X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Ffile.c;h=e43dfb5afc6278147fabbdcf7c0248302708b27f;hp=447195d989006732f73a946c08c634504d9744a9;hb=9a920a5554bd06f52b13738774d4edf2185c50d9;hpb=caae7c960c776529685b0e820bea5c623d213bd5 diff --git a/lustre/llite/file.c b/lustre/llite/file.c index 447195d..e43dfb5 100644 --- a/lustre/llite/file.c +++ b/lustre/llite/file.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -28,6 +26,8 @@ /* * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,7 +43,6 @@ #define DEBUG_SUBSYSTEM S_LLITE #include #include -#include #include #include #include "llite_internal.h" @@ -53,10 +52,11 @@ struct ll_file_data *ll_file_data_get(void) { - struct ll_file_data *fd; + struct ll_file_data *fd; - OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO); - return fd; + OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO); + fd->fd_write_failed = false; + return fd; } static void ll_file_data_put(struct ll_file_data *fd) @@ -201,15 +201,15 @@ int ll_md_real_close(struct inode *inode, int flags) och_usecount = &lli->lli_open_fd_read_count; } - cfs_down(&lli->lli_och_sem); + cfs_mutex_lock(&lli->lli_och_mutex); if (*och_usecount) { /* There are still users of this handle, so skip freeing it. */ - cfs_up(&lli->lli_och_sem); + cfs_mutex_unlock(&lli->lli_och_mutex); RETURN(0); } och=*och_p; *och_p = NULL; - cfs_up(&lli->lli_och_sem); + cfs_mutex_unlock(&lli->lli_och_mutex); if (och) { /* There might be a race and somebody have freed this och already */ @@ -241,7 +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}}; - cfs_down(&lli->lli_och_sem); + cfs_mutex_lock(&lli->lli_och_mutex); if (fd->fd_omode & FMODE_WRITE) { lockmode = LCK_CW; LASSERT(lli->lli_open_fd_write_count); @@ -255,7 +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--; } - cfs_up(&lli->lli_och_sem); + cfs_mutex_unlock(&lli->lli_och_mutex); if (!md_lock_match(md_exp, flags, ll_inode2fid(inode), LDLM_IBITS, &policy, lockmode, @@ -275,8 +275,6 @@ int ll_md_close(struct obd_export *md_exp, struct inode *inode, RETURN(rc); } -int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm); - /* While this returns an error code, fput() the caller does not, so we need * to make every effort to clean up all of our state here. Also, applications * rarely check close errors and even if an error is returned they will not @@ -287,7 +285,6 @@ int ll_file_release(struct inode *inode, struct file *file) struct ll_file_data *fd; struct ll_sb_info *sbi = ll_i2sbi(inode); struct ll_inode_info *lli = ll_i2info(inode); - struct lov_stripe_md *lsm = lli->lli_smd; int rc; ENTRY; @@ -316,7 +313,8 @@ 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 && lli->lli_opendir_pid != 0) + if (S_ISDIR(inode->i_mode) && 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) { @@ -325,13 +323,14 @@ int ll_file_release(struct inode *inode, struct file *file) RETURN(0); } - if (lsm) - lov_test_and_clear_async_rc(lsm); - lli->lli_async_rc = 0; + if (!S_ISDIR(inode->i_mode)) { + lov_read_and_clear_async_rc(lli->lli_clob); + 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)) + if (CFS_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, cfs_fail_val)) libcfs_debug_dumplog(); RETURN(rc); @@ -346,6 +345,7 @@ static int ll_intent_file_open(struct file *file, void *lmm, const int len = file->f_dentry->d_name.len; struct md_op_data *op_data; struct ptlrpc_request *req; + __u32 opc = LUSTRE_OPC_ANY; int rc; ENTRY; @@ -361,15 +361,19 @@ static int ll_intent_file_open(struct file *file, void *lmm, * makes a good candidate for using OPEN lock */ /* If lmmsize & lmm are not 0, we are just setting stripe info * parameters. No need for the open lock */ - if (!lmm && !lmmsize) + if (lmm == NULL && lmmsize == 0) { itp->it_flags |= MDS_OPEN_LOCK; + if (itp->it_flags & FMODE_WRITE) + opc = LUSTRE_OPC_CREATE; + } op_data = ll_prep_md_op_data(NULL, parent->d_inode, file->f_dentry->d_inode, name, len, - O_RDWR, LUSTRE_OPC_ANY, NULL); + O_RDWR, opc, NULL); if (IS_ERR(op_data)) RETURN(PTR_ERR(op_data)); + itp->it_flags |= MDS_OPEN_BY_FID; rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp, 0 /*unused */, &req, ll_md_blocking_ast, 0); ll_finish_md_op_data(op_data); @@ -384,6 +388,9 @@ static int ll_intent_file_open(struct file *file, void *lmm, GOTO(out, rc); } + if (it_disposition(itp, DISP_LOOKUP_NEG)) + GOTO(out, rc = -ENOENT); + if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) { rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp); CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc); @@ -392,9 +399,8 @@ static int ll_intent_file_open(struct file *file, void *lmm, 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, NULL); + ll_set_lock_data(sbi->ll_md_exp, file->f_dentry->d_inode, + itp, NULL); out: ptlrpc_req_finished(itp->d.lustre.it_data); @@ -473,9 +479,7 @@ int ll_local_open(struct file *file, struct lookup_intent *it, /* Open a file, and (for the very first open) create objects on the OSTs at * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object - * creation or open until ll_lov_setstripe() ioctl is called. We grab - * lli_open_sem to ensure no other process will create objects, send the - * stripe MD to the MDS, or try to destroy the objects if that fails. + * creation or open until ll_lov_setstripe() ioctl is called. * * If we already have the stripe MD locally then we don't request it in * md_open(), by passing a lmm_size = 0. @@ -491,10 +495,8 @@ int ll_file_open(struct inode *inode, struct file *file) struct ll_inode_info *lli = ll_i2info(inode); struct lookup_intent *it, oit = { .it_op = IT_OPEN, .it_flags = file->f_flags }; - struct lov_stripe_md *lsm; - struct ptlrpc_request *req = NULL; - struct obd_client_handle **och_p; - __u64 *och_usecount; + struct obd_client_handle **och_p = NULL; + __u64 *och_usecount = NULL; struct ll_file_data *fd; int rc = 0, opendir_set = 0; ENTRY; @@ -502,22 +504,18 @@ int ll_file_open(struct inode *inode, struct file *file) CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino, inode->i_generation, inode, file->f_flags); -#ifdef HAVE_VFS_INTENT_PATCHES - it = file->f_it; -#else it = file->private_data; /* XXX: compat macro */ file->private_data = NULL; /* prevent ll_local_open assertion */ -#endif fd = ll_file_data_get(); if (fd == NULL) - RETURN(-ENOMEM); + GOTO(out_och_free, rc = -ENOMEM); fd->fd_file = file; if (S_ISDIR(inode->i_mode)) { cfs_spin_lock(&lli->lli_sa_lock); - if (lli->lli_opendir_key == NULL && lli->lli_opendir_pid == 0) { - LASSERT(lli->lli_sai == NULL); + if (lli->lli_opendir_key == NULL && lli->lli_sai == NULL && + lli->lli_opendir_pid == 0) { lli->lli_opendir_key = fd; lli->lli_opendir_pid = cfs_curproc_pid(); opendir_set = 1; @@ -543,7 +541,7 @@ int ll_file_open(struct inode *inode, struct file *file) * dentry_open after call to open_namei that checks permissions. * Only nfsd_open call dentry_open directly without checking * permissions and because of that this code below is safe. */ - if (oit.it_flags & FMODE_WRITE) + if (oit.it_flags & (FMODE_WRITE | FMODE_READ)) oit.it_flags |= MDS_OPEN_OWNEROVERRIDE; /* We do not want O_EXCL here, presumably we opened the file @@ -572,28 +570,25 @@ restart: och_usecount = &lli->lli_open_fd_read_count; } - cfs_down(&lli->lli_och_sem); + cfs_mutex_lock(&lli->lli_och_mutex); 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); + cfs_mutex_unlock(&lli->lli_och_mutex); GOTO(out_openerr, rc); } + ll_release_openhandle(file->f_dentry, it); - lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, - LPROC_LL_OPEN); } (*och_usecount)++; rc = ll_local_open(file, it, fd, NULL); if (rc) { (*och_usecount)--; - cfs_up(&lli->lli_och_sem); - ll_file_data_put(fd); + cfs_mutex_unlock(&lli->lli_och_mutex); GOTO(out_openerr, rc); } } else { @@ -602,31 +597,22 @@ restart: /* We cannot just request lock handle now, new ELC code means that one of other OPEN locks for this file could be cancelled, and since blocking ast handler - would attempt to grab och_sem as well, that would + would attempt to grab och_mutex as well, that would result in a deadlock */ - cfs_up(&lli->lli_och_sem); + cfs_mutex_unlock(&lli->lli_och_mutex); it->it_create_mode |= M_CHECK_STALE; rc = ll_intent_file_open(file, NULL, 0, it); it->it_create_mode &= ~M_CHECK_STALE; - if (rc) { - ll_file_data_put(fd); + if (rc) GOTO(out_openerr, rc); - } - /* Got some error? Release the request */ - if (it->d.lustre.it_status < 0) { - req = it->d.lustre.it_data; - ptlrpc_req_finished(req); - } goto restart; } OBD_ALLOC(*och_p, sizeof (struct obd_client_handle)); - if (!*och_p) { - ll_file_data_put(fd); + if (!*och_p) GOTO(out_och_free, rc = -ENOMEM); - } + (*och_usecount)++; - req = it->d.lustre.it_data; /* md_intent_lock() didn't get a request ref if there was an * open error, so don't do cleanup on the request here @@ -634,53 +620,57 @@ restart: /* XXX (green): Should not we bail out on any error here, not * just open error? */ rc = it_open_error(DISP_OPEN_OPEN, it); - if (rc) { - ll_file_data_put(fd); + if (rc) GOTO(out_och_free, rc); - } - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); + LASSERT(it_disposition(it, DISP_ENQ_OPEN_REF)); + rc = ll_local_open(file, it, fd, *och_p); - if (rc) { - ll_file_data_put(fd); + if (rc) GOTO(out_och_free, rc); - } } - cfs_up(&lli->lli_och_sem); + cfs_mutex_unlock(&lli->lli_och_mutex); + fd = NULL; - /* Must do this outside lli_och_sem lock to prevent deadlock where + /* Must do this outside lli_och_mutex lock to prevent deadlock where different kind of OPEN lock for this same inode gets cancelled by ldlm_cancel_lru */ if (!S_ISREG(inode->i_mode)) - GOTO(out, rc); + GOTO(out_och_free, rc); ll_capa_open(inode); - lsm = lli->lli_smd; - if (lsm == NULL) { + if (!lli->lli_has_smd) { if (file->f_flags & O_LOV_DELAY_CREATE || !(file->f_mode & FMODE_WRITE)) { CDEBUG(D_INODE, "object creation was delayed\n"); - GOTO(out, rc); + GOTO(out_och_free, rc); } } file->f_flags &= ~O_LOV_DELAY_CREATE; - GOTO(out, rc); -out: - ptlrpc_req_finished(req); - if (req) - it_clear_disposition(it, DISP_ENQ_OPEN_REF); + GOTO(out_och_free, rc); + out_och_free: + if (it && it_disposition(it, DISP_ENQ_OPEN_REF)) { + ptlrpc_req_finished(it->d.lustre.it_data); + it_clear_disposition(it, DISP_ENQ_OPEN_REF); + } + if (rc) { - if (*och_p) { + if (och_p && *och_p) { OBD_FREE(*och_p, sizeof (struct obd_client_handle)); *och_p = NULL; /* OBD_FREE writes some magic there */ (*och_usecount)--; } - cfs_up(&lli->lli_och_sem); + cfs_mutex_unlock(&lli->lli_och_mutex); + out_openerr: if (opendir_set != 0) ll_stop_statahead(inode, lli->lli_opendir_key); + if (fd != NULL) + ll_file_data_put(fd); + } else { + ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1); } return rc; @@ -709,7 +699,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ | OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME | - OBD_MD_FLGROUP | OBD_MD_FLEPOCH; + OBD_MD_FLGROUP | OBD_MD_FLEPOCH | + OBD_MD_FLDATAVERSION; oinfo.oi_capa = capa; if (sync) { oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS; @@ -729,7 +720,8 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, 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); + OBD_MD_FLCTIME | OBD_MD_FLSIZE | + OBD_MD_FLDATAVERSION); RETURN(rc); } @@ -740,55 +732,62 @@ static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp, 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; - - 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); + struct obd_capa *capa = ll_mdscapa_get(inode); + struct lov_stripe_md *lsm; + int rc; + ENTRY; + + lsm = ccc_inode_lsm_get(inode); + rc = ll_lsm_getattr(lsm, 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 %llu, blocks %llu, blksize %lu\n", + lsm ? lsm->lsm_object_id : 0, i_size_read(inode), + (unsigned long long)inode->i_blocks, + (unsigned long)ll_inode_blksize(inode)); + } + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); } int ll_merge_lvb(struct inode *inode) { - struct ll_inode_info *lli = ll_i2info(inode); - struct ll_sb_info *sbi = ll_i2sbi(inode); - struct ost_lvb lvb; - int rc; - - ENTRY; - - ll_inode_size_lock(inode, 1); - inode_init_lvb(inode, &lvb); - - /* 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); - - 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(rc); + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct lov_stripe_md *lsm; + struct ost_lvb lvb; + int rc = 0; + + ENTRY; + + lsm = ccc_inode_lsm_get(inode); + ll_inode_size_lock(inode); + inode_init_lvb(inode, &lvb); + + /* 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; + if (lsm != NULL) { + rc = obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 0); + cl_isize_write_nolock(inode, lvb.lvb_size); + + 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); + ccc_inode_lsm_put(inode, lsm); + + RETURN(rc); } int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm, @@ -812,10 +811,11 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write) { struct inode *inode = file->f_dentry->d_inode; - 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); + if (write) { + io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND); + io->u.ci_wr.wr_sync = file->f_flags & O_SYNC || IS_SYNC(inode); + } io->ci_obj = ll_i2info(inode)->lli_clob; io->ci_lockreq = CILR_MAYBE; if (ll_file_nolock(file)) { @@ -826,22 +826,24 @@ void ll_io_init(struct cl_io *io, const struct file *file, int write) } } -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) +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; + struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + struct cl_io *io; + ssize_t result; ENTRY; - io = &ccc_env_info(env)->cti_io; + io = ccc_env_thread_io(env); 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; + int write_mutex_locked = 0; cio->cui_fd = LUSTRE_FPRIVATE(file); vio->cui_io_subtype = args->via_io_subtype; @@ -856,8 +858,12 @@ static ssize_t ll_file_io_generic(const struct lu_env *env, #endif if ((iot == CIT_WRITE) && !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) { - cfs_down(&lli->lli_write_sem); - write_sem_locked = 1; + if (cfs_mutex_lock_interruptible(&lli-> + lli_write_mutex)) + GOTO(out, result = -ERESTARTSYS); + write_mutex_locked = 1; + } else if (iot == CIT_READ) { + cfs_down_read(&lli->lli_trunc_sem); } break; case IO_SENDFILE: @@ -873,8 +879,10 @@ static ssize_t ll_file_io_generic(const struct lu_env *env, LBUG(); } result = cl_io_loop(env, io); - if (write_sem_locked) - cfs_up(&lli->lli_write_sem); + if (write_mutex_locked) + cfs_mutex_unlock(&lli->lli_write_mutex); + 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; @@ -884,8 +892,25 @@ static ssize_t ll_file_io_generic(const struct lu_env *env, result = io->ci_nob; *ppos = io->u.ci_wr.wr.crw_pos; } + GOTO(out, result); +out: cl_io_fini(env, io); - RETURN(result); + + if (iot == CIT_READ) { + if (result >= 0) + ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), + LPROC_LL_READ_BYTES, result); + } else if (iot == CIT_WRITE) { + if (result >= 0) { + ll_stats_ops_tally(ll_i2sbi(file->f_dentry->d_inode), + LPROC_LL_WRITE_BYTES, result); + fd->fd_write_failed = false; + } else { + fd->fd_write_failed = true; + } + } + + return result; } @@ -1201,26 +1226,26 @@ static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos, static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq, obd_count ost_idx) { - 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; - - OBDO_ALLOC(oa); - if (oa == NULL) - RETURN(-ENOMEM); - - ll_inode_size_lock(inode, 0); - lsm = ll_i2info(inode)->lli_smd; + 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 = NULL, *lsm2; + ENTRY; + + OBDO_ALLOC(oa); + if (oa == NULL) + RETURN(-ENOMEM); + + lsm = ccc_inode_lsm_get(inode); if (lsm == NULL) GOTO(out, rc = -ENOENT); + lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) * (lsm->lsm_stripe_count)); - OBD_ALLOC(lsm2, lsm_size); + OBD_ALLOC_LARGE(lsm2, lsm_size); if (lsm2 == NULL) GOTO(out, rc = -ENOMEM); @@ -1229,17 +1254,20 @@ static int ll_lov_recreate(struct inode *inode, obd_id id, obd_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); + obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME | + OBD_MD_FLMTIME | OBD_MD_FLCTIME); + obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid); memcpy(lsm2, lsm, lsm_size); - rc = obd_create(exp, oa, &lsm2, &oti); + ll_inode_size_lock(inode); + rc = obd_create(NULL, exp, oa, &lsm2, &oti); + ll_inode_size_unlock(inode); - OBD_FREE(lsm2, lsm_size); - GOTO(out, rc); + OBD_FREE_LARGE(lsm2, lsm_size); + GOTO(out, rc); out: - ll_inode_size_unlock(inode, 0); - OBDO_FREE(oa); - return rc; + ccc_inode_lsm_put(inode, lsm); + OBDO_FREE(oa); + return rc; } static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg) @@ -1280,25 +1308,23 @@ static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg) int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, int flags, struct lov_user_md *lum, int lum_size) { - struct lov_stripe_md *lsm; - struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; - int rc = 0; - ENTRY; - - ll_inode_size_lock(inode, 0); - lsm = ll_i2info(inode)->lli_smd; - if (lsm) { - ll_inode_size_unlock(inode, 0); - CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n", - inode->i_ino); - RETURN(-EEXIST); - } - + struct lov_stripe_md *lsm = NULL; + struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags}; + int rc = 0; + ENTRY; + + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) { + ccc_inode_lsm_put(inode, lsm); + CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n", + inode->i_ino); + RETURN(-EEXIST); + } + + ll_inode_size_lock(inode); rc = ll_intent_file_open(file, lum, lum_size, &oit); if (rc) GOTO(out, rc); - if (it_disposition(&oit, DISP_LOOKUP_NEG)) - GOTO(out_req_free, rc = -ENOENT); rc = oit.d.lustre.it_status; if (rc < 0) GOTO(out_req_free, rc); @@ -1306,12 +1332,13 @@ int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file, ll_release_openhandle(file->f_dentry, &oit); out: - ll_inode_size_unlock(inode, 0); - ll_intent_release(&oit); - RETURN(rc); + ll_inode_size_unlock(inode); + ll_intent_release(&oit); + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); out_req_free: - ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data); - goto out; + ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data); + goto out; } int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, @@ -1332,8 +1359,8 @@ int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename, op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, strlen(filename), lmmsize, LUSTRE_OPC_ANY, NULL); - if (op_data == NULL) - RETURN(-ENOMEM); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; rc = md_getattr_name(sbi->ll_md_exp, op_data, &req); @@ -1405,18 +1432,18 @@ static int ll_lov_setea(struct inode *inode, struct file *file, if (!cfs_capable(CFS_CAP_SYS_ADMIN)) RETURN(-EPERM); - OBD_ALLOC(lump, lum_size); + OBD_ALLOC_LARGE(lump, lum_size); if (lump == NULL) { RETURN(-ENOMEM); } if (cfs_copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) { - OBD_FREE(lump, lum_size); + OBD_FREE_LARGE(lump, lum_size); RETURN(-EFAULT); } rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size); - OBD_FREE(lump, lum_size); + OBD_FREE_LARGE(lump, lum_size); RETURN(rc); } @@ -1445,23 +1472,28 @@ static int ll_lov_setstripe(struct inode *inode, struct file *file, rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size); if (rc == 0) { - put_user(0, &lumv1p->lmm_stripe_count); - rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), - 0, ll_i2info(inode)->lli_smd, - (void *)arg); - } - RETURN(rc); + struct lov_stripe_md *lsm; + put_user(0, &lumv1p->lmm_stripe_count); + lsm = ccc_inode_lsm_get(inode); + rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), + 0, lsm, (void *)arg); + ccc_inode_lsm_put(inode, lsm); + } + RETURN(rc); } 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); + struct lov_stripe_md *lsm; + int rc = -ENODATA; + ENTRY; + + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) + rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, + lsm, (void *)arg); + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); } int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg) @@ -1577,9 +1609,10 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) inode, och); out: /* this one is in place of ll_file_open */ - if (it_disposition(it, DISP_ENQ_OPEN_REF)) + if (it_disposition(it, DISP_ENQ_OPEN_REF)) { ptlrpc_req_finished(it->d.lustre.it_data); - it_clear_disposition(it, DISP_ENQ_OPEN_REF); + it_clear_disposition(it, DISP_ENQ_OPEN_REF); + } RETURN(rc); } @@ -1590,8 +1623,8 @@ int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it) 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 obd_export *exp = ll_i2dtexp(inode); + struct lov_stripe_md *lsm = NULL; struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, }; int vallen = num_bytes; int rc; @@ -1610,40 +1643,52 @@ int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap, 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; + lsm = ccc_inode_lsm_get(inode); + if (lsm == NULL) + return -ENOENT; + + /* 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)) + GOTO(out, rc = -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); + obdo_from_inode(&fm_key.oa, inode, OBD_MD_FLSIZE); + obdo_set_parent_fid(&fm_key.oa, &ll_i2info(inode)->lli_fid); /* 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); + GOTO(out, rc = 0); } memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap)); - rc = obd_get_info(exp, sizeof(fm_key), &fm_key, &vallen, fiemap, lsm); + rc = obd_get_info(NULL, exp, sizeof(fm_key), &fm_key, &vallen, + fiemap, lsm); if (rc) CERROR("obd_get_info failed: rc = %d\n", rc); - RETURN(rc); +out: + ccc_inode_lsm_put(inode, lsm); + RETURN(rc); } -int ll_fid2path(struct obd_export *exp, void *arg) +int ll_fid2path(struct inode *inode, void *arg) { + struct obd_export *exp = ll_i2mdexp(inode); struct getinfo_fid2path *gfout, *gfin; int outsize, rc; ENTRY; + if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) && + !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH)) + RETURN(-EPERM); + /* Need to get the buflen */ OBD_ALLOC_PTR(gfin); if (gfin == NULL) @@ -1689,7 +1734,7 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) num_bytes = sizeof(*fiemap_s) + (extent_count * sizeof(struct ll_fiemap_extent)); - OBD_VMALLOC(fiemap_s, num_bytes); + OBD_ALLOC_LARGE(fiemap_s, num_bytes); if (fiemap_s == NULL) RETURN(-ENOMEM); @@ -1722,15 +1767,62 @@ static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg) rc = -EFAULT; error: - OBD_VFREE(fiemap_s, num_bytes); + OBD_FREE_LARGE(fiemap_s, num_bytes); RETURN(rc); } -int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, - unsigned long arg) +/* + * Read the data_version for inode. + * + * This value is computed using stripe object version on OST. + * Version is computed using server side locking. + * + * @param extent_lock Take extent lock. Not needed if a process is already + * holding the OST object group locks. + */ +static int ll_data_version(struct inode *inode, __u64 *data_version, + int extent_lock) { + struct lov_stripe_md *lsm = NULL; + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct obdo *obdo = NULL; + int rc; + ENTRY; + + /* If no stripe, we consider version is 0. */ + lsm = ccc_inode_lsm_get(inode); + if (lsm == NULL) { + *data_version = 0; + CDEBUG(D_INODE, "No object for inode\n"); + RETURN(0); + } + + OBD_ALLOC_PTR(obdo); + if (obdo == NULL) { + ccc_inode_lsm_put(inode, lsm); + RETURN(-ENOMEM); + } + + rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, obdo, 0, extent_lock); + if (!rc) { + if (!(obdo->o_valid & OBD_MD_FLDATAVERSION)) + rc = -EOPNOTSUPP; + else + *data_version = obdo->o_data_version; + } + + OBD_FREE_PTR(obdo); + ccc_inode_lsm_put(inode, lsm); + + RETURN(rc); +} + +long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + struct inode *inode = file->f_dentry->d_inode; struct ll_file_data *fd = LUSTRE_FPRIVATE(file); int flags; + ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino, @@ -1808,7 +1900,23 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, RETURN(0); } case OBD_IOC_FID2PATH: - RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg)); + RETURN(ll_fid2path(inode, (void *)arg)); + case LL_IOC_DATA_VERSION: { + struct ioc_data_version idv; + int rc; + + if (cfs_copy_from_user(&idv, (char *)arg, sizeof(idv))) + RETURN(-EFAULT); + + rc = ll_data_version(inode, &idv.idv_version, + !(idv.idv_flags & LL_DV_NOFLUSH)); + + if (rc == 0 && + cfs_copy_to_user((char *) arg, &idv, sizeof(idv))) + RETURN(-EFAULT); + + RETURN(rc); + } case LL_IOC_GET_MDTIDX: { int mdtidx; @@ -1822,7 +1930,9 @@ int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd, RETURN(0); } - + case OBD_IOC_GETDTNAME: + case OBD_IOC_GETMDNAME: + RETURN(ll_get_obd_name(inode, cmd, arg)); default: { int err; @@ -1843,18 +1953,15 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin) ENTRY; retval = offset + ((origin == 2) ? i_size_read(inode) : (origin == 1) ? file->f_pos : 0); - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%Lu=%#Lx(%s)\n", + CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%llu=%#llx(%s)\n", inode->i_ino, inode->i_generation, inode, retval, retval, origin == 2 ? "SEEK_END": origin == 1 ? "SEEK_CUR" : "SEEK_SET"); ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1); if (origin == 2) { /* SEEK_END */ - int nonblock = 0, rc; - - if (file->f_flags & O_NONBLOCK) - nonblock = LDLM_FL_BLOCK_NOWAIT; + int rc; - rc = cl_glimpse_size(inode); + rc = ll_glimpse_size(inode); if (rc != 0) RETURN(rc); @@ -1874,33 +1981,122 @@ loff_t ll_file_seek(struct file *file, loff_t offset, int origin) RETURN(retval); } +int ll_flush(struct file *file, fl_owner_t id) +{ + struct inode *inode = file->f_dentry->d_inode; + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + int rc, err; + + LASSERT(!S_ISDIR(inode->i_mode)); + + /* 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; + err = lov_read_and_clear_async_rc(lli->lli_clob); + if (rc == 0) + rc = err; + + /* The application has been told write failure already. + * Do not report failure again. */ + if (fd->fd_write_failed) + return 0; + return rc ? -EIO : 0; +} + +/** + * Called to make sure a portion of file has been written out. + * if @local_only is not true, it will send OST_SYNC RPCs to ost. + * + * Return how many pages have been written. + */ +int cl_sync_file_range(struct inode *inode, loff_t start, loff_t end, + enum cl_fsync_mode mode) +{ + struct cl_env_nest nest; + struct lu_env *env; + struct cl_io *io; + struct obd_capa *capa = NULL; + struct cl_fsync_io *fio; + int result; + ENTRY; + + if (mode != CL_FSYNC_NONE && mode != CL_FSYNC_LOCAL && + mode != CL_FSYNC_DISCARD && mode != CL_FSYNC_ALL) + RETURN(-EINVAL); + + env = cl_env_nested_get(&nest); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + capa = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE); + + io = ccc_env_thread_io(env); + io->ci_obj = cl_i2info(inode)->lli_clob; + io->ci_ignore_layout = 1; + + /* initialize parameters for sync */ + fio = &io->u.ci_fsync; + fio->fi_capa = capa; + fio->fi_start = start; + fio->fi_end = end; + fio->fi_fid = ll_inode2fid(inode); + fio->fi_mode = mode; + fio->fi_nr_written = 0; + + if (cl_io_init(env, io, CIT_FSYNC, io->ci_obj) == 0) + result = cl_io_loop(env, io); + else + result = io->ci_result; + if (result == 0) + result = fio->fi_nr_written; + cl_io_fini(env, io); + cl_env_nested_put(&nest, env); + + capa_put(capa); + + RETURN(result); +} + +#ifdef HAVE_FILE_FSYNC_4ARGS +int ll_fsync(struct file *file, loff_t start, loff_t end, int data) +#elif defined(HAVE_FILE_FSYNC_2ARGS) +int ll_fsync(struct file *file, int data) +#else int ll_fsync(struct file *file, struct dentry *dentry, int data) +#endif { - struct inode *inode = dentry->d_inode; + 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 ptlrpc_request *req; struct obd_capa *oc; + struct lov_stripe_md *lsm; int rc, err; ENTRY; CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino, inode->i_generation, inode); ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1); +#ifdef HAVE_FILE_FSYNC_4ARGS + rc = filemap_write_and_wait_range(inode->i_mapping, start, end); + mutex_lock(&inode->i_mutex); +#else /* fsync's caller has already called _fdata{sync,write}, we want * that IO to finish before calling the osc and mdc sync methods */ rc = filemap_fdatawait(inode->i_mapping); +#endif /* catch async errors that were recorded back when async writeback * failed for pages in this mapping. */ - err = lli->lli_async_rc; - lli->lli_async_rc = 0; - if (rc == 0) - rc = err; - if (lsm) { - err = lov_test_and_clear_async_rc(lsm); + if (!S_ISDIR(inode->i_mode)) { + err = lli->lli_async_rc; + lli->lli_async_rc = 0; if (rc == 0) rc = err; + err = lov_read_and_clear_async_rc(lli->lli_clob); + if (rc == 0) + rc = err; } oc = ll_mdscapa_get(inode); @@ -1912,31 +2108,25 @@ int ll_fsync(struct file *file, struct dentry *dentry, int data) if (!err) ptlrpc_req_finished(req); - if (data && lsm) { - struct obdo *oa; - - OBDO_ALLOC(oa); - if (!oa) - RETURN(rc ? rc : -ENOMEM); - - oa->o_id = lsm->lsm_object_id; - oa->o_seq = lsm->lsm_object_seq; - oa->o_valid = OBD_MD_FLID | 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, - 0, OBD_OBJECT_EOF, oc); - capa_put(oc); - if (!rc) - rc = err; - OBDO_FREE(oa); - } - - RETURN(rc); + lsm = ccc_inode_lsm_get(inode); + if (data && lsm) { + struct ll_file_data *fd = LUSTRE_FPRIVATE(file); + + err = cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, + CL_FSYNC_ALL); + if (rc == 0 && err < 0) + rc = err; + if (rc < 0) + fd->fd_write_failed = true; + else + fd->fd_write_failed = false; + } + ccc_inode_lsm_put(inode, lsm); + +#ifdef HAVE_FILE_FSYNC_4ARGS + mutex_unlock(&inode->i_mutex); +#endif + RETURN(rc); } int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) @@ -1948,7 +2138,7 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) .ei_cbdata = file_lock }; struct md_op_data *op_data; struct lustre_handle lockh = {0}; - ldlm_policy_data_t flock; + ldlm_policy_data_t flock = {{0}}; int flags = 0; int rc; ENTRY; @@ -1960,13 +2150,28 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) if (file_lock->fl_flags & FL_FLOCK) { LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK)); - /* set missing params for flock() calls */ - file_lock->fl_end = OFFSET_MAX; - file_lock->fl_pid = current->tgid; + /* flocks are whole-file locks */ + flock.l_flock.end = OFFSET_MAX; + /* For flocks owner is determined by the local file desctiptor*/ + flock.l_flock.owner = (unsigned long)file_lock->fl_file; + } else if (file_lock->fl_flags & FL_POSIX) { + flock.l_flock.owner = (unsigned long)file_lock->fl_owner; + flock.l_flock.start = file_lock->fl_start; + flock.l_flock.end = file_lock->fl_end; + } else { + RETURN(-EINVAL); } flock.l_flock.pid = file_lock->fl_pid; - flock.l_flock.start = file_lock->fl_start; - flock.l_flock.end = file_lock->fl_end; + + /* Somewhat ugly workaround for svc lockd. + * lockd installs custom fl_lmops->lm_compare_owner that checks + * for the fl_owner to be the same (which it always is on local node + * I guess between lockd processes) and then compares pid. + * As such we assign pid to the owner field to make it all work, + * conflict with normal locks is unlikely since pid space and + * pointer space for current->files are not intersecting */ + if (file_lock->fl_lmops && file_lock->fl_lmops->lm_compare_owner) + flock.l_flock.owner = (unsigned long)file_lock->fl_pid; switch (file_lock->fl_type) { case F_RDLCK: @@ -1987,8 +2192,9 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) einfo.ei_mode = LCK_PW; break; default: - CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type); - RETURN (-EINVAL); + CDEBUG(D_INFO, "Unknown fcntl lock type: %d\n", + file_lock->fl_type); + RETURN (-ENOTSUPP); } switch (cmd) { @@ -2034,13 +2240,11 @@ int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock) 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)); -#ifdef HAVE_F_OP_FLOCK + flock_lock_file_wait(file, file_lock); if ((file_lock->fl_flags & FL_POSIX) && (rc == 0 || file_lock->fl_type == F_UNLCK) && !(flags & LDLM_FL_TEST_LOCK)) posix_lock_file_wait(file, file_lock); -#endif RETURN(rc); } @@ -2052,26 +2256,55 @@ 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) +/** + * test if some locks matching bits and l_req_mode are acquired + * - bits can be in different locks + * - if found clear the common lock bits in *bits + * - the bits not found, are kept in *bits + * \param inode [IN] + * \param bits [IN] searched lock bits [IN] + * \param l_req_mode [IN] searched lock mode + * \retval boolean, true iff all bits are found + */ +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_policy_data_t policy; + 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; + int i; ENTRY; if (!inode) 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)) { - RETURN(1); + for (i = 0; i < MDS_INODELOCK_MAXSHIFT && *bits != 0; i++) { + policy.l_inodebits.bits = *bits & (1 << i); + if (policy.l_inodebits.bits == 0) + continue; + + if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, + &policy, mode, &lockh)) { + struct ldlm_lock *lock; + + lock = ldlm_handle2lock(&lockh); + if (lock) { + *bits &= + ~(lock->l_policy_data.l_inodebits.bits); + LDLM_LOCK_PUT(lock); + } else { + *bits &= ~policy.l_inodebits.bits; + } + } } - RETURN(0); + RETURN(*bits == 0); } ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits, @@ -2118,7 +2351,6 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, { struct inode *inode = dentry->d_inode; struct ptlrpc_request *req = NULL; - struct ll_sb_info *sbi; struct obd_export *exp; int rc = 0; ENTRY; @@ -2127,17 +2359,22 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, CERROR("REPORT THIS LINE TO PETER\n"); RETURN(0); } - sbi = ll_i2sbi(inode); CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n", inode->i_ino, inode->i_generation, inode, dentry->d_name.name); 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; + if (ibits == MDS_INODELOCK_LOOKUP) + oit.it_op = IT_LOOKUP; + /* Call getattr by fid, so do not provide name at all. */ op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode, dentry->d_inode, NULL, 0, 0, @@ -2168,16 +2405,11 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, do_lookup() -> ll_revalidate_it(). We cannot use d_drop 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); - } + if (!dentry->d_inode->i_nlink) + d_lustre_invalidate(dentry); ll_lookup_finish_locks(&oit, dentry); - } else if (!ll_have_md_lock(dentry->d_inode, ibits)) { + } 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 md_op_data *op_data; @@ -2193,8 +2425,8 @@ int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it, op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, ealen, LUSTRE_OPC_ANY, NULL); - if (op_data == NULL) - RETURN(-ENOMEM); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); op_data->op_valid = valid; /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one @@ -2214,28 +2446,28 @@ out: return rc; } -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; int rc; ENTRY; - rc = __ll_inode_revalidate_it(dentry, it, MDS_INODELOCK_UPDATE | - MDS_INODELOCK_LOOKUP); + rc = __ll_inode_revalidate_it(dentry, it, ibits); /* if object not yet allocated, don't validate size */ - if (rc == 0 && ll_i2info(dentry->d_inode)->lli_smd == NULL) { + if (rc == 0 && !ll_i2info(dentry->d_inode)->lli_has_smd) { 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); } - /* cl_glimpse_size will prefer locally cached writes if they extend + /* ll_glimpse_size will prefer locally cached writes if they extend * the file */ if (rc == 0) - rc = cl_glimpse_size(inode); + rc = ll_glimpse_size(inode); RETURN(rc); } @@ -2244,34 +2476,31 @@ 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_sb_info *sbi = ll_i2sbi(inode); struct ll_inode_info *lli = ll_i2info(inode); int res = 0; - res = ll_inode_revalidate_it(de, it); - ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETATTR, 1); + res = ll_inode_revalidate_it(de, it, MDS_INODELOCK_UPDATE | + MDS_INODELOCK_LOOKUP); + ll_stats_ops_tally(sbi, LPROC_LL_GETATTR, 1); if (res) return res; stat->dev = inode->i_sb->s_dev; - if (ll_need_32bit_api(ll_i2sbi(inode))) - stat->ino = cl_fid_build_ino32(&lli->lli_fid); + if (ll_need_32bit_api(sbi)) + stat->ino = cl_fid_build_ino(&lli->lli_fid, 1); else stat->ino = inode->i_ino; - stat->mode = inode->i_mode; stat->nlink = inode->i_nlink; stat->uid = inode->i_uid; stat->gid = inode->i_gid; - stat->rdev = kdev_t_to_nr(inode->i_rdev); + stat->rdev = inode->i_rdev; stat->atime = inode->i_atime; stat->mtime = inode->i_mtime; stat->ctime = inode->i_ctime; -#ifdef HAVE_INODE_BLKSIZE - stat->blksize = inode->i_blksize; -#else - stat->blksize = 1 << inode->i_blkbits; -#endif + stat->blksize = 1 << inode->i_blkbits; stat->size = i_size_read(inode); stat->blocks = inode->i_blocks; @@ -2296,7 +2525,7 @@ int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, num_bytes = sizeof(*fiemap) + (extent_count * sizeof(struct ll_fiemap_extent)); - OBD_VMALLOC(fiemap, num_bytes); + OBD_ALLOC_LARGE(fiemap, num_bytes); if (fiemap == NULL) RETURN(-ENOMEM); @@ -2315,47 +2544,67 @@ int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0], fiemap->fm_mapped_extents * sizeof(struct ll_fiemap_extent)); - OBD_VFREE(fiemap, num_bytes); + OBD_FREE_LARGE(fiemap, num_bytes); return rc; } #endif - -static -int lustre_check_acl(struct inode *inode, int mask) +#ifndef HAVE_GENERIC_PERMISSION_2ARGS +static int +# ifdef HAVE_GENERIC_PERMISSION_4ARGS +lustre_check_acl(struct inode *inode, int mask, unsigned int flags) +# else +lustre_check_acl(struct inode *inode, int mask) +# endif { -#ifdef CONFIG_FS_POSIX_ACL - struct ll_inode_info *lli = ll_i2info(inode); - struct posix_acl *acl; - int rc; - ENTRY; - - cfs_spin_lock(&lli->lli_lock); - acl = posix_acl_dup(lli->lli_posix_acl); - cfs_spin_unlock(&lli->lli_lock); - - if (!acl) - RETURN(-EAGAIN); - - rc = posix_acl_permission(inode, acl, mask); - posix_acl_release(acl); - - RETURN(rc); -#else - return -EAGAIN; -#endif +# ifdef CONFIG_FS_POSIX_ACL + struct ll_inode_info *lli = ll_i2info(inode); + struct posix_acl *acl; + int rc; + ENTRY; + +# ifdef HAVE_GENERIC_PERMISSION_4ARGS + if (flags & IPERM_FLAG_RCU) + return -ECHILD; +# endif + cfs_spin_lock(&lli->lli_lock); + acl = posix_acl_dup(lli->lli_posix_acl); + cfs_spin_unlock(&lli->lli_lock); + + if (!acl) + RETURN(-EAGAIN); + + rc = posix_acl_permission(inode, acl, mask); + posix_acl_release(acl); + + RETURN(rc); +# else /* !CONFIG_FS_POSIX_ACL */ + return -EAGAIN; +# endif /* CONFIG_FS_POSIX_ACL */ } +#endif /* HAVE_GENERIC_PERMISSION_2ARGS */ -#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) +#ifdef HAVE_GENERIC_PERMISSION_4ARGS +int ll_inode_permission(struct inode *inode, int mask, unsigned int flags) #else +# ifdef HAVE_INODE_PERMISION_2ARGS int ll_inode_permission(struct inode *inode, int mask) +# else +int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) +# endif #endif { int rc = 0; ENTRY; +#ifdef MAY_NOT_BLOCK + if (mask & MAY_NOT_BLOCK) + return -ECHILD; +#elif defined(HAVE_GENERIC_PERMISSION_4ARGS) + if (flags & IPERM_FLAG_RCU) + return -ECHILD; +#endif + /* as root inode are NOT getting validated in lookup operation, * need to do it before permission check. */ @@ -2375,61 +2624,10 @@ int ll_inode_permission(struct inode *inode, int mask) return lustre_check_remote_perm(inode, mask); ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1); - rc = generic_permission(inode, mask, lustre_check_acl); + rc = ll_generic_permission(inode, mask, flags, lustre_check_acl); RETURN(rc); } -#else -int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd) -{ - int mode = inode->i_mode; - int rc; - - CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n", - inode->i_ino, inode->i_generation, inode, 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); - - if ((mask & MAY_WRITE) && IS_RDONLY(inode) && - (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) - return -EROFS; - if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode)) - return -EACCES; - if (current->fsuid == inode->i_uid) { - mode >>= 6; - } else if (1) { - if (((mode >> 3) & mask & S_IRWXO) != mask) - goto check_groups; - rc = lustre_check_acl(inode, mask); - if (rc == -EAGAIN) - goto check_groups; - if (rc == -EACCES) - goto check_capabilities; - return rc; - } else { -check_groups: - if (cfs_curproc_is_in_groups(inode->i_gid)) - mode >>= 3; - } - if ((mode & mask & S_IRWXO) == mask) - return 0; - -check_capabilities: - if (!(mask & MAY_EXEC) || - (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode)) - if (cfs_capable(CFS_CAP_DAC_OVERRIDE)) - return 0; - - 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 @@ -2449,7 +2647,7 @@ struct file_operations ll_file_operations = { .READ_METHOD = READ_FUNCTION, .write = ll_file_write, .WRITE_METHOD = WRITE_FUNCTION, - .ioctl = ll_file_ioctl, + .unlocked_ioctl = ll_file_ioctl, .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, @@ -2461,6 +2659,7 @@ struct file_operations ll_file_operations = { .splice_read = ll_file_splice_read, #endif .fsync = ll_fsync, + .flush = ll_flush }; struct file_operations ll_file_operations_flock = { @@ -2468,7 +2667,7 @@ struct file_operations ll_file_operations_flock = { .READ_METHOD = READ_FUNCTION, .write = ll_file_write, .WRITE_METHOD = WRITE_FUNCTION, - .ioctl = ll_file_ioctl, + .unlocked_ioctl = ll_file_ioctl, .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, @@ -2480,9 +2679,8 @@ struct file_operations ll_file_operations_flock = { .splice_read = ll_file_splice_read, #endif .fsync = ll_fsync, -#ifdef HAVE_F_OP_FLOCK + .flush = ll_flush, .flock = ll_file_flock, -#endif .lock = ll_file_flock }; @@ -2492,7 +2690,7 @@ struct file_operations ll_file_operations_noflock = { .READ_METHOD = READ_FUNCTION, .write = ll_file_write, .WRITE_METHOD = WRITE_FUNCTION, - .ioctl = ll_file_ioctl, + .unlocked_ioctl = ll_file_ioctl, .open = ll_file_open, .release = ll_file_release, .mmap = ll_file_mmap, @@ -2504,16 +2702,12 @@ struct file_operations ll_file_operations_noflock = { .splice_read = ll_file_splice_read, #endif .fsync = ll_fsync, -#ifdef HAVE_F_OP_FLOCK + .flush = ll_flush, .flock = ll_file_noflock, -#endif .lock = ll_file_noflock }; struct inode_operations ll_file_inode_operations = { -#ifdef HAVE_VFS_INTENT_PATCHES - .setattr_raw = ll_setattr_raw, -#endif .setattr = ll_setattr, .truncate = ll_truncate, .getattr = ll_getattr, @@ -2626,3 +2820,141 @@ enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file, *rcp = rc; return ret; } + +int ll_layout_conf(struct inode *inode, const struct cl_object_conf *conf) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct cl_env_nest nest; + struct lu_env *env; + int result; + ENTRY; + + if (lli->lli_clob == NULL) + RETURN(0); + + env = cl_env_nested_get(&nest); + if (IS_ERR(env)) + RETURN(PTR_ERR(env)); + + result = cl_conf_set(env, lli->lli_clob, conf); + cl_env_nested_put(&nest, env); + RETURN(result); +} + +/** + * This function checks if there exists a LAYOUT lock on the client side, + * or enqueues it if it doesn't have one in cache. + * + * This function will not hold layout lock so it may be revoked any time after + * this function returns. Any operations depend on layout should be redone + * in that case. + * + * This function should be called before lov_io_init() to get an uptodate + * layout version, the caller should save the version number and after IO + * is finished, this function should be called again to verify that layout + * is not changed during IO time. + */ +int ll_layout_refresh(struct inode *inode, __u32 *gen) +{ + struct ll_inode_info *lli = ll_i2info(inode); + struct ll_sb_info *sbi = ll_i2sbi(inode); + struct md_op_data *op_data = NULL; + struct ptlrpc_request *req = NULL; + struct lookup_intent it = { .it_op = IT_LAYOUT }; + struct lustre_handle lockh; + ldlm_mode_t mode; + struct cl_object_conf conf = { .coc_inode = inode, + .coc_validate_only = true }; + int rc; + ENTRY; + + *gen = 0; + if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_LAYOUT_LOCK)) + RETURN(0); + + /* sanity checks */ + LASSERT(fid_is_sane(ll_inode2fid(inode))); + LASSERT(S_ISREG(inode->i_mode)); + + /* mostly layout lock is caching on the local side, so try to match + * it before grabbing layout lock mutex. */ + mode = ll_take_md_lock(inode, MDS_INODELOCK_LAYOUT, &lockh); + if (mode != 0) { /* hit cached lock */ + struct lov_stripe_md *lsm; + + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) + *gen = lsm->lsm_layout_gen; + ccc_inode_lsm_put(inode, lsm); + ldlm_lock_decref(&lockh, mode); + + RETURN(0); + } + + op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, + 0, 0, LUSTRE_OPC_ANY, NULL); + if (IS_ERR(op_data)) + RETURN(PTR_ERR(op_data)); + + /* take layout lock mutex to enqueue layout lock exclusively. */ + cfs_mutex_lock(&lli->lli_layout_mutex); + + /* make sure the old conf goes away */ + ll_layout_conf(inode, &conf); + + /* enqueue layout lock */ + rc = md_intent_lock(sbi->ll_md_exp, op_data, NULL, 0, &it, 0, + &req, ll_md_blocking_ast, 0); + if (rc == 0) { + /* we get a new lock, so update the lock data */ + lockh.cookie = it.d.lustre.it_lock_handle; + md_set_lock_data(sbi->ll_md_exp, &lockh.cookie, inode, NULL); + + /* req == NULL is when lock was found in client cache, without + * any request to server (but lsm can be canceled just after a + * release) */ + if (req != NULL) { + struct ldlm_lock *lock = ldlm_handle2lock(&lockh); + struct lustre_md md = { NULL }; + void *lmm; + int lmmsize; + + /* for IT_LAYOUT lock, lmm is returned in lock's lvb + * data via completion callback */ + LASSERT(lock != NULL); + lmm = lock->l_lvb_data; + lmmsize = lock->l_lvb_len; + if (lmm != NULL) + rc = obd_unpackmd(sbi->ll_dt_exp, &md.lsm, + lmm, lmmsize); + if (rc == 0) { + if (md.lsm != NULL) + *gen = md.lsm->lsm_layout_gen; + + memset(&conf, 0, sizeof conf); + conf.coc_inode = inode; + conf.u.coc_md = &md; + ll_layout_conf(inode, &conf); + /* is this racy? */ + lli->lli_has_smd = md.lsm != NULL; + } + if (md.lsm != NULL) + obd_free_memmd(sbi->ll_dt_exp, &md.lsm); + + LDLM_LOCK_PUT(lock); + ptlrpc_req_finished(req); + } else { /* hit caching lock */ + struct lov_stripe_md *lsm; + + lsm = ccc_inode_lsm_get(inode); + if (lsm != NULL) + *gen = lsm->lsm_layout_gen; + ccc_inode_lsm_put(inode, lsm); + } + ll_intent_drop_lock(&it); + } + cfs_mutex_unlock(&lli->lli_layout_mutex); + ll_finish_md_op_data(op_data); + + RETURN(rc); +}