filesetup: add direct=1 failure warning to layout
[fio.git] / filesetup.c
index eb28826..a6a94ee 100644 (file)
@@ -15,6 +15,7 @@
 #include "os/os.h"
 #include "hash.h"
 #include "lib/axmap.h"
+#include "lib/memalign.h"
 
 #ifdef CONFIG_LINUX_FALLOCATE
 #include <linux/falloc.h>
@@ -24,18 +25,90 @@ static int root_warn;
 
 static FLIST_HEAD(filename_list);
 
+/*
+ * List entry for filename_list
+ */
+struct file_name {
+       struct flist_head list;
+       char *filename;
+};
+
 static inline void clear_error(struct thread_data *td)
 {
        td->error = 0;
        td->verror[0] = '\0';
 }
 
+static inline int native_fallocate(struct thread_data *td, struct fio_file *f)
+{
+       bool success;
+
+       success = fio_fallocate(f, 0, f->real_file_size);
+       dprint(FD_FILE, "native fallocate of file %s size %llu was "
+                       "%ssuccessful\n", f->file_name,
+                       (unsigned long long) f->real_file_size,
+                       !success ? "un": "");
+
+       if (success)
+               return 0;
+
+       if (errno == ENOSYS)
+               dprint(FD_FILE, "native fallocate is not implemented\n");
+
+       return -1;
+}
+
+static void fallocate_file(struct thread_data *td, struct fio_file *f)
+{
+       int r;
+
+       if (td->o.fill_device)
+               return;
+
+       switch (td->o.fallocate_mode) {
+       case FIO_FALLOCATE_NATIVE:
+               r = native_fallocate(td, f);
+               if (r != 0 && errno != ENOSYS)
+                       log_err("fio: native_fallocate call failed: %s\n",
+                                       strerror(errno));
+               break;
+       case FIO_FALLOCATE_NONE:
+               break;
+#ifdef CONFIG_POSIX_FALLOCATE
+       case FIO_FALLOCATE_POSIX:
+               dprint(FD_FILE, "posix_fallocate file %s size %llu\n",
+                                f->file_name,
+                                (unsigned long long) f->real_file_size);
+
+               r = posix_fallocate(f->fd, 0, f->real_file_size);
+               if (r > 0)
+                       log_err("fio: posix_fallocate fails: %s\n", strerror(r));
+               break;
+#endif /* CONFIG_POSIX_FALLOCATE */
+#ifdef CONFIG_LINUX_FALLOCATE
+       case FIO_FALLOCATE_KEEP_SIZE:
+               dprint(FD_FILE, "fallocate(FALLOC_FL_KEEP_SIZE) "
+                               "file %s size %llu\n", f->file_name,
+                               (unsigned long long) f->real_file_size);
+
+               r = fallocate(f->fd, FALLOC_FL_KEEP_SIZE, 0, f->real_file_size);
+               if (r != 0)
+                       td_verror(td, errno, "fallocate");
+
+               break;
+#endif /* CONFIG_LINUX_FALLOCATE */
+       default:
+               log_err("fio: unknown fallocate mode: %d\n", td->o.fallocate_mode);
+               assert(0);
+       }
+}
+
 /*
  * Leaves f->fd open on success, caller must close
  */
 static int extend_file(struct thread_data *td, struct fio_file *f)
 {
-       int r, new_layout = 0, unlink_file = 0, flags;
+       int new_layout = 0, unlink_file = 0, flags;
        unsigned long long left;
        unsigned int bs;
        char *b = NULL;
@@ -74,6 +147,8 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
                flags |= O_CREAT;
        if (new_layout)
                flags |= O_TRUNC;
+       if (td->o.odirect)
+               flags |= OS_O_DIRECT;
 
 #ifdef WIN32
        flags |= _O_BINARY;
@@ -87,49 +162,22 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
                if (err == ENOENT && !td->o.allow_create)
                        log_err("fio: file creation disallowed by "
                                        "allow_file_create=0\n");
-               else
+               else {
+                       if (err == EINVAL && (flags & OS_O_DIRECT))
+                               log_err("fio: looks like your filesystem "
+                                       "does not support "
+                                       "direct=1/buffered=0\n");
+
                        td_verror(td, err, "open");
+               }
                return 1;
        }
 
-#ifdef CONFIG_POSIX_FALLOCATE
-       if (!td->o.fill_device) {
-               switch (td->o.fallocate_mode) {
-               case FIO_FALLOCATE_NONE:
-                       break;
-               case FIO_FALLOCATE_POSIX:
-                       dprint(FD_FILE, "posix_fallocate file %s size %llu\n",
-                                f->file_name,
-                                (unsigned long long) f->real_file_size);
-
-                       r = posix_fallocate(f->fd, 0, f->real_file_size);
-                       if (r > 0) {
-                               log_err("fio: posix_fallocate fails: %s\n",
-                                               strerror(r));
-                       }
-                       break;
-#ifdef CONFIG_LINUX_FALLOCATE
-               case FIO_FALLOCATE_KEEP_SIZE:
-                       dprint(FD_FILE,
-                               "fallocate(FALLOC_FL_KEEP_SIZE) "
-                               "file %s size %llu\n", f->file_name,
-                               (unsigned long long) f->real_file_size);
-
-                       r = fallocate(f->fd, FALLOC_FL_KEEP_SIZE, 0,
-                                       f->real_file_size);
-                       if (r != 0)
-                               td_verror(td, errno, "fallocate");
-
-                       break;
-#endif /* CONFIG_LINUX_FALLOCATE */
-               default:
-                       log_err("fio: unknown fallocate mode: %d\n",
-                               td->o.fallocate_mode);
-                       assert(0);
-               }
-       }
-#endif /* CONFIG_POSIX_FALLOCATE */
+       fallocate_file(td, f);
 
+       /*
+        * If our jobs don't require regular files initially, we're done.
+        */
        if (!new_layout)
                goto done;
 
@@ -148,11 +196,20 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
                }
        }
 
-       b = malloc(td->o.max_bs[DDIR_WRITE]);
-
        left = f->real_file_size;
+       bs = td->o.max_bs[DDIR_WRITE];
+       if (bs > left)
+               bs = left;
+
+       b = fio_memalign(page_size, bs);
+       if (!b) {
+               td_verror(td, errno, "fio_memalign");
+               goto err;
+       }
+
        while (left && !td->terminate) {
-               bs = td->o.max_bs[DDIR_WRITE];
+               ssize_t r;
+
                if (bs > left)
                        bs = left;
 
@@ -199,14 +256,14 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
                        f->io_size = f->real_file_size;
        }
 
-       free(b);
+       fio_memfree(b, bs);
 done:
        return 0;
 err:
        close(f->fd);
        f->fd = -1;
        if (b)
-               free(b);
+               fio_memfree(b, bs);
        return 1;
 }
 
@@ -217,7 +274,11 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f)
        unsigned int bs;
        char *b;
 
-       if (td_ioengine_flagged(td, FIO_PIPEIO))
+       if (td_ioengine_flagged(td, FIO_PIPEIO) ||
+           td_ioengine_flagged(td, FIO_NOIO))
+               return 0;
+
+       if (f->filetype == FIO_TYPE_CHAR)
                return 0;
 
        if (!fio_file_open(f)) {
@@ -230,8 +291,17 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f)
 
        old_runstate = td_bump_runstate(td, TD_PRE_READING);
 
+       left = f->io_size;
        bs = td->o.max_bs[DDIR_READ];
+       if (bs > left)
+               bs = left;
+
        b = malloc(bs);
+       if (!b) {
+               td_verror(td, errno, "malloc");
+               ret = 1;
+               goto error;
+       }
        memset(b, 0, bs);
 
        if (lseek(f->fd, f->file_offset, SEEK_SET) < 0) {
@@ -241,8 +311,6 @@ static int pre_read_file(struct thread_data *td, struct fio_file *f)
                goto error;
        }
 
-       left = f->io_size;
-
        while (left && !td->terminate) {
                if (bs > left)
                        bs = left;
@@ -375,11 +443,33 @@ static int get_file_size(struct thread_data *td, struct fio_file *f)
        else if (f->filetype == FIO_TYPE_CHAR)
                ret = char_size(td, f);
        else
-               f->real_file_size = -1;
+               f->real_file_size = -1ULL;
 
+       /*
+        * Leave ->real_file_size with 0 since it could be expectation
+        * of initial setup for regular files.
+        */
        if (ret)
                return ret;
 
+       /*
+        * If ->real_file_size is -1, a conditional for the message
+        * "offset extends end" is always true, but it makes no sense,
+        * so just return the same value here.
+        */
+       if (f->real_file_size == -1ULL) {
+               log_info("%s: failed to get file size of %s\n", td->o.name,
+                                       f->file_name);
+               return 1;
+       }
+
+       if (td->o.start_offset && f->file_offset == 0)
+               dprint(FD_FILE, "offset of file %s not initialized yet\n",
+                                       f->file_name);
+       /*
+        * ->file_offset normally hasn't been initialized yet, so this
+        * is basically always false.
+        */
        if (f->file_offset > f->real_file_size) {
                log_err("%s: offset extends end (%llu > %llu)\n", td->o.name,
                                        (unsigned long long) f->file_offset,
@@ -409,20 +499,22 @@ static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f,
        if (len == -1ULL || off == -1ULL)
                return 0;
 
-       dprint(FD_IO, "invalidate cache %s: %llu/%llu\n", f->file_name, off,
-                                                               len);
-
        if (td->io_ops->invalidate) {
+               dprint(FD_IO, "invalidate %s cache %s\n", td->io_ops->name,
+                       f->file_name);
                ret = td->io_ops->invalidate(td, f);
                if (ret < 0)
-                       errval = ret;
+                       errval = -ret;
        } else if (f->filetype == FIO_TYPE_FILE) {
+               dprint(FD_IO, "declare unneeded cache %s: %llu/%llu\n",
+                       f->file_name, off, len);
                ret = posix_fadvise(f->fd, off, len, POSIX_FADV_DONTNEED);
                if (ret)
                        errval = ret;
        } else if (f->filetype == FIO_TYPE_BLOCK) {
                int retry_count = 0;
 
+               dprint(FD_IO, "drop page cache %s\n", f->file_name);
                ret = blockdev_invalidate_cache(f);
                while (ret < 0 && errno == EAGAIN && retry_count++ < 25) {
                        /*
@@ -444,8 +536,11 @@ static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f,
                }
                if (ret < 0)
                        errval = errno;
-       } else if (f->filetype == FIO_TYPE_CHAR || f->filetype == FIO_TYPE_PIPE)
+       } else if (f->filetype == FIO_TYPE_CHAR ||
+                  f->filetype == FIO_TYPE_PIPE) {
+               dprint(FD_IO, "invalidate not supported %s\n", f->file_name);
                ret = 0;
+       }
 
        /*
         * Cache flushing isn't a fatal condition, and we know it will
@@ -454,7 +549,8 @@ static int __file_invalidate_cache(struct thread_data *td, struct fio_file *f,
         * continue on our way.
         */
        if (errval)
-               log_info("fio: cache invalidation of %s failed: %s\n", f->file_name, strerror(errval));
+               log_info("fio: cache invalidation of %s failed: %s\n",
+                        f->file_name, strerror(errval));
 
        return 0;
 
@@ -486,7 +582,7 @@ int generic_close_file(struct thread_data fio_unused *td, struct fio_file *f)
                f->shadow_fd = -1;
        }
 
-       f->engine_data = 0;
+       f->engine_pos = 0;
        return ret;
 }
 
@@ -498,9 +594,6 @@ int file_lookup_open(struct fio_file *f, int flags)
        __f = lookup_file_hash(f->file_name);
        if (__f) {
                dprint(FD_FILE, "found file in hash %s\n", f->file_name);
-               /*
-                * racy, need the __f->lock locked
-                */
                f->lock = __f->lock;
                from_hash = 1;
        } else {
@@ -597,7 +690,8 @@ open_again:
                        f->fd = dup(STDIN_FILENO);
                else
                        from_hash = file_lookup_open(f, flags);
-       } else { //td trim
+       } else if (td_trim(td)) {
+               assert(!td_rw(td)); /* should have matched above */
                flags |= O_RDWR;
                from_hash = file_lookup_open(f, flags);
        }
@@ -652,6 +746,10 @@ open_again:
        return 0;
 }
 
+/*
+ * This function i.e. get_file_size() is the default .get_file_size
+ * implementation of majority of I/O engines.
+ */
 int generic_get_file_size(struct thread_data *td, struct fio_file *f)
 {
        return get_file_size(td, f);
@@ -667,7 +765,7 @@ static int get_file_sizes(struct thread_data *td)
        int err = 0;
 
        for_each_file(td, f, i) {
-               dprint(FD_FILE, "get file size for %p/%d/%p\n", f, i,
+               dprint(FD_FILE, "get file size for %p/%d/%s\n", f, i,
                                                                f->file_name);
 
                if (td_io_get_file_size(td, f)) {
@@ -679,6 +777,13 @@ static int get_file_sizes(struct thread_data *td)
                        clear_error(td);
                }
 
+               /*
+                * There are corner cases where we end up with -1 for
+                * ->real_file_size due to unsupported file type, etc.
+                * We then just set to size option value divided by number
+                * of files, similar to the way file ->io_size is set.
+                * stat(2) failure doesn't set ->real_file_size to -1.
+                */
                if (f->real_file_size == -1ULL && td->o.size)
                        f->real_file_size = td->o.size / td->o.nr_files;
        }
@@ -765,12 +870,42 @@ static unsigned long long get_fs_free_counts(struct thread_data *td)
 uint64_t get_start_offset(struct thread_data *td, struct fio_file *f)
 {
        struct thread_options *o = &td->o;
+       unsigned long long align_bs;
+       unsigned long long offset;
 
        if (o->file_append && f->filetype == FIO_TYPE_FILE)
                return f->real_file_size;
 
-       return td->o.start_offset +
-               td->subjob_number * td->o.offset_increment;
+       if (o->start_offset_percent > 0) {
+               /*
+                * if blockalign is provided, find the min across read, write,
+                * and trim
+                */
+               if (fio_option_is_set(o, ba)) {
+                       align_bs = (unsigned long long) min(o->ba[DDIR_READ], o->ba[DDIR_WRITE]);
+                       align_bs = min((unsigned long long) o->ba[DDIR_TRIM], align_bs);
+               } else {
+                       /* else take the minimum block size */
+                       align_bs = td_min_bs(td);
+               }
+
+               /* calculate the raw offset */
+               offset = (f->real_file_size * o->start_offset_percent / 100) +
+                       (td->subjob_number * o->offset_increment);
+
+               /*
+                * block align the offset at the next available boundary at
+                * ceiling(offset / align_bs) * align_bs
+                */
+               offset = (offset / align_bs + (offset % align_bs != 0)) * align_bs;
+
+       } else {
+               /* start_offset_percent not set */
+               offset = o->start_offset +
+                               td->subjob_number * o->offset_increment;
+       }
+
+       return offset;
 }
 
 /*
@@ -795,7 +930,9 @@ int setup_files(struct thread_data *td)
                goto done;
 
        /*
-        * if ioengine defines a setup() method, it's responsible for
+        * Find out physical size of files or devices for this thread,
+        * before we determine I/O size and range of our targets.
+        * If ioengine defines a setup() method, it's responsible for
         * opening the files and setting f->real_file_size to indicate
         * the valid range for that file.
         */
@@ -836,7 +973,7 @@ int setup_files(struct thread_data *td)
 
        /*
         * Calculate per-file size and potential extra size for the
-        * first files, if needed.
+        * first files, if needed (i.e. if we don't have a fixed size).
         */
        if (!o->file_size_low && o->nr_files) {
                uint64_t all_fs;
@@ -858,11 +995,18 @@ int setup_files(struct thread_data *td)
        for_each_file(td, f, i) {
                f->file_offset = get_start_offset(td, f);
 
+               /*
+                * Update ->io_size depending on options specified.
+                * ->file_size_low being 0 means filesize option isn't set.
+                * Non zero ->file_size_low equals ->file_size_high means
+                * filesize option is set in a fixed size format.
+                * Non zero ->file_size_low not equals ->file_size_high means
+                * filesize option is set in a range format.
+                */
                if (!o->file_size_low) {
                        /*
-                        * no file size range given, file size is equal to
-                        * total size divided by number of files. If that is
-                        * zero, set it to the real file size. If the size
+                        * no file size or range given, file size is equal to
+                        * total size divided by number of files. If the size
                         * doesn't divide nicely with the min blocksize,
                         * make the first files bigger.
                         */
@@ -872,8 +1016,22 @@ int setup_files(struct thread_data *td)
                                f->io_size += bs;
                        }
 
-                       if (!f->io_size)
+                       /*
+                        * We normally don't come here for regular files, but
+                        * if the result is 0 for a regular file, set it to the
+                        * real file size. This could be size of the existing
+                        * one if it already exists, but otherwise will be set
+                        * to 0. A new file won't be created because
+                        * ->io_size + ->file_offset equals ->real_file_size.
+                        */
+                       if (!f->io_size) {
+                               if (f->file_offset > f->real_file_size)
+                                       goto err_offset;
                                f->io_size = f->real_file_size - f->file_offset;
+                               if (!f->io_size)
+                                       log_info("fio: file %s may be ignored\n",
+                                               f->file_name);
+                       }
                } else if (f->real_file_size < o->file_size_low ||
                           f->real_file_size > o->file_size_high) {
                        if (f->file_offset > o->file_size_low)
@@ -895,7 +1053,14 @@ int setup_files(struct thread_data *td)
                        total_size = -1ULL;
                else {
                         if (o->size_percent) {
-                               f->io_size = (f->io_size * o->size_percent) / 100;
+                               uint64_t file_size;
+
+                               file_size = f->io_size + f->file_offset;
+                               f->io_size = (file_size *
+                                             o->size_percent) / 100;
+                               if (f->io_size > (file_size - f->file_offset))
+                                       f->io_size = file_size - f->file_offset;
+
                                f->io_size -= (f->io_size % td_min_bs(td));
                        }
                        total_size += f->io_size;
@@ -907,9 +1072,9 @@ int setup_files(struct thread_data *td)
                        if (!o->create_on_open) {
                                need_extend++;
                                extend_size += (f->io_size + f->file_offset);
+                               fio_file_set_extend(f);
                        } else
                                f->real_file_size = f->io_size + f->file_offset;
-                       fio_file_set_extend(f);
                }
        }
 
@@ -943,13 +1108,21 @@ int setup_files(struct thread_data *td)
        }
 
        /*
-        * See if we need to extend some files
+        * See if we need to extend some files, typically needed when our
+        * target regular files don't exist yet, but our jobs require them
+        * initially due to read I/Os.
         */
        if (need_extend) {
                temp_stall_ts = 1;
-               if (output_format & FIO_OUTPUT_NORMAL)
-                       log_info("%s: Laying out IO file(s) (%u file(s) / %lluMiB)\n",
-                                o->name, need_extend, extend_size >> 20);
+               if (output_format & FIO_OUTPUT_NORMAL) {
+                       log_info("%s: Laying out IO file%s (%u file%s / %s%lluMiB)\n",
+                                o->name,
+                                need_extend > 1 ? "s" : "",
+                                need_extend,
+                                need_extend > 1 ? "s" : "",
+                                need_extend > 1 ? "total " : "",
+                                extend_size >> 20);
+               }
 
                for_each_file(td, f, i) {
                        unsigned long long old_len = -1ULL, extend_len = -1ULL;
@@ -996,8 +1169,8 @@ int setup_files(struct thread_data *td)
         * stored entries.
         */
        if (!o->read_iolog_file) {
-               if (o->io_limit)
-                       td->total_io_size = o->io_limit * o->loops;
+               if (o->io_size)
+                       td->total_io_size = o->io_size * o->loops;
                else
                        td->total_io_size = o->size * o->loops;
        }
@@ -1023,10 +1196,11 @@ int pre_read_files(struct thread_data *td)
        dprint(FD_FILE, "pre_read files\n");
 
        for_each_file(td, f, i) {
-               pre_read_file(td, f);
+               if (pre_read_file(td, f))
+                       return -1;
        }
 
-       return 1;
+       return 0;
 }
 
 static int __init_rand_distribution(struct thread_data *td, struct fio_file *f)