsm->lsm_oinfo[stripe]->loi_kms = kms; - unlock_res_and_lock(lock); - lov_stripe_unlock(lsm); - ll_queue_done_writing(inode, 0); - EXIT; -iput: - iput(inode); + args = vvp_env_args(env, IO_NORMAL); + args->u.normal.via_iov = (struct iovec *)iov; + args->u.normal.via_nrsegs = nr_segs; - return 0; + result = ll_file_io_generic(env, args, file, CIT_READ, ppos, count); + cl_env_put(env, &refcheck); + RETURN(result); } -#if 0 -int ll_async_completion_ast(struct ldlm_lock *lock, int flags, void *data) +static ssize_t ll_file_read(struct file *file, char *buf, size_t count, + loff_t *ppos) { - /* XXX ALLOCATE - 160 bytes */ - struct inode *inode = ll_inode_from_lock(lock); - struct ll_inode_info *lli = ll_i2info(inode); - struct lustre_handle lockh = { 0 }; - struct ost_lvb *lvb; - int stripe; + struct lu_env *env; + struct iovec *local_iov; + ssize_t result; + int refcheck; ENTRY; - if (flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED | - LDLM_FL_BLOCK_CONV)) { - LBUG(); /* not expecting any blocked async locks yet */ - LDLM_DEBUG(lock, "client-side async enqueue returned a blocked " - "lock, returning"); - ldlm_lock_dump(D_OTHER, lock, 0); - ldlm_reprocess_all(lock->l_resource); - RETURN(0); - } + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - LDLM_DEBUG(lock, "client-side async enqueue: granted/glimpsed"); + local_iov = &vvp_env_info(env)->vti_local_iov; + local_iov->iov_base = (void __user *)buf; + local_iov->iov_len = count; + result = ll_file_readv(file, local_iov, 1, ppos); + cl_env_put(env, &refcheck); + RETURN(result); +} - stripe = ll_lock_to_stripe_offset(inode, lock); - if (stripe < 0) - goto iput; +#else +static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov, + unsigned long nr_segs, loff_t pos) +{ + struct lu_env *env; + struct vvp_io_args *args; + size_t count; + ssize_t result; + int refcheck; + ENTRY; - if (lock->l_lvb_len) { - struct lov_stripe_md *lsm = lli->lli_smd; - __u64 kms; - lvb = lock->l_lvb_data; - lsm->lsm_oinfo[stripe].loi_rss = lvb->lvb_size; + result = ll_file_get_iov_count(iov, &nr_segs, &count); + if (result) + RETURN(result); - lock_res_and_lock(lock); - ll_inode_size_lock(inode, 1); - kms = MAX(lsm->lsm_oinfo[stripe].loi_kms, lvb->lvb_size); - kms = ldlm_extent_shift_kms(NULL, kms); - if (lsm->lsm_oinfo[stripe].loi_kms != kms) - LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64, - lsm->lsm_oinfo[stripe].loi_kms, kms); - lsm->lsm_oinfo[stripe].loi_kms = kms; - ll_inode_size_unlock(inode, 1); - unlock_res_and_lock(lock); - } + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); -iput: - iput(inode); - wake_up(&lock->l_waitq); + args = vvp_env_args(env, IO_NORMAL); + args->u.normal.via_iov = (struct iovec *)iov; + args->u.normal.via_nrsegs = nr_segs; + args->u.normal.via_iocb = iocb; - ldlm_lock2handle(lock, &lockh); - ldlm_lock_decref(&lockh, LCK_PR); - RETURN(0); + result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ, + &iocb->ki_pos, count); + cl_env_put(env, &refcheck); + RETURN(result); } -#endif -static int ll_glimpse_callback(struct ldlm_lock *lock, void *reqp) +static ssize_t ll_file_read(struct file *file, char *buf, size_t count, + loff_t *ppos) { - struct ptlrpc_request *req = reqp; - struct inode *inode = ll_inode_from_lock(lock); - struct ll_inode_info *lli; - struct lov_stripe_md *lsm; - struct ost_lvb *lvb; - int rc, stripe; + struct lu_env *env; + struct iovec *local_iov; + struct kiocb *kiocb; + ssize_t result; + int refcheck; ENTRY; - if (inode == NULL) - GOTO(out, rc = -ELDLM_NO_LOCK_DATA); - lli = ll_i2info(inode); - if (lli == NULL) - GOTO(iput, rc = -ELDLM_NO_LOCK_DATA); - lsm = lli->lli_smd; - if (lsm == NULL) - GOTO(iput, rc = -ELDLM_NO_LOCK_DATA); + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - /* First, find out which stripe index this lock corresponds to. */ - stripe = ll_lock_to_stripe_offset(inode, lock); - if (stripe < 0) - GOTO(iput, rc = -ELDLM_NO_LOCK_DATA); + local_iov = &vvp_env_info(env)->vti_local_iov; + kiocb = &vvp_env_info(env)->vti_kiocb; + local_iov->iov_base = (void __user *)buf; + local_iov->iov_len = count; + init_sync_kiocb(kiocb, file); + kiocb->ki_pos = *ppos; + kiocb->ki_left = count; - req_capsule_extend(&req->rq_pill, &RQF_LDLM_GL_CALLBACK); - req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, - sizeof(*lvb)); - rc = req_capsule_server_pack(&req->rq_pill); - if (rc) { - CERROR("lustre_pack_reply: %d\n", rc); - GOTO(iput, rc); - } + result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos); + *ppos = kiocb->ki_pos; + + cl_env_put(env, &refcheck); + RETURN(result); +} +#endif - lvb = req_capsule_server_get(&req->rq_pill, &RMF_DLM_LVB); - lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms; - lvb->lvb_mtime = LTIME_S(inode->i_mtime); - lvb->lvb_atime = LTIME_S(inode->i_atime); - lvb->lvb_ctime = LTIME_S(inode->i_ctime); +/* + * Write to a file (through the page cache). + */ +#ifdef HAVE_FILE_WRITEV +static ssize_t ll_file_writev(struct file *file, const struct iovec *iov, + unsigned long nr_segs, loff_t *ppos) +{ + struct lu_env *env; + struct vvp_io_args *args; + size_t count; + ssize_t result; + int refcheck; + ENTRY; - LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64 - " atime "LPU64", mtime "LPU64", ctime "LPU64, - i_size_read(inode), stripe, lvb->lvb_size, lvb->lvb_mtime, - lvb->lvb_atime, lvb->lvb_ctime); - iput: - iput(inode); + result = ll_file_get_iov_count(iov, &nr_segs, &count); + if (result) + RETURN(result); - out: - /* These errors are normal races, so we don't want to fill the console - * with messages by calling ptlrpc_error() */ - if (rc == -ELDLM_NO_LOCK_DATA) - lustre_pack_reply(req, 1, NULL, NULL); + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - req->rq_status = rc; - return rc; + args = vvp_env_args(env, IO_NORMAL); + args->u.normal.via_iov = (struct iovec *)iov; + args->u.normal.via_nrsegs = nr_segs; + + result = ll_file_io_generic(env, args, file, CIT_WRITE, ppos, count); + cl_env_put(env, &refcheck); + RETURN(result); } -static int ll_merge_lvb(struct inode *inode) +static ssize_t ll_file_write(struct file *file, const char *buf, size_t count, + loff_t *ppos) { - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ost_lvb lvb; - int rc; - + struct lu_env *env; + struct iovec *local_iov; + ssize_t result; + int refcheck; ENTRY; - ll_inode_size_lock(inode, 1); - inode_init_lvb(inode, &lvb); - rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0); - i_size_write(inode, lvb.lvb_size); - inode->i_blocks = lvb.lvb_blocks; + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - LTIME_S(inode->i_mtime) = lvb.lvb_mtime; - LTIME_S(inode->i_atime) = lvb.lvb_atime; - LTIME_S(inode->i_ctime) = lvb.lvb_ctime; - ll_inode_size_unlock(inode, 1); + local_iov = &vvp_env_info(env)->vti_local_iov; + local_iov->iov_base = (void __user *)buf; + local_iov->iov_len = count; - RETURN(rc); + result = ll_file_writev(file, local_iov, 1, ppos); + cl_env_put(env, &refcheck); + RETURN(result); } -int ll_local_size(struct inode *inode) +#else /* AIO stuff */ +static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov, + unsigned long nr_segs, loff_t pos) { - ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } }; - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct lustre_handle lockh = { 0 }; - int flags = 0; - int rc; + struct lu_env *env; + struct vvp_io_args *args; + size_t count; + ssize_t result; + int refcheck; ENTRY; - if (lli->lli_smd->lsm_stripe_count == 0) - RETURN(0); + result = ll_file_get_iov_count(iov, &nr_segs, &count); + if (result) + RETURN(result); - rc = obd_match(sbi->ll_dt_exp, lli->lli_smd, LDLM_EXTENT, - &policy, LCK_PR, &flags, inode, &lockh); - if (rc < 0) - RETURN(rc); - else if (rc == 0) - RETURN(-ENODATA); + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - rc = ll_merge_lvb(inode); - obd_cancel(sbi->ll_dt_exp, lli->lli_smd, LCK_PR, &lockh); - RETURN(rc); + args = vvp_env_args(env, IO_NORMAL); + args->u.normal.via_iov = (struct iovec *)iov; + args->u.normal.via_nrsegs = nr_segs; + args->u.normal.via_iocb = iocb; + + result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE, + &iocb->ki_pos, count); + cl_env_put(env, &refcheck); + RETURN(result); } -int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, - lstat_t *st) +static ssize_t ll_file_write(struct file *file, const char *buf, size_t count, + loff_t *ppos) { - struct lustre_handle lockh = { 0 }; - struct ldlm_enqueue_info einfo = { 0 }; - struct obd_info oinfo = { { { 0 } } }; - struct ost_lvb lvb; - int rc; - + struct lu_env *env; + struct iovec *local_iov; + struct kiocb *kiocb; + ssize_t result; + int refcheck; ENTRY; - einfo.ei_type = LDLM_EXTENT; - einfo.ei_mode = LCK_PR; - einfo.ei_cb_bl = osc_extent_blocking_cb; - einfo.ei_cb_cp = ldlm_completion_ast; - einfo.ei_cb_gl = ll_glimpse_callback; - einfo.ei_cbdata = NULL; + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF; - oinfo.oi_lockh = &lockh; - oinfo.oi_md = lsm; - oinfo.oi_flags = LDLM_FL_HAS_INTENT; + local_iov = &vvp_env_info(env)->vti_local_iov; + kiocb = &vvp_env_info(env)->vti_kiocb; + local_iov->iov_base = (void __user *)buf; + local_iov->iov_len = count; + init_sync_kiocb(kiocb, file); + kiocb->ki_pos = *ppos; + kiocb->ki_left = count; - rc = obd_enqueue_rqset(sbi->ll_dt_exp, &oinfo, &einfo); - if (rc == -ENOENT) - RETURN(rc); - if (rc != 0) { - CERROR("obd_enqueue returned rc %d, " - "returning -EIO\n", rc); - RETURN(rc > 0 ? -EIO : rc); - } - - lov_stripe_lock(lsm); - memset(&lvb, 0, sizeof(lvb)); - obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 0); - st->st_size = lvb.lvb_size; - st->st_blocks = lvb.lvb_blocks; - st->st_mtime = lvb.lvb_mtime; - st->st_atime = lvb.lvb_atime; - st->st_ctime = lvb.lvb_ctime; - lov_stripe_unlock(lsm); + result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos); + *ppos = kiocb->ki_pos; - RETURN(rc); + cl_env_put(env, &refcheck); + RETURN(result); } +#endif -/* NB: obd_merge_lvb will prefer locally cached writes if they extend the - * file (because it prefers KMS over RSS when larger) */ -int ll_glimpse_size(struct inode *inode, int ast_flags) + +#ifdef HAVE_KERNEL_SENDFILE +/* + * Send file content (through pagecache) somewhere with helper + */ +static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count, + read_actor_t actor, void *target) { - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct lustre_handle lockh = { 0 }; - struct ldlm_enqueue_info einfo = { 0 }; - struct obd_info oinfo = { { { 0 } } }; - int rc; + struct lu_env *env; + struct vvp_io_args *args; + ssize_t result; + int refcheck; ENTRY; - if (lli->lli_flags & LLIF_MDS_SIZE_LOCK) - RETURN(0); + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - CDEBUG(D_DLMTRACE, "Glimpsing inode %lu\n", inode->i_ino); + args = vvp_env_args(env, IO_SENDFILE); + args->u.sendfile.via_target = target; + args->u.sendfile.via_actor = actor; - if (!lli->lli_smd) { - CDEBUG(D_DLMTRACE, "No objects for inode %lu\n", inode->i_ino); - RETURN(0); - } + result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count); + cl_env_put(env, &refcheck); + RETURN(result); +} +#endif - /* NOTE: this looks like DLM lock request, but it may not be one. Due - * to LDLM_FL_HAS_INTENT flag, this is glimpse request, that - * won't revoke any conflicting DLM locks held. Instead, - * ll_glimpse_callback() will be called on each client - * holding a DLM lock against this file, and resulting size - * will be returned for each stripe. DLM lock on [0, EOF] is - * acquired only if there were no conflicting locks. */ - einfo.ei_type = LDLM_EXTENT; - einfo.ei_mode = LCK_PR; - einfo.ei_cb_bl = osc_extent_blocking_cb; - einfo.ei_cb_cp = ldlm_completion_ast; - einfo.ei_cb_gl = ll_glimpse_callback; - einfo.ei_cbdata = inode; - - oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF; - oinfo.oi_lockh = &lockh; - oinfo.oi_md = lli->lli_smd; - oinfo.oi_flags = ast_flags | LDLM_FL_HAS_INTENT; - - rc = obd_enqueue_rqset(sbi->ll_dt_exp, &oinfo, &einfo); - if (rc == -ENOENT) - RETURN(rc); - if (rc != 0) { - CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc); - RETURN(rc > 0 ? -EIO : rc); - } +#ifdef HAVE_KERNEL_SPLICE_READ +/* + * Send file content (through pagecache) somewhere with helper + */ +static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos, + struct pipe_inode_info *pipe, size_t count, + unsigned int flags) +{ + struct lu_env *env; + struct vvp_io_args *args; + ssize_t result; + int refcheck; + ENTRY; - rc = ll_merge_lvb(inode); + env = cl_env_get(&refcheck); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); - CDEBUG(D_DLMTRACE, "glimpse: size: %llu, blocks: %llu\n", - i_size_read(inode), (unsigned long long)inode->i_blocks); + args = vvp_env_args(env, IO_SPLICE); + args->u.splice.via_pipe = pipe; + args->u.splice.via_flags = flags; - RETURN(rc); + result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count); + cl_env_put(env, &refcheck); + RETURN(result); } +#endif -int ll_extent_lock(struct ll_file_data *fd, struct inode *inode, - struct lov_stripe_md *lsm, int mode, - ldlm_policy_data_t *policy, struct lustre_handle *lockh, - int ast_flags) +static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq, + obd_count ost_idx) { - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ost_lvb lvb; - struct ldlm_enqueue_info einfo = { 0 }; - struct obd_info oinfo = { { { 0 } } }; - int rc; + struct obd_export *exp = ll_i2dtexp(inode); + struct obd_trans_info oti = { 0 }; + struct obdo *oa = NULL; + int lsm_size; + int rc = 0; + struct lov_stripe_md *lsm, *lsm2; ENTRY; - LASSERT(!lustre_handle_is_used(lockh)); - LASSERT(lsm != NULL); - - /* don't drop the mmapped file to LRU */ - if (mapping_mapped(inode->i_mapping)) - ast_flags |= LDLM_FL_NO_LRU; + OBDO_ALLOC(oa); + if (oa == NULL) + RETURN(-ENOMEM); - /* XXX phil: can we do this? won't it screw the file size up? */ - if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) || - (sbi->ll_flags & LL_SBI_NOLCK)) - RETURN(0); + ll_inode_size_lock(inode, 0); + lsm = ll_i2info(inode)->lli_smd; + if (lsm == NULL) + GOTO(out, rc = -ENOENT); + lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * + (lsm->lsm_stripe_count)); - CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n", - inode->i_ino, policy->l_extent.start, policy->l_extent.end); + OBD_ALLOC(lsm2, lsm_size); + if (lsm2 == NULL) + GOTO(out, rc = -ENOMEM); - einfo.ei_type = LDLM_EXTENT; - einfo.ei_mode = mode; - einfo.ei_cb_bl = osc_extent_blocking_cb; - einfo.ei_cb_cp = ldlm_completion_ast; - einfo.ei_cb_gl = ll_glimpse_callback; - einfo.ei_cbdata = inode; + oa->o_id = id; + oa->o_seq = seq; + oa->o_nlink = ost_idx; + oa->o_flags |= OBD_FL_RECREATE_OBJS; + oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; + obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, OBD_MD_FLTYPE | + OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME); + memcpy(lsm2, lsm, lsm_size); + rc = obd_create(exp, oa, &lsm2, &oti); - oinfo.oi_policy = *policy; - oinfo.oi_lockh = lockh; - oinfo.oi_md = lsm; - oinfo.oi_flags = ast_flags; + OBD_FREE(lsm2, lsm_size); + GOTO(out, rc); +out: + ll_inode_size_unlock(inode, 0); + OBDO_FREE(oa); + return rc; +} - rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL); - *policy = oinfo.oi_policy; - if (rc > 0) - rc = -EIO; +static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) +{ + struct ll_recreate_obj ucreat; + ENTRY; - ll_inode_size_lock(inode, 1); - inode_init_lvb(inode, &lvb); - obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1); - - if (policy->l_extent.start == 0 && - policy->l_extent.end == OBD_OBJECT_EOF) { - /* vmtruncate()->ll_truncate() first sets the i_size and then - * the kms under both a DLM lock and the - * ll_inode_size_lock(). If we don't get the - * ll_inode_size_lock() here we can match the DLM lock and - * reset i_size from the kms before the truncating path has - * updated the kms. generic_file_write can then trust the - * stale i_size when doing appending writes and effectively - * cancel the result of the truncate. Getting the - * ll_inode_size_lock() after the enqueue maintains the DLM - * -> ll_inode_size_lock() acquiring order. */ - i_size_write(inode, lvb.lvb_size); - CDEBUG(D_INODE, "inode=%lu, updating i_size %llu\n", - inode->i_ino, i_size_read(inode)); - } + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - if (rc == 0) { - LTIME_S(inode->i_mtime) = lvb.lvb_mtime; - LTIME_S(inode->i_atime) = lvb.lvb_atime; - LTIME_S(inode->i_ctime) = lvb.lvb_ctime; - } - ll_inode_size_unlock(inode, 1); + if (cfs_copy_from_user(&ucreat, (struct ll_recreate_obj *)arg, + sizeof(struct ll_recreate_obj))) + RETURN(-EFAULT); - RETURN(rc); + RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0, + ucreat.lrc_ost_idx)); } -int ll_extent_unlock(struct ll_file_data *fd, struct inode *inode, - struct lov_stripe_md *lsm, int mode, - struct lustre_handle *lockh) +static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) { - struct ll_sb_info *sbi = ll_i2sbi(inode); - int rc; + struct lu_fid fid; + obd_id id; + obd_count ost_idx; ENTRY; - /* XXX phil: can we do this? won't it screw the file size up? */ - if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) || - (sbi->ll_flags & LL_SBI_NOLCK)) - RETURN(0); + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) + RETURN(-EPERM); - rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh); + if (cfs_copy_from_user(&fid, (struct lu_fid *)arg, + sizeof(struct lu_fid))) + RETURN(-EFAULT); - RETURN(rc); -} - -static void ll_set_file_contended(struct inode *inode) -{ - struct ll_inode_info *lli = ll_i2info(inode); - cfs_time_t now = cfs_time_current(); - - spin_lock(&lli->lli_lock); - lli->lli_contention_time = now; - lli->lli_flags |= LLIF_CONTENDED; - spin_unlock(&lli->lli_lock); -} - -void ll_clear_file_contended(struct inode *inode) -{ - struct ll_inode_info *lli = ll_i2info(inode); - - spin_lock(&lli->lli_lock); - lli->lli_flags &= ~LLIF_CONTENDED; - spin_unlock(&lli->lli_lock); -} - -static int ll_is_file_contended(struct file *file) -{ - struct inode *inode = file->f_dentry->d_inode; - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - ENTRY; - - if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SRVLOCK)) { - CDEBUG(D_INFO, "the server does not support SRVLOCK feature," - " osc connect flags = 0x"LPX64"\n", - sbi->ll_lco.lco_flags); - RETURN(0); - } - if (fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) - RETURN(1); - if (lli->lli_flags & LLIF_CONTENDED) { - cfs_time_t cur_time = cfs_time_current(); - cfs_time_t retry_time; - - retry_time = cfs_time_add( - lli->lli_contention_time, - cfs_time_seconds(sbi->ll_contention_time)); - if (cfs_time_after(cur_time, retry_time)) { - ll_clear_file_contended(inode); - RETURN(0); - } - RETURN(1); - } - RETURN(0); -} - -static int ll_file_get_tree_lock(struct ll_lock_tree *tree, struct file *file, - const char *buf, size_t count, - loff_t start, loff_t end, int rw) -{ - int append; - int tree_locked = 0; - int rc; - struct inode * inode = file->f_dentry->d_inode; - ENTRY; - - append = (rw == OBD_BRW_WRITE) && (file->f_flags & O_APPEND); - - if (append || !ll_is_file_contended(file)) { - struct ll_lock_tree_node *node; - int ast_flags; - - ast_flags = append ? 0 : LDLM_FL_DENY_ON_CONTENTION; - if (file->f_flags & O_NONBLOCK) - ast_flags |= LDLM_FL_BLOCK_NOWAIT; - node = ll_node_from_inode(inode, start, end, - (rw == OBD_BRW_WRITE) ? LCK_PW : LCK_PR); - if (IS_ERR(node)) { - rc = PTR_ERR(node); - GOTO(out, rc); - } - tree->lt_fd = LUSTRE_FPRIVATE(file); - rc = ll_tree_lock(tree, node, buf, count, ast_flags); - if (rc == 0) - tree_locked = 1; - else if (rc == -EUSERS) - ll_set_file_contended(inode); - else - GOTO(out, rc); - } - RETURN(tree_locked); -out: - return rc; -} - -/** - * Checks if requested extent lock is compatible with a lock under a page. - * - * Checks if the lock under \a page is compatible with a read or write lock - * (specified by \a rw) for an extent [\a start , \a end]. - * - * \param page the page under which lock is considered - * \param rw OBD_BRW_READ if requested for reading, - * OBD_BRW_WRITE if requested for writing - * \param start start of the requested extent - * \param end end of the requested extent - * \param cookie transparent parameter for passing locking context - * - * \post result == 1, *cookie == context, appropriate lock is referenced or - * \post result == 0 - * - * \retval 1 owned lock is reused for the request - * \retval 0 no lock reused for the request - * - * \see ll_release_short_lock - */ -static int ll_reget_short_lock(struct page *page, int rw, - obd_off start, obd_off end, - void **cookie) -{ - struct ll_async_page *llap; - struct obd_export *exp; - struct inode *inode = page->mapping->host; - - ENTRY; - - exp = ll_i2dtexp(inode); - if (exp == NULL) - RETURN(0); - - llap = llap_cast_private(page); - if (llap == NULL) - RETURN(0); - - RETURN(obd_reget_short_lock(exp, ll_i2info(inode)->lli_smd, - &llap->llap_cookie, rw, start, end, - cookie)); -} - -/** - * Releases a reference to a lock taken in a "fast" way. - * - * Releases a read or a write (specified by \a rw) lock - * referenced by \a cookie. - * - * \param inode inode to which data belong - * \param end end of the locked extent - * \param rw OBD_BRW_READ if requested for reading, - * OBD_BRW_WRITE if requested for writing - * \param cookie transparent parameter for passing locking context - * - * \post appropriate lock is dereferenced - * - * \see ll_reget_short_lock - */ -static void ll_release_short_lock(struct inode *inode, obd_off end, - void *cookie, int rw) -{ - struct obd_export *exp; - int rc; - - exp = ll_i2dtexp(inode); - if (exp == NULL) - return; - - rc = obd_release_short_lock(exp, ll_i2info(inode)->lli_smd, end, - cookie, rw); - if (rc < 0) - CERROR("unlock failed (%d)\n", rc); -} - -/** - * Checks if requested extent lock is compatible - * with a lock under a page in page cache. - * - * Checks if a lock under some \a page is compatible with a read or write lock - * (specified by \a rw) for an extent [\a start , \a end]. - * - * \param file the file under which lock is considered - * \param rw OBD_BRW_READ if requested for reading, - * OBD_BRW_WRITE if requested for writing - * \param ppos start of the requested extent - * \param end end of the requested extent - * \param cookie transparent parameter for passing locking context - * \param buf userspace buffer for the data - * - * \post result == 1, *cookie == context, appropriate lock is referenced - * \post retuls == 0 - * - * \retval 1 owned lock is reused for the request - * \retval 0 no lock reused for the request - * - * \see ll_file_put_fast_lock - */ -static inline int ll_file_get_fast_lock(struct file *file, - obd_off ppos, obd_off end, - char *buf, void **cookie, int rw) -{ - int rc = 0; - struct page *page; - - ENTRY; - - if (!ll_region_mapped((unsigned long)buf, end - ppos)) { - page = find_lock_page(file->f_dentry->d_inode->i_mapping, - ppos >> CFS_PAGE_SHIFT); - if (page) { - if (ll_reget_short_lock(page, rw, ppos, end, cookie)) - rc = 1; - - unlock_page(page); - page_cache_release(page); - } - } - - RETURN(rc); -} - -/** - * Releases a reference to a lock taken in a "fast" way. - * - * Releases a read or a write (specified by \a rw) lock - * referenced by \a cookie. - * - * \param inode inode to which data belong - * \param end end of the locked extent - * \param rw OBD_BRW_READ if requested for reading, - * OBD_BRW_WRITE if requested for writing - * \param cookie transparent parameter for passing locking context - * - * \post appropriate lock is dereferenced - * - * \see ll_file_get_fast_lock - */ -static inline void ll_file_put_fast_lock(struct inode *inode, obd_off end, - void *cookie, int rw) -{ - ll_release_short_lock(inode, end, cookie, rw); -} - -enum ll_lock_style { - LL_LOCK_STYLE_NOLOCK = 0, - LL_LOCK_STYLE_FASTLOCK = 1, - LL_LOCK_STYLE_TREELOCK = 2 -}; - -/** - * Checks if requested extent lock is compatible with a lock - * under a page cache page. - * - * Checks if the lock under \a page is compatible with a read or write lock - * (specified by \a rw) for an extent [\a start , \a end]. - * - * \param file file under which I/O is processed - * \param rw OBD_BRW_READ if requested for reading, - * OBD_BRW_WRITE if requested for writing - * \param ppos start of the requested extent - * \param end end of the requested extent - * \param cookie transparent parameter for passing locking context - * (only used with LL_LOCK_STYLE_FASTLOCK) - * \param tree lock tree (only used with LL_LOCK_STYLE_TREELOCK) - * \param buf userspace buffer for the data - * - * \retval LL_LOCK_STYLE_FASTLOCK owned lock is reused through fast lock - * \retval LL_LOCK_STYLE_TREELOCK got a lock through tree lock - * \retval LL_LOCK_STYLE_NOLOCK got no lock - * - * \see ll_file_put_lock - */ -static inline int ll_file_get_lock(struct file *file, obd_off ppos, - obd_off end, char *buf, void **cookie, - struct ll_lock_tree *tree, int rw) -{ - int rc; - - ENTRY; - - if (ll_file_get_fast_lock(file, ppos, end, buf, cookie, rw)) - RETURN(LL_LOCK_STYLE_FASTLOCK); - - rc = ll_file_get_tree_lock(tree, file, buf, ppos - end, ppos, end, rw); - /* rc: 1 for tree lock, 0 for no lock, <0 for error */ - switch (rc) { - case 1: - RETURN(LL_LOCK_STYLE_TREELOCK); - case 0: - RETURN(LL_LOCK_STYLE_NOLOCK); - } - - /* an error happened if we reached this point, rc = -errno here */ - RETURN(rc); -} - -/** - * Drops the lock taken by ll_file_get_lock. - * - * Releases a read or a write (specified by \a rw) lock - * referenced by \a tree or \a cookie. - * - * \param inode inode to which data belong - * \param end end of the locked extent - * \param lockstyle facility through which the lock was taken - * \param rw OBD_BRW_READ if requested for reading, - * OBD_BRW_WRITE if requested for writing - * \param cookie transparent parameter for passing locking context - * (only used with LL_LOCK_STYLE_FASTLOCK) - * \param tree lock tree (only used with LL_LOCK_STYLE_TREELOCK) - * - * \post appropriate lock is dereferenced - * - * \see ll_file_get_lock - */ -static inline void ll_file_put_lock(struct inode *inode, obd_off end, - enum ll_lock_style lock_style, - void *cookie, struct ll_lock_tree *tree, - int rw) - -{ - switch (lock_style) { - case LL_LOCK_STYLE_TREELOCK: - ll_tree_unlock(tree); - break; - case LL_LOCK_STYLE_FASTLOCK: - ll_file_put_fast_lock(inode, end, cookie, rw); - break; - default: - CERROR("invalid locking style (%d)\n", lock_style); - } -} - -static ssize_t ll_file_read(struct file *file, char *buf, size_t count, - loff_t *ppos) -{ - struct inode *inode = file->f_dentry->d_inode; - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ll_lock_tree tree; - struct ost_lvb lvb; - struct ll_ra_read bead; - int ra = 0; - obd_off end; - ssize_t retval, chunk, sum = 0; - int lock_style; - void *cookie; - - __u64 kms; - ENTRY; - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n", - inode->i_ino, inode->i_generation, inode, count, *ppos); - /* "If nbyte is 0, read() will return 0 and have no other results." - * -- Single Unix Spec */ - if (count == 0) - RETURN(0); - - ll_stats_ops_tally(sbi, LPROC_LL_READ_BYTES, count); - - if (!lsm) { - /* Read on file with no objects should return zero-filled - * buffers up to file size (we can get non-zero sizes with - * mknod + truncate, then opening file for read. This is a - * common pattern in NFS case, it seems). Bug 6243 */ - int notzeroed; - /* Since there are no objects on OSTs, we have nothing to get - * lock on and so we are forced to access inode->i_size - * unguarded */ - - /* Read beyond end of file */ - if (*ppos >= i_size_read(inode)) - RETURN(0); - - if (count > i_size_read(inode) - *ppos) - count = i_size_read(inode) - *ppos; - /* Make sure to correctly adjust the file pos pointer for - * EFAULT case */ - notzeroed = clear_user(buf, count); - count -= notzeroed; - *ppos += count; - if (!count) - RETURN(-EFAULT); - RETURN(count); - } -repeat: - if (sbi->ll_max_rw_chunk != 0) { - /* first, let's know the end of the current stripe */ - end = *ppos; - obd_extent_calc(sbi->ll_dt_exp, lsm, OBD_CALC_STRIPE_END, &end); - - /* correct, the end is beyond the request */ - if (end > *ppos + count - 1) - end = *ppos + count - 1; - - /* and chunk shouldn't be too large even if striping is wide */ - if (end - *ppos > sbi->ll_max_rw_chunk) - end = *ppos + sbi->ll_max_rw_chunk - 1; - } else { - end = *ppos + count - 1; - } - - lock_style = ll_file_get_lock(file, (obd_off)(*ppos), end, - buf, &cookie, &tree, OBD_BRW_READ); - if (lock_style < 0) - GOTO(out, retval = lock_style); - - ll_inode_size_lock(inode, 1); - /* - * Consistency guarantees: following possibilities exist for the - * relation between region being read and real file size at this - * moment: - * - * (A): the region is completely inside of the file; - * - * (B-x): x bytes of region are inside of the file, the rest is - * outside; - * - * (C): the region is completely outside of the file. - * - * This classification is stable under DLM lock acquired by - * ll_tree_lock() above, because to change class, other client has to - * take DLM lock conflicting with our lock. Also, any updates to - * ->i_size by other threads on this client are serialized by - * ll_inode_size_lock(). This guarantees that short reads are handled - * correctly in the face of concurrent writes and truncates. - */ - inode_init_lvb(inode, &lvb); - obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1); - kms = lvb.lvb_size; - if (*ppos + count - 1 > kms) { - /* A glimpse is necessary to determine whether we return a - * short read (B) or some zeroes at the end of the buffer (C) */ - ll_inode_size_unlock(inode, 1); - retval = ll_glimpse_size(inode, LDLM_FL_BLOCK_GRANTED); - if (retval) { - if (lock_style != LL_LOCK_STYLE_NOLOCK) - ll_file_put_lock(inode, end, lock_style, - cookie, &tree, OBD_BRW_READ); - goto out; - } - } else { - /* region is within kms and, hence, within real file size (A). - * We need to increase i_size to cover the read region so that - * generic_file_read() will do its job, but that doesn't mean - * the kms size is _correct_, it is only the _minimum_ size. - * If someone does a stat they will get the correct size which - * will always be >= the kms value here. b=11081 */ - if (i_size_read(inode) < kms) - i_size_write(inode, kms); - ll_inode_size_unlock(inode, 1); - } - - chunk = end - *ppos + 1; - CDEBUG(D_INODE, "Read ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n", - inode->i_ino, chunk, *ppos, i_size_read(inode)); - - if (lock_style != LL_LOCK_STYLE_NOLOCK) { - /* turn off the kernel's read-ahead */ - file->f_ra.ra_pages = 0; - - /* initialize read-ahead window once per syscall */ - if (ra == 0) { - ra = 1; - bead.lrr_start = *ppos >> CFS_PAGE_SHIFT; - bead.lrr_count = (count + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT; - ll_ra_read_in(file, &bead); - } - - /* BUG: 5972 */ - file_accessed(file); - retval = generic_file_read(file, buf, chunk, ppos); - ll_file_put_lock(inode, end, lock_style, cookie, &tree, - OBD_BRW_READ); - } else { - retval = ll_file_lockless_io(file, buf, chunk, ppos, READ); - } - - ll_rw_stats_tally(sbi, current->pid, file, chunk, 0); - - if (retval > 0) { - buf += retval; - count -= retval; - sum += retval; - if (retval == chunk && count > 0) - goto repeat; - } - - out: - if (ra != 0) - ll_ra_read_ex(file, &bead); - retval = (sum > 0) ? sum : retval; - RETURN(retval); -} - -/* - * Write to a file (through the page cache). - */ -static ssize_t ll_file_write(struct file *file, const char *buf, size_t count, - loff_t *ppos) -{ - struct inode *inode = file->f_dentry->d_inode; - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; - struct ll_lock_tree tree; - loff_t maxbytes = ll_file_maxbytes(inode); - loff_t lock_start, lock_end, end; - ssize_t retval, chunk, sum = 0; - int tree_locked; - ENTRY; - - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n", - inode->i_ino, inode->i_generation, inode, count, *ppos); - - SIGNAL_MASK_ASSERT(); /* XXX BUG 1511 */ - - /* POSIX, but surprised the VFS doesn't check this already */ - if (count == 0) - RETURN(0); - - /* If file was opened for LL_IOC_LOV_SETSTRIPE but the ioctl wasn't - * called on the file, don't fail the below assertion (bug 2388). */ - if (file->f_flags & O_LOV_DELAY_CREATE && - ll_i2info(inode)->lli_smd == NULL) - RETURN(-EBADF); - - LASSERT(ll_i2info(inode)->lli_smd != NULL); - - down(&ll_i2info(inode)->lli_write_sem); - -repeat: - chunk = 0; /* just to fix gcc's warning */ - end = *ppos + count - 1; - - if (file->f_flags & O_APPEND) { - lock_start = 0; - lock_end = OBD_OBJECT_EOF; - } else if (sbi->ll_max_rw_chunk != 0) { - /* first, let's know the end of the current stripe */ - end = *ppos; - obd_extent_calc(sbi->ll_dt_exp, lsm, OBD_CALC_STRIPE_END, - (obd_off *)&end); - - /* correct, the end is beyond the request */ - if (end > *ppos + count - 1) - end = *ppos + count - 1; - - /* and chunk shouldn't be too large even if striping is wide */ - if (end - *ppos > sbi->ll_max_rw_chunk) - end = *ppos + sbi->ll_max_rw_chunk - 1; - lock_start = *ppos; - lock_end = end; - } else { - lock_start = *ppos; - lock_end = *ppos + count - 1; - } - - tree_locked = ll_file_get_tree_lock(&tree, file, buf, count, - lock_start, lock_end, OBD_BRW_WRITE); - if (tree_locked < 0) - GOTO(out, retval = tree_locked); - - /* This is ok, g_f_w will overwrite this under i_sem if it races - * with a local truncate, it just makes our maxbyte checking easier. - * The i_size value gets updated in ll_extent_lock() as a consequence - * of the [0,EOF] extent lock we requested above. */ - if (file->f_flags & O_APPEND) { - *ppos = i_size_read(inode); - end = *ppos + count - 1; - } - - if (*ppos >= maxbytes) { - send_sig(SIGXFSZ, current, 0); - GOTO(out_unlock, retval = -EFBIG); - } - if (end > maxbytes - 1) - end = maxbytes - 1; - - /* generic_file_write handles O_APPEND after getting i_mutex */ - chunk = end - *ppos + 1; - CDEBUG(D_INFO, "Writing inode %lu, "LPSZ" bytes, offset %Lu\n", - inode->i_ino, chunk, *ppos); - if (tree_locked) - retval = generic_file_write(file, buf, chunk, ppos); - else - retval = ll_file_lockless_io(file, (char*)buf, chunk, - ppos, WRITE); - ll_rw_stats_tally(ll_i2sbi(inode), current->pid, file, chunk, 1); - -out_unlock: - if (tree_locked) - ll_tree_unlock(&tree); - -out: - if (retval > 0) { - buf += retval; - count -= retval; - sum += retval; - if (retval == chunk && count > 0) - goto repeat; - } - - up(&ll_i2info(inode)->lli_write_sem); - - retval = (sum > 0) ? sum : retval; - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES, - retval > 0 ? retval : 0); - RETURN(retval); -} - -/* - * Send file content (through pagecache) somewhere with helper - */ -static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count, - read_actor_t actor, void *target) -{ - struct inode *inode = in_file->f_dentry->d_inode; - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; - struct ll_lock_tree tree; - struct ll_lock_tree_node *node; - struct ost_lvb lvb; - struct ll_ra_read bead; - int rc; - ssize_t retval; - __u64 kms; - ENTRY; - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n", - inode->i_ino, inode->i_generation, inode, count, *ppos); - - /* "If nbyte is 0, read() will return 0 and have no other results." - * -- Single Unix Spec */ - if (count == 0) - RETURN(0); - - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_READ_BYTES, count); - /* turn off the kernel's read-ahead */ - in_file->f_ra.ra_pages = 0; - - /* File with no objects, nothing to lock */ - if (!lsm) - RETURN(generic_file_sendfile(in_file, ppos, count, actor, target)); - - node = ll_node_from_inode(inode, *ppos, *ppos + count - 1, LCK_PR); - if (IS_ERR(node)) - RETURN(PTR_ERR(node)); - - tree.lt_fd = LUSTRE_FPRIVATE(in_file); - rc = ll_tree_lock(&tree, node, NULL, count, - in_file->f_flags & O_NONBLOCK?LDLM_FL_BLOCK_NOWAIT:0); - if (rc != 0) - RETURN(rc); - - ll_clear_file_contended(inode); - ll_inode_size_lock(inode, 1); - /* - * Consistency guarantees: following possibilities exist for the - * relation between region being read and real file size at this - * moment: - * - * (A): the region is completely inside of the file; - * - * (B-x): x bytes of region are inside of the file, the rest is - * outside; - * - * (C): the region is completely outside of the file. - * - * This classification is stable under DLM lock acquired by - * ll_tree_lock() above, because to change class, other client has to - * take DLM lock conflicting with our lock. Also, any updates to - * ->i_size by other threads on this client are serialized by - * ll_inode_size_lock(). This guarantees that short reads are handled - * correctly in the face of concurrent writes and truncates. - */ - inode_init_lvb(inode, &lvb); - obd_merge_lvb(ll_i2sbi(inode)->ll_dt_exp, lsm, &lvb, 1); - kms = lvb.lvb_size; - if (*ppos + count - 1 > kms) { - /* A glimpse is necessary to determine whether we return a - * short read (B) or some zeroes at the end of the buffer (C) */ - ll_inode_size_unlock(inode, 1); - retval = ll_glimpse_size(inode, LDLM_FL_BLOCK_GRANTED); - if (retval) - goto out; - } else { - /* region is within kms and, hence, within real file size (A) */ - i_size_write(inode, kms); - ll_inode_size_unlock(inode, 1); - } - - CDEBUG(D_INFO, "Send ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n", - inode->i_ino, count, *ppos, i_size_read(inode)); - - bead.lrr_start = *ppos >> CFS_PAGE_SHIFT; - bead.lrr_count = (count + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT; - ll_ra_read_in(in_file, &bead); - /* BUG: 5972 */ - file_accessed(in_file); - retval = generic_file_sendfile(in_file, ppos, count, actor, target); - ll_ra_read_ex(in_file, &bead); - - out: - ll_tree_unlock(&tree); - RETURN(retval); -} - -static int ll_lov_recreate_obj(struct inode *inode, struct file *file, - unsigned long arg) -{ - struct ll_inode_info *lli = ll_i2info(inode); - struct obd_export *exp = ll_i2dtexp(inode); - struct ll_recreate_obj ucreatp; - struct obd_trans_info oti = { 0 }; - struct obdo *oa = NULL; - int lsm_size; - int rc = 0; - struct lov_stripe_md *lsm, *lsm2; - ENTRY; - - if (!capable (CAP_SYS_ADMIN)) - RETURN(-EPERM); - - rc = copy_from_user(&ucreatp, (struct ll_recreate_obj *)arg, - sizeof(struct ll_recreate_obj)); - if (rc) { - RETURN(-EFAULT); - } - OBDO_ALLOC(oa); - if (oa == NULL) - RETURN(-ENOMEM); - - down(&lli->lli_size_sem); - lsm = lli->lli_smd; - if (lsm == NULL) - GOTO(out, rc = -ENOENT); - lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * - (lsm->lsm_stripe_count)); - - OBD_ALLOC(lsm2, lsm_size); - if (lsm2 == NULL) - GOTO(out, rc = -ENOMEM); - - oa->o_id = ucreatp.lrc_id; - oa->o_gr = ucreatp.lrc_group; - oa->o_nlink = ucreatp.lrc_ost_idx; - oa->o_flags |= OBD_FL_RECREATE_OBJS; - oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP; - obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | - OBD_MD_FLMTIME | OBD_MD_FLCTIME); - - memcpy(lsm2, lsm, lsm_size); - rc = obd_create(exp, oa, &lsm2, &oti); - - OBD_FREE(lsm2, lsm_size); - GOTO(out, rc); -out: - up(&lli->lli_size_sem); - OBDO_FREE(oa); - return rc; + id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32); + ost_idx = (fid_seq(&fid) >> 16) & 0xffff; + RETURN(ll_lov_recreate(inode, id, 0, ost_idx)); } int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, int flags, struct lov_user_md *lum, int lum_size) { - struct ll_inode_info *lli = ll_i2info(inode); struct lov_stripe_md *lsm; struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; int rc = 0; ENTRY; - down(&lli->lli_size_sem); - lsm = lli->lli_smd; + ll_inode_size_lock(inode, 0); + lsm = ll_i2info(inode)->lli_smd; if (lsm) { - up(&lli->lli_size_sem); + ll_inode_size_unlock(inode, 0); CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n", inode->i_ino); RETURN(-EEXIST); @@ -2043,7 +1313,7 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, ll_release_openhandle(file->f_dentry, &oit); out: - up(&lli->lli_size_sem); + ll_inode_size_unlock(inode, 0); ll_intent_release(&oit); RETURN(rc); out_req_free: @@ -2051,27 +1321,30 @@ out_req_free: goto out; } -int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, - struct lov_mds_md **lmmp, int *lmm_size, +int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, + struct lov_mds_md **lmmp, int *lmm_size, struct ptlrpc_request **request) { struct ll_sb_info *sbi = ll_i2sbi(inode); struct mdt_body *body; struct lov_mds_md *lmm = NULL; struct ptlrpc_request *req = NULL; - struct obd_capa *oc; + struct md_op_data *op_data; int rc, lmmsize; rc = ll_get_max_mdsize(sbi, &lmmsize); if (rc) RETURN(rc); - oc = ll_mdscapa_get(inode); - rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode), - oc, filename, strlen(filename) + 1, - OBD_MD_FLEASIZE | OBD_MD_FLDIREA, lmmsize, - ll_i2suppgid(inode), &req); - capa_put(oc); + op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, + strlen(filename), lmmsize, + LUSTRE_OPC_ANY, NULL); + if (op_data == NULL) + RETURN(-ENOMEM); + + op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; + rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); + ll_finish_md_op_data(op_data); if (rc < 0) { CDEBUG(D_INFO, "md_getattr_name failed " "on %s: rc %d\n", filename, rc); @@ -2091,69 +1364,34 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize); LASSERT(lmm != NULL); + if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) && + (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) { + GOTO(out, rc = -EPROTO); + } + /* * This is coming from the MDS, so is probably in * little endian. We convert it to host endian before * passing it to userspace. */ - if (lmm->lmm_magic == __swab32(LOV_MAGIC)) { - lustre_swab_lov_user_md((struct lov_user_md *)lmm); - lustre_swab_lov_user_md_objects((struct lov_user_md *)lmm); - } else if (lmm->lmm_magic == __swab32(LOV_MAGIC_JOIN)) { - lustre_swab_lov_user_md_join((struct lov_user_md_join *)lmm); - } - - if (lmm->lmm_magic == LOV_MAGIC_JOIN) { - struct lov_stripe_md *lsm; - struct lov_user_md_join *lmj; - int lmj_size, i, aindex = 0; - - rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize); - if (rc < 0) - GOTO(out, rc = -ENOMEM); - rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm); - if (rc) - GOTO(out_free_memmd, rc); - - lmj_size = sizeof(struct lov_user_md_join) + - lsm->lsm_stripe_count * - sizeof(struct lov_user_ost_data_join); - OBD_ALLOC(lmj, lmj_size); - if (!lmj) - GOTO(out_free_memmd, rc = -ENOMEM); - - memcpy(lmj, lmm, sizeof(struct lov_user_md_join)); - for (i = 0; i < lsm->lsm_stripe_count; i++) { - struct lov_extent *lex = - &lsm->lsm_array->lai_ext_array[aindex]; - - if (lex->le_loi_idx + lex->le_stripe_count <= i) - aindex ++; - CDEBUG(D_INFO, "aindex %d i %d l_extent_start " - LPU64" len %d\n", aindex, i, - lex->le_start, (int)lex->le_len); - lmj->lmm_objects[i].l_extent_start = - lex->le_start; - - if ((int)lex->le_len == -1) - lmj->lmm_objects[i].l_extent_end = -1; - else - lmj->lmm_objects[i].l_extent_end = - lex->le_start + lex->le_len; - lmj->lmm_objects[i].l_object_id = - lsm->lsm_oinfo[i]->loi_id; - lmj->lmm_objects[i].l_object_gr = - lsm->lsm_oinfo[i]->loi_gr; - lmj->lmm_objects[i].l_ost_gen = - lsm->lsm_oinfo[i]->loi_ost_gen; - lmj->lmm_objects[i].l_ost_idx = - lsm->lsm_oinfo[i]->loi_ost_idx; + if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) { + /* if function called for directory - we should + * avoid swab not existent lsm objects */ + if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) { + lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm); + if (S_ISREG(body->mode)) + lustre_swab_lov_user_md_objects( + ((struct lov_user_md_v1 *)lmm)->lmm_objects, + ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count); + } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) { + lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm); + if (S_ISREG(body->mode)) + lustre_swab_lov_user_md_objects( + ((struct lov_user_md_v3 *)lmm)->lmm_objects, + ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count); } - lmm = (struct lov_mds_md *)lmj; - lmmsize = lmj_size; -out_free_memmd: - obd_free_memmd(sbi->ll_dt_exp, &lsm); } + out: *lmmp = lmm; *lmm_size = lmmsize; @@ -2171,15 +1409,14 @@ static int ll_lov_setea(struct inode *inode, struct file *file, int rc; ENTRY; - if (!capable (CAP_SYS_ADMIN)) + if (!cfs_capable(CFS_CAP_SYS_ADMIN)) RETURN(-EPERM); OBD_ALLOC(lump, lum_size); if (lump == NULL) { RETURN(-ENOMEM); } - rc = copy_from_user(lump, (struct lov_user_md *)arg, lum_size); - if (rc) { + if (cfs_copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { OBD_FREE(lump, lum_size); RETURN(-EFAULT); } @@ -2193,23 +1430,32 @@ static int ll_lov_setea(struct inode *inode, struct file *file, static int ll_lov_setstripe(struct inode *inode, struct file *file, unsigned long arg) { - struct lov_user_md lum, *lump = (struct lov_user_md *)arg; + struct lov_user_md_v3 lumv3; + struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3; + struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg; + struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg; + int lum_size; int rc; int flags = FMODE_WRITE; ENTRY; - /* Bug 1152: copy properly when this is no longer true */ - LASSERT(sizeof(lum) == sizeof(*lump)); - LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0])); - rc = copy_from_user(&lum, lump, sizeof(lum)); - if (rc) + /* first try with v1 which is smaller than v3 */ + lum_size = sizeof(struct lov_user_md_v1); + if (cfs_copy_from_user(lumv1, lumv1p, lum_size)) RETURN(-EFAULT); - rc = ll_lov_setstripe_ea_info(inode, file, flags, &lum, sizeof(lum)); + if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) { + lum_size = sizeof(struct lov_user_md_v3); + if (cfs_copy_from_user(&lumv3, lumv3p, lum_size)) + RETURN(-EFAULT); + } + + rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); if (rc == 0) { - put_user(0, &lump->lmm_stripe_count); + put_user(0, &lumv1p->lmm_stripe_count); rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), - 0, ll_i2info(inode)->lli_smd, lump); + 0, ll_i2info(inode)->lli_smd, + (void *)arg); } RETURN(rc); } @@ -2218,242 +1464,96 @@ static int ll_lov_getstripe(struct inode *inode, unsigned long arg) { struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; - if (!lsm) - RETURN(-ENODATA); - - return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, lsm, - (void *)arg); -} - -static int ll_get_grouplock(struct inode *inode, struct file *file, - unsigned long arg) -{ - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - ldlm_policy_data_t policy = { .l_extent = { .start = 0, - .end = OBD_OBJECT_EOF}}; - struct lustre_handle lockh = { 0 }; - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; - int flags = 0, rc; - ENTRY; - - if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { - RETURN(-EINVAL); - } - - policy.l_extent.gid = arg; - if (file->f_flags & O_NONBLOCK) - flags = LDLM_FL_BLOCK_NOWAIT; - - rc = ll_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh, flags); - if (rc) - RETURN(rc); - - fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK; - fd->fd_gid = arg; - memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh)); + if (!lsm) + RETURN(-ENODATA); - RETURN(0); + return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, lsm, + (void *)arg); } -static int ll_put_grouplock(struct inode *inode, struct file *file, - unsigned long arg) +int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) { - struct ll_file_data *fd = LUSTRE_FPRIVATE(file); - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; - int rc; + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + struct ccc_grouplock grouplock; + int rc; ENTRY; - if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - /* Ugh, it's already unlocked. */ - RETURN(-EINVAL); - } + if (ll_file_nolock(file)) + RETURN(-EOPNOTSUPP); - if (fd->fd_gid != arg) /* Ugh? Unlocking with different gid? */ + cfs_spin_lock(&lli->lli_lock); + if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { + CWARN("group lock already existed with gid %lu\n", + fd->fd_grouplock.cg_gid); + cfs_spin_unlock(&lli->lli_lock); RETURN(-EINVAL); + } + LASSERT(fd->fd_grouplock.cg_lock == NULL); + cfs_spin_unlock(&lli->lli_lock); - fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK); - - rc = ll_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh); + rc = cl_get_grouplock(cl_i2info(inode)->lli_clob, + arg, (file->f_flags & O_NONBLOCK), &grouplock); if (rc) RETURN(rc); - fd->fd_gid = 0; - memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh)); - - RETURN(0); -} - -static int join_sanity_check(struct inode *head, struct inode *tail) -{ - ENTRY; - if ((ll_i2sbi(head)->ll_flags & LL_SBI_JOIN) == 0) { - CERROR("server do not support join \n"); - RETURN(-EINVAL); - } - if (!S_ISREG(tail->i_mode) || !S_ISREG(head->i_mode)) { - CERROR("tail ino %lu and ino head %lu must be regular\n", - head->i_ino, tail->i_ino); - RETURN(-EINVAL); - } - if (head->i_ino == tail->i_ino) { - CERROR("file %lu can not be joined to itself \n", head->i_ino); - RETURN(-EINVAL); - } - if (i_size_read(head) % JOIN_FILE_ALIGN) { - CERROR("hsize %llu must be times of 64K\n", i_size_read(head)); + cfs_spin_lock(&lli->lli_lock); + if (fd->fd_flags & LL_FILE_GROUP_LOCKED) { + cfs_spin_unlock(&lli->lli_lock); + CERROR("another thread just won the race\n"); + cl_put_grouplock(&grouplock); RETURN(-EINVAL); } + + fd->fd_flags |= LL_FILE_GROUP_LOCKED; + fd->fd_grouplock = grouplock; + cfs_spin_unlock(&lli->lli_lock); + + CDEBUG(D_INFO, "group lock %lu obtained\n", arg); RETURN(0); } -static int join_file(struct inode *head_inode, struct file *head_filp, - struct file *tail_filp) +int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg) { - struct dentry *tail_dentry = tail_filp->f_dentry; - struct lookup_intent oit = {.it_op = IT_OPEN, - .it_flags = head_filp->f_flags|O_JOIN_FILE}; - struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CW, - ll_md_blocking_ast, ldlm_completion_ast, NULL, NULL }; - - struct lustre_handle lockh; - struct md_op_data *op_data; - int rc; - loff_t data; + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + struct ccc_grouplock grouplock; ENTRY; - tail_dentry = tail_filp->f_dentry; - - data = i_size_read(head_inode); - op_data = ll_prep_md_op_data(NULL, head_inode, - tail_dentry->d_parent->d_inode, - tail_dentry->d_name.name, - tail_dentry->d_name.len, 0, - LUSTRE_OPC_ANY, &data); - if (IS_ERR(op_data)) - RETURN(PTR_ERR(op_data)); - - rc = md_enqueue(ll_i2mdexp(head_inode), &einfo, &oit, - op_data, &lockh, NULL, 0, 0); - - ll_finish_md_op_data(op_data); - if (rc < 0) - GOTO(out, rc); - - rc = oit.d.lustre.it_status; - - if (rc < 0 || it_open_error(DISP_OPEN_OPEN, &oit)) { - rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, &oit); - ptlrpc_req_finished((struct ptlrpc_request *) - oit.d.lustre.it_data); - GOTO(out, rc); + cfs_spin_lock(&lli->lli_lock); + if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) { + cfs_spin_unlock(&lli->lli_lock); + CWARN("no group lock held\n"); + RETURN(-EINVAL); } + LASSERT(fd->fd_grouplock.cg_lock != NULL); - if (oit.d.lustre.it_lock_mode) { /* If we got lock - release it right - * away */ - ldlm_lock_decref(&lockh, oit.d.lustre.it_lock_mode); - oit.d.lustre.it_lock_mode = 0; + if (fd->fd_grouplock.cg_gid != arg) { + CWARN("group lock %lu doesn't match current id %lu\n", + arg, fd->fd_grouplock.cg_gid); + cfs_spin_unlock(&lli->lli_lock); + RETURN(-EINVAL); } - ll_release_openhandle(head_filp->f_dentry, &oit); -out: - ll_intent_release(&oit); - RETURN(rc); -} - -static int ll_file_join(struct inode *head, struct file *filp, - char *filename_tail) -{ - struct inode *tail = NULL, *first = NULL, *second = NULL; - struct dentry *tail_dentry; - struct file *tail_filp, *first_filp, *second_filp; - struct ll_lock_tree first_tree, second_tree; - struct ll_lock_tree_node *first_node, *second_node; - struct ll_inode_info *hlli = ll_i2info(head), *tlli; - int rc = 0, cleanup_phase = 0; - ENTRY; - CDEBUG(D_VFSTRACE, "VFS Op:head=%lu/%u(%p) tail %s\n", - head->i_ino, head->i_generation, head, filename_tail); - - tail_filp = filp_open(filename_tail, O_WRONLY, 0644); - if (IS_ERR(tail_filp)) { - CERROR("Can not open tail file %s", filename_tail); - rc = PTR_ERR(tail_filp); - GOTO(cleanup, rc); - } - tail = igrab(tail_filp->f_dentry->d_inode); - - tlli = ll_i2info(tail); - tail_dentry = tail_filp->f_dentry; - LASSERT(tail_dentry); - cleanup_phase = 1; - - /*reorder the inode for lock sequence*/ - first = head->i_ino > tail->i_ino ? head : tail; - second = head->i_ino > tail->i_ino ? tail : head; - first_filp = head->i_ino > tail->i_ino ? filp : tail_filp; - second_filp = head->i_ino > tail->i_ino ? tail_filp : filp; - - CDEBUG(D_INFO, "reorder object from %lu:%lu to %lu:%lu \n", - head->i_ino, tail->i_ino, first->i_ino, second->i_ino); - first_node = ll_node_from_inode(first, 0, OBD_OBJECT_EOF, LCK_EX); - if (IS_ERR(first_node)){ - rc = PTR_ERR(first_node); - GOTO(cleanup, rc); - } - first_tree.lt_fd = first_filp->private_data; - rc = ll_tree_lock(&first_tree, first_node, NULL, 0, 0); - if (rc != 0) - GOTO(cleanup, rc); - cleanup_phase = 2; - - second_node = ll_node_from_inode(second, 0, OBD_OBJECT_EOF, LCK_EX); - if (IS_ERR(second_node)){ - rc = PTR_ERR(second_node); - GOTO(cleanup, rc); - } - second_tree.lt_fd = second_filp->private_data; - rc = ll_tree_lock(&second_tree, second_node, NULL, 0, 0); - if (rc != 0) - GOTO(cleanup, rc); - cleanup_phase = 3; - - rc = join_sanity_check(head, tail); - if (rc) - GOTO(cleanup, rc); + grouplock = fd->fd_grouplock; + memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock)); + fd->fd_flags &= ~LL_FILE_GROUP_LOCKED; + cfs_spin_unlock(&lli->lli_lock); - rc = join_file(head, filp, tail_filp); - if (rc) - GOTO(cleanup, rc); -cleanup: - switch (cleanup_phase) { - case 3: - ll_tree_unlock(&second_tree); - obd_cancel_unused(ll_i2dtexp(second), - ll_i2info(second)->lli_smd, 0, NULL); - case 2: - ll_tree_unlock(&first_tree); - obd_cancel_unused(ll_i2dtexp(first), - ll_i2info(first)->lli_smd, 0, NULL); - case 1: - filp_close(tail_filp, 0); - if (tail) - iput(tail); - if (head && rc == 0) { - obd_free_memmd(ll_i2sbi(head)->ll_dt_exp, - &hlli->lli_smd); - hlli->lli_smd = NULL; - } - case 0: - break; - default: - CERROR("invalid cleanup_phase %d\n", cleanup_phase); - LBUG(); - } - RETURN(rc); + cl_put_grouplock(&grouplock); + CDEBUG(D_INFO, "group lock %lu released\n", arg); + RETURN(0); } +/** + * Close inode open handle + * + * \param dentry [in] dentry which contains the inode + * \param it [in,out] intent which contains open info and result + * + * \retval 0 success + * \retval <0 failure + */ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) { struct inode *inode = dentry->d_inode; @@ -2484,11 +1584,155 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) inode, och); out: /* this one is in place of ll_file_open */ - ptlrpc_req_finished(it->d.lustre.it_data); + if (it_disposition(it, DISP_ENQ_OPEN_REF)) + ptlrpc_req_finished(it->d.lustre.it_data); it_clear_disposition(it, DISP_ENQ_OPEN_REF); RETURN(rc); } +/** + * Get size for inode for which FIEMAP mapping is requested. + * Make the FIEMAP get_info call and returns the result. + */ +int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, + int num_bytes) +{ + struct obd_export *exp = ll_i2dtexp(inode); + struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd; + struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; + int vallen = num_bytes; + int rc; + ENTRY; + + /* Checks for fiemap flags */ + if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) { + fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT; + return -EBADR; + } + + /* Check for FIEMAP_FLAG_SYNC */ + if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) { + rc = filemap_fdatawrite(inode->i_mapping); + if (rc) + return rc; + } + + /* If the stripe_count > 1 and the application does not understand + * DEVICE_ORDER flag, then it cannot interpret the extents correctly. + */ + if (lsm->lsm_stripe_count > 1 && + !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER)) + return -EOPNOTSUPP; + + fm_key.oa.o_id = lsm->lsm_object_id; + fm_key.oa.o_seq = lsm->lsm_object_seq; + fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; + + obdo_from_inode(&fm_key.oa, inode, &ll_i2info(inode)->lli_fid, + OBD_MD_FLSIZE); + /* If filesize is 0, then there would be no objects for mapping */ + if (fm_key.oa.o_size == 0) { + fiemap->fm_mapped_extents = 0; + RETURN(0); + } + + memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); + + rc = obd_get_info(exp, sizeof(fm_key), &fm_key, &vallen, fiemap, lsm); + if (rc) + CERROR("obd_get_info failed: rc = %d\n", rc); + + RETURN(rc); +} + +int ll_fid2path(struct obd_export *exp, void *arg) +{ + struct getinfo_fid2path *gfout, *gfin; + int outsize, rc; + ENTRY; + + /* Need to get the buflen */ + OBD_ALLOC_PTR(gfin); + if (gfin == NULL) + RETURN(-ENOMEM); + if (cfs_copy_from_user(gfin, arg, sizeof(*gfin))) { + OBD_FREE_PTR(gfin); + RETURN(-EFAULT); + } + + outsize = sizeof(*gfout) + gfin->gf_pathlen; + OBD_ALLOC(gfout, outsize); + if (gfout == NULL) { + OBD_FREE_PTR(gfin); + RETURN(-ENOMEM); + } + memcpy(gfout, gfin, sizeof(*gfout)); + OBD_FREE_PTR(gfin); + + /* Call mdc_iocontrol */ + rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL); + if (rc) + GOTO(gf_free, rc); + if (cfs_copy_to_user(arg, gfout, outsize)) + rc = -EFAULT; + +gf_free: + OBD_FREE(gfout, outsize); + RETURN(rc); +} + +static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) +{ + struct ll_user_fiemap *fiemap_s; + size_t num_bytes, ret_bytes; + unsigned int extent_count; + int rc = 0; + + /* Get the extent count so we can calculate the size of + * required fiemap buffer */ + if (get_user(extent_count, + &((struct ll_user_fiemap __user *)arg)->fm_extent_count)) + RETURN(-EFAULT); + num_bytes = sizeof(*fiemap_s) + (extent_count * + sizeof(struct ll_fiemap_extent)); + + OBD_VMALLOC(fiemap_s, num_bytes); + if (fiemap_s == NULL) + RETURN(-ENOMEM); + + /* get the fiemap value */ + if (copy_from_user(fiemap_s,(struct ll_user_fiemap __user *)arg, + sizeof(*fiemap_s))) + GOTO(error, rc = -EFAULT); + + /* If fm_extent_count is non-zero, read the first extent since + * it is used to calculate end_offset and device from previous + * fiemap call. */ + if (extent_count) { + if (copy_from_user(&fiemap_s->fm_extents[0], + (char __user *)arg + sizeof(*fiemap_s), + sizeof(struct ll_fiemap_extent))) + GOTO(error, rc = -EFAULT); + } + + rc = ll_do_fiemap(inode, fiemap_s, num_bytes); + if (rc) + GOTO(error, rc); + + ret_bytes = sizeof(struct ll_user_fiemap); + + if (extent_count != 0) + ret_bytes += (fiemap_s->fm_mapped_extents * + sizeof(struct ll_fiemap_extent)); + + if (copy_to_user((void *)arg, fiemap_s, ret_bytes)) + rc = -EFAULT; + +error: + OBD_VFREE(fiemap_s, num_bytes); + RETURN(rc); +} + int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg) { @@ -2537,24 +1781,17 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, case LL_IOC_LOV_GETSTRIPE: RETURN(ll_lov_getstripe(inode, arg)); case LL_IOC_RECREATE_OBJ: - RETURN(ll_lov_recreate_obj(inode, file, arg)); - case EXT3_IOC_GETFLAGS: - case EXT3_IOC_SETFLAGS: + RETURN(ll_lov_recreate_obj(inode, arg)); + case LL_IOC_RECREATE_FID: + RETURN(ll_lov_recreate_fid(inode, arg)); + case FSFILT_IOC_FIEMAP: + RETURN(ll_ioctl_fiemap(inode, arg)); + case FSFILT_IOC_GETFLAGS: + case FSFILT_IOC_SETFLAGS: RETURN(ll_iocontrol(inode, file, cmd, arg)); - case EXT3_IOC_GETVERSION_OLD: - case EXT3_IOC_GETVERSION: + case FSFILT_IOC_GETVERSION_OLD: + case FSFILT_IOC_GETVERSION: RETURN(put_user(inode->i_generation, (int *)arg)); - case LL_IOC_JOIN: { - char *ftail; - int rc; - - ftail = getname((const char *)arg); - if (IS_ERR(ftail)) - RETURN(PTR_ERR(ftail)); - rc = ll_file_join(inode, file, ftail); - putname(ftail); - RETURN(rc); - } case LL_IOC_GROUP_LOCK: RETURN(ll_get_grouplock(inode, file, arg)); case LL_IOC_GROUP_UNLOCK: @@ -2565,15 +1802,38 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, /* We need to special case any other ioctls we want to handle, * to send them to the MDS/OST as appropriate and to properly * network encode the arg field. - case EXT3_IOC_SETVERSION_OLD: - case EXT3_IOC_SETVERSION: + case FSFILT_IOC_SETVERSION_OLD: + case FSFILT_IOC_SETVERSION: */ case LL_IOC_FLUSHCTX: RETURN(ll_flush_ctx(inode)); + case LL_IOC_PATH2FID: { + if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode), + sizeof(struct lu_fid))) + RETURN(-EFAULT); + + RETURN(0); + } + case OBD_IOC_FID2PATH: + RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg)); + + case LL_IOC_GET_MDTIDX: { + int mdtidx; + + mdtidx = ll_get_mdt_idx(inode); + if (mdtidx < 0) + RETURN(mdtidx); + + if (put_user((int)mdtidx, (int*)arg)) + RETURN(-EFAULT); + + RETURN(0); + } + default: { int err; - if (LLIOC_STOP == + if (LLIOC_STOP == ll_iocontrol_call(inode, file, cmd, arg, &err)) RETURN(err); @@ -2586,8 +1846,6 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, loff_t ll_file_seek(struct file *file, loff_t offset, int origin) { struct inode *inode = file->f_dentry->d_inode; - struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; loff_t retval; ENTRY; retval = offset + ((origin == 2) ? i_size_read(inode) : @@ -2603,15 +1861,11 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin) if (file->f_flags & O_NONBLOCK) nonblock = LDLM_FL_BLOCK_NOWAIT; - if (lsm != NULL) { - rc = ll_glimpse_size(inode, nonblock); - if (rc != 0) - RETURN(rc); - } + rc = cl_glimpse_size(inode); + if (rc != 0) + RETURN(rc); - ll_inode_size_lock(inode, 0); offset += i_size_read(inode); - ll_inode_size_unlock(inode, 0); } else if (origin == 1) { /* SEEK_CUR */ offset += file->f_pos; } @@ -2620,17 +1874,37 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin) if (offset >= 0 && offset <= ll_file_maxbytes(inode)) { if (offset != file->f_pos) { file->f_pos = offset; -#if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)) - file->f_reada = 0; - file->f_version = ++event; -#endif } retval = offset; } - + RETURN(retval); } +#ifdef HAVE_FLUSH_OWNER_ID +int ll_flush(struct file *file, fl_owner_t id) +#else +int ll_flush(struct file *file) +#endif +{ + struct inode *inode = file->f_dentry->d_inode; + struct ll_inode_info *lli = ll_i2info(inode); + struct lov_stripe_md *lsm = lli->lli_smd; + int rc, err; + + /* catch async errors that were recorded back when async writeback + * failed for pages in this mapping. */ + rc = lli->lli_async_rc; + lli->lli_async_rc = 0; + if (lsm) { + err = lov_test_and_clear_async_rc(lsm); + if (rc == 0) + rc = err; + } + + return rc ? -EIO : 0; +} + int ll_fsync(struct file *file, struct dentry *dentry, int data) { struct inode *inode = dentry->d_inode; @@ -2671,17 +1945,18 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data) if (data && lsm) { struct obdo *oa; - + OBDO_ALLOC(oa); if (!oa) RETURN(rc ? rc : -ENOMEM); oa->o_id = lsm->lsm_object_id; - oa->o_gr = lsm->lsm_object_gr; + oa->o_seq = lsm->lsm_object_seq; oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; - obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | - OBD_MD_FLMTIME | OBD_MD_FLCTIME | - OBD_MD_FLGROUP); + obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, + OBD_MD_FLTYPE | OBD_MD_FLATIME | + OBD_MD_FLMTIME | OBD_MD_FLCTIME | + OBD_MD_FLGROUP); oc = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE); err = obd_sync(ll_i2sbi(inode)->ll_dt_exp, oa, lsm, @@ -2699,13 +1974,10 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) { struct inode *inode = file->f_dentry->d_inode; struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ldlm_res_id res_id = - { .name = { fid_seq(ll_inode2fid(inode)), - fid_oid(ll_inode2fid(inode)), - fid_ver(ll_inode2fid(inode)), - LDLM_FLOCK} }; - struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL, - ldlm_flock_completion_ast, NULL, file_lock }; + struct ldlm_enqueue_info einfo = { .ei_type = LDLM_FLOCK, + .ei_cb_cp =ldlm_flock_completion_ast, + .ei_cbdata = file_lock }; + struct md_op_data *op_data; struct lustre_handle lockh = {0}; ldlm_policy_data_t flock; int flags = 0; @@ -2716,7 +1988,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) inode->i_ino, file_lock); ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1); - + if (file_lock->fl_flags & FL_FLOCK) { LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK)); /* set missing params for flock() calls */ @@ -2747,7 +2019,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) break; default: CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type); - LBUG(); + RETURN (-EINVAL); } switch (cmd) { @@ -2774,15 +2046,23 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) break; default: CERROR("unknown fcntl lock command: %d\n", cmd); - LBUG(); + RETURN (-EINVAL); } + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0, + LUSTRE_OPC_ANY, NULL); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); + CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, " "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid, flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end); - rc = ldlm_cli_enqueue(sbi->ll_md_exp, NULL, &einfo, &res_id, - &flock, &flags, NULL, 0, NULL, &lockh, 0); + rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL, + op_data, &lockh, &flock, 0, NULL /* req */, flags); + + ll_finish_md_op_data(op_data); + if ((file_lock->fl_flags & FL_FLOCK) && (rc == 0 || file_lock->fl_type == F_UNLCK)) ll_flock_lock_file_wait(file, file_lock, (cmd == F_SETLKW)); @@ -2803,10 +2083,12 @@ int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock) RETURN(-ENOSYS); } -int ll_have_md_lock(struct inode *inode, __u64 bits) +int ll_have_md_lock(struct inode *inode, __u64 bits, ldlm_mode_t l_req_mode) { struct lustre_handle lockh; ldlm_policy_data_t policy = { .l_inodebits = {bits}}; + ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ? + (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode; struct lu_fid *fid; int flags; ENTRY; @@ -2815,11 +2097,12 @@ int ll_have_md_lock(struct inode *inode, __u64 bits) RETURN(0); fid = &ll_i2info(inode)->lli_fid; - CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid)); + CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid), + ldlm_lockname[mode]); flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK; if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy, - LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) { + mode, &lockh)) { RETURN(1); } RETURN(0); @@ -2864,13 +2147,14 @@ static int ll_inode_revalidate_fini(struct inode *inode, int rc) { return 0; } -int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it) +int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, + __u64 ibits) { struct inode *inode = dentry->d_inode; struct ptlrpc_request *req = NULL; struct ll_sb_info *sbi; struct obd_export *exp; - int rc; + int rc = 0; ENTRY; if (!inode) { @@ -2884,6 +2168,9 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it) exp = ll_i2mdexp(inode); + /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC. + * But under CMD case, it caused some lock issues, should be fixed + * with new CMD ibits lock. See bug 12718 */ if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) { struct lookup_intent oit = { .it_op = IT_GETATTR }; struct md_op_data *op_data; @@ -2895,14 +2182,14 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it) if (IS_ERR(op_data)) RETURN(PTR_ERR(op_data)); - oit.it_flags |= O_CHECK_STALE; + oit.it_create_mode |= M_CHECK_STALE; rc = md_intent_lock(exp, op_data, NULL, 0, /* we are not interested in name based lookup */ &oit, 0, &req, ll_md_blocking_ast, 0); ll_finish_md_op_data(op_data); - oit.it_flags &= ~O_CHECK_STALE; + oit.it_create_mode &= ~M_CHECK_STALE; if (rc < 0) { rc = ll_inode_revalidate_fini(inode, rc); GOTO (out, rc); @@ -2919,17 +2206,18 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it) here to preserve get_cwd functionality on 2.6. Bug 10503 */ if (!dentry->d_inode->i_nlink) { + cfs_spin_lock(&ll_lookup_lock); spin_lock(&dcache_lock); ll_drop_dentry(dentry); spin_unlock(&dcache_lock); + cfs_spin_unlock(&ll_lookup_lock); } ll_lookup_finish_locks(&oit, dentry); - } else if (!ll_have_md_lock(dentry->d_inode, MDS_INODELOCK_UPDATE | - MDS_INODELOCK_LOOKUP)) { + } else if (!ll_have_md_lock(dentry->d_inode, ibits, LCK_MINMODE)) { struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode); obd_valid valid = OBD_MD_FLGETATTR; - struct obd_capa *oc; + struct md_op_data *op_data; int ealen = 0; if (S_ISREG(inode->i_mode)) { @@ -2938,40 +2226,62 @@ int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it) RETURN(rc); valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE; } + + op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, + 0, ealen, LUSTRE_OPC_ANY, + NULL); + if (op_data == NULL) + RETURN(-ENOMEM); + + op_data->op_valid = valid; /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one * capa for this inode. Because we only keep capas of dirs * fresh. */ - oc = ll_mdscapa_get(inode); - rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc, valid, - ealen, &req); - capa_put(oc); + rc = md_getattr(sbi->ll_md_exp, op_data, &req); + ll_finish_md_op_data(op_data); if (rc) { rc = ll_inode_revalidate_fini(inode, rc); RETURN(rc); } rc = ll_prep_inode(&inode, req, NULL); - if (rc) - GOTO(out, rc); } +out: + ptlrpc_req_finished(req); + return rc; +} + +int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it) +{ + struct inode *inode = dentry->d_inode; + int rc; + ENTRY; + + rc = __ll_inode_revalidate_it(dentry, it, MDS_INODELOCK_UPDATE | + MDS_INODELOCK_LOOKUP); /* if object not yet allocated, don't validate size */ - if (ll_i2info(inode)->lli_smd == NULL) - GOTO(out, rc = 0); + if (rc == 0 && ll_i2info(dentry->d_inode)->lli_smd == NULL) { + LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime; + LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime; + LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime; + RETURN(0); + } - /* ll_glimpse_size will prefer locally cached writes if they extend + /* cl_glimpse_size will prefer locally cached writes if they extend * the file */ - rc = ll_glimpse_size(inode, 0); - EXIT; -out: - ptlrpc_req_finished(req); - return rc; + + if (rc == 0) + rc = cl_glimpse_size(inode); + + RETURN(rc); } int ll_getattr_it(struct vfsmount *mnt, struct dentry *de, struct lookup_intent *it, struct kstat *stat) { struct inode *inode = de->d_inode; + struct ll_inode_info *lli = ll_i2info(inode); int res = 0; res = ll_inode_revalidate_it(de, it); @@ -2981,7 +2291,11 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de, return res; stat->dev = inode->i_sb->s_dev; - stat->ino = inode->i_ino; + if (ll_need_32bit_api(ll_i2sbi(inode))) + stat->ino = cl_fid_build_ino32(&lli->lli_fid); + else + stat->ino = inode->i_ino; + stat->mode = inode->i_mode; stat->nlink = inode->i_nlink; stat->uid = inode->i_uid; @@ -2996,10 +2310,8 @@ int ll_getattr_it(struct vfsmount *mnt, struct dentry *de, stat->blksize = 1 << inode->i_blkbits; #endif - ll_inode_size_lock(inode, 0); stat->size = i_size_read(inode); stat->blocks = inode->i_blocks; - ll_inode_size_unlock(inode, 0); return 0; } @@ -3010,6 +2322,42 @@ int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat) return ll_getattr_it(mnt, de, &it, stat); } +#ifdef HAVE_LINUX_FIEMAP_H +int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, + __u64 start, __u64 len) +{ + int rc; + size_t num_bytes; + struct ll_user_fiemap *fiemap; + unsigned int extent_count = fieinfo->fi_extents_max; + + num_bytes = sizeof(*fiemap) + (extent_count * + sizeof(struct ll_fiemap_extent)); + OBD_VMALLOC(fiemap, num_bytes); + + if (fiemap == NULL) + RETURN(-ENOMEM); + + fiemap->fm_flags = fieinfo->fi_flags; + fiemap->fm_extent_count = fieinfo->fi_extents_max; + fiemap->fm_start = start; + fiemap->fm_length = len; + memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start, + sizeof(struct ll_fiemap_extent)); + + rc = ll_do_fiemap(inode, fiemap, num_bytes); + + fieinfo->fi_flags = fiemap->fm_flags; + fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents; + memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0], + fiemap->fm_mapped_extents * sizeof(struct ll_fiemap_extent)); + + OBD_VFREE(fiemap, num_bytes); + return rc; +} +#endif + + static int lustre_check_acl(struct inode *inode, int mask) { @@ -3019,9 +2367,9 @@ int lustre_check_acl(struct inode *inode, int mask) int rc; ENTRY; - spin_lock(&lli->lli_lock); + cfs_spin_lock(&lli->lli_lock); acl = posix_acl_dup(lli->lli_posix_acl); - spin_unlock(&lli->lli_lock); + cfs_spin_unlock(&lli->lli_lock); if (!acl) RETURN(-EAGAIN); @@ -3036,15 +2384,37 @@ int lustre_check_acl(struct inode *inode, int mask) } #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10)) +#ifndef HAVE_INODE_PERMISION_2ARGS int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) +#else +int ll_inode_permission(struct inode *inode, int mask) +#endif { - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n", - inode->i_ino, inode->i_generation, inode, mask); + int rc = 0; + ENTRY; + + /* as root inode are NOT getting validated in lookup operation, + * need to do it before permission check. */ + + if (inode == inode->i_sb->s_root->d_inode) { + struct lookup_intent it = { .it_op = IT_LOOKUP }; + + rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it, + MDS_INODELOCK_LOOKUP); + if (rc) + RETURN(rc); + } + + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n", + inode->i_ino, inode->i_generation, inode, inode->i_mode, mask); + if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT) return lustre_check_remote_perm(inode, mask); - + ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); - return generic_permission(inode, mask, lustre_check_acl); + rc = generic_permission(inode, mask, lustre_check_acl); + + RETURN(rc); } #else int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) @@ -3065,7 +2435,7 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) return -EROFS; if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode)) return -EACCES; - if (current->fsuid == inode->i_uid) { + if (cfs_curproc_fsuid() == inode->i_uid) { mode >>= 6; } else if (1) { if (((mode >> 3) & mask & S_IRWXO) != mask) @@ -3078,7 +2448,7 @@ int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) return rc; } else { check_groups: - if (in_group_p(inode->i_gid)) + if (cfs_curproc_is_in_groups(inode->i_gid)) mode >>= 3; } if ((mode & mask & S_IRWXO) == mask) @@ -3087,40 +2457,68 @@ check_groups: check_capabilities: if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) - if (capable(CAP_DAC_OVERRIDE)) + if (cfs_capable(CFS_CAP_DAC_OVERRIDE)) return 0; - if (capable(CAP_DAC_READ_SEARCH) && ((mask == MAY_READ) || + if (cfs_capable(CFS_CAP_DAC_READ_SEARCH) && ((mask == MAY_READ) || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))) return 0; - + return -EACCES; } #endif +#ifdef HAVE_FILE_READV +#define READ_METHOD readv +#define READ_FUNCTION ll_file_readv +#define WRITE_METHOD writev +#define WRITE_FUNCTION ll_file_writev +#else +#define READ_METHOD aio_read +#define READ_FUNCTION ll_file_aio_read +#define WRITE_METHOD aio_write +#define WRITE_FUNCTION ll_file_aio_write +#endif + /* -o localflock - only provides locally consistent flock locks */ struct file_operations ll_file_operations = { .read = ll_file_read, + .READ_METHOD = READ_FUNCTION, .write = ll_file_write, + .WRITE_METHOD = WRITE_FUNCTION, .ioctl = ll_file_ioctl, .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, .llseek = ll_file_seek, +#ifdef HAVE_KERNEL_SENDFILE .sendfile = ll_file_sendfile, +#endif +#ifdef HAVE_KERNEL_SPLICE_READ + .splice_read = ll_file_splice_read, +#endif .fsync = ll_fsync, + .flush = ll_flush }; struct file_operations ll_file_operations_flock = { .read = ll_file_read, + .READ_METHOD = READ_FUNCTION, .write = ll_file_write, + .WRITE_METHOD = WRITE_FUNCTION, .ioctl = ll_file_ioctl, .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, .llseek = ll_file_seek, +#ifdef HAVE_KERNEL_SENDFILE .sendfile = ll_file_sendfile, +#endif +#ifdef HAVE_KERNEL_SPLICE_READ + .splice_read = ll_file_splice_read, +#endif .fsync = ll_fsync, + .flush = ll_flush, #ifdef HAVE_F_OP_FLOCK .flock = ll_file_flock, #endif @@ -3130,14 +2528,22 @@ struct file_operations ll_file_operations_flock = { /* These are for -o noflock - to return ENOSYS on flock calls */ struct file_operations ll_file_operations_noflock = { .read = ll_file_read, + .READ_METHOD = READ_FUNCTION, .write = ll_file_write, + .WRITE_METHOD = WRITE_FUNCTION, .ioctl = ll_file_ioctl, .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, .llseek = ll_file_seek, +#ifdef HAVE_KERNEL_SENDFILE .sendfile = ll_file_sendfile, +#endif +#ifdef HAVE_KERNEL_SPLICE_READ + .splice_read = ll_file_splice_read, +#endif .fsync = ll_fsync, + .flush = ll_flush, #ifdef HAVE_F_OP_FLOCK .flock = ll_file_noflock, #endif @@ -3156,20 +2562,23 @@ struct inode_operations ll_file_inode_operations = { .getxattr = ll_getxattr, .listxattr = ll_listxattr, .removexattr = ll_removexattr, +#ifdef HAVE_LINUX_FIEMAP_H + .fiemap = ll_fiemap, +#endif }; /* dynamic ioctl number support routins */ static struct llioc_ctl_data { - struct rw_semaphore ioc_sem; - struct list_head ioc_head; -} llioc = { - __RWSEM_INITIALIZER(llioc.ioc_sem), + cfs_rw_semaphore_t ioc_sem; + cfs_list_t ioc_head; +} llioc = { + __RWSEM_INITIALIZER(llioc.ioc_sem), CFS_LIST_HEAD_INIT(llioc.ioc_head) }; struct llioc_data { - struct list_head iocd_list; + cfs_list_t iocd_list; unsigned int iocd_size; llioc_callback_t iocd_cb; unsigned int iocd_count; @@ -3197,9 +2606,9 @@ void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd) in_data->iocd_count = count; memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count); - down_write(&llioc.ioc_sem); - list_add_tail(&in_data->iocd_list, &llioc.ioc_head); - up_write(&llioc.ioc_sem); + cfs_down_write(&llioc.ioc_sem); + cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head); + cfs_up_write(&llioc.ioc_sem); RETURN(in_data); } @@ -3211,19 +2620,19 @@ void ll_iocontrol_unregister(void *magic) if (magic == NULL) return; - down_write(&llioc.ioc_sem); - list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { + cfs_down_write(&llioc.ioc_sem); + cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) { if (tmp == magic) { unsigned int size = tmp->iocd_size; - list_del(&tmp->iocd_list); - up_write(&llioc.ioc_sem); + cfs_list_del(&tmp->iocd_list); + cfs_up_write(&llioc.ioc_sem); OBD_FREE(tmp, size); return; } } - up_write(&llioc.ioc_sem); + cfs_up_write(&llioc.ioc_sem); CWARN("didn't find iocontrol register block with magic: %p\n", magic); } @@ -3231,17 +2640,17 @@ void ll_iocontrol_unregister(void *magic) EXPORT_SYMBOL(ll_iocontrol_register); EXPORT_SYMBOL(ll_iocontrol_unregister); -enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, +enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg, int *rcp) { enum llioc_iter ret = LLIOC_CONT; struct llioc_data *data; int rc = -EINVAL, i; - down_read(&llioc.ioc_sem); - list_for_each_entry(data, &llioc.ioc_head, iocd_list) { + cfs_down_read(&llioc.ioc_sem); + cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) { for (i = 0; i < data->iocd_count; i++) { - if (cmd != data->iocd_cmd[i]) + if (cmd != data->iocd_cmd[i]) continue; ret = data->iocd_cb(inode, file, cmd, arg, data, &rc); @@ -3251,7 +2660,7 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, if (ret == LLIOC_STOP) break; } - up_read(&llioc.ioc_sem); + cfs_up_read(&llioc.ioc_sem); if (rcp) *rcp = rc;